vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-08 09:59:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge branch 'vulkan_test' into concedo_experimental

Browse source 
# Conflicts:
#	CMakeLists.txt
#	Makefile
#	llama.cpp
This commit is contained in:
Concedo 2024-01-25 23:01:44 +08:00
commit 2a4a7241e6
117 changed files with 394082 additions and 68 deletions
Download patch file Download diff file Expand all files Collapse all files

2
.gitignore vendored
View file

@ -106,6 +106,7 @@ tests/test-tokenizer-1-bpe
/koboldcpp_clblast.so
/koboldcpp_clblast_noavx2.so
/koboldcpp_cublas.so
/koboldcpp_vulkan.so
/koboldcpp_default.dll
/koboldcpp_failsafe.dll
/koboldcpp_openblas.dll
@ -113,6 +114,7 @@ tests/test-tokenizer-1-bpe
/koboldcpp_clblast.dll
/koboldcpp_clblast_noavx2.dll
/koboldcpp_cublas.dll
/koboldcpp_vulkan.dll
/cublas64_11.dll
/cublasLt64_11.dll
/rocblas/

58
Makefile
View file

@ -1,4 +1,4 @@
default: koboldcpp_default koboldcpp_failsafe koboldcpp_openblas koboldcpp_noavx2 koboldcpp_clblast koboldcpp_clblast_noavx2 koboldcpp_cublas koboldcpp_hipblas
default: koboldcpp_default koboldcpp_failsafe koboldcpp_openblas koboldcpp_noavx2 koboldcpp_clblast koboldcpp_clblast_noavx2 koboldcpp_cublas koboldcpp_hipblas koboldcpp_vulkan
tools: quantize_gpt2 quantize_gptj quantize_llama quantize_neox quantize_mpt
dev: koboldcpp_openblas
dev2: koboldcpp_clblast
@ -39,8 +39,8 @@ endif
#
# keep standard at C11 and C++11
CFLAGS = -I. -I./include -I./include/CL -I./otherarch -I./otherarch/tools -O3 -DNDEBUG -std=c11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
CXXFLAGS = -I. -I./common -I./include -I./include/CL -I./otherarch -I./otherarch/tools -O3 -DNDEBUG -std=c++11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
CFLAGS = -I. -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./include/vulkan -O3 -DNDEBUG -std=c11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
CXXFLAGS = -I. -I./common -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./include/vulkan -O3 -DNDEBUG -std=c++11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
LDFLAGS =
# these are used on windows, to build some libraries with extra old device compatibility
@ -51,6 +51,7 @@ NONECFLAGS =
OPENBLAS_FLAGS = -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
CLBLAST_FLAGS = -DGGML_USE_CLBLAST
FAILSAFE_FLAGS = -DUSE_FAILSAFE
VULKAN_FLAGS = -DGGML_USE_VULKAN
ifdef LLAMA_CUBLAS
CUBLAS_FLAGS = -DGGML_USE_CUBLAS
else
@ -300,6 +301,7 @@ NOAVX2_BUILD =
CLBLAST_BUILD =
CUBLAS_BUILD =
HIPBLAS_BUILD =
VULKAN_BUILD =
ifeq ($(OS),Windows_NT)
DEFAULT_BUILD = $(CXX) $(CXXFLAGS) $^ -shared -o $@.dll $(LDFLAGS)
@ -307,6 +309,7 @@ ifeq ($(OS),Windows_NT)
OPENBLAS_BUILD = $(CXX) $(CXXFLAGS) $^ lib/libopenblas.lib -shared -o $@.dll $(LDFLAGS)
NOAVX2_BUILD = $(CXX) $(CXXFLAGS) $^ -shared -o $@.dll $(LDFLAGS)
CLBLAST_BUILD = $(CXX) $(CXXFLAGS) $^ lib/OpenCL.lib lib/clblast.lib -shared -o $@.dll $(LDFLAGS)
VULKAN_BUILD = $(CXX) $(CXXFLAGS) $^ lib/vulkan-1.lib -shared -o $@.dll $(LDFLAGS)
ifdef LLAMA_CUBLAS
CUBLAS_BUILD = $(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $^ -shared -o $@.dll $(CUBLASLD_FLAGS) $(LDFLAGS)
@ -337,16 +340,21 @@ else
ifdef LLAMA_HIPBLAS
HIPBLAS_BUILD = $(HCXX) $(CXXFLAGS) $(HIPFLAGS) $^ -shared -o $@.so $(HIPLDFLAGS) $(LDFLAGS)
endif
ifdef LLAMA_VULKAN
VULKAN_BUILD = $(CXX) $(CXXFLAGS) $^ -lvulkan -shared -o $@.so $(LDFLAGS)
endif
ifndef LLAMA_OPENBLAS
ifndef LLAMA_CLBLAST
ifndef LLAMA_CUBLAS
ifndef LLAMA_HIPBLAS
ifndef LLAMA_VULKAN
OPENBLAS_BUILD = @echo 'Your OS $(OS) does not appear to be Windows. For faster speeds, install and link a BLAS library. Set LLAMA_OPENBLAS=1 to compile with OpenBLAS support or LLAMA_CLBLAST=1 to compile with ClBlast support. This is just a reminder, not an error.'
endif
endif
endif
endif
endif
endif
CCV := $(shell $(CC) --version | head -n 1)
@ -373,18 +381,20 @@ $(info )
ggml.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
ggml_openblas.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_openblas.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
ggml_failsafe.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_failsafe.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(NONECFLAGS) -c $< -o $@
ggml_noavx2.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_noavx2.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(SIMPLECFLAGS) -c $< -o $@
ggml_clblast.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_clblast.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
ggml_cublas.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_cublas.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
ggml_clblast_noavx2.o: ggml.c ggml.h ggml-cuda.h
ggml_v4_clblast_noavx2.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
ggml_v4_vulkan.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) $(FULLCFLAGS) $(VULKAN_FLAGS) -c $< -o $@
#quants
ggml-quants.o: ggml-quants.c ggml.h ggml-quants.h ggml-cuda.h
@ -449,6 +459,10 @@ ggml_v2-opencl-legacy.o: otherarch/ggml_v2-opencl-legacy.c otherarch/ggml_v2-ope
ggml_v3-opencl.o: otherarch/ggml_v3-opencl.cpp otherarch/ggml_v3-opencl.h
$(CXX) $(CXXFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
#vulkan
ggml-vulkan.o: ggml-vulkan.cpp ggml-vulkan.h
$(CXX) $(CXXFLAGS) $(VULKAN_FLAGS) -c $< -o $@
# intermediate objects
llama.o: llama.cpp ggml.h ggml-alloc.h ggml-backend.h ggml-cuda.h ggml-metal.h llama.h otherarch/llama-util.h
$(CXX) $(CXXFLAGS) -c $< -o $@
@ -473,9 +487,11 @@ gpttype_adapter_cublas.o: $(GPTTYPE_ADAPTER)
$(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
gpttype_adapter_clblast_noavx2.o: $(GPTTYPE_ADAPTER)
$(CXX) $(CXXFLAGS) $(FAILSAFE_FLAGS) $(CLBLAST_FLAGS) -c $< -o $@
gpttype_adapter_vulkan.o: $(GPTTYPE_ADAPTER)
$(CXX) $(CXXFLAGS) $(VULKAN_FLAGS) -c $< -o $@
clean:
rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize_neox quantize_mpt quantize-stats perplexity embedding benchmark-matmult save-load-state gguf gguf.exe main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe quantize_neox.exe quantize_mpt.exe koboldcpp_default.dll koboldcpp_openblas.dll koboldcpp_failsafe.dll koboldcpp_noavx2.dll koboldcpp_clblast.dll koboldcpp_clblast_noavx2.dll koboldcpp_cublas.dll koboldcpp_hipblas.dll koboldcpp_default.so koboldcpp_openblas.so koboldcpp_failsafe.so koboldcpp_noavx2.so koboldcpp_clblast.so koboldcpp_clblast_noavx2.so koboldcpp_cublas.so koboldcpp_hipblas.so
rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize_neox quantize_mpt quantize-stats perplexity embedding benchmark-matmult save-load-state gguf gguf.exe main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe quantize_neox.exe quantize_mpt.exe koboldcpp_default.dll koboldcpp_openblas.dll koboldcpp_failsafe.dll koboldcpp_noavx2.dll koboldcpp_clblast.dll koboldcpp_clblast_noavx2.dll koboldcpp_cublas.dll koboldcpp_hipblas.dll koboldcpp_vulkan.dll koboldcpp_default.so koboldcpp_openblas.so koboldcpp_failsafe.so koboldcpp_noavx2.so koboldcpp_clblast.so koboldcpp_clblast_noavx2.so koboldcpp_cublas.so koboldcpp_hipblas.so koboldcpp_vulkan.so
main: examples/main/main.cpp common/sampling.cpp build-info.h ggml.o ggml-quants.o ggml-alloc.o ggml-backend.o llama.o common.o console.o grammar-parser.o $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
@ -492,7 +508,7 @@ koboldcpp_default: ggml.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o common.o gpttyp
$(DEFAULT_BUILD)
ifdef OPENBLAS_BUILD
koboldcpp_openblas: ggml_openblas.o ggml_v3_openblas.o ggml_v2_openblas.o ggml_v1.o expose.o common.o gpttype_adapter.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
koboldcpp_openblas: ggml_v4_openblas.o ggml_v3_openblas.o ggml_v2_openblas.o ggml_v1.o expose.o common.o gpttype_adapter.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(OPENBLAS_BUILD)
else
koboldcpp_openblas:
@ -500,7 +516,7 @@ koboldcpp_openblas:
endif
ifdef FAILSAFE_BUILD
koboldcpp_failsafe: ggml_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_failsafe.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
koboldcpp_failsafe: ggml_v4_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_failsafe.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(FAILSAFE_BUILD)
else
koboldcpp_failsafe:
@ -508,7 +524,7 @@ koboldcpp_failsafe:
endif
ifdef NOAVX2_BUILD
koboldcpp_noavx2: ggml_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
koboldcpp_noavx2: ggml_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(NOAVX2_BUILD)
else
koboldcpp_noavx2:
@ -516,10 +532,10 @@ koboldcpp_noavx2:
endif
ifdef CLBLAST_BUILD
koboldcpp_clblast: ggml_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o common.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
koboldcpp_clblast: ggml_v4_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o common.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(CLBLAST_BUILD)
ifdef NOAVX2_BUILD
koboldcpp_clblast_noavx2: ggml_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
koboldcpp_clblast_noavx2: ggml_v4_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(CLBLAST_BUILD)
else
koboldcpp_clblast_noavx2:
@ -533,7 +549,7 @@ koboldcpp_clblast_noavx2:
endif
ifdef CUBLAS_BUILD
koboldcpp_cublas: ggml_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(CUBLAS_OBJS) $(OBJS)
koboldcpp_cublas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(CUBLAS_OBJS) $(OBJS)
$(CUBLAS_BUILD)
else
koboldcpp_cublas:
@ -541,13 +557,21 @@ koboldcpp_cublas:
endif
ifdef HIPBLAS_BUILD
koboldcpp_hipblas: ggml_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(HIP_OBJS) $(OBJS)
koboldcpp_hipblas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(HIP_OBJS) $(OBJS)
$(HIPBLAS_BUILD)
else
koboldcpp_hipblas:
$(DONOTHING)
endif
ifdef VULKAN_BUILD
koboldcpp_vulkan: ggml_v4_vulkan.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o common.o gpttype_adapter_vulkan.o ggml-vulkan.o ggml-quants.o ggml-alloc.o ggml-backend.o grammar-parser.o $(OBJS)
$(VULKAN_BUILD)
else
koboldcpp_vulkan:
$(DONOTHING)
endif
# tools
quantize_llama: examples/quantize/quantize.cpp ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)

11
examples/llama-bench/llama-bench.cpp
View file

@ -563,6 +563,7 @@ struct test {
static const int build_number;
static const bool cuda;
static const bool opencl;
static const bool vulkan;
static const bool metal;
static const bool gpu_blas;
static const bool blas;
@ -644,6 +645,9 @@ struct test {
if (opencl) {
return "OpenCL";
}
if (vulkan) {
return "Vulkan";
}
if (metal) {
return "Metal";
}
@ -659,7 +663,7 @@ struct test {
static const std::vector<std::string> & get_fields() {
static const std::vector<std::string> fields = {
"build_commit", "build_number",
"cuda", "opencl", "metal", "gpu_blas", "blas",
"cuda", "opencl", "vulkan", "metal", "gpu_blas", "blas",
"cpu_info", "gpu_info",
"model_filename", "model_type", "model_size", "model_n_params",
"n_batch", "n_threads", "type_k", "type_v",
@ -683,7 +687,7 @@ struct test {
field == "avg_ns" || field == "stddev_ns") {
return INT;
}
if (field == "cuda" || field == "opencl" || field == "metal" || field == "gpu_blas" || field == "blas" ||
if (field == "cuda" || field == "opencl" || field == "vulkan"|| field == "metal" || field == "gpu_blas" || field == "blas" ||
field == "f16_kv" || field == "no_kv_offload" || field == "mul_mat_q") {
return BOOL;
}
@ -711,7 +715,7 @@ struct test {
}
std::vector<std::string> values = {
build_commit, std::to_string(build_number),
std::to_string(cuda), std::to_string(opencl), std::to_string(metal), std::to_string(gpu_blas), std::to_string(blas),
std::to_string(cuda), std::to_string(opencl), std::to_string(vulkan), std::to_string(metal), std::to_string(gpu_blas), std::to_string(blas),
cpu_info, gpu_info,
model_filename, model_type, std::to_string(model_size), std::to_string(model_n_params),
std::to_string(n_batch), std::to_string(n_threads), ggml_type_name(type_k), ggml_type_name(type_v),
@ -739,6 +743,7 @@ const std::string test::build_commit = LLAMA_COMMIT;
const int test::build_number = LLAMA_BUILD_NUMBER;
const bool test::cuda = !!ggml_cpu_has_cublas();
const bool test::opencl = !!ggml_cpu_has_clblast();
const bool test::vulkan = !!ggml_cpu_has_vulkan();
const bool test::metal = !!ggml_cpu_has_metal();
const bool test::gpu_blas = !!ggml_cpu_has_gpublas();
const bool test::blas = !!ggml_cpu_has_blas();

7
expose.cpp
View file

@ -23,7 +23,7 @@
extern "C"
{
std::string platformenv, deviceenv;
std::string platformenv, deviceenv, vulkandeviceenv;
//return val: 0=fail, 1=(original ggml, alpaca), 2=(ggmf), 3=(ggjt)
static FileFormat file_format = FileFormat::BADFORMAT;
@ -58,6 +58,11 @@ extern "C"
deviceenv = "GGML_OPENCL_DEVICE="+std::to_string(devices);
putenv((char*)platformenv.c_str());
putenv((char*)deviceenv.c_str());
int vulkan_info = inputs.vulkan_info;
vulkandeviceenv = "GGML_VULKAN_DEVICE="+std::to_string(vulkan_info);
putenv((char*)vulkandeviceenv.c_str());
executable_path = inputs.executable_path;
if(file_format==FileFormat::GPTJ_1 || file_format==FileFormat::GPTJ_2 || file_format==FileFormat::GPTJ_3 || file_format==FileFormat::GPTJ_4 || file_format==FileFormat::GPTJ_5)

1
expose.h
View file

@ -44,6 +44,7 @@ struct load_model_inputs
const bool use_contextshift;
const int clblast_info = 0;
const int cublas_info = 0;
const int vulkan_info = 0;
const int blasbatchsize = 512;
const int debugmode = 0;
const int forceversion = 0;

106
ggml-alloc.c
View file

@ -776,38 +776,26 @@ size_t ggml_allocr_alloc_graph(ggml_allocr_t alloc, struct ggml_cgraph * graph)
}
// utils
ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, ggml_backend_buffer_type_t buft) {
GGML_ASSERT(ggml_get_no_alloc(ctx) == true);
size_t alignment = ggml_backend_buft_get_alignment(buft);
size_t nbytes = 0;
for (struct ggml_tensor * t = ggml_get_first_tensor(ctx); t != NULL; t = ggml_get_next_tensor(ctx, t)) {
if (t->data == NULL && t->view_src == NULL) {
nbytes += GGML_PAD(ggml_backend_buft_get_alloc_size(buft, t), alignment);
}
}
if (nbytes == 0) {
// all the tensors in the context are already allocated
#ifndef NDEBUG
fprintf(stderr, "%s: all tensors in the context are already allocated\n", __func__);
#endif
return NULL;
}
ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, nbytes);
static bool alloc_tensor_range(struct ggml_context * ctx,
struct ggml_tensor * first, struct ggml_tensor * last,
ggml_backend_buffer_type_t buft, size_t size,
ggml_backend_buffer_t ** buffers, size_t * n_buffers) {
ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, size);
if (buffer == NULL) {
// failed to allocate buffer
#ifndef NDEBUG
fprintf(stderr, "%s: failed to allocate buffer\n", __func__);
fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size);
#endif
return NULL;
for (size_t i = 0; i < *n_buffers; i++) {
ggml_backend_buffer_free(*buffers[i]);
}
free(buffers);
return false;
}
ggml_tallocr_t tallocr = ggml_tallocr_new_from_buffer(buffer);
for (struct ggml_tensor * t = ggml_get_first_tensor(ctx); t != NULL; t = ggml_get_next_tensor(ctx, t)) {
for (struct ggml_tensor * t = first; t != last; t = ggml_get_next_tensor(ctx, t)) {
if (t->data == NULL) {
if (t->view_src == NULL) {
ggml_tallocr_alloc(tallocr, t);
@ -824,6 +812,76 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte
ggml_tallocr_free(tallocr);
*buffers = realloc(*buffers, sizeof(ggml_backend_buffer_t) * (*n_buffers + 1));
(*buffers)[(*n_buffers)++] = buffer;
return true;
}
ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_context * ctx, ggml_backend_buffer_type_t buft) {
GGML_ASSERT(ggml_get_no_alloc(ctx) == true);
size_t alignment = ggml_backend_buft_get_alignment(buft);
size_t max_size = ggml_backend_buft_get_max_size(buft);
ggml_backend_buffer_t * buffers = NULL;
size_t n_buffers = 0;
size_t cur_buf_size = 0;
struct ggml_tensor * first = ggml_get_first_tensor(ctx);
for (struct ggml_tensor * t = first; t != NULL; t = ggml_get_next_tensor(ctx, t)) {
size_t this_size = 0;
if (t->data == NULL && t->view_src == NULL) {
this_size = GGML_PAD(ggml_backend_buft_get_alloc_size(buft, t), alignment);
}
if (this_size > max_size) {
// tensor is too large to fit in a single buffer
fprintf(stderr, "%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n",
__func__, t->name,
ggml_backend_buft_name(buft),
this_size, max_size);
for (size_t i = 0; i < n_buffers; i++) {
ggml_backend_buffer_free(buffers[i]);
}
free(buffers);
return NULL;
}
if ((cur_buf_size + this_size) > max_size) {
// allocate tensors in the current buffer
if (!alloc_tensor_range(ctx, first, t, buft, cur_buf_size, &buffers, &n_buffers)) {
return NULL;
}
first = t;
cur_buf_size = this_size;
} else {
cur_buf_size += this_size;
}
}
// allocate remaining tensors
if (cur_buf_size > 0) {
if (!alloc_tensor_range(ctx, first, NULL, buft, cur_buf_size, &buffers, &n_buffers)) {
return NULL;
}
}
if (n_buffers == 0) {
// all the tensors in the context are already allocated
#ifndef NDEBUG
fprintf(stderr, "%s: all tensors in the context are already allocated\n", __func__);
#endif
return NULL;
}
ggml_backend_buffer_t buffer;
if (n_buffers == 1) {
buffer = buffers[0];
} else {
buffer = ggml_backend_multi_buffer_alloc_buffer(buffers, n_buffers);
}
free(buffers);
return buffer;
}

6
ggml-backend-impl.h
View file

@ -19,6 +19,7 @@ extern "C" {
const char * (*GGML_CALL get_name) (ggml_backend_buffer_type_t buft);
ggml_backend_buffer_t (*GGML_CALL alloc_buffer) (ggml_backend_buffer_type_t buft, size_t size);
size_t (*GGML_CALL get_alignment) (ggml_backend_buffer_type_t buft); // tensor alignment
size_t (*GGML_CALL get_max_size) (ggml_backend_buffer_type_t buft); // allocation max size
size_t (*GGML_CALL get_alloc_size) (ggml_backend_buffer_type_t buft, const struct ggml_tensor * tensor); // data size needed to allocate the tensor, including padding
bool (*GGML_CALL supports_backend)(ggml_backend_buffer_type_t buft, ggml_backend_t backend); // check if the buffer type is usable by the backend
// check if tensor data is in host memory
@ -63,6 +64,11 @@ extern "C" {
// do not use directly, use ggml_backend_tensor_copy instead
bool ggml_backend_buffer_copy_tensor(const struct ggml_tensor * src, struct ggml_tensor * dst);
// buffer that contains a collection of buffers
GGML_CALL ggml_backend_buffer_t ggml_backend_multi_buffer_alloc_buffer(ggml_backend_buffer_t * buffers, size_t n_buffers);
GGML_CALL bool ggml_backend_buffer_is_multi_buffer(ggml_backend_buffer_t buffer);
GGML_CALL void ggml_backend_multi_buffer_set_usage(ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage);
//
// Backend
//

104
ggml-backend.c
View file

@ -27,6 +27,14 @@ size_t ggml_backend_buft_get_alignment(ggml_backend_buffer_type_t buft) {
return buft->iface.get_alignment(buft);
}
size_t ggml_backend_buft_get_max_size(ggml_backend_buffer_type_t buft) {
// get_max_size is optional, defaults to SIZE_MAX
if (buft->iface.get_max_size) {
return buft->iface.get_max_size(buft);
}
return SIZE_MAX;
}
GGML_CALL size_t ggml_backend_buft_get_alloc_size(ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor) {
// get_alloc_size is optional, defaults to ggml_nbytes
if (buft->iface.get_alloc_size) {
@ -55,8 +63,6 @@ GGML_CALL ggml_backend_buffer_t ggml_backend_buffer_init(
size_t size) {
ggml_backend_buffer_t buffer = malloc(sizeof(struct ggml_backend_buffer));
GGML_ASSERT(iface.get_base != NULL);
(*buffer) = (struct ggml_backend_buffer) {
/* .interface = */ iface,
/* .buft = */ buft,
@ -106,6 +112,10 @@ size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer) {
return ggml_backend_buft_get_alignment(ggml_backend_buffer_get_type(buffer));
}
size_t ggml_backend_buffer_get_max_size(ggml_backend_buffer_t buffer) {
return ggml_backend_buft_get_max_size(ggml_backend_buffer_get_type(buffer));
}
size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor) {
return ggml_backend_buft_get_alloc_size(ggml_backend_buffer_get_type(buffer), tensor);
}
@ -120,6 +130,11 @@ bool ggml_backend_buffer_is_host(ggml_backend_buffer_t buffer) {
void ggml_backend_buffer_set_usage(ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage) {
buffer->usage = usage;
// FIXME: add a generic callback to the buffer interface
if (ggml_backend_buffer_is_multi_buffer(buffer)) {
ggml_backend_multi_buffer_set_usage(buffer, usage);
}
}
ggml_backend_buffer_type_t ggml_backend_buffer_get_type(ggml_backend_buffer_t buffer) {
@ -169,6 +184,10 @@ size_t ggml_backend_get_alignment(ggml_backend_t backend) {
return ggml_backend_buft_get_alignment(ggml_backend_get_default_buffer_type(backend));
}
size_t ggml_backend_get_max_size(ggml_backend_t backend) {
return ggml_backend_buft_get_max_size(ggml_backend_get_default_buffer_type(backend));
}
void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
GGML_ASSERT(tensor->data != NULL && "tensor not allocated");
GGML_ASSERT(offset + size <= ggml_nbytes(tensor) && "tensor write out of bounds");
@ -342,6 +361,11 @@ GGML_CALL static void ggml_backend_registry_init(void) {
extern GGML_CALL ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void);
ggml_backend_register("Metal", ggml_backend_reg_metal_init, ggml_backend_metal_buffer_type(), NULL);
#endif
#ifdef GGML_USE_VULKAN
extern GGML_CALL int ggml_backend_vk_reg_devices(void);
ggml_backend_vk_reg_devices();
#endif
}
GGML_CALL void ggml_backend_register(const char * name, ggml_backend_init_fn init_fn, ggml_backend_buffer_type_t default_buffer_type, void * user_data) {
@ -545,6 +569,7 @@ GGML_CALL ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void) {
/* .get_name = */ ggml_backend_cpu_buffer_type_get_name,
/* .alloc_buffer = */ ggml_backend_cpu_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cpu_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ NULL, // defaults to ggml_nbytes
/* .supports_backend = */ ggml_backend_cpu_buffer_type_supports_backend,
/* .is_host = */ ggml_backend_cpu_buffer_type_is_host,
@ -600,6 +625,7 @@ ggml_backend_buffer_type_t ggml_backend_cpu_hbm_buffer_type(void) {
/* .get_name = */ ggml_backend_cpu_hbm_buffer_type_get_name,
/* .alloc_buffer = */ ggml_backend_cpu_hbm_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cpu_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ NULL, // defaults to ggml_nbytes
/* .supports_backend = */ ggml_backend_cpu_buffer_type_supports_backend,
/* .is_host = */ ggml_backend_cpu_buffer_type_is_host,
@ -756,6 +782,80 @@ GGML_CALL static ggml_backend_t ggml_backend_reg_cpu_init(const char * params, v
GGML_UNUSED(user_data);
}
// multi-buffer buffer
struct ggml_backend_multi_buffer_context {
ggml_backend_buffer_t * buffers;
size_t n_buffers;
};
typedef struct ggml_backend_multi_buffer_context * ggml_backend_multi_buffer_context_t;
GGML_CALL static const char * ggml_backend_multi_buffer_get_name(ggml_backend_buffer_t buffer) {
ggml_backend_multi_buffer_context_t ctx = (ggml_backend_multi_buffer_context_t) buffer->context;
return ctx->buffers[0]->iface.get_name(ctx->buffers[0]);
}
GGML_CALL static void ggml_backend_multi_buffer_free_buffer(ggml_backend_buffer_t buffer) {
ggml_backend_multi_buffer_context_t ctx = (ggml_backend_multi_buffer_context_t) buffer->context;
for (size_t i = 0; i < ctx->n_buffers; i++) {
ggml_backend_buffer_free(ctx->buffers[i]);
}
free(ctx->buffers);
free(ctx);
}
GGML_CALL static void ggml_backend_multi_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
ggml_backend_multi_buffer_context_t ctx = (ggml_backend_multi_buffer_context_t) buffer->context;
for (size_t i = 0; i < ctx->n_buffers; i++) {
ggml_backend_buffer_clear(ctx->buffers[i], value);
}
}
static struct ggml_backend_buffer_i ggml_backend_multi_buffer_context_interface(void) {
static struct ggml_backend_buffer_i multi_backend_buffer_i = {
/* .get_name = */ ggml_backend_multi_buffer_get_name,
/* .free_buffer = */ ggml_backend_multi_buffer_free_buffer,
/* .get_base = */ NULL,
/* .init_tensor = */ NULL,
/* .set_tensor = */ NULL,
/* .get_tensor = */ NULL,
/* .cpy_tensor = */ NULL,
/* .clear = */ ggml_backend_multi_buffer_clear,
/* .reset = */ NULL,
};
return multi_backend_buffer_i;
}
GGML_CALL ggml_backend_buffer_t ggml_backend_multi_buffer_alloc_buffer(ggml_backend_buffer_t * buffers, size_t n_buffers) {
ggml_backend_multi_buffer_context_t ctx = (ggml_backend_multi_buffer_context_t) malloc(sizeof(struct ggml_backend_multi_buffer_context));
ctx->n_buffers = n_buffers;
ctx->buffers = (ggml_backend_buffer_t *) malloc(n_buffers * sizeof(ggml_backend_buffer_t));
size_t total_size = 0;
for (size_t i = 0; i < n_buffers; i++) {
ctx->buffers[i] = buffers[i];
total_size += ggml_backend_buffer_get_size(buffers[i]);
}
return ggml_backend_buffer_init(buffers[0]->buft, ggml_backend_multi_buffer_context_interface(), ctx, total_size);
}
GGML_CALL bool ggml_backend_buffer_is_multi_buffer(ggml_backend_buffer_t buffer) {
return buffer->iface.get_name == ggml_backend_multi_buffer_get_name;
}
GGML_CALL void ggml_backend_multi_buffer_set_usage(ggml_backend_buffer_t buffer, enum ggml_backend_buffer_usage usage) {
GGML_ASSERT(ggml_backend_buffer_is_multi_buffer(buffer));
ggml_backend_multi_buffer_context_t ctx = (ggml_backend_multi_buffer_context_t) buffer->context;
for (size_t i = 0; i < ctx->n_buffers; i++) {
ggml_backend_buffer_set_usage(ctx->buffers[i], usage);
}
}
// scheduler

3
ggml-backend.h
View file

@ -20,6 +20,7 @@ extern "C" {
GGML_API const char * ggml_backend_buft_name (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL ggml_backend_buffer_t ggml_backend_buft_alloc_buffer (ggml_backend_buffer_type_t buft, size_t size);
GGML_API size_t ggml_backend_buft_get_alignment (ggml_backend_buffer_type_t buft);
GGML_API size_t ggml_backend_buft_get_max_size (ggml_backend_buffer_type_t buft);
GGML_API GGML_CALL size_t ggml_backend_buft_get_alloc_size (ggml_backend_buffer_type_t buft, struct ggml_tensor * tensor);
GGML_API bool ggml_backend_buft_supports_backend(ggml_backend_buffer_type_t buft, ggml_backend_t backend);
GGML_API bool ggml_backend_buft_is_host (ggml_backend_buffer_type_t buft);
@ -36,6 +37,7 @@ extern "C" {
GGML_API size_t ggml_backend_buffer_get_size (ggml_backend_buffer_t buffer);
GGML_API GGML_CALL void ggml_backend_buffer_init_tensor (ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API size_t ggml_backend_buffer_get_alignment (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_max_size (ggml_backend_buffer_t buffer);
GGML_API size_t ggml_backend_buffer_get_alloc_size(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor);
GGML_API void ggml_backend_buffer_clear (ggml_backend_buffer_t buffer, uint8_t value);
GGML_API bool ggml_backend_buffer_is_host (ggml_backend_buffer_t buffer);
@ -54,6 +56,7 @@ extern "C" {
GGML_API ggml_backend_buffer_type_t ggml_backend_get_default_buffer_type(ggml_backend_t backend);
GGML_API ggml_backend_buffer_t ggml_backend_alloc_buffer(ggml_backend_t backend, size_t size);
GGML_API size_t ggml_backend_get_alignment(ggml_backend_t backend);
GGML_API size_t ggml_backend_get_max_size(ggml_backend_t backend);
GGML_API void ggml_backend_tensor_set_async(ggml_backend_t backend, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
GGML_API void ggml_backend_tensor_get_async(ggml_backend_t backend, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size);

9
ggml-cuda.cu
View file

@ -4284,7 +4284,7 @@ static __device__ __forceinline__ float vec_dot_iq2_xxs_q8_1(
q8 += 8;
aux32 >>= 7;
}
const float d = (float)bq2->d * (0.5f + aux32) * (float)bq8_1[ib32].ds.x * 0.25f;
const float d = (float)bq2->d * (0.5f + aux32) * __low2float(bq8_1[ib32].ds) * 0.25f;
return d * sumi;
#else
// iqs is 0...15
@ -4295,7 +4295,7 @@ static __device__ __forceinline__ float vec_dot_iq2_xxs_q8_1(
const uint8_t * grid1 = (const uint8_t *)(iq2xxs_grid + aux8[2*il+0]);
const uint8_t * grid2 = (const uint8_t *)(iq2xxs_grid + aux8[2*il+1]);
const uint32_t aux32 = q2[2] | (q2[3] << 16);
const float d = (float)bq2->d * (0.5f + (aux32 >> 28)) * (float)bq8_1[ib32].ds.x * 0.25f;
const float d = (float)bq2->d * (0.5f + (aux32 >> 28)) * __low2float(bq8_1[ib32].ds) * 0.25f;
const uint8_t signs1 = ksigns_iq2xs[(aux32 >> 14*il) & 127];
const uint8_t signs2 = ksigns_iq2xs[(aux32 >> (14*il + 7)) & 127];
const int8_t * q8 = bq8_1[ib32].qs + 16*il;
@ -4340,7 +4340,7 @@ static __device__ __forceinline__ float vec_dot_iq2_xs_q8_1(
}
q8 += 8;
}
const float d = (float)bq2->d * (float)bq8_1[ib32].ds.x * 0.25f;
const float d = (float)bq2->d * __low2float(bq8_1[ib32].ds) * 0.25f;
return d * ((0.5f + ls1) * sumi1 + (0.5f + ls2) * sumi2);
#else
assert(false);
@ -10450,6 +10450,7 @@ static ggml_backend_buffer_type_i ggml_backend_cuda_buffer_type_interface = {
/* .get_name = */ ggml_backend_cuda_buffer_type_name,
/* .alloc_buffer = */ ggml_backend_cuda_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cuda_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ ggml_backend_cuda_buffer_type_get_alloc_size,
/* .supports_backend = */ ggml_backend_cuda_buffer_type_supports_backend,
/* .is_host = */ NULL,
@ -10725,6 +10726,7 @@ static ggml_backend_buffer_type_i ggml_backend_cuda_split_buffer_type_interface
/* .get_name = */ ggml_backend_cuda_split_buffer_type_name,
/* .alloc_buffer = */ ggml_backend_cuda_split_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cuda_split_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ ggml_backend_cuda_split_buffer_type_get_alloc_size,
/* .supports_backend = */ ggml_backend_cuda_split_buffer_type_supports_backend,
/* .is_host = */ ggml_backend_cuda_split_buffer_type_is_host,
@ -10804,6 +10806,7 @@ GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_host_buffer_type() {
/* .get_name = */ ggml_backend_cuda_host_buffer_type_name,
/* .alloc_buffer = */ ggml_backend_cuda_host_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cpu_buffer_type()->iface.get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
/* .supports_backend = */ ggml_backend_cpu_buffer_type()->iface.supports_backend,
/* .is_host = */ ggml_backend_cpu_buffer_type()->iface.is_host,

14
ggml-metal.m
View file

@ -277,6 +277,10 @@ static struct ggml_metal_context * ggml_metal_init(int n_cb) {
NSURL * libURL = [NSURL fileURLWithPath:libPath];
GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [libPath UTF8String]);
ctx->library = [ctx->device newLibraryWithURL:libURL error:&error];
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
} else {
GGML_METAL_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
@ -315,13 +319,12 @@ static struct ggml_metal_context * ggml_metal_init(int n_cb) {
//[options setFastMathEnabled:false];
ctx->library = [ctx->device newLibraryWithSource:src options:options error:&error];
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
}
}
if (error) {
GGML_METAL_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
return NULL;
}
}
// print MTL GPU family:
@ -2438,6 +2441,7 @@ GGML_CALL ggml_backend_buffer_type_t ggml_backend_metal_buffer_type(void) {
/* .get_name = */ ggml_backend_metal_buffer_type_get_name,
/* .alloc_buffer = */ ggml_backend_metal_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_metal_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // TODO: return device.maxBufferLength
/* .get_alloc_size = */ NULL, // defaults to ggml_nbytes
/* .supports_backend = */ ggml_backend_metal_buffer_type_supports_backend,
/* .is_host = */ ggml_backend_metal_buffer_type_is_host,

2
ggml-opencl.cpp
View file

@ -2066,6 +2066,7 @@ static ggml_backend_buffer_type_i ggml_backend_opencl_buffer_type_interface = {
/* .get_name = */ ggml_backend_opencl_buffer_type_name,
/* .alloc_buffer = */ ggml_backend_opencl_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_opencl_buffer_type_get_alignment,
/* .get_max_size = */ NULL, // TODO: return from device info
/* .get_alloc_size = */ NULL,
/* .supports_backend = */ ggml_backend_opencl_buffer_type_supports_backend,
/* .is_host = */ NULL,
@ -2122,6 +2123,7 @@ ggml_backend_buffer_type_t ggml_backend_opencl_host_buffer_type() {
/* .get_name = */ ggml_backend_opencl_host_buffer_type_name,
/* .alloc_buffer = */ ggml_backend_opencl_host_buffer_type_alloc_buffer,
/* .get_alignment = */ ggml_backend_cpu_buffer_type()->iface.get_alignment,
/* .get_max_size = */ NULL, // defaults to SIZE_MAX
/* .get_alloc_size = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
/* .supports_backend = */ ggml_backend_cpu_buffer_type()->iface.supports_backend,
/* .is_host = */ ggml_backend_cpu_buffer_type()->iface.is_host,

61447
ggml-vulkan-shaders.hpp Normal file
View file

File diff suppressed because it is too large Load diff

5141
ggml-vulkan.cpp Normal file
View file

File diff suppressed because it is too large Load diff

46
ggml-vulkan.h Normal file
View file

@ -0,0 +1,46 @@
#pragma once
#include "ggml.h"
#include "ggml-backend.h"
#ifdef __cplusplus
extern "C" {
#endif
#define GGML_VK_NAME "Vulkan"
GGML_API void ggml_vk_init(void);
GGML_API void ggml_vk_preallocate_buffers_graph(struct ggml_tensor * node);
GGML_API void ggml_vk_preallocate_buffers(void);
GGML_API void ggml_vk_build_graph(struct ggml_tensor * node, bool last_node);
GGML_API bool ggml_vk_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
#ifdef GGML_VULKAN_CHECK_RESULTS
void ggml_vk_check_results_0(struct ggml_compute_params * params, struct ggml_tensor * tensor);
void ggml_vk_check_results_1(struct ggml_compute_params * params, struct ggml_tensor * tensor);
#endif
GGML_API void ggml_vk_graph_cleanup(void);
GGML_API void * ggml_vk_host_malloc(size_t size);
GGML_API void ggml_vk_host_free(void * ptr);
GGML_API void ggml_vk_transform_tensor_temporary(const void * data, struct ggml_tensor * tensor);
GGML_API void ggml_vk_transform_tensor_static(const void * data, struct ggml_tensor * tensor);
GGML_API void ggml_vk_assign_buffer(struct ggml_tensor * tensor);
GGML_API void ggml_vk_prepare_tensor(struct ggml_tensor * tensor);
GGML_API void ggml_vk_cleanup(void);
GGML_API bool ggml_vk_can_mul_mat(const struct ggml_tensor * src0, const struct ggml_tensor * src1, const struct ggml_tensor * dst);
// backend API
GGML_API GGML_CALL ggml_backend_t ggml_backend_vk_init(void);
GGML_API GGML_CALL bool ggml_backend_is_vk(ggml_backend_t backend);
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(void);
// pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
GGML_API GGML_CALL ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
#ifdef __cplusplus
}
#endif

45
ggml.c
View file

@ -248,6 +248,8 @@ inline static void * ggml_aligned_malloc(size_t size) {
#include "ggml-cuda.h"
#elif defined(GGML_USE_CLBLAST)
#include "ggml-opencl.h"
#elif defined(GGML_USE_VULKAN)
#include "ggml-vulkan.h"
#endif
// floating point type used to accumulate sums
@ -2293,6 +2295,8 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
ggml_init_cublas();
#elif defined(GGML_USE_CLBLAST)
ggml_cl_init();
#elif defined(GGML_USE_VULKAN)
ggml_vk_init();
#endif
ggml_setup_op_has_task_pass();
@ -7999,7 +8003,7 @@ static void ggml_compute_forward_mul_f32(
const int ith = params->ith;
const int nth = params->nth;
#ifdef GGML_USE_CLBLAST
#if defined(GGML_USE_CLBLAST)
if (src1->backend == GGML_BACKEND_GPU) {
// TODO: OpenCL kernel support full broadcast
GGML_ASSERT(ggml_can_repeat_rows(src1, src0));
@ -14683,6 +14687,18 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
}
GGML_ASSERT(tensor->src[0] == NULL || tensor->src[0]->backend == GGML_BACKEND_CPU);
GGML_ASSERT(tensor->src[1] == NULL || tensor->src[1]->backend == GGML_BACKEND_CPU);
#elif defined(GGML_USE_VULKAN)
const bool skip_cpu = ggml_vk_compute_forward(params, tensor);
#ifdef GGML_VULKAN_CHECK_RESULTS
if (skip_cpu) {
ggml_vk_check_results_1(params, tensor);
}
#endif
if (skip_cpu) {
return;
}
GGML_ASSERT(tensor->src[0] == NULL || tensor->src[0]->backend == GGML_BACKEND_CPU);
GGML_ASSERT(tensor->src[1] == NULL || tensor->src[1]->backend == GGML_BACKEND_CPU);
#endif // GGML_USE_CUBLAS
switch (tensor->op) {
@ -17079,6 +17095,17 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
}
}
#ifdef GGML_USE_VULKAN
for (int i = 0; i < cgraph->n_nodes; i++) {
ggml_vk_preallocate_buffers_graph(cgraph->nodes[i]);
}
ggml_vk_preallocate_buffers();
for (int i = 0; i < cgraph->n_nodes; i++) {
ggml_vk_build_graph(cgraph->nodes[i], i == cgraph->n_nodes - 1);
}
#endif
const int n_threads = cplan->n_threads;
struct ggml_compute_state_shared state_shared = {
@ -17130,6 +17157,10 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
}
}
#ifdef GGML_USE_VULKAN
ggml_vk_graph_cleanup();
#endif
// performance stats (graph)
{
int64_t perf_cycles_cur = ggml_perf_cycles() - perf_start_cycles;
@ -20309,7 +20340,7 @@ int ggml_cpu_has_wasm_simd(void) {
}
int ggml_cpu_has_blas(void) {
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUBLAS) || defined(GGML_USE_CLBLAST)
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUBLAS) || defined(GGML_USE_VULKAN) || defined(GGML_USE_CLBLAST)
return 1;
#else
return 0;
@ -20332,8 +20363,16 @@ int ggml_cpu_has_clblast(void) {
#endif
}
int ggml_cpu_has_vulkan(void) {
#if defined(GGML_USE_VULKAN)
return 1;
#else
return 0;
#endif
}
int ggml_cpu_has_gpublas(void) {
return ggml_cpu_has_cublas() || ggml_cpu_has_clblast();
return ggml_cpu_has_cublas() || ggml_cpu_has_clblast() || ggml_cpu_has_vulkan();
}
int ggml_cpu_has_sse3(void) {

1
ggml.h
View file

@ -2270,6 +2270,7 @@ extern "C" {
GGML_API int ggml_cpu_has_blas (void);
GGML_API int ggml_cpu_has_cublas (void);
GGML_API int ggml_cpu_has_clblast (void);
GGML_API int ggml_cpu_has_vulkan (void);
GGML_API int ggml_cpu_has_gpublas (void);
GGML_API int ggml_cpu_has_sse3 (void);
GGML_API int ggml_cpu_has_ssse3 (void);

2363
ggml_vk_generate_shaders.py Normal file
View file

File diff suppressed because it is too large Load diff

BIN
glslc.exe Normal file
View file

Binary file not shown.

1099
include/vulkan/dxc/WinAdapter.h Normal file
View file

File diff suppressed because it is too large Load diff

1218
include/vulkan/dxc/dxcapi.h Normal file
View file

File diff suppressed because it is too large Load diff

623
include/vulkan/glslang/Include/BaseTypes.h Normal file
View file

@ -0,0 +1,623 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2012-2013 LunarG, Inc.
// Copyright (C) 2017 ARM Limited.
// Modifications Copyright (C) 2020 Advanced Micro Devices, Inc. All rights reserved.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _BASICTYPES_INCLUDED_
#define _BASICTYPES_INCLUDED_
namespace glslang {
//
// Basic type. Arrays, vectors, sampler details, etc., are orthogonal to this.
//
enum TBasicType {
EbtVoid,
EbtFloat,
EbtDouble,
EbtFloat16,
EbtInt8,
EbtUint8,
EbtInt16,
EbtUint16,
EbtInt,
EbtUint,
EbtInt64,
EbtUint64,
EbtBool,
EbtAtomicUint,
EbtSampler,
EbtStruct,
EbtBlock,
EbtAccStruct,
EbtReference,
EbtRayQuery,
EbtHitObjectNV,
#ifndef GLSLANG_WEB
// SPIR-V type defined by spirv_type
EbtSpirvType,
#endif
// HLSL types that live only temporarily.
EbtString,
EbtNumTypes
};
//
// Storage qualifiers. Should align with different kinds of storage or
// resource or GLSL storage qualifier. Expansion is deprecated.
//
// N.B.: You probably DON'T want to add anything here, but rather just add it
// to the built-in variables. See the comment above TBuiltInVariable.
//
// A new built-in variable will normally be an existing qualifier, like 'in', 'out', etc.
// DO NOT follow the design pattern of, say EvqInstanceId, etc.
//
enum TStorageQualifier {
EvqTemporary, // For temporaries (within a function), read/write
EvqGlobal, // For globals read/write
EvqConst, // User-defined constant values, will be semantically constant and constant folded
EvqVaryingIn, // pipeline input, read only, also supercategory for all built-ins not included in this enum (see TBuiltInVariable)
EvqVaryingOut, // pipeline output, read/write, also supercategory for all built-ins not included in this enum (see TBuiltInVariable)
EvqUniform, // read only, shared with app
EvqBuffer, // read/write, shared with app
EvqShared, // compute shader's read/write 'shared' qualifier
#ifndef GLSLANG_WEB
EvqSpirvStorageClass, // spirv_storage_class
#endif
EvqPayload,
EvqPayloadIn,
EvqHitAttr,
EvqCallableData,
EvqCallableDataIn,
EvqHitObjectAttrNV,
EvqtaskPayloadSharedEXT,
// parameters
EvqIn, // also, for 'in' in the grammar before we know if it's a pipeline input or an 'in' parameter
EvqOut, // also, for 'out' in the grammar before we know if it's a pipeline output or an 'out' parameter
EvqInOut,
EvqConstReadOnly, // input; also other read-only types having neither a constant value nor constant-value semantics
// built-ins read by vertex shader
EvqVertexId,
EvqInstanceId,
// built-ins written by vertex shader
EvqPosition,
EvqPointSize,
EvqClipVertex,
// built-ins read by fragment shader
EvqFace,
EvqFragCoord,
EvqPointCoord,
// built-ins written by fragment shader
EvqFragColor,
EvqFragDepth,
EvqFragStencil,
EvqTileImageEXT,
// end of list
EvqLast
};
//
// Subcategories of the TStorageQualifier, simply to give a direct mapping
// between built-in variable names and an numerical value (the enum).
//
// For backward compatibility, there is some redundancy between the
// TStorageQualifier and these. Existing members should both be maintained accurately.
// However, any new built-in variable (and any existing non-redundant one)
// must follow the pattern that the specific built-in is here, and only its
// general qualifier is in TStorageQualifier.
//
// Something like gl_Position, which is sometimes 'in' and sometimes 'out'
// shows up as two different built-in variables in a single stage, but
// only has a single enum in TBuiltInVariable, so both the
// TStorageQualifier and the TBuitinVariable are needed to distinguish
// between them.
//
enum TBuiltInVariable {
EbvNone,
EbvNumWorkGroups,
EbvWorkGroupSize,
EbvWorkGroupId,
EbvLocalInvocationId,
EbvGlobalInvocationId,
EbvLocalInvocationIndex,
EbvNumSubgroups,
EbvSubgroupID,
EbvSubGroupSize,
EbvSubGroupInvocation,
EbvSubGroupEqMask,
EbvSubGroupGeMask,
EbvSubGroupGtMask,
EbvSubGroupLeMask,
EbvSubGroupLtMask,
EbvSubgroupSize2,
EbvSubgroupInvocation2,
EbvSubgroupEqMask2,
EbvSubgroupGeMask2,
EbvSubgroupGtMask2,
EbvSubgroupLeMask2,
EbvSubgroupLtMask2,
EbvVertexId,
EbvInstanceId,
EbvVertexIndex,
EbvInstanceIndex,
EbvBaseVertex,
EbvBaseInstance,
EbvDrawId,
EbvPosition,
EbvPointSize,
EbvClipVertex,
EbvClipDistance,
EbvCullDistance,
EbvNormal,
EbvVertex,
EbvMultiTexCoord0,
EbvMultiTexCoord1,
EbvMultiTexCoord2,
EbvMultiTexCoord3,
EbvMultiTexCoord4,
EbvMultiTexCoord5,
EbvMultiTexCoord6,
EbvMultiTexCoord7,
EbvFrontColor,
EbvBackColor,
EbvFrontSecondaryColor,
EbvBackSecondaryColor,
EbvTexCoord,
EbvFogFragCoord,
EbvInvocationId,
EbvPrimitiveId,
EbvLayer,
EbvViewportIndex,
EbvPatchVertices,
EbvTessLevelOuter,
EbvTessLevelInner,
EbvBoundingBox,
EbvTessCoord,
EbvColor,
EbvSecondaryColor,
EbvFace,
EbvFragCoord,
EbvPointCoord,
EbvFragColor,
EbvFragData,
EbvFragDepth,
EbvFragStencilRef,
EbvSampleId,
EbvSamplePosition,
EbvSampleMask,
EbvHelperInvocation,
EbvBaryCoordNoPersp,
EbvBaryCoordNoPerspCentroid,
EbvBaryCoordNoPerspSample,
EbvBaryCoordSmooth,
EbvBaryCoordSmoothCentroid,
EbvBaryCoordSmoothSample,
EbvBaryCoordPullModel,
EbvViewIndex,
EbvDeviceIndex,
EbvShadingRateKHR,
EbvPrimitiveShadingRateKHR,
EbvFragSizeEXT,
EbvFragInvocationCountEXT,
EbvSecondaryFragDataEXT,
EbvSecondaryFragColorEXT,
EbvViewportMaskNV,
EbvSecondaryPositionNV,
EbvSecondaryViewportMaskNV,
EbvPositionPerViewNV,
EbvViewportMaskPerViewNV,
EbvFragFullyCoveredNV,
EbvFragmentSizeNV,
EbvInvocationsPerPixelNV,
// ray tracing
EbvLaunchId,
EbvLaunchSize,
EbvInstanceCustomIndex,
EbvGeometryIndex,
EbvWorldRayOrigin,
EbvWorldRayDirection,
EbvObjectRayOrigin,
EbvObjectRayDirection,
EbvRayTmin,
EbvRayTmax,
EbvCullMask,
EbvHitT,
EbvHitKind,
EbvObjectToWorld,
EbvObjectToWorld3x4,
EbvWorldToObject,
EbvWorldToObject3x4,
EbvIncomingRayFlags,
EbvCurrentRayTimeNV,
// barycentrics
EbvBaryCoordNV,
EbvBaryCoordNoPerspNV,
EbvBaryCoordEXT,
EbvBaryCoordNoPerspEXT,
// mesh shaders
EbvTaskCountNV,
EbvPrimitiveCountNV,
EbvPrimitiveIndicesNV,
EbvClipDistancePerViewNV,
EbvCullDistancePerViewNV,
EbvLayerPerViewNV,
EbvMeshViewCountNV,
EbvMeshViewIndicesNV,
//GL_EXT_mesh_shader
EbvPrimitivePointIndicesEXT,
EbvPrimitiveLineIndicesEXT,
EbvPrimitiveTriangleIndicesEXT,
EbvCullPrimitiveEXT,
// sm builtins
EbvWarpsPerSM,
EbvSMCount,
EbvWarpID,
EbvSMID,
// HLSL built-ins that live only temporarily, until they get remapped
// to one of the above.
EbvFragDepthGreater,
EbvFragDepthLesser,
EbvGsOutputStream,
EbvOutputPatch,
EbvInputPatch,
// structbuffer types
EbvAppendConsume, // no need to differentiate append and consume
EbvRWStructuredBuffer,
EbvStructuredBuffer,
EbvByteAddressBuffer,
EbvRWByteAddressBuffer,
// ARM specific core builtins
EbvCoreCountARM,
EbvCoreIDARM,
EbvCoreMaxIDARM,
EbvWarpIDARM,
EbvWarpMaxIDARM,
EbvPositionFetch,
EbvLast
};
// In this enum, order matters; users can assume higher precision is a bigger value
// and EpqNone is 0.
enum TPrecisionQualifier {
EpqNone = 0,
EpqLow,
EpqMedium,
EpqHigh
};
#ifdef GLSLANG_WEB
__inline const char* GetStorageQualifierString(TStorageQualifier q) { return ""; }
__inline const char* GetPrecisionQualifierString(TPrecisionQualifier p) { return ""; }
#else
// These will show up in error messages
__inline const char* GetStorageQualifierString(TStorageQualifier q)
{
switch (q) {
case EvqTemporary: return "temp"; break;
case EvqGlobal: return "global"; break;
case EvqConst: return "const"; break;
case EvqConstReadOnly: return "const (read only)"; break;
#ifndef GLSLANG_WEB
case EvqSpirvStorageClass: return "spirv_storage_class"; break;
#endif
case EvqVaryingIn: return "in"; break;
case EvqVaryingOut: return "out"; break;
case EvqUniform: return "uniform"; break;
case EvqBuffer: return "buffer"; break;
case EvqShared: return "shared"; break;
case EvqIn: return "in"; break;
case EvqOut: return "out"; break;
case EvqInOut: return "inout"; break;
case EvqVertexId: return "gl_VertexId"; break;
case EvqInstanceId: return "gl_InstanceId"; break;
case EvqPosition: return "gl_Position"; break;
case EvqPointSize: return "gl_PointSize"; break;
case EvqClipVertex: return "gl_ClipVertex"; break;
case EvqFace: return "gl_FrontFacing"; break;
case EvqFragCoord: return "gl_FragCoord"; break;
case EvqPointCoord: return "gl_PointCoord"; break;
case EvqFragColor: return "fragColor"; break;
case EvqFragDepth: return "gl_FragDepth"; break;
case EvqFragStencil: return "gl_FragStencilRefARB"; break;
case EvqPayload: return "rayPayloadNV"; break;
case EvqPayloadIn: return "rayPayloadInNV"; break;
case EvqHitAttr: return "hitAttributeNV"; break;
case EvqCallableData: return "callableDataNV"; break;
case EvqCallableDataIn: return "callableDataInNV"; break;
case EvqtaskPayloadSharedEXT: return "taskPayloadSharedEXT"; break;
case EvqHitObjectAttrNV:return "hitObjectAttributeNV"; break;
default: return "unknown qualifier";
}
}
__inline const char* GetBuiltInVariableString(TBuiltInVariable v)
{
switch (v) {
case EbvNone: return "";
case EbvNumWorkGroups: return "NumWorkGroups";
case EbvWorkGroupSize: return "WorkGroupSize";
case EbvWorkGroupId: return "WorkGroupID";
case EbvLocalInvocationId: return "LocalInvocationID";
case EbvGlobalInvocationId: return "GlobalInvocationID";
case EbvLocalInvocationIndex: return "LocalInvocationIndex";
case EbvNumSubgroups: return "NumSubgroups";
case EbvSubgroupID: return "SubgroupID";
case EbvSubGroupSize: return "SubGroupSize";
case EbvSubGroupInvocation: return "SubGroupInvocation";
case EbvSubGroupEqMask: return "SubGroupEqMask";
case EbvSubGroupGeMask: return "SubGroupGeMask";
case EbvSubGroupGtMask: return "SubGroupGtMask";
case EbvSubGroupLeMask: return "SubGroupLeMask";
case EbvSubGroupLtMask: return "SubGroupLtMask";
case EbvSubgroupSize2: return "SubgroupSize";
case EbvSubgroupInvocation2: return "SubgroupInvocationID";
case EbvSubgroupEqMask2: return "SubgroupEqMask";
case EbvSubgroupGeMask2: return "SubgroupGeMask";
case EbvSubgroupGtMask2: return "SubgroupGtMask";
case EbvSubgroupLeMask2: return "SubgroupLeMask";
case EbvSubgroupLtMask2: return "SubgroupLtMask";
case EbvVertexId: return "VertexId";
case EbvInstanceId: return "InstanceId";
case EbvVertexIndex: return "VertexIndex";
case EbvInstanceIndex: return "InstanceIndex";
case EbvBaseVertex: return "BaseVertex";
case EbvBaseInstance: return "BaseInstance";
case EbvDrawId: return "DrawId";
case EbvPosition: return "Position";
case EbvPointSize: return "PointSize";
case EbvClipVertex: return "ClipVertex";
case EbvClipDistance: return "ClipDistance";
case EbvCullDistance: return "CullDistance";
case EbvNormal: return "Normal";
case EbvVertex: return "Vertex";
case EbvMultiTexCoord0: return "MultiTexCoord0";
case EbvMultiTexCoord1: return "MultiTexCoord1";
case EbvMultiTexCoord2: return "MultiTexCoord2";
case EbvMultiTexCoord3: return "MultiTexCoord3";
case EbvMultiTexCoord4: return "MultiTexCoord4";
case EbvMultiTexCoord5: return "MultiTexCoord5";
case EbvMultiTexCoord6: return "MultiTexCoord6";
case EbvMultiTexCoord7: return "MultiTexCoord7";
case EbvFrontColor: return "FrontColor";
case EbvBackColor: return "BackColor";
case EbvFrontSecondaryColor: return "FrontSecondaryColor";
case EbvBackSecondaryColor: return "BackSecondaryColor";
case EbvTexCoord: return "TexCoord";
case EbvFogFragCoord: return "FogFragCoord";
case EbvInvocationId: return "InvocationID";
case EbvPrimitiveId: return "PrimitiveID";
case EbvLayer: return "Layer";
case EbvViewportIndex: return "ViewportIndex";
case EbvPatchVertices: return "PatchVertices";
case EbvTessLevelOuter: return "TessLevelOuter";
case EbvTessLevelInner: return "TessLevelInner";
case EbvBoundingBox: return "BoundingBox";
case EbvTessCoord: return "TessCoord";
case EbvColor: return "Color";
case EbvSecondaryColor: return "SecondaryColor";
case EbvFace: return "Face";
case EbvFragCoord: return "FragCoord";
case EbvPointCoord: return "PointCoord";
case EbvFragColor: return "FragColor";
case EbvFragData: return "FragData";
case EbvFragDepth: return "FragDepth";
case EbvFragStencilRef: return "FragStencilRef";
case EbvSampleId: return "SampleId";
case EbvSamplePosition: return "SamplePosition";
case EbvSampleMask: return "SampleMaskIn";
case EbvHelperInvocation: return "HelperInvocation";
case EbvBaryCoordNoPersp: return "BaryCoordNoPersp";
case EbvBaryCoordNoPerspCentroid: return "BaryCoordNoPerspCentroid";
case EbvBaryCoordNoPerspSample: return "BaryCoordNoPerspSample";
case EbvBaryCoordSmooth: return "BaryCoordSmooth";
case EbvBaryCoordSmoothCentroid: return "BaryCoordSmoothCentroid";
case EbvBaryCoordSmoothSample: return "BaryCoordSmoothSample";
case EbvBaryCoordPullModel: return "BaryCoordPullModel";
case EbvViewIndex: return "ViewIndex";
case EbvDeviceIndex: return "DeviceIndex";
case EbvFragSizeEXT: return "FragSizeEXT";
case EbvFragInvocationCountEXT: return "FragInvocationCountEXT";
case EbvSecondaryFragDataEXT: return "SecondaryFragDataEXT";
case EbvSecondaryFragColorEXT: return "SecondaryFragColorEXT";
case EbvViewportMaskNV: return "ViewportMaskNV";
case EbvSecondaryPositionNV: return "SecondaryPositionNV";
case EbvSecondaryViewportMaskNV: return "SecondaryViewportMaskNV";
case EbvPositionPerViewNV: return "PositionPerViewNV";
case EbvViewportMaskPerViewNV: return "ViewportMaskPerViewNV";
case EbvFragFullyCoveredNV: return "FragFullyCoveredNV";
case EbvFragmentSizeNV: return "FragmentSizeNV";
case EbvInvocationsPerPixelNV: return "InvocationsPerPixelNV";
case EbvLaunchId: return "LaunchIdNV";
case EbvLaunchSize: return "LaunchSizeNV";
case EbvInstanceCustomIndex: return "InstanceCustomIndexNV";
case EbvGeometryIndex: return "GeometryIndexEXT";
case EbvWorldRayOrigin: return "WorldRayOriginNV";
case EbvWorldRayDirection: return "WorldRayDirectionNV";
case EbvObjectRayOrigin: return "ObjectRayOriginNV";
case EbvObjectRayDirection: return "ObjectRayDirectionNV";
case EbvRayTmin: return "ObjectRayTminNV";
case EbvRayTmax: return "ObjectRayTmaxNV";
case EbvHitT: return "HitTNV";
case EbvHitKind: return "HitKindNV";
case EbvIncomingRayFlags: return "IncomingRayFlagsNV";
case EbvObjectToWorld: return "ObjectToWorldNV";
case EbvWorldToObject: return "WorldToObjectNV";
case EbvCurrentRayTimeNV: return "CurrentRayTimeNV";
case EbvBaryCoordEXT:
case EbvBaryCoordNV: return "BaryCoordKHR";
case EbvBaryCoordNoPerspEXT:
case EbvBaryCoordNoPerspNV: return "BaryCoordNoPerspKHR";
case EbvTaskCountNV: return "TaskCountNV";
case EbvPrimitiveCountNV: return "PrimitiveCountNV";
case EbvPrimitiveIndicesNV: return "PrimitiveIndicesNV";
case EbvClipDistancePerViewNV: return "ClipDistancePerViewNV";
case EbvCullDistancePerViewNV: return "CullDistancePerViewNV";
case EbvLayerPerViewNV: return "LayerPerViewNV";
case EbvMeshViewCountNV: return "MeshViewCountNV";
case EbvMeshViewIndicesNV: return "MeshViewIndicesNV";
// GL_EXT_mesh_shader
case EbvPrimitivePointIndicesEXT: return "PrimitivePointIndicesEXT";
case EbvPrimitiveLineIndicesEXT: return "PrimitiveLineIndicesEXT";
case EbvPrimitiveTriangleIndicesEXT: return "PrimitiveTriangleIndicesEXT";
case EbvCullPrimitiveEXT: return "CullPrimitiveEXT";
case EbvWarpsPerSM: return "WarpsPerSMNV";
case EbvSMCount: return "SMCountNV";
case EbvWarpID: return "WarpIDNV";
case EbvSMID: return "SMIDNV";
case EbvShadingRateKHR: return "ShadingRateKHR";
case EbvPrimitiveShadingRateKHR: return "PrimitiveShadingRateKHR";
default: return "unknown built-in variable";
}
}
__inline const char* GetPrecisionQualifierString(TPrecisionQualifier p)
{
switch (p) {
case EpqNone: return ""; break;
case EpqLow: return "lowp"; break;
case EpqMedium: return "mediump"; break;
case EpqHigh: return "highp"; break;
default: return "unknown precision qualifier";
}
}
#endif
__inline bool isTypeSignedInt(TBasicType type)
{
switch (type) {
case EbtInt8:
case EbtInt16:
case EbtInt:
case EbtInt64:
return true;
default:
return false;
}
}
__inline bool isTypeUnsignedInt(TBasicType type)
{
switch (type) {
case EbtUint8:
case EbtUint16:
case EbtUint:
case EbtUint64:
return true;
default:
return false;
}
}
__inline bool isTypeInt(TBasicType type)
{
return isTypeSignedInt(type) || isTypeUnsignedInt(type);
}
__inline bool isTypeFloat(TBasicType type)
{
switch (type) {
case EbtFloat:
case EbtDouble:
case EbtFloat16:
return true;
default:
return false;
}
}
__inline int getTypeRank(TBasicType type)
{
int res = -1;
switch(type) {
case EbtInt8:
case EbtUint8:
res = 0;
break;
case EbtInt16:
case EbtUint16:
res = 1;
break;
case EbtInt:
case EbtUint:
res = 2;
break;
case EbtInt64:
case EbtUint64:
res = 3;
break;
default:
assert(false);
break;
}
return res;
}
} // end namespace glslang
#endif // _BASICTYPES_INCLUDED_

325
include/vulkan/glslang/Include/Common.h Normal file
View file

@ -0,0 +1,325 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2012-2013 LunarG, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _COMMON_INCLUDED_
#define _COMMON_INCLUDED_
#include <algorithm>
#include <cassert>
#ifdef _MSC_VER
#include <cfloat>
#else
#include <cmath>
#endif
#include <cstdint>
#include <cstdio>
#include <cstdlib>
#include <list>
#include <map>
#include <set>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <vector>
#if defined(__ANDROID__)
#include <sstream>
namespace std {
template<typename T>
std::string to_string(const T& val) {
std::ostringstream os;
os << val;
return os.str();
}
}
#endif
#if defined(MINGW_HAS_SECURE_API) && MINGW_HAS_SECURE_API
#include <basetsd.h>
#ifndef snprintf
#define snprintf sprintf_s
#endif
#define safe_vsprintf(buf,max,format,args) vsnprintf_s((buf), (max), (max), (format), (args))
#elif defined (solaris)
#define safe_vsprintf(buf,max,format,args) vsnprintf((buf), (max), (format), (args))
#include <sys/int_types.h>
#define UINT_PTR uintptr_t
#else
#define safe_vsprintf(buf,max,format,args) vsnprintf((buf), (max), (format), (args))
#include <stdint.h>
#define UINT_PTR uintptr_t
#endif
#if defined(_MSC_VER)
#define strdup _strdup
#endif
/* windows only pragma */
#ifdef _MSC_VER
#pragma warning(disable : 4786) // Don't warn about too long identifiers
#pragma warning(disable : 4514) // unused inline method
#pragma warning(disable : 4201) // nameless union
#endif
#include "PoolAlloc.h"
//
// Put POOL_ALLOCATOR_NEW_DELETE in base classes to make them use this scheme.
//
#define POOL_ALLOCATOR_NEW_DELETE(A) \
void* operator new(size_t s) { return (A).allocate(s); } \
void* operator new(size_t, void *_Where) { return (_Where); } \
void operator delete(void*) { } \
void operator delete(void *, void *) { } \
void* operator new[](size_t s) { return (A).allocate(s); } \
void* operator new[](size_t, void *_Where) { return (_Where); } \
void operator delete[](void*) { } \
void operator delete[](void *, void *) { }
namespace glslang {
//
// Pool version of string.
//
typedef pool_allocator<char> TStringAllocator;
typedef std::basic_string <char, std::char_traits<char>, TStringAllocator> TString;
} // end namespace glslang
// Repackage the std::hash for use by unordered map/set with a TString key.
namespace std {
template<> struct hash<glslang::TString> {
std::size_t operator()(const glslang::TString& s) const
{
const unsigned _FNV_offset_basis = 2166136261U;
const unsigned _FNV_prime = 16777619U;
unsigned _Val = _FNV_offset_basis;
size_t _Count = s.size();
const char* _First = s.c_str();
for (size_t _Next = 0; _Next < _Count; ++_Next)
{
_Val ^= (unsigned)_First[_Next];
_Val *= _FNV_prime;
}
return _Val;
}
};
}
namespace glslang {
inline TString* NewPoolTString(const char* s)
{
void* memory = GetThreadPoolAllocator().allocate(sizeof(TString));
return new(memory) TString(s);
}
template<class T> inline T* NewPoolObject(T*)
{
return new(GetThreadPoolAllocator().allocate(sizeof(T))) T;
}
template<class T> inline T* NewPoolObject(T, int instances)
{
return new(GetThreadPoolAllocator().allocate(instances * sizeof(T))) T[instances];
}
//
// Pool allocator versions of vectors, lists, and maps
//
template <class T> class TVector : public std::vector<T, pool_allocator<T> > {
public:
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
typedef typename std::vector<T, pool_allocator<T> >::size_type size_type;
TVector() : std::vector<T, pool_allocator<T> >() {}
TVector(const pool_allocator<T>& a) : std::vector<T, pool_allocator<T> >(a) {}
TVector(size_type i) : std::vector<T, pool_allocator<T> >(i) {}
TVector(size_type i, const T& val) : std::vector<T, pool_allocator<T> >(i, val) {}
};
template <class T> class TList : public std::list<T, pool_allocator<T> > {
};
template <class K, class D, class CMP = std::less<K> >
class TMap : public std::map<K, D, CMP, pool_allocator<std::pair<K const, D> > > {
};
template <class K, class D, class HASH = std::hash<K>, class PRED = std::equal_to<K> >
class TUnorderedMap : public std::unordered_map<K, D, HASH, PRED, pool_allocator<std::pair<K const, D> > > {
};
template <class K, class CMP = std::less<K> >
class TSet : public std::set<K, CMP, pool_allocator<K> > {
};
//
// Persistent string memory. Should only be used for strings that survive
// across compiles/links.
//
typedef std::basic_string<char> TPersistString;
//
// templatized min and max functions.
//
template <class T> T Min(const T a, const T b) { return a < b ? a : b; }
template <class T> T Max(const T a, const T b) { return a > b ? a : b; }
//
// Create a TString object from an integer.
//
#if defined(_MSC_VER) || (defined(MINGW_HAS_SECURE_API) && MINGW_HAS_SECURE_API)
inline const TString String(const int i, const int base = 10)
{
char text[16]; // 32 bit ints are at most 10 digits in base 10
_itoa_s(i, text, sizeof(text), base);
return text;
}
#else
inline const TString String(const int i, const int /*base*/ = 10)
{
char text[16]; // 32 bit ints are at most 10 digits in base 10
// we assume base 10 for all cases
snprintf(text, sizeof(text), "%d", i);
return text;
}
#endif
struct TSourceLoc {
void init()
{
name = nullptr; string = 0; line = 0; column = 0;
}
void init(int stringNum) { init(); string = stringNum; }
// Returns the name if it exists. Otherwise, returns the string number.
std::string getStringNameOrNum(bool quoteStringName = true) const
{
if (name != nullptr) {
TString qstr = quoteStringName ? ("\"" + *name + "\"") : *name;
std::string ret_str(qstr.c_str());
return ret_str;
}
return std::to_string((long long)string);
}
const char* getFilename() const
{
if (name == nullptr)
return nullptr;
return name->c_str();
}
const char* getFilenameStr() const { return name == nullptr ? "" : name->c_str(); }
TString* name; // descriptive name for this string, when a textual name is available, otherwise nullptr
int string;
int line;
int column;
};
class TPragmaTable : public TMap<TString, TString> {
public:
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
};
const int MaxTokenLength = 1024;
template <class T> bool IsPow2(T powerOf2)
{
if (powerOf2 <= 0)
return false;
return (powerOf2 & (powerOf2 - 1)) == 0;
}
// Round number up to a multiple of the given powerOf2, which is not
// a power, just a number that must be a power of 2.
template <class T> void RoundToPow2(T& number, int powerOf2)
{
assert(IsPow2(powerOf2));
number = (number + powerOf2 - 1) & ~(powerOf2 - 1);
}
template <class T> bool IsMultipleOfPow2(T number, int powerOf2)
{
assert(IsPow2(powerOf2));
return ! (number & (powerOf2 - 1));
}
// Returns log2 of an integer power of 2.
// T should be integral.
template <class T> int IntLog2(T n)
{
assert(IsPow2(n));
int result = 0;
while ((T(1) << result) != n) {
result++;
}
return result;
}
inline bool IsInfinity(double x) {
#ifdef _MSC_VER
switch (_fpclass(x)) {
case _FPCLASS_NINF:
case _FPCLASS_PINF:
return true;
default:
return false;
}
#else
return std::isinf(x);
#endif
}
inline bool IsNan(double x) {
#ifdef _MSC_VER
switch (_fpclass(x)) {
case _FPCLASS_SNAN:
case _FPCLASS_QNAN:
return true;
default:
return false;
}
#else
return std::isnan(x);
#endif
}
} // end namespace glslang
#endif // _COMMON_INCLUDED_

974
include/vulkan/glslang/Include/ConstantUnion.h Normal file
View file

@ -0,0 +1,974 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2013 LunarG, Inc.
// Copyright (C) 2017 ARM Limited.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _CONSTANT_UNION_INCLUDED_
#define _CONSTANT_UNION_INCLUDED_
#include "../Include/Common.h"
#include "../Include/BaseTypes.h"
namespace glslang {
class TConstUnion {
public:
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TConstUnion() : iConst(0), type(EbtInt) { }
void setI8Const(signed char i)
{
i8Const = i;
type = EbtInt8;
}
void setU8Const(unsigned char u)
{
u8Const = u;
type = EbtUint8;
}
void setI16Const(signed short i)
{
i16Const = i;
type = EbtInt16;
}
void setU16Const(unsigned short u)
{
u16Const = u;
type = EbtUint16;
}
void setIConst(int i)
{
iConst = i;
type = EbtInt;
}
void setUConst(unsigned int u)
{
uConst = u;
type = EbtUint;
}
void setI64Const(long long i64)
{
i64Const = i64;
type = EbtInt64;
}
void setU64Const(unsigned long long u64)
{
u64Const = u64;
type = EbtUint64;
}
void setDConst(double d)
{
dConst = d;
type = EbtDouble;
}
void setBConst(bool b)
{
bConst = b;
type = EbtBool;
}
void setSConst(const TString* s)
{
sConst = s;
type = EbtString;
}
signed char getI8Const() const { return i8Const; }
unsigned char getU8Const() const { return u8Const; }
signed short getI16Const() const { return i16Const; }
unsigned short getU16Const() const { return u16Const; }
int getIConst() const { return iConst; }
unsigned int getUConst() const { return uConst; }
long long getI64Const() const { return i64Const; }
unsigned long long getU64Const() const { return u64Const; }
double getDConst() const { return dConst; }
bool getBConst() const { return bConst; }
const TString* getSConst() const { return sConst; }
bool operator==(const signed char i) const
{
if (i == i8Const)
return true;
return false;
}
bool operator==(const unsigned char u) const
{
if (u == u8Const)
return true;
return false;
}
bool operator==(const signed short i) const
{
if (i == i16Const)
return true;
return false;
}
bool operator==(const unsigned short u) const
{
if (u == u16Const)
return true;
return false;
}
bool operator==(const int i) const
{
if (i == iConst)
return true;
return false;
}
bool operator==(const unsigned int u) const
{
if (u == uConst)
return true;
return false;
}
bool operator==(const long long i64) const
{
if (i64 == i64Const)
return true;
return false;
}
bool operator==(const unsigned long long u64) const
{
if (u64 == u64Const)
return true;
return false;
}
bool operator==(const double d) const
{
if (d == dConst)
return true;
return false;
}
bool operator==(const bool b) const
{
if (b == bConst)
return true;
return false;
}
bool operator==(const TConstUnion& constant) const
{
if (constant.type != type)
return false;
switch (type) {
case EbtInt:
if (constant.iConst == iConst)
return true;
break;
case EbtUint:
if (constant.uConst == uConst)
return true;
break;
case EbtBool:
if (constant.bConst == bConst)
return true;
break;
case EbtDouble:
if (constant.dConst == dConst)
return true;
break;
#ifndef GLSLANG_WEB
case EbtInt16:
if (constant.i16Const == i16Const)
return true;
break;
case EbtUint16:
if (constant.u16Const == u16Const)
return true;
break;
case EbtInt8:
if (constant.i8Const == i8Const)
return true;
break;
case EbtUint8:
if (constant.u8Const == u8Const)
return true;
break;
case EbtInt64:
if (constant.i64Const == i64Const)
return true;
break;
case EbtUint64:
if (constant.u64Const == u64Const)
return true;
break;
#endif
default:
assert(false && "Default missing");
}
return false;
}
bool operator!=(const signed char i) const
{
return !operator==(i);
}
bool operator!=(const unsigned char u) const
{
return !operator==(u);
}
bool operator!=(const signed short i) const
{
return !operator==(i);
}
bool operator!=(const unsigned short u) const
{
return !operator==(u);
}
bool operator!=(const int i) const
{
return !operator==(i);
}
bool operator!=(const unsigned int u) const
{
return !operator==(u);
}
bool operator!=(const long long i) const
{
return !operator==(i);
}
bool operator!=(const unsigned long long u) const
{
return !operator==(u);
}
bool operator!=(const float f) const
{
return !operator==(f);
}
bool operator!=(const bool b) const
{
return !operator==(b);
}
bool operator!=(const TConstUnion& constant) const
{
return !operator==(constant);
}
bool operator>(const TConstUnion& constant) const
{
assert(type == constant.type);
switch (type) {
case EbtInt:
if (iConst > constant.iConst)
return true;
return false;
case EbtUint:
if (uConst > constant.uConst)
return true;
return false;
case EbtDouble:
if (dConst > constant.dConst)
return true;
return false;
#ifndef GLSLANG_WEB
case EbtInt8:
if (i8Const > constant.i8Const)
return true;
return false;
case EbtUint8:
if (u8Const > constant.u8Const)
return true;
return false;
case EbtInt16:
if (i16Const > constant.i16Const)
return true;
return false;
case EbtUint16:
if (u16Const > constant.u16Const)
return true;
return false;
case EbtInt64:
if (i64Const > constant.i64Const)
return true;
return false;
case EbtUint64:
if (u64Const > constant.u64Const)
return true;
return false;
#endif
default:
assert(false && "Default missing");
return false;
}
}
bool operator<(const TConstUnion& constant) const
{
assert(type == constant.type);
switch (type) {
#ifndef GLSLANG_WEB
case EbtInt8:
if (i8Const < constant.i8Const)
return true;
return false;
case EbtUint8:
if (u8Const < constant.u8Const)
return true;
return false;
case EbtInt16:
if (i16Const < constant.i16Const)
return true;
return false;
case EbtUint16:
if (u16Const < constant.u16Const)
return true;
return false;
case EbtInt64:
if (i64Const < constant.i64Const)
return true;
return false;
case EbtUint64:
if (u64Const < constant.u64Const)
return true;
return false;
#endif
case EbtDouble:
if (dConst < constant.dConst)
return true;
return false;
case EbtInt:
if (iConst < constant.iConst)
return true;
return false;
case EbtUint:
if (uConst < constant.uConst)
return true;
return false;
default:
assert(false && "Default missing");
return false;
}
}
TConstUnion operator+(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst + constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst + constant.uConst); break;
case EbtDouble: returnValue.setDConst(dConst + constant.dConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const + constant.i8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const + constant.i16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const + constant.i64Const); break;
case EbtUint8: returnValue.setU8Const(u8Const + constant.u8Const); break;
case EbtUint16: returnValue.setU16Const(u16Const + constant.u16Const); break;
case EbtUint64: returnValue.setU64Const(u64Const + constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator-(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst - constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst - constant.uConst); break;
case EbtDouble: returnValue.setDConst(dConst - constant.dConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const - constant.i8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const - constant.i16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const - constant.i64Const); break;
case EbtUint8: returnValue.setU8Const(u8Const - constant.u8Const); break;
case EbtUint16: returnValue.setU16Const(u16Const - constant.u16Const); break;
case EbtUint64: returnValue.setU64Const(u64Const - constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator*(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst * constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst * constant.uConst); break;
case EbtDouble: returnValue.setDConst(dConst * constant.dConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const * constant.i8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const * constant.i16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const * constant.i64Const); break;
case EbtUint8: returnValue.setU8Const(u8Const * constant.u8Const); break;
case EbtUint16: returnValue.setU16Const(u16Const * constant.u16Const); break;
case EbtUint64: returnValue.setU64Const(u64Const * constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator%(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst % constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst % constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const % constant.i8Const); break;
case EbtInt16: returnValue.setI8Const(i8Const % constant.i16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const % constant.i64Const); break;
case EbtUint8: returnValue.setU8Const(u8Const % constant.u8Const); break;
case EbtUint16: returnValue.setU16Const(u16Const % constant.u16Const); break;
case EbtUint64: returnValue.setU64Const(u64Const % constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator>>(const TConstUnion& constant) const
{
TConstUnion returnValue;
switch (type) {
#ifndef GLSLANG_WEB
case EbtInt8:
switch (constant.type) {
case EbtInt8: returnValue.setI8Const(i8Const >> constant.i8Const); break;
case EbtUint8: returnValue.setI8Const(i8Const >> constant.u8Const); break;
case EbtInt16: returnValue.setI8Const(i8Const >> constant.i16Const); break;
case EbtUint16: returnValue.setI8Const(i8Const >> constant.u16Const); break;
case EbtInt: returnValue.setI8Const(i8Const >> constant.iConst); break;
case EbtUint: returnValue.setI8Const(i8Const >> constant.uConst); break;
case EbtInt64: returnValue.setI8Const(i8Const >> constant.i64Const); break;
case EbtUint64: returnValue.setI8Const(i8Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint8:
switch (constant.type) {
case EbtInt8: returnValue.setU8Const(u8Const >> constant.i8Const); break;
case EbtUint8: returnValue.setU8Const(u8Const >> constant.u8Const); break;
case EbtInt16: returnValue.setU8Const(u8Const >> constant.i16Const); break;
case EbtUint16: returnValue.setU8Const(u8Const >> constant.u16Const); break;
case EbtInt: returnValue.setU8Const(u8Const >> constant.iConst); break;
case EbtUint: returnValue.setU8Const(u8Const >> constant.uConst); break;
case EbtInt64: returnValue.setU8Const(u8Const >> constant.i64Const); break;
case EbtUint64: returnValue.setU8Const(u8Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtInt16:
switch (constant.type) {
case EbtInt8: returnValue.setI16Const(i16Const >> constant.i8Const); break;
case EbtUint8: returnValue.setI16Const(i16Const >> constant.u8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const >> constant.i16Const); break;
case EbtUint16: returnValue.setI16Const(i16Const >> constant.u16Const); break;
case EbtInt: returnValue.setI16Const(i16Const >> constant.iConst); break;
case EbtUint: returnValue.setI16Const(i16Const >> constant.uConst); break;
case EbtInt64: returnValue.setI16Const(i16Const >> constant.i64Const); break;
case EbtUint64: returnValue.setI16Const(i16Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint16:
switch (constant.type) {
case EbtInt8: returnValue.setU16Const(u16Const >> constant.i8Const); break;
case EbtUint8: returnValue.setU16Const(u16Const >> constant.u8Const); break;
case EbtInt16: returnValue.setU16Const(u16Const >> constant.i16Const); break;
case EbtUint16: returnValue.setU16Const(u16Const >> constant.u16Const); break;
case EbtInt: returnValue.setU16Const(u16Const >> constant.iConst); break;
case EbtUint: returnValue.setU16Const(u16Const >> constant.uConst); break;
case EbtInt64: returnValue.setU16Const(u16Const >> constant.i64Const); break;
case EbtUint64: returnValue.setU16Const(u16Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
#endif
case EbtInt:
switch (constant.type) {
case EbtInt: returnValue.setIConst(iConst >> constant.iConst); break;
case EbtUint: returnValue.setIConst(iConst >> constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setIConst(iConst >> constant.i8Const); break;
case EbtUint8: returnValue.setIConst(iConst >> constant.u8Const); break;
case EbtInt16: returnValue.setIConst(iConst >> constant.i16Const); break;
case EbtUint16: returnValue.setIConst(iConst >> constant.u16Const); break;
case EbtInt64: returnValue.setIConst(iConst >> constant.i64Const); break;
case EbtUint64: returnValue.setIConst(iConst >> constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
break;
case EbtUint:
switch (constant.type) {
case EbtInt: returnValue.setUConst(uConst >> constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst >> constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setUConst(uConst >> constant.i8Const); break;
case EbtUint8: returnValue.setUConst(uConst >> constant.u8Const); break;
case EbtInt16: returnValue.setUConst(uConst >> constant.i16Const); break;
case EbtUint16: returnValue.setUConst(uConst >> constant.u16Const); break;
case EbtInt64: returnValue.setUConst(uConst >> constant.i64Const); break;
case EbtUint64: returnValue.setUConst(uConst >> constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
break;
#ifndef GLSLANG_WEB
case EbtInt64:
switch (constant.type) {
case EbtInt8: returnValue.setI64Const(i64Const >> constant.i8Const); break;
case EbtUint8: returnValue.setI64Const(i64Const >> constant.u8Const); break;
case EbtInt16: returnValue.setI64Const(i64Const >> constant.i16Const); break;
case EbtUint16: returnValue.setI64Const(i64Const >> constant.u16Const); break;
case EbtInt: returnValue.setI64Const(i64Const >> constant.iConst); break;
case EbtUint: returnValue.setI64Const(i64Const >> constant.uConst); break;
case EbtInt64: returnValue.setI64Const(i64Const >> constant.i64Const); break;
case EbtUint64: returnValue.setI64Const(i64Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint64:
switch (constant.type) {
case EbtInt8: returnValue.setU64Const(u64Const >> constant.i8Const); break;
case EbtUint8: returnValue.setU64Const(u64Const >> constant.u8Const); break;
case EbtInt16: returnValue.setU64Const(u64Const >> constant.i16Const); break;
case EbtUint16: returnValue.setU64Const(u64Const >> constant.u16Const); break;
case EbtInt: returnValue.setU64Const(u64Const >> constant.iConst); break;
case EbtUint: returnValue.setU64Const(u64Const >> constant.uConst); break;
case EbtInt64: returnValue.setU64Const(u64Const >> constant.i64Const); break;
case EbtUint64: returnValue.setU64Const(u64Const >> constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator<<(const TConstUnion& constant) const
{
TConstUnion returnValue;
switch (type) {
#ifndef GLSLANG_WEB
case EbtInt8:
switch (constant.type) {
case EbtInt8: returnValue.setI8Const(i8Const << constant.i8Const); break;
case EbtUint8: returnValue.setI8Const(i8Const << constant.u8Const); break;
case EbtInt16: returnValue.setI8Const(i8Const << constant.i16Const); break;
case EbtUint16: returnValue.setI8Const(i8Const << constant.u16Const); break;
case EbtInt: returnValue.setI8Const(i8Const << constant.iConst); break;
case EbtUint: returnValue.setI8Const(i8Const << constant.uConst); break;
case EbtInt64: returnValue.setI8Const(i8Const << constant.i64Const); break;
case EbtUint64: returnValue.setI8Const(i8Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint8:
switch (constant.type) {
case EbtInt8: returnValue.setU8Const(u8Const << constant.i8Const); break;
case EbtUint8: returnValue.setU8Const(u8Const << constant.u8Const); break;
case EbtInt16: returnValue.setU8Const(u8Const << constant.i16Const); break;
case EbtUint16: returnValue.setU8Const(u8Const << constant.u16Const); break;
case EbtInt: returnValue.setU8Const(u8Const << constant.iConst); break;
case EbtUint: returnValue.setU8Const(u8Const << constant.uConst); break;
case EbtInt64: returnValue.setU8Const(u8Const << constant.i64Const); break;
case EbtUint64: returnValue.setU8Const(u8Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtInt16:
switch (constant.type) {
case EbtInt8: returnValue.setI16Const(i16Const << constant.i8Const); break;
case EbtUint8: returnValue.setI16Const(i16Const << constant.u8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const << constant.i16Const); break;
case EbtUint16: returnValue.setI16Const(i16Const << constant.u16Const); break;
case EbtInt: returnValue.setI16Const(i16Const << constant.iConst); break;
case EbtUint: returnValue.setI16Const(i16Const << constant.uConst); break;
case EbtInt64: returnValue.setI16Const(i16Const << constant.i64Const); break;
case EbtUint64: returnValue.setI16Const(i16Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint16:
switch (constant.type) {
case EbtInt8: returnValue.setU16Const(u16Const << constant.i8Const); break;
case EbtUint8: returnValue.setU16Const(u16Const << constant.u8Const); break;
case EbtInt16: returnValue.setU16Const(u16Const << constant.i16Const); break;
case EbtUint16: returnValue.setU16Const(u16Const << constant.u16Const); break;
case EbtInt: returnValue.setU16Const(u16Const << constant.iConst); break;
case EbtUint: returnValue.setU16Const(u16Const << constant.uConst); break;
case EbtInt64: returnValue.setU16Const(u16Const << constant.i64Const); break;
case EbtUint64: returnValue.setU16Const(u16Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtInt64:
switch (constant.type) {
case EbtInt8: returnValue.setI64Const(i64Const << constant.i8Const); break;
case EbtUint8: returnValue.setI64Const(i64Const << constant.u8Const); break;
case EbtInt16: returnValue.setI64Const(i64Const << constant.i16Const); break;
case EbtUint16: returnValue.setI64Const(i64Const << constant.u16Const); break;
case EbtInt: returnValue.setI64Const(i64Const << constant.iConst); break;
case EbtUint: returnValue.setI64Const(i64Const << constant.uConst); break;
case EbtInt64: returnValue.setI64Const(i64Const << constant.i64Const); break;
case EbtUint64: returnValue.setI64Const(i64Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
case EbtUint64:
switch (constant.type) {
case EbtInt8: returnValue.setU64Const(u64Const << constant.i8Const); break;
case EbtUint8: returnValue.setU64Const(u64Const << constant.u8Const); break;
case EbtInt16: returnValue.setU64Const(u64Const << constant.i16Const); break;
case EbtUint16: returnValue.setU64Const(u64Const << constant.u16Const); break;
case EbtInt: returnValue.setU64Const(u64Const << constant.iConst); break;
case EbtUint: returnValue.setU64Const(u64Const << constant.uConst); break;
case EbtInt64: returnValue.setU64Const(u64Const << constant.i64Const); break;
case EbtUint64: returnValue.setU64Const(u64Const << constant.u64Const); break;
default: assert(false && "Default missing");
}
break;
#endif
case EbtInt:
switch (constant.type) {
case EbtInt: returnValue.setIConst(iConst << constant.iConst); break;
case EbtUint: returnValue.setIConst(iConst << constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setIConst(iConst << constant.i8Const); break;
case EbtUint8: returnValue.setIConst(iConst << constant.u8Const); break;
case EbtInt16: returnValue.setIConst(iConst << constant.i16Const); break;
case EbtUint16: returnValue.setIConst(iConst << constant.u16Const); break;
case EbtInt64: returnValue.setIConst(iConst << constant.i64Const); break;
case EbtUint64: returnValue.setIConst(iConst << constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
break;
case EbtUint:
switch (constant.type) {
case EbtInt: returnValue.setUConst(uConst << constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst << constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setUConst(uConst << constant.i8Const); break;
case EbtUint8: returnValue.setUConst(uConst << constant.u8Const); break;
case EbtInt16: returnValue.setUConst(uConst << constant.i16Const); break;
case EbtUint16: returnValue.setUConst(uConst << constant.u16Const); break;
case EbtInt64: returnValue.setUConst(uConst << constant.i64Const); break;
case EbtUint64: returnValue.setUConst(uConst << constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
break;
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator&(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst & constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst & constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const & constant.i8Const); break;
case EbtUint8: returnValue.setU8Const(u8Const & constant.u8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const & constant.i16Const); break;
case EbtUint16: returnValue.setU16Const(u16Const & constant.u16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const & constant.i64Const); break;
case EbtUint64: returnValue.setU64Const(u64Const & constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator|(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst | constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst | constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const | constant.i8Const); break;
case EbtUint8: returnValue.setU8Const(u8Const | constant.u8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const | constant.i16Const); break;
case EbtUint16: returnValue.setU16Const(u16Const | constant.u16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const | constant.i64Const); break;
case EbtUint64: returnValue.setU64Const(u64Const | constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator^(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtInt: returnValue.setIConst(iConst ^ constant.iConst); break;
case EbtUint: returnValue.setUConst(uConst ^ constant.uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(i8Const ^ constant.i8Const); break;
case EbtUint8: returnValue.setU8Const(u8Const ^ constant.u8Const); break;
case EbtInt16: returnValue.setI16Const(i16Const ^ constant.i16Const); break;
case EbtUint16: returnValue.setU16Const(u16Const ^ constant.u16Const); break;
case EbtInt64: returnValue.setI64Const(i64Const ^ constant.i64Const); break;
case EbtUint64: returnValue.setU64Const(u64Const ^ constant.u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator~() const
{
TConstUnion returnValue;
switch (type) {
case EbtInt: returnValue.setIConst(~iConst); break;
case EbtUint: returnValue.setUConst(~uConst); break;
#ifndef GLSLANG_WEB
case EbtInt8: returnValue.setI8Const(~i8Const); break;
case EbtUint8: returnValue.setU8Const(~u8Const); break;
case EbtInt16: returnValue.setI16Const(~i16Const); break;
case EbtUint16: returnValue.setU16Const(~u16Const); break;
case EbtInt64: returnValue.setI64Const(~i64Const); break;
case EbtUint64: returnValue.setU64Const(~u64Const); break;
#endif
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator&&(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtBool: returnValue.setBConst(bConst && constant.bConst); break;
default: assert(false && "Default missing");
}
return returnValue;
}
TConstUnion operator||(const TConstUnion& constant) const
{
TConstUnion returnValue;
assert(type == constant.type);
switch (type) {
case EbtBool: returnValue.setBConst(bConst || constant.bConst); break;
default: assert(false && "Default missing");
}
return returnValue;
}
TBasicType getType() const { return type; }
private:
union {
signed char i8Const; // used for i8vec, scalar int8s
unsigned char u8Const; // used for u8vec, scalar uint8s
signed short i16Const; // used for i16vec, scalar int16s
unsigned short u16Const; // used for u16vec, scalar uint16s
int iConst; // used for ivec, scalar ints
unsigned int uConst; // used for uvec, scalar uints
long long i64Const; // used for i64vec, scalar int64s
unsigned long long u64Const; // used for u64vec, scalar uint64s
bool bConst; // used for bvec, scalar bools
double dConst; // used for vec, dvec, mat, dmat, scalar floats and doubles
const TString* sConst; // string constant
};
TBasicType type;
};
// Encapsulate having a pointer to an array of TConstUnion,
// which only needs to be allocated if its size is going to be
// bigger than 0.
//
// One convenience is being able to use [] to go inside the array, instead
// of C++ assuming it as an array of pointers to vectors.
//
// General usage is that the size is known up front, and it is
// created once with the proper size.
//
class TConstUnionArray {
public:
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TConstUnionArray() : unionArray(nullptr) { }
virtual ~TConstUnionArray() { }
explicit TConstUnionArray(int size)
{
if (size == 0)
unionArray = nullptr;
else
unionArray = new TConstUnionVector(size);
}
TConstUnionArray(const TConstUnionArray& a) = default;
TConstUnionArray(const TConstUnionArray& a, int start, int size)
{
unionArray = new TConstUnionVector(size);
for (int i = 0; i < size; ++i)
(*unionArray)[i] = a[start + i];
}
// Use this constructor for a smear operation
TConstUnionArray(int size, const TConstUnion& val)
{
unionArray = new TConstUnionVector(size, val);
}
int size() const { return unionArray ? (int)unionArray->size() : 0; }
TConstUnion& operator[](size_t index) { return (*unionArray)[index]; }
const TConstUnion& operator[](size_t index) const { return (*unionArray)[index]; }
bool operator==(const TConstUnionArray& rhs) const
{
// this includes the case that both are unallocated
if (unionArray == rhs.unionArray)
return true;
if (! unionArray || ! rhs.unionArray)
return false;
return *unionArray == *rhs.unionArray;
}
bool operator!=(const TConstUnionArray& rhs) const { return ! operator==(rhs); }
double dot(const TConstUnionArray& rhs)
{
assert(rhs.unionArray->size() == unionArray->size());
double sum = 0.0;
for (size_t comp = 0; comp < unionArray->size(); ++comp)
sum += (*this)[comp].getDConst() * rhs[comp].getDConst();
return sum;
}
bool empty() const { return unionArray == nullptr; }
protected:
typedef TVector<TConstUnion> TConstUnionVector;
TConstUnionVector* unionArray;
};
} // end namespace glslang
#endif // _CONSTANT_UNION_INCLUDED_

318
include/vulkan/glslang/Include/PoolAlloc.h Normal file
View file

@ -0,0 +1,318 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2012-2013 LunarG, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _POOLALLOC_INCLUDED_
#define _POOLALLOC_INCLUDED_
#ifndef NDEBUG
# define GUARD_BLOCKS // define to enable guard block sanity checking
#endif
//
// This header defines an allocator that can be used to efficiently
// allocate a large number of small requests for heap memory, with the
// intention that they are not individually deallocated, but rather
// collectively deallocated at one time.
//
// This simultaneously
//
// * Makes each individual allocation much more efficient; the
// typical allocation is trivial.
// * Completely avoids the cost of doing individual deallocation.
// * Saves the trouble of tracking down and plugging a large class of leaks.
//
// Individual classes can use this allocator by supplying their own
// new and delete methods.
//
// STL containers can use this allocator by using the pool_allocator
// class as the allocator (second) template argument.
//
#include <cstddef>
#include <cstring>
#include <vector>
namespace glslang {
// If we are using guard blocks, we must track each individual
// allocation. If we aren't using guard blocks, these
// never get instantiated, so won't have any impact.
//
class TAllocation {
public:
TAllocation(size_t size, unsigned char* mem, TAllocation* prev = nullptr) :
size(size), mem(mem), prevAlloc(prev) {
// Allocations are bracketed:
// [allocationHeader][initialGuardBlock][userData][finalGuardBlock]
// This would be cleaner with if (guardBlockSize)..., but that
// makes the compiler print warnings about 0 length memsets,
// even with the if() protecting them.
# ifdef GUARD_BLOCKS
memset(preGuard(), guardBlockBeginVal, guardBlockSize);
memset(data(), userDataFill, size);
memset(postGuard(), guardBlockEndVal, guardBlockSize);
# endif
}
void check() const {
checkGuardBlock(preGuard(), guardBlockBeginVal, "before");
checkGuardBlock(postGuard(), guardBlockEndVal, "after");
}
void checkAllocList() const;
// Return total size needed to accommodate user buffer of 'size',
// plus our tracking data.
inline static size_t allocationSize(size_t size) {
return size + 2 * guardBlockSize + headerSize();
}
// Offset from surrounding buffer to get to user data buffer.
inline static unsigned char* offsetAllocation(unsigned char* m) {
return m + guardBlockSize + headerSize();
}
private:
void checkGuardBlock(unsigned char* blockMem, unsigned char val, const char* locText) const;
// Find offsets to pre and post guard blocks, and user data buffer
unsigned char* preGuard() const { return mem + headerSize(); }
unsigned char* data() const { return preGuard() + guardBlockSize; }
unsigned char* postGuard() const { return data() + size; }
size_t size; // size of the user data area
unsigned char* mem; // beginning of our allocation (pts to header)
TAllocation* prevAlloc; // prior allocation in the chain
const static unsigned char guardBlockBeginVal;
const static unsigned char guardBlockEndVal;
const static unsigned char userDataFill;
const static size_t guardBlockSize;
# ifdef GUARD_BLOCKS
inline static size_t headerSize() { return sizeof(TAllocation); }
# else
inline static size_t headerSize() { return 0; }
# endif
};
//
// There are several stacks. One is to track the pushing and popping
// of the user, and not yet implemented. The others are simply a
// repositories of free pages or used pages.
//
// Page stacks are linked together with a simple header at the beginning
// of each allocation obtained from the underlying OS. Multi-page allocations
// are returned to the OS. Individual page allocations are kept for future
// re-use.
//
// The "page size" used is not, nor must it match, the underlying OS
// page size. But, having it be about that size or equal to a set of
// pages is likely most optimal.
//
class TPoolAllocator {
public:
TPoolAllocator(int growthIncrement = 8*1024, int allocationAlignment = 16);
//
// Don't call the destructor just to free up the memory, call pop()
//
~TPoolAllocator();
//
// Call push() to establish a new place to pop memory too. Does not
// have to be called to get things started.
//
void push();
//
// Call pop() to free all memory allocated since the last call to push(),
// or if no last call to push, frees all memory since first allocation.
//
void pop();
//
// Call popAll() to free all memory allocated.
//
void popAll();
//
// Call allocate() to actually acquire memory. Returns nullptr if no memory
// available, otherwise a properly aligned pointer to 'numBytes' of memory.
//
void* allocate(size_t numBytes);
//
// There is no deallocate. The point of this class is that
// deallocation can be skipped by the user of it, as the model
// of use is to simultaneously deallocate everything at once
// by calling pop(), and to not have to solve memory leak problems.
//
protected:
friend struct tHeader;
struct tHeader {
tHeader(tHeader* nextPage, size_t pageCount) :
#ifdef GUARD_BLOCKS
lastAllocation(nullptr),
#endif
nextPage(nextPage), pageCount(pageCount) { }
~tHeader() {
#ifdef GUARD_BLOCKS
if (lastAllocation)
lastAllocation->checkAllocList();
#endif
}
#ifdef GUARD_BLOCKS
TAllocation* lastAllocation;
#endif
tHeader* nextPage;
size_t pageCount;
};
struct tAllocState {
size_t offset;
tHeader* page;
};
typedef std::vector<tAllocState> tAllocStack;
// Track allocations if and only if we're using guard blocks
#ifndef GUARD_BLOCKS
void* initializeAllocation(tHeader*, unsigned char* memory, size_t) {
#else
void* initializeAllocation(tHeader* block, unsigned char* memory, size_t numBytes) {
new(memory) TAllocation(numBytes, memory, block->lastAllocation);
block->lastAllocation = reinterpret_cast<TAllocation*>(memory);
#endif
// This is optimized entirely away if GUARD_BLOCKS is not defined.
return TAllocation::offsetAllocation(memory);
}
size_t pageSize; // granularity of allocation from the OS
size_t alignment; // all returned allocations will be aligned at
// this granularity, which will be a power of 2
size_t alignmentMask;
size_t headerSkip; // amount of memory to skip to make room for the
// header (basically, size of header, rounded
// up to make it aligned
size_t currentPageOffset; // next offset in top of inUseList to allocate from
tHeader* freeList; // list of popped memory
tHeader* inUseList; // list of all memory currently being used
tAllocStack stack; // stack of where to allocate from, to partition pool
int numCalls; // just an interesting statistic
size_t totalBytes; // just an interesting statistic
private:
TPoolAllocator& operator=(const TPoolAllocator&); // don't allow assignment operator
TPoolAllocator(const TPoolAllocator&); // don't allow default copy constructor
};
//
// There could potentially be many pools with pops happening at
// different times. But a simple use is to have a global pop
// with everyone using the same global allocator.
//
extern TPoolAllocator& GetThreadPoolAllocator();
void SetThreadPoolAllocator(TPoolAllocator* poolAllocator);
//
// This STL compatible allocator is intended to be used as the allocator
// parameter to templatized STL containers, like vector and map.
//
// It will use the pools for allocation, and not
// do any deallocation, but will still do destruction.
//
template<class T>
class pool_allocator {
public:
typedef size_t size_type;
typedef ptrdiff_t difference_type;
typedef T *pointer;
typedef const T *const_pointer;
typedef T& reference;
typedef const T& const_reference;
typedef T value_type;
template<class Other>
struct rebind {
typedef pool_allocator<Other> other;
};
pointer address(reference x) const { return &x; }
const_pointer address(const_reference x) const { return &x; }
pool_allocator() : allocator(GetThreadPoolAllocator()) { }
pool_allocator(TPoolAllocator& a) : allocator(a) { }
pool_allocator(const pool_allocator<T>& p) : allocator(p.allocator) { }
template<class Other>
pool_allocator(const pool_allocator<Other>& p) : allocator(p.getAllocator()) { }
pointer allocate(size_type n) {
return reinterpret_cast<pointer>(getAllocator().allocate(n * sizeof(T))); }
pointer allocate(size_type n, const void*) {
return reinterpret_cast<pointer>(getAllocator().allocate(n * sizeof(T))); }
void deallocate(void*, size_type) { }
void deallocate(pointer, size_type) { }
pointer _Charalloc(size_t n) {
return reinterpret_cast<pointer>(getAllocator().allocate(n)); }
void construct(pointer p, const T& val) { new ((void *)p) T(val); }
void destroy(pointer p) { p->T::~T(); }
bool operator==(const pool_allocator& rhs) const { return &getAllocator() == &rhs.getAllocator(); }
bool operator!=(const pool_allocator& rhs) const { return &getAllocator() != &rhs.getAllocator(); }
size_type max_size() const { return static_cast<size_type>(-1) / sizeof(T); }
size_type max_size(int size) const { return static_cast<size_type>(-1) / size; }
TPoolAllocator& getAllocator() const { return allocator; }
pool_allocator select_on_container_copy_construction() const { return pool_allocator{}; }
protected:
pool_allocator& operator=(const pool_allocator&) { return *this; }
TPoolAllocator& allocator;
};
} // end namespace glslang
#endif // _POOLALLOC_INCLUDED_

159
include/vulkan/glslang/Include/ResourceLimits.h Normal file
View file

@ -0,0 +1,159 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2013 LunarG, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _RESOURCE_LIMITS_INCLUDED_
#define _RESOURCE_LIMITS_INCLUDED_
struct TLimits {
bool nonInductiveForLoops;
bool whileLoops;
bool doWhileLoops;
bool generalUniformIndexing;
bool generalAttributeMatrixVectorIndexing;
bool generalVaryingIndexing;
bool generalSamplerIndexing;
bool generalVariableIndexing;
bool generalConstantMatrixVectorIndexing;
};
struct TBuiltInResource {
int maxLights;
int maxClipPlanes;
int maxTextureUnits;
int maxTextureCoords;
int maxVertexAttribs;
int maxVertexUniformComponents;
int maxVaryingFloats;
int maxVertexTextureImageUnits;
int maxCombinedTextureImageUnits;
int maxTextureImageUnits;
int maxFragmentUniformComponents;
int maxDrawBuffers;
int maxVertexUniformVectors;
int maxVaryingVectors;
int maxFragmentUniformVectors;
int maxVertexOutputVectors;
int maxFragmentInputVectors;
int minProgramTexelOffset;
int maxProgramTexelOffset;
int maxClipDistances;
int maxComputeWorkGroupCountX;
int maxComputeWorkGroupCountY;
int maxComputeWorkGroupCountZ;
int maxComputeWorkGroupSizeX;
int maxComputeWorkGroupSizeY;
int maxComputeWorkGroupSizeZ;
int maxComputeUniformComponents;
int maxComputeTextureImageUnits;
int maxComputeImageUniforms;
int maxComputeAtomicCounters;
int maxComputeAtomicCounterBuffers;
int maxVaryingComponents;
int maxVertexOutputComponents;
int maxGeometryInputComponents;
int maxGeometryOutputComponents;
int maxFragmentInputComponents;
int maxImageUnits;
int maxCombinedImageUnitsAndFragmentOutputs;
int maxCombinedShaderOutputResources;
int maxImageSamples;
int maxVertexImageUniforms;
int maxTessControlImageUniforms;
int maxTessEvaluationImageUniforms;
int maxGeometryImageUniforms;
int maxFragmentImageUniforms;
int maxCombinedImageUniforms;
int maxGeometryTextureImageUnits;
int maxGeometryOutputVertices;
int maxGeometryTotalOutputComponents;
int maxGeometryUniformComponents;
int maxGeometryVaryingComponents;
int maxTessControlInputComponents;
int maxTessControlOutputComponents;
int maxTessControlTextureImageUnits;
int maxTessControlUniformComponents;
int maxTessControlTotalOutputComponents;
int maxTessEvaluationInputComponents;
int maxTessEvaluationOutputComponents;
int maxTessEvaluationTextureImageUnits;
int maxTessEvaluationUniformComponents;
int maxTessPatchComponents;
int maxPatchVertices;
int maxTessGenLevel;
int maxViewports;
int maxVertexAtomicCounters;
int maxTessControlAtomicCounters;
int maxTessEvaluationAtomicCounters;
int maxGeometryAtomicCounters;
int maxFragmentAtomicCounters;
int maxCombinedAtomicCounters;
int maxAtomicCounterBindings;
int maxVertexAtomicCounterBuffers;
int maxTessControlAtomicCounterBuffers;
int maxTessEvaluationAtomicCounterBuffers;
int maxGeometryAtomicCounterBuffers;
int maxFragmentAtomicCounterBuffers;
int maxCombinedAtomicCounterBuffers;
int maxAtomicCounterBufferSize;
int maxTransformFeedbackBuffers;
int maxTransformFeedbackInterleavedComponents;
int maxCullDistances;
int maxCombinedClipAndCullDistances;
int maxSamples;
int maxMeshOutputVerticesNV;
int maxMeshOutputPrimitivesNV;
int maxMeshWorkGroupSizeX_NV;
int maxMeshWorkGroupSizeY_NV;
int maxMeshWorkGroupSizeZ_NV;
int maxTaskWorkGroupSizeX_NV;
int maxTaskWorkGroupSizeY_NV;
int maxTaskWorkGroupSizeZ_NV;
int maxMeshViewCountNV;
int maxMeshOutputVerticesEXT;
int maxMeshOutputPrimitivesEXT;
int maxMeshWorkGroupSizeX_EXT;
int maxMeshWorkGroupSizeY_EXT;
int maxMeshWorkGroupSizeZ_EXT;
int maxTaskWorkGroupSizeX_EXT;
int maxTaskWorkGroupSizeY_EXT;
int maxTaskWorkGroupSizeZ_EXT;
int maxMeshViewCountEXT;
int maxDualSourceDrawBuffersEXT;
TLimits limits;
};
#endif // _RESOURCE_LIMITS_INCLUDED_

128
include/vulkan/glslang/Include/SpirvIntrinsics.h Normal file
View file

@ -0,0 +1,128 @@
//
// Copyright(C) 2021 Advanced Micro Devices, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#pragma once
#ifndef GLSLANG_WEB
//
// GL_EXT_spirv_intrinsics
//
#include "Common.h"
namespace glslang {
class TIntermTyped;
class TIntermConstantUnion;
class TType;
// SPIR-V requirements
struct TSpirvRequirement {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
// capability = [..]
TSet<TString> extensions;
// extension = [..]
TSet<int> capabilities;
};
// SPIR-V execution modes
struct TSpirvExecutionMode {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
// spirv_execution_mode
TMap<int, TVector<const TIntermConstantUnion*>> modes;
// spirv_execution_mode_id
TMap<int, TVector<const TIntermTyped*> > modeIds;
};
// SPIR-V decorations
struct TSpirvDecorate {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
// spirv_decorate
TMap<int, TVector<const TIntermConstantUnion*> > decorates;
// spirv_decorate_id
TMap<int, TVector<const TIntermTyped*>> decorateIds;
// spirv_decorate_string
TMap<int, TVector<const TIntermConstantUnion*> > decorateStrings;
};
// SPIR-V instruction
struct TSpirvInstruction {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TSpirvInstruction() { set = ""; id = -1; }
bool operator==(const TSpirvInstruction& rhs) const { return set == rhs.set && id == rhs.id; }
bool operator!=(const TSpirvInstruction& rhs) const { return !operator==(rhs); }
// spirv_instruction
TString set;
int id;
};
// SPIR-V type parameter
struct TSpirvTypeParameter {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TSpirvTypeParameter(const TIntermConstantUnion* arg) { constant = arg; }
bool operator==(const TSpirvTypeParameter& rhs) const { return constant == rhs.constant; }
bool operator!=(const TSpirvTypeParameter& rhs) const { return !operator==(rhs); }
const TIntermConstantUnion* constant;
};
typedef TVector<TSpirvTypeParameter> TSpirvTypeParameters;
// SPIR-V type
struct TSpirvType {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
bool operator==(const TSpirvType& rhs) const
{
return spirvInst == rhs.spirvInst && typeParams == rhs.typeParams;
}
bool operator!=(const TSpirvType& rhs) const { return !operator==(rhs); }
// spirv_type
TSpirvInstruction spirvInst;
TSpirvTypeParameters typeParams;
};
} // end namespace glslang
#endif // GLSLANG_WEB

2985
include/vulkan/glslang/Include/Types.h Normal file
View file

File diff suppressed because it is too large Load diff

352
include/vulkan/glslang/Include/arrays.h Normal file
View file

@ -0,0 +1,352 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2012-2013 LunarG, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
//
// Implement types for tracking GLSL arrays, arrays of arrays, etc.
//
#ifndef _ARRAYS_INCLUDED
#define _ARRAYS_INCLUDED
#include <algorithm>
namespace glslang {
// This is used to mean there is no size yet (unsized), it is waiting to get a size from somewhere else.
const int UnsizedArraySize = 0;
class TIntermTyped;
extern bool SameSpecializationConstants(TIntermTyped*, TIntermTyped*);
// Specialization constants need both a nominal size and a node that defines
// the specialization constant being used. Array types are the same when their
// size and specialization constant nodes are the same.
struct TArraySize {
unsigned int size;
TIntermTyped* node; // nullptr means no specialization constant node
bool operator==(const TArraySize& rhs) const
{
if (size != rhs.size)
return false;
if (node == nullptr || rhs.node == nullptr)
return node == rhs.node;
return SameSpecializationConstants(node, rhs.node);
}
};
//
// TSmallArrayVector is used as the container for the set of sizes in TArraySizes.
// It has generic-container semantics, while TArraySizes has array-of-array semantics.
// That is, TSmallArrayVector should be more focused on mechanism and TArraySizes on policy.
//
struct TSmallArrayVector {
//
// TODO: memory: TSmallArrayVector is intended to be smaller.
// Almost all arrays could be handled by two sizes each fitting
// in 16 bits, needing a real vector only in the cases where there
// are more than 3 sizes or a size needing more than 16 bits.
//
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TSmallArrayVector() : sizes(nullptr) { }
virtual ~TSmallArrayVector() { dealloc(); }
// For breaking into two non-shared copies, independently modifiable.
TSmallArrayVector& operator=(const TSmallArrayVector& from)
{
if (from.sizes == nullptr)
sizes = nullptr;
else {
alloc();
*sizes = *from.sizes;
}
return *this;
}
int size() const
{
if (sizes == nullptr)
return 0;
return (int)sizes->size();
}
unsigned int frontSize() const
{
assert(sizes != nullptr && sizes->size() > 0);
return sizes->front().size;
}
TIntermTyped* frontNode() const
{
assert(sizes != nullptr && sizes->size() > 0);
return sizes->front().node;
}
void changeFront(unsigned int s)
{
assert(sizes != nullptr);
// this should only happen for implicitly sized arrays, not specialization constants
assert(sizes->front().node == nullptr);
sizes->front().size = s;
}
void push_back(unsigned int e, TIntermTyped* n)
{
alloc();
TArraySize pair = { e, n };
sizes->push_back(pair);
}
void push_back(const TSmallArrayVector& newDims)
{
alloc();
sizes->insert(sizes->end(), newDims.sizes->begin(), newDims.sizes->end());
}
void pop_front()
{
assert(sizes != nullptr && sizes->size() > 0);
if (sizes->size() == 1)
dealloc();
else
sizes->erase(sizes->begin());
}
// 'this' should currently not be holding anything, and copyNonFront
// will make it hold a copy of all but the first element of rhs.
// (This would be useful for making a type that is dereferenced by
// one dimension.)
void copyNonFront(const TSmallArrayVector& rhs)
{
assert(sizes == nullptr);
if (rhs.size() > 1) {
alloc();
sizes->insert(sizes->begin(), rhs.sizes->begin() + 1, rhs.sizes->end());
}
}
unsigned int getDimSize(int i) const
{
assert(sizes != nullptr && (int)sizes->size() > i);
return (*sizes)[i].size;
}
void setDimSize(int i, unsigned int size) const
{
assert(sizes != nullptr && (int)sizes->size() > i);
assert((*sizes)[i].node == nullptr);
(*sizes)[i].size = size;
}
TIntermTyped* getDimNode(int i) const
{
assert(sizes != nullptr && (int)sizes->size() > i);
return (*sizes)[i].node;
}
bool operator==(const TSmallArrayVector& rhs) const
{
if (sizes == nullptr && rhs.sizes == nullptr)
return true;
if (sizes == nullptr || rhs.sizes == nullptr)
return false;
return *sizes == *rhs.sizes;
}
bool operator!=(const TSmallArrayVector& rhs) const { return ! operator==(rhs); }
protected:
TSmallArrayVector(const TSmallArrayVector&);
void alloc()
{
if (sizes == nullptr)
sizes = new TVector<TArraySize>;
}
void dealloc()
{
delete sizes;
sizes = nullptr;
}
TVector<TArraySize>* sizes; // will either hold such a pointer, or in the future, hold the two array sizes
};
//
// Represent an array, or array of arrays, to arbitrary depth. This is not
// done through a hierarchy of types in a type tree, rather all contiguous arrayness
// in the type hierarchy is localized into this single cumulative object.
//
// The arrayness in TTtype is a pointer, so that it can be non-allocated and zero
// for the vast majority of types that are non-array types.
//
// Order Policy: these are all identical:
// - left to right order within a contiguous set of ...[..][..][..]... in the source language
// - index order 0, 1, 2, ... within the 'sizes' member below
// - outer-most to inner-most
//
struct TArraySizes {
POOL_ALLOCATOR_NEW_DELETE(GetThreadPoolAllocator())
TArraySizes() : implicitArraySize(0), implicitlySized(true), variablyIndexed(false){ }
// For breaking into two non-shared copies, independently modifiable.
TArraySizes& operator=(const TArraySizes& from)
{
implicitArraySize = from.implicitArraySize;
variablyIndexed = from.variablyIndexed;
sizes = from.sizes;
implicitlySized = from.implicitlySized;
return *this;
}
// translate from array-of-array semantics to container semantics
int getNumDims() const { return sizes.size(); }
int getDimSize(int dim) const { return sizes.getDimSize(dim); }
TIntermTyped* getDimNode(int dim) const { return sizes.getDimNode(dim); }
void setDimSize(int dim, int size) { sizes.setDimSize(dim, size); }
int getOuterSize() const { return sizes.frontSize(); }
TIntermTyped* getOuterNode() const { return sizes.frontNode(); }
int getCumulativeSize() const
{
int size = 1;
for (int d = 0; d < sizes.size(); ++d) {
// this only makes sense in paths that have a known array size
assert(sizes.getDimSize(d) != UnsizedArraySize);
size *= sizes.getDimSize(d);
}
return size;
}
void addInnerSize() { addInnerSize((unsigned)UnsizedArraySize); }
void addInnerSize(int s) { addInnerSize((unsigned)s, nullptr); }
void addInnerSize(int s, TIntermTyped* n) { sizes.push_back((unsigned)s, n); }
void addInnerSize(TArraySize pair) {
sizes.push_back(pair.size, pair.node);
implicitlySized = false;
}
void addInnerSizes(const TArraySizes& s) { sizes.push_back(s.sizes); }
void changeOuterSize(int s) {
sizes.changeFront((unsigned)s);
implicitlySized = false;
}
int getImplicitSize() const { return implicitArraySize > 0 ? implicitArraySize : 1; }
void updateImplicitSize(int s) {
implicitArraySize = (std::max)(implicitArraySize, s);
}
bool isInnerUnsized() const
{
for (int d = 1; d < sizes.size(); ++d) {
if (sizes.getDimSize(d) == (unsigned)UnsizedArraySize)
return true;
}
return false;
}
bool clearInnerUnsized()
{
for (int d = 1; d < sizes.size(); ++d) {
if (sizes.getDimSize(d) == (unsigned)UnsizedArraySize)
setDimSize(d, 1);
}
return false;
}
bool isInnerSpecialization() const
{
for (int d = 1; d < sizes.size(); ++d) {
if (sizes.getDimNode(d) != nullptr)
return true;
}
return false;
}
bool isOuterSpecialization()
{
return sizes.getDimNode(0) != nullptr;
}
bool hasUnsized() const { return getOuterSize() == UnsizedArraySize || isInnerUnsized(); }
bool isSized() const { return getOuterSize() != UnsizedArraySize; }
bool isImplicitlySized() const { return implicitlySized; }
bool isDefaultImplicitlySized() const { return implicitlySized && implicitArraySize == 0; }
void setImplicitlySized(bool isImplicitSizing) { implicitlySized = isImplicitSizing; }
void dereference() { sizes.pop_front(); }
void copyDereferenced(const TArraySizes& rhs)
{
assert(sizes.size() == 0);
if (rhs.sizes.size() > 1)
sizes.copyNonFront(rhs.sizes);
}
bool sameInnerArrayness(const TArraySizes& rhs) const
{
if (sizes.size() != rhs.sizes.size())
return false;
for (int d = 1; d < sizes.size(); ++d) {
if (sizes.getDimSize(d) != rhs.sizes.getDimSize(d) ||
sizes.getDimNode(d) != rhs.sizes.getDimNode(d))
return false;
}
return true;
}
void setVariablyIndexed() { variablyIndexed = true; }
bool isVariablyIndexed() const { return variablyIndexed; }
bool operator==(const TArraySizes& rhs) const { return sizes == rhs.sizes; }
bool operator!=(const TArraySizes& rhs) const { return sizes != rhs.sizes; }
protected:
TSmallArrayVector sizes;
TArraySizes(const TArraySizes&);
// For tracking maximum referenced compile-time constant index.
// Applies only to the outer-most dimension. Potentially becomes
// the implicit size of the array, if not variably indexed and
// otherwise legal.
int implicitArraySize;
bool implicitlySized;
bool variablyIndexed; // true if array is indexed with a non compile-time constant
};
} // end namespace glslang
#endif // _ARRAYS_INCLUDED_

1898
include/vulkan/glslang/Include/intermediate.h Normal file
View file

File diff suppressed because it is too large Load diff

367
include/vulkan/glslang/MachineIndependent/Versions.h Normal file
View file

@ -0,0 +1,367 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2012-2013 LunarG, Inc.
// Copyright (C) 2017 ARM Limited.
// Copyright (C) 2015-2018 Google, Inc.
// Modifications Copyright (C) 2020 Advanced Micro Devices, Inc. All rights reserved.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _VERSIONS_INCLUDED_
#define _VERSIONS_INCLUDED_
#define LAST_ELEMENT_MARKER(x) x
//
// Help manage multiple profiles, versions, extensions etc.
//
//
// Profiles are set up for masking operations, so queries can be done on multiple
// profiles at the same time.
//
// Don't maintain an ordinal set of enums (0,1,2,3...) to avoid all possible
// defects from mixing the two different forms.
//
typedef enum : unsigned {
EBadProfile = 0,
ENoProfile = (1 << 0), // only for desktop, before profiles showed up
ECoreProfile = (1 << 1),
ECompatibilityProfile = (1 << 2),
EEsProfile = (1 << 3),
LAST_ELEMENT_MARKER(EProfileCount),
} EProfile;
namespace glslang {
//
// Map from profile enum to externally readable text name.
//
inline const char* ProfileName(EProfile profile)
{
switch (profile) {
case ENoProfile: return "none";
case ECoreProfile: return "core";
case ECompatibilityProfile: return "compatibility";
case EEsProfile: return "es";
default: return "unknown profile";
}
}
//
// What source rules, validation rules, target language, etc. are needed or
// desired for SPIR-V?
//
// 0 means a target or rule set is not enabled (ignore rules from that entity).
// Non-0 means to apply semantic rules arising from that version of its rule set.
// The union of all requested rule sets will be applied.
//
struct SpvVersion {
SpvVersion() : spv(0), vulkanGlsl(0), vulkan(0), openGl(0), vulkanRelaxed(false) {}
unsigned int spv; // the version of SPIR-V to target, as defined by "word 1" of the SPIR-V binary header
int vulkanGlsl; // the version of GLSL semantics for Vulkan, from GL_KHR_vulkan_glsl, for "#define VULKAN XXX"
int vulkan; // the version of Vulkan, for which SPIR-V execution environment rules to use
int openGl; // the version of GLSL semantics for OpenGL, from GL_ARB_gl_spirv, for "#define GL_SPIRV XXX"
bool vulkanRelaxed; // relax changes to GLSL for Vulkan, allowing some GL-specific to be compiled to Vulkan SPIR-V target
};
//
// The behaviors from the GLSL "#extension extension_name : behavior"
//
typedef enum {
EBhMissing = 0,
EBhRequire,
EBhEnable,
EBhWarn,
EBhDisable,
EBhDisablePartial // use as initial state of an extension that is only partially implemented
} TExtensionBehavior;
//
// Symbolic names for extensions. Strings may be directly used when calling the
// functions, but better to have the compiler do spelling checks.
//
const char* const E_GL_OES_texture_3D = "GL_OES_texture_3D";
const char* const E_GL_OES_standard_derivatives = "GL_OES_standard_derivatives";
const char* const E_GL_EXT_frag_depth = "GL_EXT_frag_depth";
const char* const E_GL_OES_EGL_image_external = "GL_OES_EGL_image_external";
const char* const E_GL_OES_EGL_image_external_essl3 = "GL_OES_EGL_image_external_essl3";
const char* const E_GL_EXT_YUV_target = "GL_EXT_YUV_target";
const char* const E_GL_EXT_shader_texture_lod = "GL_EXT_shader_texture_lod";
const char* const E_GL_EXT_shadow_samplers = "GL_EXT_shadow_samplers";
const char* const E_GL_ARB_texture_rectangle = "GL_ARB_texture_rectangle";
const char* const E_GL_3DL_array_objects = "GL_3DL_array_objects";
const char* const E_GL_ARB_shading_language_420pack = "GL_ARB_shading_language_420pack";
const char* const E_GL_ARB_texture_gather = "GL_ARB_texture_gather";
const char* const E_GL_ARB_gpu_shader5 = "GL_ARB_gpu_shader5";
const char* const E_GL_ARB_separate_shader_objects = "GL_ARB_separate_shader_objects";
const char* const E_GL_ARB_compute_shader = "GL_ARB_compute_shader";
const char* const E_GL_ARB_tessellation_shader = "GL_ARB_tessellation_shader";
const char* const E_GL_ARB_enhanced_layouts = "GL_ARB_enhanced_layouts";
const char* const E_GL_ARB_texture_cube_map_array = "GL_ARB_texture_cube_map_array";
const char* const E_GL_ARB_texture_multisample = "GL_ARB_texture_multisample";
const char* const E_GL_ARB_shader_texture_lod = "GL_ARB_shader_texture_lod";
const char* const E_GL_ARB_explicit_attrib_location = "GL_ARB_explicit_attrib_location";
const char* const E_GL_ARB_explicit_uniform_location = "GL_ARB_explicit_uniform_location";
const char* const E_GL_ARB_shader_image_load_store = "GL_ARB_shader_image_load_store";
const char* const E_GL_ARB_shader_atomic_counters = "GL_ARB_shader_atomic_counters";
const char* const E_GL_ARB_shader_atomic_counter_ops = "GL_ARB_shader_atomic_counter_ops";
const char* const E_GL_ARB_shader_draw_parameters = "GL_ARB_shader_draw_parameters";
const char* const E_GL_ARB_shader_group_vote = "GL_ARB_shader_group_vote";
const char* const E_GL_ARB_derivative_control = "GL_ARB_derivative_control";
const char* const E_GL_ARB_shader_texture_image_samples = "GL_ARB_shader_texture_image_samples";
const char* const E_GL_ARB_viewport_array = "GL_ARB_viewport_array";
const char* const E_GL_ARB_gpu_shader_int64 = "GL_ARB_gpu_shader_int64";
const char* const E_GL_ARB_gpu_shader_fp64 = "GL_ARB_gpu_shader_fp64";
const char* const E_GL_ARB_shader_ballot = "GL_ARB_shader_ballot";
const char* const E_GL_ARB_sparse_texture2 = "GL_ARB_sparse_texture2";
const char* const E_GL_ARB_sparse_texture_clamp = "GL_ARB_sparse_texture_clamp";
const char* const E_GL_ARB_shader_stencil_export = "GL_ARB_shader_stencil_export";
// const char* const E_GL_ARB_cull_distance = "GL_ARB_cull_distance"; // present for 4.5, but need extension control over block members
const char* const E_GL_ARB_post_depth_coverage = "GL_ARB_post_depth_coverage";
const char* const E_GL_ARB_shader_viewport_layer_array = "GL_ARB_shader_viewport_layer_array";
const char* const E_GL_ARB_fragment_shader_interlock = "GL_ARB_fragment_shader_interlock";
const char* const E_GL_ARB_shader_clock = "GL_ARB_shader_clock";
const char* const E_GL_ARB_uniform_buffer_object = "GL_ARB_uniform_buffer_object";
const char* const E_GL_ARB_sample_shading = "GL_ARB_sample_shading";
const char* const E_GL_ARB_shader_bit_encoding = "GL_ARB_shader_bit_encoding";
const char* const E_GL_ARB_shader_image_size = "GL_ARB_shader_image_size";
const char* const E_GL_ARB_shader_storage_buffer_object = "GL_ARB_shader_storage_buffer_object";
const char* const E_GL_ARB_shading_language_packing = "GL_ARB_shading_language_packing";
const char* const E_GL_ARB_texture_query_lod = "GL_ARB_texture_query_lod";
const char* const E_GL_ARB_vertex_attrib_64bit = "GL_ARB_vertex_attrib_64bit";
const char* const E_GL_ARB_draw_instanced = "GL_ARB_draw_instanced";
const char* const E_GL_ARB_fragment_coord_conventions = "GL_ARB_fragment_coord_conventions";
const char* const E_GL_ARB_bindless_texture = "GL_ARB_bindless_texture";
const char* const E_GL_KHR_shader_subgroup_basic = "GL_KHR_shader_subgroup_basic";
const char* const E_GL_KHR_shader_subgroup_vote = "GL_KHR_shader_subgroup_vote";
const char* const E_GL_KHR_shader_subgroup_arithmetic = "GL_KHR_shader_subgroup_arithmetic";
const char* const E_GL_KHR_shader_subgroup_ballot = "GL_KHR_shader_subgroup_ballot";
const char* const E_GL_KHR_shader_subgroup_shuffle = "GL_KHR_shader_subgroup_shuffle";
const char* const E_GL_KHR_shader_subgroup_shuffle_relative = "GL_KHR_shader_subgroup_shuffle_relative";
const char* const E_GL_KHR_shader_subgroup_clustered = "GL_KHR_shader_subgroup_clustered";
const char* const E_GL_KHR_shader_subgroup_quad = "GL_KHR_shader_subgroup_quad";
const char* const E_GL_KHR_memory_scope_semantics = "GL_KHR_memory_scope_semantics";
const char* const E_GL_EXT_shader_atomic_int64 = "GL_EXT_shader_atomic_int64";
const char* const E_GL_EXT_shader_non_constant_global_initializers = "GL_EXT_shader_non_constant_global_initializers";
const char* const E_GL_EXT_shader_image_load_formatted = "GL_EXT_shader_image_load_formatted";
const char* const E_GL_EXT_shader_16bit_storage = "GL_EXT_shader_16bit_storage";
const char* const E_GL_EXT_shader_8bit_storage = "GL_EXT_shader_8bit_storage";
// EXT extensions
const char* const E_GL_EXT_device_group = "GL_EXT_device_group";
const char* const E_GL_EXT_multiview = "GL_EXT_multiview";
const char* const E_GL_EXT_post_depth_coverage = "GL_EXT_post_depth_coverage";
const char* const E_GL_EXT_control_flow_attributes = "GL_EXT_control_flow_attributes";
const char* const E_GL_EXT_nonuniform_qualifier = "GL_EXT_nonuniform_qualifier";
const char* const E_GL_EXT_samplerless_texture_functions = "GL_EXT_samplerless_texture_functions";
const char* const E_GL_EXT_scalar_block_layout = "GL_EXT_scalar_block_layout";
const char* const E_GL_EXT_fragment_invocation_density = "GL_EXT_fragment_invocation_density";
const char* const E_GL_EXT_buffer_reference = "GL_EXT_buffer_reference";
const char* const E_GL_EXT_buffer_reference2 = "GL_EXT_buffer_reference2";
const char* const E_GL_EXT_buffer_reference_uvec2 = "GL_EXT_buffer_reference_uvec2";
const char* const E_GL_EXT_demote_to_helper_invocation = "GL_EXT_demote_to_helper_invocation";
const char* const E_GL_EXT_shader_realtime_clock = "GL_EXT_shader_realtime_clock";
const char* const E_GL_EXT_debug_printf = "GL_EXT_debug_printf";
const char* const E_GL_EXT_ray_tracing = "GL_EXT_ray_tracing";
const char* const E_GL_EXT_ray_query = "GL_EXT_ray_query";
const char* const E_GL_EXT_ray_flags_primitive_culling = "GL_EXT_ray_flags_primitive_culling";
const char* const E_GL_EXT_ray_cull_mask = "GL_EXT_ray_cull_mask";
const char* const E_GL_EXT_blend_func_extended = "GL_EXT_blend_func_extended";
const char* const E_GL_EXT_shader_implicit_conversions = "GL_EXT_shader_implicit_conversions";
const char* const E_GL_EXT_fragment_shading_rate = "GL_EXT_fragment_shading_rate";
const char* const E_GL_EXT_shader_image_int64 = "GL_EXT_shader_image_int64";
const char* const E_GL_EXT_null_initializer = "GL_EXT_null_initializer";
const char* const E_GL_EXT_shared_memory_block = "GL_EXT_shared_memory_block";
const char* const E_GL_EXT_subgroup_uniform_control_flow = "GL_EXT_subgroup_uniform_control_flow";
const char* const E_GL_EXT_spirv_intrinsics = "GL_EXT_spirv_intrinsics";
const char* const E_GL_EXT_fragment_shader_barycentric = "GL_EXT_fragment_shader_barycentric";
const char* const E_GL_EXT_mesh_shader = "GL_EXT_mesh_shader";
const char* const E_GL_EXT_opacity_micromap = "GL_EXT_opacity_micromap";
// Arrays of extensions for the above viewportEXTs duplications
const char* const post_depth_coverageEXTs[] = { E_GL_ARB_post_depth_coverage, E_GL_EXT_post_depth_coverage };
const int Num_post_depth_coverageEXTs = sizeof(post_depth_coverageEXTs) / sizeof(post_depth_coverageEXTs[0]);
// Array of extensions to cover both extensions providing ray tracing capabilities.
const char* const ray_tracing_EXTs[] = { E_GL_EXT_ray_query, E_GL_EXT_ray_tracing };
const int Num_ray_tracing_EXTs = sizeof(ray_tracing_EXTs) / sizeof(ray_tracing_EXTs[0]);
// OVR extensions
const char* const E_GL_OVR_multiview = "GL_OVR_multiview";
const char* const E_GL_OVR_multiview2 = "GL_OVR_multiview2";
const char* const OVR_multiview_EXTs[] = { E_GL_OVR_multiview, E_GL_OVR_multiview2 };
const int Num_OVR_multiview_EXTs = sizeof(OVR_multiview_EXTs) / sizeof(OVR_multiview_EXTs[0]);
// #line and #include
const char* const E_GL_GOOGLE_cpp_style_line_directive = "GL_GOOGLE_cpp_style_line_directive";
const char* const E_GL_GOOGLE_include_directive = "GL_GOOGLE_include_directive";
const char* const E_GL_AMD_shader_ballot = "GL_AMD_shader_ballot";
const char* const E_GL_AMD_shader_trinary_minmax = "GL_AMD_shader_trinary_minmax";
const char* const E_GL_AMD_shader_explicit_vertex_parameter = "GL_AMD_shader_explicit_vertex_parameter";
const char* const E_GL_AMD_gcn_shader = "GL_AMD_gcn_shader";
const char* const E_GL_AMD_gpu_shader_half_float = "GL_AMD_gpu_shader_half_float";
const char* const E_GL_AMD_texture_gather_bias_lod = "GL_AMD_texture_gather_bias_lod";
const char* const E_GL_AMD_gpu_shader_int16 = "GL_AMD_gpu_shader_int16";
const char* const E_GL_AMD_shader_image_load_store_lod = "GL_AMD_shader_image_load_store_lod";
const char* const E_GL_AMD_shader_fragment_mask = "GL_AMD_shader_fragment_mask";
const char* const E_GL_AMD_gpu_shader_half_float_fetch = "GL_AMD_gpu_shader_half_float_fetch";
const char* const E_GL_AMD_shader_early_and_late_fragment_tests = "GL_AMD_shader_early_and_late_fragment_tests";
const char* const E_GL_INTEL_shader_integer_functions2 = "GL_INTEL_shader_integer_functions2";
const char* const E_GL_NV_sample_mask_override_coverage = "GL_NV_sample_mask_override_coverage";
const char* const E_SPV_NV_geometry_shader_passthrough = "GL_NV_geometry_shader_passthrough";
const char* const E_GL_NV_viewport_array2 = "GL_NV_viewport_array2";
const char* const E_GL_NV_stereo_view_rendering = "GL_NV_stereo_view_rendering";
const char* const E_GL_NVX_multiview_per_view_attributes = "GL_NVX_multiview_per_view_attributes";
const char* const E_GL_NV_shader_atomic_int64 = "GL_NV_shader_atomic_int64";
const char* const E_GL_NV_conservative_raster_underestimation = "GL_NV_conservative_raster_underestimation";
const char* const E_GL_NV_shader_noperspective_interpolation = "GL_NV_shader_noperspective_interpolation";
const char* const E_GL_NV_shader_subgroup_partitioned = "GL_NV_shader_subgroup_partitioned";
const char* const E_GL_NV_shading_rate_image = "GL_NV_shading_rate_image";
const char* const E_GL_NV_ray_tracing = "GL_NV_ray_tracing";
const char* const E_GL_NV_ray_tracing_motion_blur = "GL_NV_ray_tracing_motion_blur";
const char* const E_GL_NV_fragment_shader_barycentric = "GL_NV_fragment_shader_barycentric";
const char* const E_GL_NV_compute_shader_derivatives = "GL_NV_compute_shader_derivatives";
const char* const E_GL_NV_shader_texture_footprint = "GL_NV_shader_texture_footprint";
const char* const E_GL_NV_mesh_shader = "GL_NV_mesh_shader";
const char* const E_GL_EXT_ray_tracing_position_fetch = "GL_EXT_ray_tracing_position_fetch";
// ARM
const char* const E_GL_ARM_shader_core_builtins = "GL_ARM_shader_core_builtins";
// Arrays of extensions for the above viewportEXTs duplications
const char* const viewportEXTs[] = { E_GL_ARB_shader_viewport_layer_array, E_GL_NV_viewport_array2 };
const int Num_viewportEXTs = sizeof(viewportEXTs) / sizeof(viewportEXTs[0]);
const char* const E_GL_NV_cooperative_matrix = "GL_NV_cooperative_matrix";
const char* const E_GL_NV_shader_sm_builtins = "GL_NV_shader_sm_builtins";
const char* const E_GL_NV_integer_cooperative_matrix = "GL_NV_integer_cooperative_matrix";
const char* const E_GL_NV_shader_invocation_reorder = "GL_NV_shader_invocation_reorder";
// AEP
const char* const E_GL_ANDROID_extension_pack_es31a = "GL_ANDROID_extension_pack_es31a";
const char* const E_GL_KHR_blend_equation_advanced = "GL_KHR_blend_equation_advanced";
const char* const E_GL_OES_sample_variables = "GL_OES_sample_variables";
const char* const E_GL_OES_shader_image_atomic = "GL_OES_shader_image_atomic";
const char* const E_GL_OES_shader_multisample_interpolation = "GL_OES_shader_multisample_interpolation";
const char* const E_GL_OES_texture_storage_multisample_2d_array = "GL_OES_texture_storage_multisample_2d_array";
const char* const E_GL_EXT_geometry_shader = "GL_EXT_geometry_shader";
const char* const E_GL_EXT_geometry_point_size = "GL_EXT_geometry_point_size";
const char* const E_GL_EXT_gpu_shader5 = "GL_EXT_gpu_shader5";
const char* const E_GL_EXT_primitive_bounding_box = "GL_EXT_primitive_bounding_box";
const char* const E_GL_EXT_shader_io_blocks = "GL_EXT_shader_io_blocks";
const char* const E_GL_EXT_tessellation_shader = "GL_EXT_tessellation_shader";
const char* const E_GL_EXT_tessellation_point_size = "GL_EXT_tessellation_point_size";
const char* const E_GL_EXT_texture_buffer = "GL_EXT_texture_buffer";
const char* const E_GL_EXT_texture_cube_map_array = "GL_EXT_texture_cube_map_array";
const char* const E_GL_EXT_shader_integer_mix = "GL_EXT_shader_integer_mix";
// OES matching AEP
const char* const E_GL_OES_geometry_shader = "GL_OES_geometry_shader";
const char* const E_GL_OES_geometry_point_size = "GL_OES_geometry_point_size";
const char* const E_GL_OES_gpu_shader5 = "GL_OES_gpu_shader5";
const char* const E_GL_OES_primitive_bounding_box = "GL_OES_primitive_bounding_box";
const char* const E_GL_OES_shader_io_blocks = "GL_OES_shader_io_blocks";
const char* const E_GL_OES_tessellation_shader = "GL_OES_tessellation_shader";
const char* const E_GL_OES_tessellation_point_size = "GL_OES_tessellation_point_size";
const char* const E_GL_OES_texture_buffer = "GL_OES_texture_buffer";
const char* const E_GL_OES_texture_cube_map_array = "GL_OES_texture_cube_map_array";
// EXT
const char* const E_GL_EXT_shader_explicit_arithmetic_types = "GL_EXT_shader_explicit_arithmetic_types";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_int8 = "GL_EXT_shader_explicit_arithmetic_types_int8";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_int16 = "GL_EXT_shader_explicit_arithmetic_types_int16";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_int32 = "GL_EXT_shader_explicit_arithmetic_types_int32";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_int64 = "GL_EXT_shader_explicit_arithmetic_types_int64";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_float16 = "GL_EXT_shader_explicit_arithmetic_types_float16";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_float32 = "GL_EXT_shader_explicit_arithmetic_types_float32";
const char* const E_GL_EXT_shader_explicit_arithmetic_types_float64 = "GL_EXT_shader_explicit_arithmetic_types_float64";
const char* const E_GL_EXT_shader_subgroup_extended_types_int8 = "GL_EXT_shader_subgroup_extended_types_int8";
const char* const E_GL_EXT_shader_subgroup_extended_types_int16 = "GL_EXT_shader_subgroup_extended_types_int16";
const char* const E_GL_EXT_shader_subgroup_extended_types_int64 = "GL_EXT_shader_subgroup_extended_types_int64";
const char* const E_GL_EXT_shader_subgroup_extended_types_float16 = "GL_EXT_shader_subgroup_extended_types_float16";
const char* const E_GL_EXT_terminate_invocation = "GL_EXT_terminate_invocation";
const char* const E_GL_EXT_shader_atomic_float = "GL_EXT_shader_atomic_float";
const char* const E_GL_EXT_shader_atomic_float2 = "GL_EXT_shader_atomic_float2";
const char* const E_GL_EXT_shader_tile_image = "GL_EXT_shader_tile_image";
// Arrays of extensions for the above AEP duplications
const char* const AEP_geometry_shader[] = { E_GL_EXT_geometry_shader, E_GL_OES_geometry_shader };
const int Num_AEP_geometry_shader = sizeof(AEP_geometry_shader)/sizeof(AEP_geometry_shader[0]);
const char* const AEP_geometry_point_size[] = { E_GL_EXT_geometry_point_size, E_GL_OES_geometry_point_size };
const int Num_AEP_geometry_point_size = sizeof(AEP_geometry_point_size)/sizeof(AEP_geometry_point_size[0]);
const char* const AEP_gpu_shader5[] = { E_GL_EXT_gpu_shader5, E_GL_OES_gpu_shader5 };
const int Num_AEP_gpu_shader5 = sizeof(AEP_gpu_shader5)/sizeof(AEP_gpu_shader5[0]);
const char* const AEP_primitive_bounding_box[] = { E_GL_EXT_primitive_bounding_box, E_GL_OES_primitive_bounding_box };
const int Num_AEP_primitive_bounding_box = sizeof(AEP_primitive_bounding_box)/sizeof(AEP_primitive_bounding_box[0]);
const char* const AEP_shader_io_blocks[] = { E_GL_EXT_shader_io_blocks, E_GL_OES_shader_io_blocks };
const int Num_AEP_shader_io_blocks = sizeof(AEP_shader_io_blocks)/sizeof(AEP_shader_io_blocks[0]);
const char* const AEP_tessellation_shader[] = { E_GL_EXT_tessellation_shader, E_GL_OES_tessellation_shader };
const int Num_AEP_tessellation_shader = sizeof(AEP_tessellation_shader)/sizeof(AEP_tessellation_shader[0]);
const char* const AEP_tessellation_point_size[] = { E_GL_EXT_tessellation_point_size, E_GL_OES_tessellation_point_size };
const int Num_AEP_tessellation_point_size = sizeof(AEP_tessellation_point_size)/sizeof(AEP_tessellation_point_size[0]);
const char* const AEP_texture_buffer[] = { E_GL_EXT_texture_buffer, E_GL_OES_texture_buffer };
const int Num_AEP_texture_buffer = sizeof(AEP_texture_buffer)/sizeof(AEP_texture_buffer[0]);
const char* const AEP_texture_cube_map_array[] = { E_GL_EXT_texture_cube_map_array, E_GL_OES_texture_cube_map_array };
const int Num_AEP_texture_cube_map_array = sizeof(AEP_texture_cube_map_array)/sizeof(AEP_texture_cube_map_array[0]);
const char* const AEP_mesh_shader[] = { E_GL_NV_mesh_shader, E_GL_EXT_mesh_shader };
const int Num_AEP_mesh_shader = sizeof(AEP_mesh_shader)/sizeof(AEP_mesh_shader[0]);
} // end namespace glslang
#endif // _VERSIONS_INCLUDED_

1310
include/vulkan/glslang/MachineIndependent/localintermediate.h Normal file
View file

File diff suppressed because it is too large Load diff

57
include/vulkan/glslang/Public/ResourceLimits.h Normal file
View file

@ -0,0 +1,57 @@
//
// Copyright (C) 2016 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
#ifndef _STAND_ALONE_RESOURCE_LIMITS_INCLUDED_
#define _STAND_ALONE_RESOURCE_LIMITS_INCLUDED_
#include <string>
#include "../Include/ResourceLimits.h"
// Return pointer to user-writable Resource to pass through API in
// future-proof way.
extern TBuiltInResource* GetResources();
// These are the default resources for TBuiltInResources, used for both
// - parsing this string for the case where the user didn't supply one,
// - dumping out a template for user construction of a config file.
extern const TBuiltInResource* GetDefaultResources();
// Returns the DefaultTBuiltInResource as a human-readable string.
std::string GetDefaultTBuiltInResourceString();
// Decodes the resource limits from |config| to |resources|.
void DecodeResourceLimits(TBuiltInResource* resources, char* config);
#endif // _STAND_ALONE_RESOURCE_LIMITS_INCLUDED_

987
include/vulkan/glslang/Public/ShaderLang.h Normal file
View file

@ -0,0 +1,987 @@
//
// Copyright (C) 2002-2005 3Dlabs Inc. Ltd.
// Copyright (C) 2013-2016 LunarG, Inc.
// Copyright (C) 2015-2018 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef _COMPILER_INTERFACE_INCLUDED_
#define _COMPILER_INTERFACE_INCLUDED_
#include "../Include/ResourceLimits.h"
#include "../MachineIndependent/Versions.h"
#include <cstring>
#include <vector>
#ifdef _WIN32
#define C_DECL __cdecl
#else
#define C_DECL
#endif
#ifdef GLSLANG_IS_SHARED_LIBRARY
#ifdef _WIN32
#ifdef GLSLANG_EXPORTING
#define GLSLANG_EXPORT __declspec(dllexport)
#else
#define GLSLANG_EXPORT __declspec(dllimport)
#endif
#elif __GNUC__ >= 4
#define GLSLANG_EXPORT __attribute__((visibility("default")))
#endif
#endif // GLSLANG_IS_SHARED_LIBRARY
#ifndef GLSLANG_EXPORT
#define GLSLANG_EXPORT
#endif
//
// This is the platform independent interface between an OGL driver
// and the shading language compiler/linker.
//
#ifdef __cplusplus
extern "C" {
#endif
//
// Call before doing any other compiler/linker operations.
//
// (Call once per process, not once per thread.)
//
GLSLANG_EXPORT int ShInitialize();
//
// Call this at process shutdown to clean up memory.
//
GLSLANG_EXPORT int ShFinalize();
//
// Types of languages the compiler can consume.
//
typedef enum {
EShLangVertex,
EShLangTessControl,
EShLangTessEvaluation,
EShLangGeometry,
EShLangFragment,
EShLangCompute,
EShLangRayGen,
EShLangRayGenNV = EShLangRayGen,
EShLangIntersect,
EShLangIntersectNV = EShLangIntersect,
EShLangAnyHit,
EShLangAnyHitNV = EShLangAnyHit,
EShLangClosestHit,
EShLangClosestHitNV = EShLangClosestHit,
EShLangMiss,
EShLangMissNV = EShLangMiss,
EShLangCallable,
EShLangCallableNV = EShLangCallable,
EShLangTask,
EShLangTaskNV = EShLangTask,
EShLangMesh,
EShLangMeshNV = EShLangMesh,
LAST_ELEMENT_MARKER(EShLangCount),
} EShLanguage; // would be better as stage, but this is ancient now
typedef enum : unsigned {
EShLangVertexMask = (1 << EShLangVertex),
EShLangTessControlMask = (1 << EShLangTessControl),
EShLangTessEvaluationMask = (1 << EShLangTessEvaluation),
EShLangGeometryMask = (1 << EShLangGeometry),
EShLangFragmentMask = (1 << EShLangFragment),
EShLangComputeMask = (1 << EShLangCompute),
EShLangRayGenMask = (1 << EShLangRayGen),
EShLangRayGenNVMask = EShLangRayGenMask,
EShLangIntersectMask = (1 << EShLangIntersect),
EShLangIntersectNVMask = EShLangIntersectMask,
EShLangAnyHitMask = (1 << EShLangAnyHit),
EShLangAnyHitNVMask = EShLangAnyHitMask,
EShLangClosestHitMask = (1 << EShLangClosestHit),
EShLangClosestHitNVMask = EShLangClosestHitMask,
EShLangMissMask = (1 << EShLangMiss),
EShLangMissNVMask = EShLangMissMask,
EShLangCallableMask = (1 << EShLangCallable),
EShLangCallableNVMask = EShLangCallableMask,
EShLangTaskMask = (1 << EShLangTask),
EShLangTaskNVMask = EShLangTaskMask,
EShLangMeshMask = (1 << EShLangMesh),
EShLangMeshNVMask = EShLangMeshMask,
LAST_ELEMENT_MARKER(EShLanguageMaskCount),
} EShLanguageMask;
namespace glslang {
class TType;
typedef enum {
EShSourceNone,
EShSourceGlsl, // GLSL, includes ESSL (OpenGL ES GLSL)
EShSourceHlsl, // HLSL
LAST_ELEMENT_MARKER(EShSourceCount),
} EShSource; // if EShLanguage were EShStage, this could be EShLanguage instead
typedef enum {
EShClientNone, // use when there is no client, e.g. for validation
EShClientVulkan, // as GLSL dialect, specifies KHR_vulkan_glsl extension
EShClientOpenGL, // as GLSL dialect, specifies ARB_gl_spirv extension
LAST_ELEMENT_MARKER(EShClientCount),
} EShClient;
typedef enum {
EShTargetNone,
EShTargetSpv, // SPIR-V (preferred spelling)
EshTargetSpv = EShTargetSpv, // legacy spelling
LAST_ELEMENT_MARKER(EShTargetCount),
} EShTargetLanguage;
typedef enum {
EShTargetVulkan_1_0 = (1 << 22), // Vulkan 1.0
EShTargetVulkan_1_1 = (1 << 22) | (1 << 12), // Vulkan 1.1
EShTargetVulkan_1_2 = (1 << 22) | (2 << 12), // Vulkan 1.2
EShTargetVulkan_1_3 = (1 << 22) | (3 << 12), // Vulkan 1.3
EShTargetOpenGL_450 = 450, // OpenGL
LAST_ELEMENT_MARKER(EShTargetClientVersionCount = 5),
} EShTargetClientVersion;
typedef EShTargetClientVersion EshTargetClientVersion;
typedef enum {
EShTargetSpv_1_0 = (1 << 16), // SPIR-V 1.0
EShTargetSpv_1_1 = (1 << 16) | (1 << 8), // SPIR-V 1.1
EShTargetSpv_1_2 = (1 << 16) | (2 << 8), // SPIR-V 1.2
EShTargetSpv_1_3 = (1 << 16) | (3 << 8), // SPIR-V 1.3
EShTargetSpv_1_4 = (1 << 16) | (4 << 8), // SPIR-V 1.4
EShTargetSpv_1_5 = (1 << 16) | (5 << 8), // SPIR-V 1.5
EShTargetSpv_1_6 = (1 << 16) | (6 << 8), // SPIR-V 1.6
LAST_ELEMENT_MARKER(EShTargetLanguageVersionCount = 7),
} EShTargetLanguageVersion;
struct TInputLanguage {
EShSource languageFamily; // redundant information with other input, this one overrides when not EShSourceNone
EShLanguage stage; // redundant information with other input, this one overrides when not EShSourceNone
EShClient dialect;
int dialectVersion; // version of client's language definition, not the client (when not EShClientNone)
bool vulkanRulesRelaxed;
};
struct TClient {
EShClient client;
EShTargetClientVersion version; // version of client itself (not the client's input dialect)
};
struct TTarget {
EShTargetLanguage language;
EShTargetLanguageVersion version; // version to target, if SPIR-V, defined by "word 1" of the SPIR-V header
bool hlslFunctionality1; // can target hlsl_functionality1 extension(s)
};
// All source/client/target versions and settings.
// Can override previous methods of setting, when items are set here.
// Expected to grow, as more are added, rather than growing parameter lists.
struct TEnvironment {
TInputLanguage input; // definition of the input language
TClient client; // what client is the overall compilation being done for?
TTarget target; // what to generate
};
GLSLANG_EXPORT const char* StageName(EShLanguage);
} // end namespace glslang
//
// Types of output the linker will create.
//
typedef enum {
EShExVertexFragment,
EShExFragment
} EShExecutable;
//
// Optimization level for the compiler.
//
typedef enum {
EShOptNoGeneration,
EShOptNone,
EShOptSimple, // Optimizations that can be done quickly
EShOptFull, // Optimizations that will take more time
LAST_ELEMENT_MARKER(EshOptLevelCount),
} EShOptimizationLevel;
//
// Texture and Sampler transformation mode.
//
typedef enum {
EShTexSampTransKeep, // keep textures and samplers as is (default)
EShTexSampTransUpgradeTextureRemoveSampler, // change texture w/o embeded sampler into sampled texture and throw away all samplers
LAST_ELEMENT_MARKER(EShTexSampTransCount),
} EShTextureSamplerTransformMode;
//
// Message choices for what errors and warnings are given.
//
enum EShMessages : unsigned {
EShMsgDefault = 0, // default is to give all required errors and extra warnings
EShMsgRelaxedErrors = (1 << 0), // be liberal in accepting input
EShMsgSuppressWarnings = (1 << 1), // suppress all warnings, except those required by the specification
EShMsgAST = (1 << 2), // print the AST intermediate representation
EShMsgSpvRules = (1 << 3), // issue messages for SPIR-V generation
EShMsgVulkanRules = (1 << 4), // issue messages for Vulkan-requirements of GLSL for SPIR-V
EShMsgOnlyPreprocessor = (1 << 5), // only print out errors produced by the preprocessor
EShMsgReadHlsl = (1 << 6), // use HLSL parsing rules and semantics
EShMsgCascadingErrors = (1 << 7), // get cascading errors; risks error-recovery issues, instead of an early exit
EShMsgKeepUncalled = (1 << 8), // for testing, don't eliminate uncalled functions
EShMsgHlslOffsets = (1 << 9), // allow block offsets to follow HLSL rules instead of GLSL rules
EShMsgDebugInfo = (1 << 10), // save debug information
EShMsgHlslEnable16BitTypes = (1 << 11), // enable use of 16-bit types in SPIR-V for HLSL
EShMsgHlslLegalization = (1 << 12), // enable HLSL Legalization messages
EShMsgHlslDX9Compatible = (1 << 13), // enable HLSL DX9 compatible mode (for samplers and semantics)
EShMsgBuiltinSymbolTable = (1 << 14), // print the builtin symbol table
EShMsgEnhanced = (1 << 15), // enhanced message readability
LAST_ELEMENT_MARKER(EShMsgCount),
};
//
// Options for building reflection
//
typedef enum {
EShReflectionDefault = 0, // default is original behaviour before options were added
EShReflectionStrictArraySuffix = (1 << 0), // reflection will follow stricter rules for array-of-structs suffixes
EShReflectionBasicArraySuffix = (1 << 1), // arrays of basic types will be appended with [0] as in GL reflection
EShReflectionIntermediateIO = (1 << 2), // reflect inputs and outputs to program, even with no vertex shader
EShReflectionSeparateBuffers = (1 << 3), // buffer variables and buffer blocks are reflected separately
EShReflectionAllBlockVariables = (1 << 4), // reflect all variables in blocks, even if they are inactive
EShReflectionUnwrapIOBlocks = (1 << 5), // unwrap input/output blocks the same as with uniform blocks
EShReflectionAllIOVariables = (1 << 6), // reflect all input/output variables, even if they are inactive
EShReflectionSharedStd140SSBO = (1 << 7), // Apply std140/shared rules for ubo to ssbo
EShReflectionSharedStd140UBO = (1 << 8), // Apply std140/shared rules for ubo to ssbo
LAST_ELEMENT_MARKER(EShReflectionCount),
} EShReflectionOptions;
//
// Build a table for bindings. This can be used for locating
// attributes, uniforms, globals, etc., as needed.
//
typedef struct {
const char* name;
int binding;
} ShBinding;
typedef struct {
int numBindings;
ShBinding* bindings; // array of bindings
} ShBindingTable;
//
// ShHandle held by but opaque to the driver. It is allocated,
// managed, and de-allocated by the compiler/linker. Its contents
// are defined by and used by the compiler and linker. For example,
// symbol table information and object code passed from the compiler
// to the linker can be stored where ShHandle points.
//
// If handle creation fails, 0 will be returned.
//
typedef void* ShHandle;
//
// Driver calls these to create and destroy compiler/linker
// objects.
//
GLSLANG_EXPORT ShHandle ShConstructCompiler(const EShLanguage, int debugOptions); // one per shader
GLSLANG_EXPORT ShHandle ShConstructLinker(const EShExecutable, int debugOptions); // one per shader pair
GLSLANG_EXPORT ShHandle ShConstructUniformMap(); // one per uniform namespace (currently entire program object)
GLSLANG_EXPORT void ShDestruct(ShHandle);
//
// The return value of ShCompile is boolean, non-zero indicating
// success.
//
// The info-log should be written by ShCompile into
// ShHandle, so it can answer future queries.
//
GLSLANG_EXPORT int ShCompile(
const ShHandle,
const char* const shaderStrings[],
const int numStrings,
const int* lengths,
const EShOptimizationLevel,
const TBuiltInResource *resources,
int debugOptions,
int defaultVersion = 110, // use 100 for ES environment, overridden by #version in shader
bool forwardCompatible = false, // give errors for use of deprecated features
EShMessages messages = EShMsgDefault // warnings and errors
);
GLSLANG_EXPORT int ShLinkExt(
const ShHandle, // linker object
const ShHandle h[], // compiler objects to link together
const int numHandles);
//
// ShSetEncrpytionMethod is a place-holder for specifying
// how source code is encrypted.
//
GLSLANG_EXPORT void ShSetEncryptionMethod(ShHandle);
//
// All the following return 0 if the information is not
// available in the object passed down, or the object is bad.
//
GLSLANG_EXPORT const char* ShGetInfoLog(const ShHandle);
GLSLANG_EXPORT const void* ShGetExecutable(const ShHandle);
GLSLANG_EXPORT int ShSetVirtualAttributeBindings(const ShHandle, const ShBindingTable*); // to detect user aliasing
GLSLANG_EXPORT int ShSetFixedAttributeBindings(const ShHandle, const ShBindingTable*); // to force any physical mappings
//
// Tell the linker to never assign a vertex attribute to this list of physical attributes
//
GLSLANG_EXPORT int ShExcludeAttributes(const ShHandle, int *attributes, int count);
//
// Returns the location ID of the named uniform.
// Returns -1 if error.
//
GLSLANG_EXPORT int ShGetUniformLocation(const ShHandle uniformMap, const char* name);
#ifdef __cplusplus
} // end extern "C"
#endif
////////////////////////////////////////////////////////////////////////////////////////////
//
// Deferred-Lowering C++ Interface
// -----------------------------------
//
// Below is a new alternate C++ interface, which deprecates the above
// opaque handle-based interface.
//
// The below is further designed to handle multiple compilation units per stage, where
// the intermediate results, including the parse tree, are preserved until link time,
// rather than the above interface which is designed to have each compilation unit
// lowered at compile time. In the above model, linking occurs on the lowered results,
// whereas in this model intra-stage linking can occur at the parse tree
// (treeRoot in TIntermediate) level, and then a full stage can be lowered.
//
#include <list>
#include <string>
#include <utility>
class TCompiler;
class TInfoSink;
namespace glslang {
struct Version {
int major;
int minor;
int patch;
const char* flavor;
};
GLSLANG_EXPORT Version GetVersion();
GLSLANG_EXPORT const char* GetEsslVersionString();
GLSLANG_EXPORT const char* GetGlslVersionString();
GLSLANG_EXPORT int GetKhronosToolId();
class TIntermediate;
class TProgram;
class TPoolAllocator;
// Call this exactly once per process before using anything else
GLSLANG_EXPORT bool InitializeProcess();
// Call once per process to tear down everything
GLSLANG_EXPORT void FinalizeProcess();
// Resource type for IO resolver
enum TResourceType {
EResSampler,
EResTexture,
EResImage,
EResUbo,
EResSsbo,
EResUav,
EResCount
};
enum TBlockStorageClass
{
EbsUniform = 0,
EbsStorageBuffer,
EbsPushConstant,
EbsNone, // not a uniform or buffer variable
EbsCount,
};
// Make one TShader per shader that you will link into a program. Then
// - provide the shader through setStrings() or setStringsWithLengths()
// - optionally call setEnv*(), see below for more detail
// - optionally use setPreamble() to set a special shader string that will be
// processed before all others but won't affect the validity of #version
// - optionally call addProcesses() for each setting/transform,
// see comment for class TProcesses
// - call parse(): source language and target environment must be selected
// either by correct setting of EShMessages sent to parse(), or by
// explicitly calling setEnv*()
// - query the info logs
//
// N.B.: Does not yet support having the same TShader instance being linked into
// multiple programs.
//
// N.B.: Destruct a linked program *before* destructing the shaders linked into it.
//
class TShader {
public:
GLSLANG_EXPORT explicit TShader(EShLanguage);
GLSLANG_EXPORT virtual ~TShader();
GLSLANG_EXPORT void setStrings(const char* const* s, int n);
GLSLANG_EXPORT void setStringsWithLengths(
const char* const* s, const int* l, int n);
GLSLANG_EXPORT void setStringsWithLengthsAndNames(
const char* const* s, const int* l, const char* const* names, int n);
void setPreamble(const char* s) { preamble = s; }
GLSLANG_EXPORT void setEntryPoint(const char* entryPoint);
GLSLANG_EXPORT void setSourceEntryPoint(const char* sourceEntryPointName);
GLSLANG_EXPORT void addProcesses(const std::vector<std::string>&);
GLSLANG_EXPORT void setUniqueId(unsigned long long id);
GLSLANG_EXPORT void setOverrideVersion(int version);
GLSLANG_EXPORT void setDebugInfo(bool debugInfo);
// IO resolver binding data: see comments in ShaderLang.cpp
GLSLANG_EXPORT void setShiftBinding(TResourceType res, unsigned int base);
GLSLANG_EXPORT void setShiftSamplerBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftTextureBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftImageBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftUboBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftUavBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftCbufferBinding(unsigned int base); // synonym for setShiftUboBinding
GLSLANG_EXPORT void setShiftSsboBinding(unsigned int base); // DEPRECATED: use setShiftBinding
GLSLANG_EXPORT void setShiftBindingForSet(TResourceType res, unsigned int base, unsigned int set);
GLSLANG_EXPORT void setResourceSetBinding(const std::vector<std::string>& base);
GLSLANG_EXPORT void setAutoMapBindings(bool map);
GLSLANG_EXPORT void setAutoMapLocations(bool map);
GLSLANG_EXPORT void addUniformLocationOverride(const char* name, int loc);
GLSLANG_EXPORT void setUniformLocationBase(int base);
GLSLANG_EXPORT void setInvertY(bool invert);
GLSLANG_EXPORT void setDxPositionW(bool dxPosW);
GLSLANG_EXPORT void setEnhancedMsgs();
#ifdef ENABLE_HLSL
GLSLANG_EXPORT void setHlslIoMapping(bool hlslIoMap);
GLSLANG_EXPORT void setFlattenUniformArrays(bool flatten);
#endif
GLSLANG_EXPORT void setNoStorageFormat(bool useUnknownFormat);
GLSLANG_EXPORT void setNanMinMaxClamp(bool nanMinMaxClamp);
GLSLANG_EXPORT void setTextureSamplerTransformMode(EShTextureSamplerTransformMode mode);
GLSLANG_EXPORT void addBlockStorageOverride(const char* nameStr, glslang::TBlockStorageClass backing);
GLSLANG_EXPORT void setGlobalUniformBlockName(const char* name);
GLSLANG_EXPORT void setAtomicCounterBlockName(const char* name);
GLSLANG_EXPORT void setGlobalUniformSet(unsigned int set);
GLSLANG_EXPORT void setGlobalUniformBinding(unsigned int binding);
GLSLANG_EXPORT void setAtomicCounterBlockSet(unsigned int set);
GLSLANG_EXPORT void setAtomicCounterBlockBinding(unsigned int binding);
// For setting up the environment (cleared to nothingness in the constructor).
// These must be called so that parsing is done for the right source language and
// target environment, either indirectly through TranslateEnvironment() based on
// EShMessages et. al., or directly by the user.
//
// setEnvInput: The input source language and stage. If generating code for a
// specific client, the input client semantics to use and the
// version of that client's input semantics to use, otherwise
// use EShClientNone and version of 0, e.g. for validation mode.
// Note 'version' does not describe the target environment,
// just the version of the source dialect to compile under.
// For example, to choose the Vulkan dialect of GLSL defined by
// version 100 of the KHR_vulkan_glsl extension: lang = EShSourceGlsl,
// dialect = EShClientVulkan, and version = 100.
//
// See the definitions of TEnvironment, EShSource, EShLanguage,
// and EShClient for choices and more detail.
//
// setEnvClient: The client that will be hosting the execution, and its version.
// Note 'version' is not the version of the languages involved, but
// the version of the client environment.
// Use EShClientNone and version of 0 if there is no client, e.g.
// for validation mode.
//
// See EShTargetClientVersion for choices.
//
// setEnvTarget: The language to translate to when generating code, and that
// language's version.
// Use EShTargetNone and version of 0 if there is no client, e.g.
// for validation mode.
//
void setEnvInput(EShSource lang, EShLanguage envStage, EShClient client, int version)
{
environment.input.languageFamily = lang;
environment.input.stage = envStage;
environment.input.dialect = client;
environment.input.dialectVersion = version;
}
void setEnvClient(EShClient client, EShTargetClientVersion version)
{
environment.client.client = client;
environment.client.version = version;
}
void setEnvTarget(EShTargetLanguage lang, EShTargetLanguageVersion version)
{
environment.target.language = lang;
environment.target.version = version;
}
void getStrings(const char* const* &s, int& n) { s = strings; n = numStrings; }
#ifdef ENABLE_HLSL
void setEnvTargetHlslFunctionality1() { environment.target.hlslFunctionality1 = true; }
bool getEnvTargetHlslFunctionality1() const { return environment.target.hlslFunctionality1; }
#else
bool getEnvTargetHlslFunctionality1() const { return false; }
#endif
void setEnvInputVulkanRulesRelaxed() { environment.input.vulkanRulesRelaxed = true; }
bool getEnvInputVulkanRulesRelaxed() const { return environment.input.vulkanRulesRelaxed; }
// Interface to #include handlers.
//
// To support #include, a client of Glslang does the following:
// 1. Call setStringsWithNames to set the source strings and associated
// names. For example, the names could be the names of the files
// containing the shader sources.
// 2. Call parse with an Includer.
//
// When the Glslang parser encounters an #include directive, it calls
// the Includer's include method with the requested include name
// together with the current string name. The returned IncludeResult
// contains the fully resolved name of the included source, together
// with the source text that should replace the #include directive
// in the source stream. After parsing that source, Glslang will
// release the IncludeResult object.
class Includer {
public:
// An IncludeResult contains the resolved name and content of a source
// inclusion.
struct IncludeResult {
IncludeResult(const std::string& headerName, const char* const headerData, const size_t headerLength, void* userData) :
headerName(headerName), headerData(headerData), headerLength(headerLength), userData(userData) { }
// For a successful inclusion, the fully resolved name of the requested
// include. For example, in a file system-based includer, full resolution
// should convert a relative path name into an absolute path name.
// For a failed inclusion, this is an empty string.
const std::string headerName;
// The content and byte length of the requested inclusion. The
// Includer producing this IncludeResult retains ownership of the
// storage.
// For a failed inclusion, the header
// field points to a string containing error details.
const char* const headerData;
const size_t headerLength;
// Include resolver's context.
void* userData;
protected:
IncludeResult& operator=(const IncludeResult&);
IncludeResult();
};
// For both include methods below:
//
// Resolves an inclusion request by name, current source name,
// and include depth.
// On success, returns an IncludeResult containing the resolved name
// and content of the include.
// On failure, returns a nullptr, or an IncludeResult
// with an empty string for the headerName and error details in the
// header field.
// The Includer retains ownership of the contents
// of the returned IncludeResult value, and those contents must
// remain valid until the releaseInclude method is called on that
// IncludeResult object.
//
// Note "local" vs. "system" is not an "either/or": "local" is an
// extra thing to do over "system". Both might get called, as per
// the C++ specification.
// For the "system" or <>-style includes; search the "system" paths.
virtual IncludeResult* includeSystem(const char* /*headerName*/,
const char* /*includerName*/,
size_t /*inclusionDepth*/) { return nullptr; }
// For the "local"-only aspect of a "" include. Should not search in the
// "system" paths, because on returning a failure, the parser will
// call includeSystem() to look in the "system" locations.
virtual IncludeResult* includeLocal(const char* /*headerName*/,
const char* /*includerName*/,
size_t /*inclusionDepth*/) { return nullptr; }
// Signals that the parser will no longer use the contents of the
// specified IncludeResult.
virtual void releaseInclude(IncludeResult*) = 0;
virtual ~Includer() {}
};
// Fail all Includer searches
class ForbidIncluder : public Includer {
public:
virtual void releaseInclude(IncludeResult*) override { }
};
GLSLANG_EXPORT bool parse(
const TBuiltInResource*, int defaultVersion, EProfile defaultProfile,
bool forceDefaultVersionAndProfile, bool forwardCompatible,
EShMessages, Includer&);
bool parse(const TBuiltInResource* res, int defaultVersion, EProfile defaultProfile, bool forceDefaultVersionAndProfile,
bool forwardCompatible, EShMessages messages)
{
TShader::ForbidIncluder includer;
return parse(res, defaultVersion, defaultProfile, forceDefaultVersionAndProfile, forwardCompatible, messages, includer);
}
// Equivalent to parse() without a default profile and without forcing defaults.
bool parse(const TBuiltInResource* builtInResources, int defaultVersion, bool forwardCompatible, EShMessages messages)
{
return parse(builtInResources, defaultVersion, ENoProfile, false, forwardCompatible, messages);
}
bool parse(const TBuiltInResource* builtInResources, int defaultVersion, bool forwardCompatible, EShMessages messages,
Includer& includer)
{
return parse(builtInResources, defaultVersion, ENoProfile, false, forwardCompatible, messages, includer);
}
// NOTE: Doing just preprocessing to obtain a correct preprocessed shader string
// is not an officially supported or fully working path.
GLSLANG_EXPORT bool preprocess(
const TBuiltInResource* builtInResources, int defaultVersion,
EProfile defaultProfile, bool forceDefaultVersionAndProfile,
bool forwardCompatible, EShMessages message, std::string* outputString,
Includer& includer);
GLSLANG_EXPORT const char* getInfoLog();
GLSLANG_EXPORT const char* getInfoDebugLog();
EShLanguage getStage() const { return stage; }
TIntermediate* getIntermediate() const { return intermediate; }
protected:
TPoolAllocator* pool;
EShLanguage stage;
TCompiler* compiler;
TIntermediate* intermediate;
TInfoSink* infoSink;
// strings and lengths follow the standard for glShaderSource:
// strings is an array of numStrings pointers to string data.
// lengths can be null, but if not it is an array of numStrings
// integers containing the length of the associated strings.
// if lengths is null or lengths[n] < 0 the associated strings[n] is
// assumed to be null-terminated.
// stringNames is the optional names for all the strings. If stringNames
// is null, then none of the strings has name. If a certain element in
// stringNames is null, then the corresponding string does not have name.
const char* const* strings; // explicit code to compile, see previous comment
const int* lengths;
const char* const* stringNames;
int numStrings; // size of the above arrays
const char* preamble; // string of implicit code to compile before the explicitly provided code
// a function in the source string can be renamed FROM this TO the name given in setEntryPoint.
std::string sourceEntryPointName;
// overrides #version in shader source or default version if #version isn't present
int overrideVersion;
TEnvironment environment;
friend class TProgram;
private:
TShader& operator=(TShader&);
};
#if !defined(GLSLANG_WEB)
//
// A reflection database and its interface, consistent with the OpenGL API reflection queries.
//
// Data needed for just a single object at the granularity exchanged by the reflection API
class TObjectReflection {
public:
GLSLANG_EXPORT TObjectReflection(const std::string& pName, const TType& pType, int pOffset, int pGLDefineType, int pSize, int pIndex);
const TType* getType() const { return type; }
GLSLANG_EXPORT int getBinding() const;
GLSLANG_EXPORT void dump() const;
static TObjectReflection badReflection() { return TObjectReflection(); }
std::string name;
int offset;
int glDefineType;
int size; // data size in bytes for a block, array size for a (non-block) object that's an array
int index;
int counterIndex;
int numMembers;
int arrayStride; // stride of an array variable
int topLevelArraySize; // size of the top-level variable in a storage buffer member
int topLevelArrayStride; // stride of the top-level variable in a storage buffer member
EShLanguageMask stages;
protected:
TObjectReflection()
: offset(-1), glDefineType(-1), size(-1), index(-1), counterIndex(-1), numMembers(-1), arrayStride(0),
topLevelArrayStride(0), stages(EShLanguageMask(0)), type(nullptr)
{
}
const TType* type;
};
class TReflection;
class TIoMapper;
struct TVarEntryInfo;
// Allows to customize the binding layout after linking.
// All used uniform variables will invoke at least validateBinding.
// If validateBinding returned true then the other resolveBinding,
// resolveSet, and resolveLocation are invoked to resolve the binding
// and descriptor set index respectively.
//
// Invocations happen in a particular order:
// 1) all shader inputs
// 2) all shader outputs
// 3) all uniforms with binding and set already defined
// 4) all uniforms with binding but no set defined
// 5) all uniforms with set but no binding defined
// 6) all uniforms with no binding and no set defined
//
// mapIO will use this resolver in two phases. The first
// phase is a notification phase, calling the corresponging
// notifiy callbacks, this phase ends with a call to endNotifications.
// Phase two starts directly after the call to endNotifications
// and calls all other callbacks to validate and to get the
// bindings, sets, locations, component and color indices.
//
// NOTE: that still limit checks are applied to bindings and sets
// and may result in an error.
class TIoMapResolver
{
public:
virtual ~TIoMapResolver() {}
// Should return true if the resulting/current binding would be okay.
// Basic idea is to do aliasing binding checks with this.
virtual bool validateBinding(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current binding should be overridden.
// Return -1 if the current binding (including no binding) should be kept.
virtual int resolveBinding(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current set should be overridden.
// Return -1 if the current set (including no set) should be kept.
virtual int resolveSet(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current location should be overridden.
// Return -1 if the current location (including no location) should be kept.
virtual int resolveUniformLocation(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return true if the resulting/current setup would be okay.
// Basic idea is to do aliasing checks and reject invalid semantic names.
virtual bool validateInOut(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current location should be overridden.
// Return -1 if the current location (including no location) should be kept.
virtual int resolveInOutLocation(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current component index should be overridden.
// Return -1 if the current component index (including no index) should be kept.
virtual int resolveInOutComponent(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Should return a value >= 0 if the current color index should be overridden.
// Return -1 if the current color index (including no index) should be kept.
virtual int resolveInOutIndex(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Notification of a uniform variable
virtual void notifyBinding(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Notification of a in or out variable
virtual void notifyInOut(EShLanguage stage, TVarEntryInfo& ent) = 0;
// Called by mapIO when it starts its notify pass for the given stage
virtual void beginNotifications(EShLanguage stage) = 0;
// Called by mapIO when it has finished the notify pass
virtual void endNotifications(EShLanguage stage) = 0;
// Called by mipIO when it starts its resolve pass for the given stage
virtual void beginResolve(EShLanguage stage) = 0;
// Called by mapIO when it has finished the resolve pass
virtual void endResolve(EShLanguage stage) = 0;
// Called by mapIO when it starts its symbol collect for teh given stage
virtual void beginCollect(EShLanguage stage) = 0;
// Called by mapIO when it has finished the symbol collect
virtual void endCollect(EShLanguage stage) = 0;
// Called by TSlotCollector to resolve storage locations or bindings
virtual void reserverStorageSlot(TVarEntryInfo& ent, TInfoSink& infoSink) = 0;
// Called by TSlotCollector to resolve resource locations or bindings
virtual void reserverResourceSlot(TVarEntryInfo& ent, TInfoSink& infoSink) = 0;
// Called by mapIO.addStage to set shader stage mask to mark a stage be added to this pipeline
virtual void addStage(EShLanguage stage, TIntermediate& stageIntermediate) = 0;
};
#endif // !GLSLANG_WEB
// Make one TProgram per set of shaders that will get linked together. Add all
// the shaders that are to be linked together. After calling shader.parse()
// for all shaders, call link().
//
// N.B.: Destruct a linked program *before* destructing the shaders linked into it.
//
class TProgram {
public:
GLSLANG_EXPORT TProgram();
GLSLANG_EXPORT virtual ~TProgram();
void addShader(TShader* shader) { stages[shader->stage].push_back(shader); }
std::list<TShader*>& getShaders(EShLanguage stage) { return stages[stage]; }
// Link Validation interface
GLSLANG_EXPORT bool link(EShMessages);
GLSLANG_EXPORT const char* getInfoLog();
GLSLANG_EXPORT const char* getInfoDebugLog();
TIntermediate* getIntermediate(EShLanguage stage) const { return intermediate[stage]; }
#if !defined(GLSLANG_WEB)
// Reflection Interface
// call first, to do liveness analysis, index mapping, etc.; returns false on failure
GLSLANG_EXPORT bool buildReflection(int opts = EShReflectionDefault);
GLSLANG_EXPORT unsigned getLocalSize(int dim) const; // return dim'th local size
GLSLANG_EXPORT int getReflectionIndex(const char *name) const;
GLSLANG_EXPORT int getReflectionPipeIOIndex(const char* name, const bool inOrOut) const;
GLSLANG_EXPORT int getNumUniformVariables() const;
GLSLANG_EXPORT const TObjectReflection& getUniform(int index) const;
GLSLANG_EXPORT int getNumUniformBlocks() const;
GLSLANG_EXPORT const TObjectReflection& getUniformBlock(int index) const;
GLSLANG_EXPORT int getNumPipeInputs() const;
GLSLANG_EXPORT const TObjectReflection& getPipeInput(int index) const;
GLSLANG_EXPORT int getNumPipeOutputs() const;
GLSLANG_EXPORT const TObjectReflection& getPipeOutput(int index) const;
GLSLANG_EXPORT int getNumBufferVariables() const;
GLSLANG_EXPORT const TObjectReflection& getBufferVariable(int index) const;
GLSLANG_EXPORT int getNumBufferBlocks() const;
GLSLANG_EXPORT const TObjectReflection& getBufferBlock(int index) const;
GLSLANG_EXPORT int getNumAtomicCounters() const;
GLSLANG_EXPORT const TObjectReflection& getAtomicCounter(int index) const;
// Legacy Reflection Interface - expressed in terms of above interface
// can be used for glGetProgramiv(GL_ACTIVE_UNIFORMS)
int getNumLiveUniformVariables() const { return getNumUniformVariables(); }
// can be used for glGetProgramiv(GL_ACTIVE_UNIFORM_BLOCKS)
int getNumLiveUniformBlocks() const { return getNumUniformBlocks(); }
// can be used for glGetProgramiv(GL_ACTIVE_ATTRIBUTES)
int getNumLiveAttributes() const { return getNumPipeInputs(); }
// can be used for glGetUniformIndices()
int getUniformIndex(const char *name) const { return getReflectionIndex(name); }
int getPipeIOIndex(const char *name, const bool inOrOut) const
{ return getReflectionPipeIOIndex(name, inOrOut); }
// can be used for "name" part of glGetActiveUniform()
const char *getUniformName(int index) const { return getUniform(index).name.c_str(); }
// returns the binding number
int getUniformBinding(int index) const { return getUniform(index).getBinding(); }
// returns Shaders Stages where a Uniform is present
EShLanguageMask getUniformStages(int index) const { return getUniform(index).stages; }
// can be used for glGetActiveUniformsiv(GL_UNIFORM_BLOCK_INDEX)
int getUniformBlockIndex(int index) const { return getUniform(index).index; }
// can be used for glGetActiveUniformsiv(GL_UNIFORM_TYPE)
int getUniformType(int index) const { return getUniform(index).glDefineType; }
// can be used for glGetActiveUniformsiv(GL_UNIFORM_OFFSET)
int getUniformBufferOffset(int index) const { return getUniform(index).offset; }
// can be used for glGetActiveUniformsiv(GL_UNIFORM_SIZE)
int getUniformArraySize(int index) const { return getUniform(index).size; }
// returns a TType*
const TType *getUniformTType(int index) const { return getUniform(index).getType(); }
// can be used for glGetActiveUniformBlockName()
const char *getUniformBlockName(int index) const { return getUniformBlock(index).name.c_str(); }
// can be used for glGetActiveUniformBlockiv(UNIFORM_BLOCK_DATA_SIZE)
int getUniformBlockSize(int index) const { return getUniformBlock(index).size; }
// returns the block binding number
int getUniformBlockBinding(int index) const { return getUniformBlock(index).getBinding(); }
// returns block index of associated counter.
int getUniformBlockCounterIndex(int index) const { return getUniformBlock(index).counterIndex; }
// returns a TType*
const TType *getUniformBlockTType(int index) const { return getUniformBlock(index).getType(); }
// can be used for glGetActiveAttrib()
const char *getAttributeName(int index) const { return getPipeInput(index).name.c_str(); }
// can be used for glGetActiveAttrib()
int getAttributeType(int index) const { return getPipeInput(index).glDefineType; }
// returns a TType*
const TType *getAttributeTType(int index) const { return getPipeInput(index).getType(); }
GLSLANG_EXPORT void dumpReflection();
// I/O mapping: apply base offsets and map live unbound variables
// If resolver is not provided it uses the previous approach
// and respects auto assignment and offsets.
GLSLANG_EXPORT bool mapIO(TIoMapResolver* pResolver = nullptr, TIoMapper* pIoMapper = nullptr);
#endif // !GLSLANG_WEB
protected:
GLSLANG_EXPORT bool linkStage(EShLanguage, EShMessages);
GLSLANG_EXPORT bool crossStageCheck(EShMessages);
TPoolAllocator* pool;
std::list<TShader*> stages[EShLangCount];
TIntermediate* intermediate[EShLangCount];
bool newedIntermediate[EShLangCount]; // track which intermediate were "new" versus reusing a singleton unit in a stage
TInfoSink* infoSink;
#if !defined(GLSLANG_WEB)
TReflection* reflection;
#endif
bool linked;
private:
TProgram(TProgram&);
TProgram& operator=(TProgram&);
};
} // end namespace glslang
#endif // _COMPILER_INTERFACE_INCLUDED_

57
include/vulkan/glslang/Public/resource_limits_c.h Normal file
View file

@ -0,0 +1,57 @@
/**
BSD 2-Clause License
Copyright (c) 2020, Travis Fort
All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
1. Redistributions of source code must retain the above copyright notice, this
list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice,
this list of conditions and the following disclaimer in the documentation
and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**/
#ifndef _STAND_ALONE_RESOURCE_LIMITS_C_INCLUDED_
#define _STAND_ALONE_RESOURCE_LIMITS_C_INCLUDED_
#include "../Include/glslang_c_interface.h"
#ifdef __cplusplus
extern "C" {
#endif
// Returns a struct that can be use to create custom resource values.
glslang_resource_t* glslang_resource(void);
// These are the default resources for TBuiltInResources, used for both
// - parsing this string for the case where the user didn't supply one,
// - dumping out a template for user construction of a config file.
const glslang_resource_t* glslang_default_resource(void);
// Returns the DefaultTBuiltInResource as a human-readable string.
// NOTE: User is responsible for freeing this string.
const char* glslang_default_resource_string();
// Decodes the resource limits from |config| to |resources|.
void glslang_decode_resource_limits(glslang_resource_t* resources, char* config);
#ifdef __cplusplus
}
#endif
#endif // _STAND_ALONE_RESOURCE_LIMITS_C_INCLUDED_

61
include/vulkan/glslang/SPIRV/GlslangToSpv.h Normal file
View file

@ -0,0 +1,61 @@
//
// Copyright (C) 2014 LunarG, Inc.
// Copyright (C) 2015-2018 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
#pragma once
#if defined(_MSC_VER) && _MSC_VER >= 1900
#pragma warning(disable : 4464) // relative include path contains '..'
#endif
#include "SpvTools.h"
#include "glslang/Include/intermediate.h"
#include <string>
#include <vector>
#include "Logger.h"
namespace glslang {
void GetSpirvVersion(std::string&);
int GetSpirvGeneratorVersion();
void GlslangToSpv(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
SpvOptions* options = nullptr);
void GlslangToSpv(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger* logger, SpvOptions* options = nullptr);
void OutputSpvBin(const std::vector<unsigned int>& spirv, const char* baseName);
void OutputSpvHex(const std::vector<unsigned int>& spirv, const char* baseName, const char* varName);
}

83
include/vulkan/glslang/SPIRV/Logger.h Normal file
View file

@ -0,0 +1,83 @@
//
// Copyright (C) 2016 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
#ifndef GLSLANG_SPIRV_LOGGER_H
#define GLSLANG_SPIRV_LOGGER_H
#include <string>
#include <vector>
namespace spv {
// A class for holding all SPIR-V build status messages, including
// missing/TBD functionalities, warnings, and errors.
class SpvBuildLogger {
public:
SpvBuildLogger() {}
#ifdef GLSLANG_WEB
void tbdFunctionality(const std::string& f) { }
void missingFunctionality(const std::string& f) { }
void warning(const std::string& w) { }
void error(const std::string& e) { errors.push_back(e); }
std::string getAllMessages() { return ""; }
#else
// Registers a TBD functionality.
void tbdFunctionality(const std::string& f);
// Registers a missing functionality.
void missingFunctionality(const std::string& f);
// Logs a warning.
void warning(const std::string& w) { warnings.push_back(w); }
// Logs an error.
void error(const std::string& e) { errors.push_back(e); }
// Returns all messages accumulated in the order of:
// TBD functionalities, missing functionalities, warnings, errors.
std::string getAllMessages() const;
#endif
private:
SpvBuildLogger(const SpvBuildLogger&);
std::vector<std::string> tbdFeatures;
std::vector<std::string> missingFeatures;
std::vector<std::string> warnings;
std::vector<std::string> errors;
};
} // end spv namespace
#endif // GLSLANG_SPIRV_LOGGER_H

284
include/vulkan/glslang/SPIRV/SPVRemapper.h Normal file
View file

@ -0,0 +1,284 @@
//
// Copyright (C) 2015 LunarG, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
#ifndef SPIRVREMAPPER_H
#define SPIRVREMAPPER_H
#include <string>
#include <vector>
#include <cstdlib>
#include <exception>
namespace spv {
class spirvbin_base_t
{
public:
enum Options {
NONE = 0,
STRIP = (1<<0),
MAP_TYPES = (1<<1),
MAP_NAMES = (1<<2),
MAP_FUNCS = (1<<3),
DCE_FUNCS = (1<<4),
DCE_VARS = (1<<5),
DCE_TYPES = (1<<6),
OPT_LOADSTORE = (1<<7),
OPT_FWD_LS = (1<<8), // EXPERIMENTAL: PRODUCES INVALID SCHEMA-0 SPIRV
MAP_ALL = (MAP_TYPES | MAP_NAMES | MAP_FUNCS),
DCE_ALL = (DCE_FUNCS | DCE_VARS | DCE_TYPES),
OPT_ALL = (OPT_LOADSTORE),
ALL_BUT_STRIP = (MAP_ALL | DCE_ALL | OPT_ALL),
DO_EVERYTHING = (STRIP | ALL_BUT_STRIP)
};
};
} // namespace SPV
#include <functional>
#include <cstdint>
#include <unordered_map>
#include <unordered_set>
#include <map>
#include <set>
#include <cassert>
#include "spirv.hpp"
#include "spvIR.h"
namespace spv {
// class to hold SPIR-V binary data for remapping, DCE, and debug stripping
class spirvbin_t : public spirvbin_base_t
{
public:
spirvbin_t(int verbose = 0) : entryPoint(spv::NoResult), largestNewId(0), verbose(verbose), errorLatch(false)
{ }
virtual ~spirvbin_t() { }
// remap on an existing binary in memory
void remap(std::vector<std::uint32_t>& spv, const std::vector<std::string>& whiteListStrings,
std::uint32_t opts = DO_EVERYTHING);
// remap on an existing binary in memory - legacy interface without white list
void remap(std::vector<std::uint32_t>& spv, std::uint32_t opts = DO_EVERYTHING);
// Type for error/log handler functions
typedef std::function<void(const std::string&)> errorfn_t;
typedef std::function<void(const std::string&)> logfn_t;
// Register error/log handling functions (can be lambda fn / functor / etc)
static void registerErrorHandler(errorfn_t handler) { errorHandler = handler; }
static void registerLogHandler(logfn_t handler) { logHandler = handler; }
protected:
// This can be overridden to provide other message behavior if needed
virtual void msg(int minVerbosity, int indent, const std::string& txt) const;
private:
// Local to global, or global to local ID map
typedef std::unordered_map<spv::Id, spv::Id> idmap_t;
typedef std::unordered_set<spv::Id> idset_t;
typedef std::unordered_map<spv::Id, int> blockmap_t;
void remap(std::uint32_t opts = DO_EVERYTHING);
// Map of names to IDs
typedef std::unordered_map<std::string, spv::Id> namemap_t;
typedef std::uint32_t spirword_t;
typedef std::pair<unsigned, unsigned> range_t;
typedef std::function<void(spv::Id&)> idfn_t;
typedef std::function<bool(spv::Op, unsigned start)> instfn_t;
// Special Values for ID map:
static const spv::Id unmapped; // unchanged from default value
static const spv::Id unused; // unused ID
static const int header_size; // SPIR header = 5 words
class id_iterator_t;
// For mapping type entries between different shaders
typedef std::vector<spirword_t> typeentry_t;
typedef std::map<spv::Id, typeentry_t> globaltypes_t;
// A set that preserves position order, and a reverse map
typedef std::set<int> posmap_t;
typedef std::unordered_map<spv::Id, int> posmap_rev_t;
// Maps and ID to the size of its base type, if known.
typedef std::unordered_map<spv::Id, unsigned> typesize_map_t;
// handle error
void error(const std::string& txt) const { errorLatch = true; errorHandler(txt); }
bool isConstOp(spv::Op opCode) const;
bool isTypeOp(spv::Op opCode) const;
bool isStripOp(spv::Op opCode) const;
bool isFlowCtrl(spv::Op opCode) const;
range_t literalRange(spv::Op opCode) const;
range_t typeRange(spv::Op opCode) const;
range_t constRange(spv::Op opCode) const;
unsigned typeSizeInWords(spv::Id id) const;
unsigned idTypeSizeInWords(spv::Id id) const;
bool isStripOp(spv::Op opCode, unsigned start) const;
spv::Id& asId(unsigned word) { return spv[word]; }
const spv::Id& asId(unsigned word) const { return spv[word]; }
spv::Op asOpCode(unsigned word) const { return opOpCode(spv[word]); }
std::uint32_t asOpCodeHash(unsigned word);
spv::Decoration asDecoration(unsigned word) const { return spv::Decoration(spv[word]); }
unsigned asWordCount(unsigned word) const { return opWordCount(spv[word]); }
spv::Id asTypeConstId(unsigned word) const { return asId(word + (isTypeOp(asOpCode(word)) ? 1 : 2)); }
unsigned idPos(spv::Id id) const;
static unsigned opWordCount(spirword_t data) { return data >> spv::WordCountShift; }
static spv::Op opOpCode(spirword_t data) { return spv::Op(data & spv::OpCodeMask); }
// Header access & set methods
spirword_t magic() const { return spv[0]; } // return magic number
spirword_t bound() const { return spv[3]; } // return Id bound from header
spirword_t bound(spirword_t b) { return spv[3] = b; }
spirword_t genmagic() const { return spv[2]; } // generator magic
spirword_t genmagic(spirword_t m) { return spv[2] = m; }
spirword_t schemaNum() const { return spv[4]; } // schema number from header
// Mapping fns: get
spv::Id localId(spv::Id id) const { return idMapL[id]; }
// Mapping fns: set
inline spv::Id localId(spv::Id id, spv::Id newId);
void countIds(spv::Id id);
// Return next unused new local ID.
// NOTE: boost::dynamic_bitset would be more efficient due to find_next(),
// which std::vector<bool> doens't have.
inline spv::Id nextUnusedId(spv::Id id);
void buildLocalMaps();
std::string literalString(unsigned word) const; // Return literal as a std::string
int literalStringWords(const std::string& str) const { return (int(str.size())+4)/4; }
bool isNewIdMapped(spv::Id newId) const { return isMapped(newId); }
bool isOldIdUnmapped(spv::Id oldId) const { return localId(oldId) == unmapped; }
bool isOldIdUnused(spv::Id oldId) const { return localId(oldId) == unused; }
bool isOldIdMapped(spv::Id oldId) const { return !isOldIdUnused(oldId) && !isOldIdUnmapped(oldId); }
bool isFunction(spv::Id oldId) const { return fnPos.find(oldId) != fnPos.end(); }
// bool matchType(const globaltypes_t& globalTypes, spv::Id lt, spv::Id gt) const;
// spv::Id findType(const globaltypes_t& globalTypes, spv::Id lt) const;
std::uint32_t hashType(unsigned typeStart) const;
spirvbin_t& process(instfn_t, idfn_t, unsigned begin = 0, unsigned end = 0);
int processInstruction(unsigned word, instfn_t, idfn_t);
void validate() const;
void mapTypeConst();
void mapFnBodies();
void optLoadStore();
void dceFuncs();
void dceVars();
void dceTypes();
void mapNames();
void foldIds(); // fold IDs to smallest space
void forwardLoadStores(); // load store forwarding (EXPERIMENTAL)
void offsetIds(); // create relative offset IDs
void applyMap(); // remap per local name map
void mapRemainder(); // map any IDs we haven't touched yet
void stripDebug(); // strip all debug info
void stripDeadRefs(); // strips debug info for now-dead references after DCE
void strip(); // remove debug symbols
std::vector<spirword_t> spv; // SPIR words
std::vector<std::string> stripWhiteList;
namemap_t nameMap; // ID names from OpName
// Since we want to also do binary ops, we can't use std::vector<bool>. we could use
// boost::dynamic_bitset, but we're trying to avoid a boost dependency.
typedef std::uint64_t bits_t;
std::vector<bits_t> mapped; // which new IDs have been mapped
static const int mBits = sizeof(bits_t) * 4;
bool isMapped(spv::Id id) const { return id < maxMappedId() && ((mapped[id/mBits] & (1LL<<(id%mBits))) != 0); }
void setMapped(spv::Id id) { resizeMapped(id); mapped[id/mBits] |= (1LL<<(id%mBits)); }
void resizeMapped(spv::Id id) { if (id >= maxMappedId()) mapped.resize(id/mBits+1, 0); }
size_t maxMappedId() const { return mapped.size() * mBits; }
// Add a strip range for a given instruction starting at 'start'
// Note: avoiding brace initializers to please older versions os MSVC.
void stripInst(unsigned start) { stripRange.push_back(range_t(start, start + asWordCount(start))); }
// Function start and end. use unordered_map because we'll have
// many fewer functions than IDs.
std::unordered_map<spv::Id, range_t> fnPos;
// Which functions are called, anywhere in the module, with a call count
std::unordered_map<spv::Id, int> fnCalls;
posmap_t typeConstPos; // word positions that define types & consts (ordered)
posmap_rev_t idPosR; // reverse map from IDs to positions
typesize_map_t idTypeSizeMap; // maps each ID to its type size, if known.
std::vector<spv::Id> idMapL; // ID {M}ap from {L}ocal to {G}lobal IDs
spv::Id entryPoint; // module entry point
spv::Id largestNewId; // biggest new ID we have mapped anything to
// Sections of the binary to strip, given as [begin,end)
std::vector<range_t> stripRange;
// processing options:
std::uint32_t options;
int verbose; // verbosity level
// Error latch: this is set if the error handler is ever executed. It would be better to
// use a try/catch block and throw, but that's not desired for certain environments, so
// this is the alternative.
mutable bool errorLatch;
static errorfn_t errorHandler;
static logfn_t logHandler;
};
} // namespace SPV
#endif // SPIRVREMAPPER_H

112
include/vulkan/glslang/SPIRV/SpvTools.h Normal file
View file

@ -0,0 +1,112 @@
//
// Copyright (C) 2014-2016 LunarG, Inc.
// Copyright (C) 2018 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
// Call into SPIRV-Tools to disassemble, validate, and optimize.
//
#pragma once
#ifndef GLSLANG_SPV_TOOLS_H
#define GLSLANG_SPV_TOOLS_H
#if ENABLE_OPT
#include <vector>
#include <ostream>
#include "spirv-tools/libspirv.h"
#endif
#include "glslang/MachineIndependent/localintermediate.h"
#include "Logger.h"
namespace glslang {
struct SpvOptions {
bool generateDebugInfo {false};
bool stripDebugInfo {false};
bool disableOptimizer {true};
bool optimizeSize {false};
bool disassemble {false};
bool validate {false};
bool emitNonSemanticShaderDebugInfo {false};
bool emitNonSemanticShaderDebugSource{ false };
};
#if ENABLE_OPT
// Translate glslang's view of target versioning to what SPIRV-Tools uses.
spv_target_env MapToSpirvToolsEnv(const SpvVersion& spvVersion, spv::SpvBuildLogger* logger);
// Use the SPIRV-Tools disassembler to print SPIR-V using a SPV_ENV_UNIVERSAL_1_3 environment.
void SpirvToolsDisassemble(std::ostream& out, const std::vector<unsigned int>& spirv);
// Use the SPIRV-Tools disassembler to print SPIR-V with a provided SPIR-V environment.
void SpirvToolsDisassemble(std::ostream& out, const std::vector<unsigned int>& spirv,
spv_target_env requested_context);
// Apply the SPIRV-Tools validator to generated SPIR-V.
void SpirvToolsValidate(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger*, bool prelegalization);
// Apply the SPIRV-Tools optimizer to generated SPIR-V. HLSL SPIR-V is legalized in the process.
void SpirvToolsTransform(const glslang::TIntermediate& intermediate, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger*, const SpvOptions*);
// Apply the SPIRV-Tools EliminateDeadInputComponents pass to generated SPIR-V. Put result in |spirv|.
void SpirvToolsEliminateDeadInputComponents(spv_target_env target_env, std::vector<unsigned int>& spirv,
spv::SpvBuildLogger*);
// Apply the SPIRV-Tools AnalyzeDeadOutputStores pass to generated SPIR-V. Put result in |live_locs|.
// Return true if the result is valid.
bool SpirvToolsAnalyzeDeadOutputStores(spv_target_env target_env, std::vector<unsigned int>& spirv,
std::unordered_set<uint32_t>* live_locs,
std::unordered_set<uint32_t>* live_builtins, spv::SpvBuildLogger*);
// Apply the SPIRV-Tools EliminateDeadOutputStores and AggressiveDeadCodeElimination passes to generated SPIR-V using
// |live_locs|. Put result in |spirv|.
void SpirvToolsEliminateDeadOutputStores(spv_target_env target_env, std::vector<unsigned int>& spirv,
std::unordered_set<uint32_t>* live_locs,
std::unordered_set<uint32_t>* live_builtins, spv::SpvBuildLogger*);
// Apply the SPIRV-Tools optimizer to strip debug info from SPIR-V. This is implicitly done by
// SpirvToolsTransform if spvOptions->stripDebugInfo is set, but can be called separately if
// optimization is disabled.
void SpirvToolsStripDebugInfo(const glslang::TIntermediate& intermediate,
std::vector<unsigned int>& spirv, spv::SpvBuildLogger*);
#endif
} // end namespace glslang
#endif // GLSLANG_SPV_TOOLS_H

2729
include/vulkan/glslang/SPIRV/spirv.hpp Normal file
View file

File diff suppressed because it is too large Load diff

520
include/vulkan/glslang/SPIRV/spvIR.h Normal file
View file

@ -0,0 +1,520 @@
//
// Copyright (C) 2014 LunarG, Inc.
// Copyright (C) 2015-2018 Google, Inc.
//
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions
// are met:
//
// Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
//
// Neither the name of 3Dlabs Inc. Ltd. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
// SPIRV-IR
//
// Simple in-memory representation (IR) of SPIRV. Just for holding
// Each function's CFG of blocks. Has this hierarchy:
// - Module, which is a list of
// - Function, which is a list of
// - Block, which is a list of
// - Instruction
//
#pragma once
#ifndef spvIR_H
#define spvIR_H
#include "spirv.hpp"
#include <algorithm>
#include <cassert>
#include <functional>
#include <iostream>
#include <memory>
#include <vector>
#include <set>
namespace spv {
class Block;
class Function;
class Module;
const Id NoResult = 0;
const Id NoType = 0;
const Decoration NoPrecision = DecorationMax;
#ifdef __GNUC__
# define POTENTIALLY_UNUSED __attribute__((unused))
#else
# define POTENTIALLY_UNUSED
#endif
POTENTIALLY_UNUSED
const MemorySemanticsMask MemorySemanticsAllMemory =
(MemorySemanticsMask)(MemorySemanticsUniformMemoryMask |
MemorySemanticsWorkgroupMemoryMask |
MemorySemanticsAtomicCounterMemoryMask |
MemorySemanticsImageMemoryMask);
struct IdImmediate {
bool isId; // true if word is an Id, false if word is an immediate
unsigned word;
IdImmediate(bool i, unsigned w) : isId(i), word(w) {}
};
//
// SPIR-V IR instruction.
//
class Instruction {
public:
Instruction(Id resultId, Id typeId, Op opCode) : resultId(resultId), typeId(typeId), opCode(opCode), block(nullptr) { }
explicit Instruction(Op opCode) : resultId(NoResult), typeId(NoType), opCode(opCode), block(nullptr) { }
virtual ~Instruction() {}
void addIdOperand(Id id) {
operands.push_back(id);
idOperand.push_back(true);
}
void addImmediateOperand(unsigned int immediate) {
operands.push_back(immediate);
idOperand.push_back(false);
}
void setImmediateOperand(unsigned idx, unsigned int immediate) {
assert(!idOperand[idx]);
operands[idx] = immediate;
}
void addStringOperand(const char* str)
{
unsigned int word = 0;
unsigned int shiftAmount = 0;
char c;
do {
c = *(str++);
word |= ((unsigned int)c) << shiftAmount;
shiftAmount += 8;
if (shiftAmount == 32) {
addImmediateOperand(word);
word = 0;
shiftAmount = 0;
}
} while (c != 0);
// deal with partial last word
if (shiftAmount > 0) {
addImmediateOperand(word);
}
}
bool isIdOperand(int op) const { return idOperand[op]; }
void setBlock(Block* b) { block = b; }
Block* getBlock() const { return block; }
Op getOpCode() const { return opCode; }
int getNumOperands() const
{
assert(operands.size() == idOperand.size());
return (int)operands.size();
}
Id getResultId() const { return resultId; }
Id getTypeId() const { return typeId; }
Id getIdOperand(int op) const {
assert(idOperand[op]);
return operands[op];
}
unsigned int getImmediateOperand(int op) const {
assert(!idOperand[op]);
return operands[op];
}
// Write out the binary form.
void dump(std::vector<unsigned int>& out) const
{
// Compute the wordCount
unsigned int wordCount = 1;
if (typeId)
++wordCount;
if (resultId)
++wordCount;
wordCount += (unsigned int)operands.size();
// Write out the beginning of the instruction
out.push_back(((wordCount) << WordCountShift) | opCode);
if (typeId)
out.push_back(typeId);
if (resultId)
out.push_back(resultId);
// Write out the operands
for (int op = 0; op < (int)operands.size(); ++op)
out.push_back(operands[op]);
}
protected:
Instruction(const Instruction&);
Id resultId;
Id typeId;
Op opCode;
std::vector<Id> operands; // operands, both <id> and immediates (both are unsigned int)
std::vector<bool> idOperand; // true for operands that are <id>, false for immediates
Block* block;
};
//
// SPIR-V IR block.
//
class Block {
public:
Block(Id id, Function& parent);
virtual ~Block()
{
}
Id getId() { return instructions.front()->getResultId(); }
Function& getParent() const { return parent; }
void addInstruction(std::unique_ptr<Instruction> inst);
void addPredecessor(Block* pred) { predecessors.push_back(pred); pred->successors.push_back(this);}
void addLocalVariable(std::unique_ptr<Instruction> inst) { localVariables.push_back(std::move(inst)); }
const std::vector<Block*>& getPredecessors() const { return predecessors; }
const std::vector<Block*>& getSuccessors() const { return successors; }
const std::vector<std::unique_ptr<Instruction> >& getInstructions() const {
return instructions;
}
const std::vector<std::unique_ptr<Instruction> >& getLocalVariables() const { return localVariables; }
void setUnreachable() { unreachable = true; }
bool isUnreachable() const { return unreachable; }
// Returns the block's merge instruction, if one exists (otherwise null).
const Instruction* getMergeInstruction() const {
if (instructions.size() < 2) return nullptr;
const Instruction* nextToLast = (instructions.cend() - 2)->get();
switch (nextToLast->getOpCode()) {
case OpSelectionMerge:
case OpLoopMerge:
return nextToLast;
default:
return nullptr;
}
return nullptr;
}
// Change this block into a canonical dead merge block. Delete instructions
// as necessary. A canonical dead merge block has only an OpLabel and an
// OpUnreachable.
void rewriteAsCanonicalUnreachableMerge() {
assert(localVariables.empty());
// Delete all instructions except for the label.
assert(instructions.size() > 0);
instructions.resize(1);
successors.clear();
addInstruction(std::unique_ptr<Instruction>(new Instruction(OpUnreachable)));
}
// Change this block into a canonical dead continue target branching to the
// given header ID. Delete instructions as necessary. A canonical dead continue
// target has only an OpLabel and an unconditional branch back to the corresponding
// header.
void rewriteAsCanonicalUnreachableContinue(Block* header) {
assert(localVariables.empty());
// Delete all instructions except for the label.
assert(instructions.size() > 0);
instructions.resize(1);
successors.clear();
// Add OpBranch back to the header.
assert(header != nullptr);
Instruction* branch = new Instruction(OpBranch);
branch->addIdOperand(header->getId());
addInstruction(std::unique_ptr<Instruction>(branch));
successors.push_back(header);
}
bool isTerminated() const
{
switch (instructions.back()->getOpCode()) {
case OpBranch:
case OpBranchConditional:
case OpSwitch:
case OpKill:
case OpTerminateInvocation:
case OpReturn:
case OpReturnValue:
case OpUnreachable:
return true;
default:
return false;
}
}
void dump(std::vector<unsigned int>& out) const
{
instructions[0]->dump(out);
for (int i = 0; i < (int)localVariables.size(); ++i)
localVariables[i]->dump(out);
for (int i = 1; i < (int)instructions.size(); ++i)
instructions[i]->dump(out);
}
protected:
Block(const Block&);
Block& operator=(Block&);
// To enforce keeping parent and ownership in sync:
friend Function;
std::vector<std::unique_ptr<Instruction> > instructions;
std::vector<Block*> predecessors, successors;
std::vector<std::unique_ptr<Instruction> > localVariables;
Function& parent;
// track whether this block is known to be uncreachable (not necessarily
// true for all unreachable blocks, but should be set at least
// for the extraneous ones introduced by the builder).
bool unreachable;
};
// The different reasons for reaching a block in the inReadableOrder traversal.
enum ReachReason {
// Reachable from the entry block via transfers of control, i.e. branches.
ReachViaControlFlow = 0,
// A continue target that is not reachable via control flow.
ReachDeadContinue,
// A merge block that is not reachable via control flow.
ReachDeadMerge
};
// Traverses the control-flow graph rooted at root in an order suited for
// readable code generation. Invokes callback at every node in the traversal
// order. The callback arguments are:
// - the block,
// - the reason we reached the block,
// - if the reason was that block is an unreachable continue or unreachable merge block
// then the last parameter is the corresponding header block.
void inReadableOrder(Block* root, std::function<void(Block*, ReachReason, Block* header)> callback);
//
// SPIR-V IR Function.
//
class Function {
public:
Function(Id id, Id resultType, Id functionType, Id firstParam, Module& parent);
virtual ~Function()
{
for (int i = 0; i < (int)parameterInstructions.size(); ++i)
delete parameterInstructions[i];
for (int i = 0; i < (int)blocks.size(); ++i)
delete blocks[i];
}
Id getId() const { return functionInstruction.getResultId(); }
Id getParamId(int p) const { return parameterInstructions[p]->getResultId(); }
Id getParamType(int p) const { return parameterInstructions[p]->getTypeId(); }
void addBlock(Block* block) { blocks.push_back(block); }
void removeBlock(Block* block)
{
auto found = find(blocks.begin(), blocks.end(), block);
assert(found != blocks.end());
blocks.erase(found);
delete block;
}
Module& getParent() const { return parent; }
Block* getEntryBlock() const { return blocks.front(); }
Block* getLastBlock() const { return blocks.back(); }
const std::vector<Block*>& getBlocks() const { return blocks; }
void addLocalVariable(std::unique_ptr<Instruction> inst);
Id getReturnType() const { return functionInstruction.getTypeId(); }
Id getFuncId() const { return functionInstruction.getResultId(); }
void setReturnPrecision(Decoration precision)
{
if (precision == DecorationRelaxedPrecision)
reducedPrecisionReturn = true;
}
Decoration getReturnPrecision() const
{ return reducedPrecisionReturn ? DecorationRelaxedPrecision : NoPrecision; }
void setDebugLineInfo(Id fileName, int line, int column) {
lineInstruction = std::unique_ptr<Instruction>{new Instruction(OpLine)};
lineInstruction->addIdOperand(fileName);
lineInstruction->addImmediateOperand(line);
lineInstruction->addImmediateOperand(column);
}
bool hasDebugLineInfo() const { return lineInstruction != nullptr; }
void setImplicitThis() { implicitThis = true; }
bool hasImplicitThis() const { return implicitThis; }
void addParamPrecision(unsigned param, Decoration precision)
{
if (precision == DecorationRelaxedPrecision)
reducedPrecisionParams.insert(param);
}
Decoration getParamPrecision(unsigned param) const
{
return reducedPrecisionParams.find(param) != reducedPrecisionParams.end() ?
DecorationRelaxedPrecision : NoPrecision;
}
void dump(std::vector<unsigned int>& out) const
{
// OpLine
if (lineInstruction != nullptr) {
lineInstruction->dump(out);
}
// OpFunction
functionInstruction.dump(out);
// OpFunctionParameter
for (int p = 0; p < (int)parameterInstructions.size(); ++p)
parameterInstructions[p]->dump(out);
// Blocks
inReadableOrder(blocks[0], [&out](const Block* b, ReachReason, Block*) { b->dump(out); });
Instruction end(0, 0, OpFunctionEnd);
end.dump(out);
}
protected:
Function(const Function&);
Function& operator=(Function&);
Module& parent;
std::unique_ptr<Instruction> lineInstruction;
Instruction functionInstruction;
std::vector<Instruction*> parameterInstructions;
std::vector<Block*> blocks;
bool implicitThis; // true if this is a member function expecting to be passed a 'this' as the first argument
bool reducedPrecisionReturn;
std::set<int> reducedPrecisionParams; // list of parameter indexes that need a relaxed precision arg
};
//
// SPIR-V IR Module.
//
class Module {
public:
Module() {}
virtual ~Module()
{
// TODO delete things
}
void addFunction(Function *fun) { functions.push_back(fun); }
void mapInstruction(Instruction *instruction)
{
spv::Id resultId = instruction->getResultId();
// map the instruction's result id
if (resultId >= idToInstruction.size())
idToInstruction.resize(resultId + 16);
idToInstruction[resultId] = instruction;
}
Instruction* getInstruction(Id id) const { return idToInstruction[id]; }
const std::vector<Function*>& getFunctions() const { return functions; }
spv::Id getTypeId(Id resultId) const {
return idToInstruction[resultId] == nullptr ? NoType : idToInstruction[resultId]->getTypeId();
}
StorageClass getStorageClass(Id typeId) const
{
assert(idToInstruction[typeId]->getOpCode() == spv::OpTypePointer);
return (StorageClass)idToInstruction[typeId]->getImmediateOperand(0);
}
void dump(std::vector<unsigned int>& out) const
{
for (int f = 0; f < (int)functions.size(); ++f)
functions[f]->dump(out);
}
protected:
Module(const Module&);
std::vector<Function*> functions;
// map from result id to instruction having that result id
std::vector<Instruction*> idToInstruction;
// map from a result id to its type id
};
//
// Implementation (it's here due to circular type definitions).
//
// Add both
// - the OpFunction instruction
// - all the OpFunctionParameter instructions
__inline Function::Function(Id id, Id resultType, Id functionType, Id firstParamId, Module& parent)
: parent(parent), lineInstruction(nullptr),
functionInstruction(id, resultType, OpFunction), implicitThis(false),
reducedPrecisionReturn(false)
{
// OpFunction
functionInstruction.addImmediateOperand(FunctionControlMaskNone);
functionInstruction.addIdOperand(functionType);
parent.mapInstruction(&functionInstruction);
parent.addFunction(this);
// OpFunctionParameter
Instruction* typeInst = parent.getInstruction(functionType);
int numParams = typeInst->getNumOperands() - 1;
for (int p = 0; p < numParams; ++p) {
Instruction* param = new Instruction(firstParamId + p, typeInst->getIdOperand(p + 1), OpFunctionParameter);
parent.mapInstruction(param);
parameterInstructions.push_back(param);
}
}
__inline void Function::addLocalVariable(std::unique_ptr<Instruction> inst)
{
Instruction* raw_instruction = inst.get();
blocks[0]->addLocalVariable(std::move(inst));
parent.mapInstruction(raw_instruction);
}
__inline Block::Block(Id id, Function& parent) : parent(parent), unreachable(false)
{
instructions.push_back(std::unique_ptr<Instruction>(new Instruction(id, NoType, OpLabel)));
instructions.back()->setBlock(this);
parent.getParent().mapInstruction(instructions.back().get());
}
__inline void Block::addInstruction(std::unique_ptr<Instruction> inst)
{
Instruction* raw_instruction = inst.get();
instructions.push_back(std::move(inst));
raw_instruction->setBlock(this);
if (raw_instruction->getResultId())
parent.getParent().mapInstruction(raw_instruction);
}
} // end spv namespace
#endif // spvIR_H

72
include/vulkan/shaderc/env.h Normal file
View file

@ -0,0 +1,72 @@
// Copyright 2018 The Shaderc Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef SHADERC_ENV_H_
#define SHADERC_ENV_H_
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef enum {
shaderc_target_env_vulkan, // SPIR-V under Vulkan semantics
shaderc_target_env_opengl, // SPIR-V under OpenGL semantics
// NOTE: SPIR-V code generation is not supported for shaders under OpenGL
// compatibility profile.
shaderc_target_env_opengl_compat, // SPIR-V under OpenGL semantics,
// including compatibility profile
// functions
shaderc_target_env_webgpu, // Deprecated, SPIR-V under WebGPU
// semantics
shaderc_target_env_default = shaderc_target_env_vulkan
} shaderc_target_env;
typedef enum {
// For Vulkan, use Vulkan's mapping of version numbers to integers.
// See vulkan.h
shaderc_env_version_vulkan_1_0 = ((1u << 22)),
shaderc_env_version_vulkan_1_1 = ((1u << 22) | (1 << 12)),
shaderc_env_version_vulkan_1_2 = ((1u << 22) | (2 << 12)),
shaderc_env_version_vulkan_1_3 = ((1u << 22) | (3 << 12)),
// For OpenGL, use the number from #version in shaders.
// TODO(dneto): Currently no difference between OpenGL 4.5 and 4.6.
// See glslang/Standalone/Standalone.cpp
// TODO(dneto): Glslang doesn't accept a OpenGL client version of 460.
shaderc_env_version_opengl_4_5 = 450,
shaderc_env_version_webgpu, // Deprecated, WebGPU env never defined versions
} shaderc_env_version;
// The known versions of SPIR-V.
typedef enum {
// Use the values used for word 1 of a SPIR-V binary:
// - bits 24 to 31: zero
// - bits 16 to 23: major version number
// - bits 8 to 15: minor version number
// - bits 0 to 7: zero
shaderc_spirv_version_1_0 = 0x010000u,
shaderc_spirv_version_1_1 = 0x010100u,
shaderc_spirv_version_1_2 = 0x010200u,
shaderc_spirv_version_1_3 = 0x010300u,
shaderc_spirv_version_1_4 = 0x010400u,
shaderc_spirv_version_1_5 = 0x010500u,
shaderc_spirv_version_1_6 = 0x010600u
} shaderc_spirv_version;
#ifdef __cplusplus
}
#endif // __cplusplus
#endif // SHADERC_ENV_H_

613
include/vulkan/shaderc/shaderc.h Normal file
View file

@ -0,0 +1,613 @@
// Copyright 2015 The Shaderc Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef SHADERC_SHADERC_H_
#define SHADERC_SHADERC_H_
#ifdef __cplusplus
extern "C" {
#endif
#include <stdbool.h>
#include <stddef.h>
#include <stdint.h>
#include "shaderc/env.h"
#include "shaderc/status.h"
#include "shaderc/visibility.h"
// Source language kind.
typedef enum {
shaderc_source_language_glsl,
shaderc_source_language_hlsl,
} shaderc_source_language;
typedef enum {
// Forced shader kinds. These shader kinds force the compiler to compile the
// source code as the specified kind of shader.
shaderc_vertex_shader,
shaderc_fragment_shader,
shaderc_compute_shader,
shaderc_geometry_shader,
shaderc_tess_control_shader,
shaderc_tess_evaluation_shader,
shaderc_glsl_vertex_shader = shaderc_vertex_shader,
shaderc_glsl_fragment_shader = shaderc_fragment_shader,
shaderc_glsl_compute_shader = shaderc_compute_shader,
shaderc_glsl_geometry_shader = shaderc_geometry_shader,
shaderc_glsl_tess_control_shader = shaderc_tess_control_shader,
shaderc_glsl_tess_evaluation_shader = shaderc_tess_evaluation_shader,
// Deduce the shader kind from #pragma annotation in the source code. Compiler
// will emit error if #pragma annotation is not found.
shaderc_glsl_infer_from_source,
// Default shader kinds. Compiler will fall back to compile the source code as
// the specified kind of shader when #pragma annotation is not found in the
// source code.
shaderc_glsl_default_vertex_shader,
shaderc_glsl_default_fragment_shader,
shaderc_glsl_default_compute_shader,
shaderc_glsl_default_geometry_shader,
shaderc_glsl_default_tess_control_shader,
shaderc_glsl_default_tess_evaluation_shader,
shaderc_spirv_assembly,
shaderc_raygen_shader,
shaderc_anyhit_shader,
shaderc_closesthit_shader,
shaderc_miss_shader,
shaderc_intersection_shader,
shaderc_callable_shader,
shaderc_glsl_raygen_shader = shaderc_raygen_shader,
shaderc_glsl_anyhit_shader = shaderc_anyhit_shader,
shaderc_glsl_closesthit_shader = shaderc_closesthit_shader,
shaderc_glsl_miss_shader = shaderc_miss_shader,
shaderc_glsl_intersection_shader = shaderc_intersection_shader,
shaderc_glsl_callable_shader = shaderc_callable_shader,
shaderc_glsl_default_raygen_shader,
shaderc_glsl_default_anyhit_shader,
shaderc_glsl_default_closesthit_shader,
shaderc_glsl_default_miss_shader,
shaderc_glsl_default_intersection_shader,
shaderc_glsl_default_callable_shader,
shaderc_task_shader,
shaderc_mesh_shader,
shaderc_glsl_task_shader = shaderc_task_shader,
shaderc_glsl_mesh_shader = shaderc_mesh_shader,
shaderc_glsl_default_task_shader,
shaderc_glsl_default_mesh_shader,
} shaderc_shader_kind;
typedef enum {
shaderc_profile_none, // Used if and only if GLSL version did not specify
// profiles.
shaderc_profile_core,
shaderc_profile_compatibility, // Disabled. This generates an error
shaderc_profile_es,
} shaderc_profile;
// Optimization level.
typedef enum {
shaderc_optimization_level_zero, // no optimization
shaderc_optimization_level_size, // optimize towards reducing code size
shaderc_optimization_level_performance, // optimize towards performance
} shaderc_optimization_level;
// Resource limits.
typedef enum {
shaderc_limit_max_lights,
shaderc_limit_max_clip_planes,
shaderc_limit_max_texture_units,
shaderc_limit_max_texture_coords,
shaderc_limit_max_vertex_attribs,
shaderc_limit_max_vertex_uniform_components,
shaderc_limit_max_varying_floats,
shaderc_limit_max_vertex_texture_image_units,
shaderc_limit_max_combined_texture_image_units,
shaderc_limit_max_texture_image_units,
shaderc_limit_max_fragment_uniform_components,
shaderc_limit_max_draw_buffers,
shaderc_limit_max_vertex_uniform_vectors,
shaderc_limit_max_varying_vectors,
shaderc_limit_max_fragment_uniform_vectors,
shaderc_limit_max_vertex_output_vectors,
shaderc_limit_max_fragment_input_vectors,
shaderc_limit_min_program_texel_offset,
shaderc_limit_max_program_texel_offset,
shaderc_limit_max_clip_distances,
shaderc_limit_max_compute_work_group_count_x,
shaderc_limit_max_compute_work_group_count_y,
shaderc_limit_max_compute_work_group_count_z,
shaderc_limit_max_compute_work_group_size_x,
shaderc_limit_max_compute_work_group_size_y,
shaderc_limit_max_compute_work_group_size_z,
shaderc_limit_max_compute_uniform_components,
shaderc_limit_max_compute_texture_image_units,
shaderc_limit_max_compute_image_uniforms,
shaderc_limit_max_compute_atomic_counters,
shaderc_limit_max_compute_atomic_counter_buffers,
shaderc_limit_max_varying_components,
shaderc_limit_max_vertex_output_components,
shaderc_limit_max_geometry_input_components,
shaderc_limit_max_geometry_output_components,
shaderc_limit_max_fragment_input_components,
shaderc_limit_max_image_units,
shaderc_limit_max_combined_image_units_and_fragment_outputs,
shaderc_limit_max_combined_shader_output_resources,
shaderc_limit_max_image_samples,
shaderc_limit_max_vertex_image_uniforms,
shaderc_limit_max_tess_control_image_uniforms,
shaderc_limit_max_tess_evaluation_image_uniforms,
shaderc_limit_max_geometry_image_uniforms,
shaderc_limit_max_fragment_image_uniforms,
shaderc_limit_max_combined_image_uniforms,
shaderc_limit_max_geometry_texture_image_units,
shaderc_limit_max_geometry_output_vertices,
shaderc_limit_max_geometry_total_output_components,
shaderc_limit_max_geometry_uniform_components,
shaderc_limit_max_geometry_varying_components,
shaderc_limit_max_tess_control_input_components,
shaderc_limit_max_tess_control_output_components,
shaderc_limit_max_tess_control_texture_image_units,
shaderc_limit_max_tess_control_uniform_components,
shaderc_limit_max_tess_control_total_output_components,
shaderc_limit_max_tess_evaluation_input_components,
shaderc_limit_max_tess_evaluation_output_components,
shaderc_limit_max_tess_evaluation_texture_image_units,
shaderc_limit_max_tess_evaluation_uniform_components,
shaderc_limit_max_tess_patch_components,
shaderc_limit_max_patch_vertices,
shaderc_limit_max_tess_gen_level,
shaderc_limit_max_viewports,
shaderc_limit_max_vertex_atomic_counters,
shaderc_limit_max_tess_control_atomic_counters,
shaderc_limit_max_tess_evaluation_atomic_counters,
shaderc_limit_max_geometry_atomic_counters,
shaderc_limit_max_fragment_atomic_counters,
shaderc_limit_max_combined_atomic_counters,
shaderc_limit_max_atomic_counter_bindings,
shaderc_limit_max_vertex_atomic_counter_buffers,
shaderc_limit_max_tess_control_atomic_counter_buffers,
shaderc_limit_max_tess_evaluation_atomic_counter_buffers,
shaderc_limit_max_geometry_atomic_counter_buffers,
shaderc_limit_max_fragment_atomic_counter_buffers,
shaderc_limit_max_combined_atomic_counter_buffers,
shaderc_limit_max_atomic_counter_buffer_size,
shaderc_limit_max_transform_feedback_buffers,
shaderc_limit_max_transform_feedback_interleaved_components,
shaderc_limit_max_cull_distances,
shaderc_limit_max_combined_clip_and_cull_distances,
shaderc_limit_max_samples,
shaderc_limit_max_mesh_output_vertices_nv,
shaderc_limit_max_mesh_output_primitives_nv,
shaderc_limit_max_mesh_work_group_size_x_nv,
shaderc_limit_max_mesh_work_group_size_y_nv,
shaderc_limit_max_mesh_work_group_size_z_nv,
shaderc_limit_max_task_work_group_size_x_nv,
shaderc_limit_max_task_work_group_size_y_nv,
shaderc_limit_max_task_work_group_size_z_nv,
shaderc_limit_max_mesh_view_count_nv,
shaderc_limit_max_mesh_output_vertices_ext,
shaderc_limit_max_mesh_output_primitives_ext,
shaderc_limit_max_mesh_work_group_size_x_ext,
shaderc_limit_max_mesh_work_group_size_y_ext,
shaderc_limit_max_mesh_work_group_size_z_ext,
shaderc_limit_max_task_work_group_size_x_ext,
shaderc_limit_max_task_work_group_size_y_ext,
shaderc_limit_max_task_work_group_size_z_ext,
shaderc_limit_max_mesh_view_count_ext,
shaderc_limit_max_dual_source_draw_buffers_ext,
} shaderc_limit;
// Uniform resource kinds.
// In Vulkan, uniform resources are bound to the pipeline via descriptors
// with numbered bindings and sets.
typedef enum {
// Image and image buffer.
shaderc_uniform_kind_image,
// Pure sampler.
shaderc_uniform_kind_sampler,
// Sampled texture in GLSL, and Shader Resource View in HLSL.
shaderc_uniform_kind_texture,
// Uniform Buffer Object (UBO) in GLSL. Cbuffer in HLSL.
shaderc_uniform_kind_buffer,
// Shader Storage Buffer Object (SSBO) in GLSL.
shaderc_uniform_kind_storage_buffer,
// Unordered Access View, in HLSL. (Writable storage image or storage
// buffer.)
shaderc_uniform_kind_unordered_access_view,
} shaderc_uniform_kind;
// Usage examples:
//
// Aggressively release compiler resources, but spend time in initialization
// for each new use.
// shaderc_compiler_t compiler = shaderc_compiler_initialize();
// shaderc_compilation_result_t result = shaderc_compile_into_spv(
// compiler, "#version 450\nvoid main() {}", 27,
// shaderc_glsl_vertex_shader, "main.vert", "main", nullptr);
// // Do stuff with compilation results.
// shaderc_result_release(result);
// shaderc_compiler_release(compiler);
//
// Keep the compiler object around for a long time, but pay for extra space
// occupied.
// shaderc_compiler_t compiler = shaderc_compiler_initialize();
// // On the same, other or multiple simultaneous threads.
// shaderc_compilation_result_t result = shaderc_compile_into_spv(
// compiler, "#version 450\nvoid main() {}", 27,
// shaderc_glsl_vertex_shader, "main.vert", "main", nullptr);
// // Do stuff with compilation results.
// shaderc_result_release(result);
// // Once no more compilations are to happen.
// shaderc_compiler_release(compiler);
// An opaque handle to an object that manages all compiler state.
typedef struct shaderc_compiler* shaderc_compiler_t;
// Returns a shaderc_compiler_t that can be used to compile modules.
// A return of NULL indicates that there was an error initializing the compiler.
// Any function operating on shaderc_compiler_t must offer the basic
// thread-safety guarantee.
// [http://herbsutter.com/2014/01/13/gotw-95-solution-thread-safety-and-synchronization/]
// That is: concurrent invocation of these functions on DIFFERENT objects needs
// no synchronization; concurrent invocation of these functions on the SAME
// object requires synchronization IF AND ONLY IF some of them take a non-const
// argument.
SHADERC_EXPORT shaderc_compiler_t shaderc_compiler_initialize(void);
// Releases the resources held by the shaderc_compiler_t.
// After this call it is invalid to make any future calls to functions
// involving this shaderc_compiler_t.
SHADERC_EXPORT void shaderc_compiler_release(shaderc_compiler_t);
// An opaque handle to an object that manages options to a single compilation
// result.
typedef struct shaderc_compile_options* shaderc_compile_options_t;
// Returns a default-initialized shaderc_compile_options_t that can be used
// to modify the functionality of a compiled module.
// A return of NULL indicates that there was an error initializing the options.
// Any function operating on shaderc_compile_options_t must offer the
// basic thread-safety guarantee.
SHADERC_EXPORT shaderc_compile_options_t
shaderc_compile_options_initialize(void);
// Returns a copy of the given shaderc_compile_options_t.
// If NULL is passed as the parameter the call is the same as
// shaderc_compile_options_init.
SHADERC_EXPORT shaderc_compile_options_t shaderc_compile_options_clone(
const shaderc_compile_options_t options);
// Releases the compilation options. It is invalid to use the given
// shaderc_compile_options_t object in any future calls. It is safe to pass
// NULL to this function, and doing such will have no effect.
SHADERC_EXPORT void shaderc_compile_options_release(
shaderc_compile_options_t options);
// Adds a predefined macro to the compilation options. This has the same
// effect as passing -Dname=value to the command-line compiler. If value
// is NULL, it has the same effect as passing -Dname to the command-line
// compiler. If a macro definition with the same name has previously been
// added, the value is replaced with the new value. The macro name and
// value are passed in with char pointers, which point to their data, and
// the lengths of their data. The strings that the name and value pointers
// point to must remain valid for the duration of the call, but can be
// modified or deleted after this function has returned. In case of adding
// a valueless macro, the value argument should be a null pointer or the
// value_length should be 0u.
SHADERC_EXPORT void shaderc_compile_options_add_macro_definition(
shaderc_compile_options_t options, const char* name, size_t name_length,
const char* value, size_t value_length);
// Sets the source language. The default is GLSL.
SHADERC_EXPORT void shaderc_compile_options_set_source_language(
shaderc_compile_options_t options, shaderc_source_language lang);
// Sets the compiler mode to generate debug information in the output.
SHADERC_EXPORT void shaderc_compile_options_set_generate_debug_info(
shaderc_compile_options_t options);
// Sets the compiler optimization level to the given level. Only the last one
// takes effect if multiple calls of this function exist.
SHADERC_EXPORT void shaderc_compile_options_set_optimization_level(
shaderc_compile_options_t options, shaderc_optimization_level level);
// Forces the GLSL language version and profile to a given pair. The version
// number is the same as would appear in the #version annotation in the source.
// Version and profile specified here overrides the #version annotation in the
// source. Use profile: 'shaderc_profile_none' for GLSL versions that do not
// define profiles, e.g. versions below 150.
SHADERC_EXPORT void shaderc_compile_options_set_forced_version_profile(
shaderc_compile_options_t options, int version, shaderc_profile profile);
// Source text inclusion via #include is supported with a pair of callbacks
// to an "includer" on the client side. The first callback processes an
// inclusion request, and returns an include result. The includer owns
// the contents of the result, and those contents must remain valid until the
// second callback is invoked to release the result. Both callbacks take a
// user_data argument to specify the client context.
// To return an error, set the source_name to an empty string and put your
// error message in content.
// An include result.
typedef struct shaderc_include_result {
// The name of the source file. The name should be fully resolved
// in the sense that it should be a unique name in the context of the
// includer. For example, if the includer maps source names to files in
// a filesystem, then this name should be the absolute path of the file.
// For a failed inclusion, this string is empty.
const char* source_name;
size_t source_name_length;
// The text contents of the source file in the normal case.
// For a failed inclusion, this contains the error message.
const char* content;
size_t content_length;
// User data to be passed along with this request.
void* user_data;
} shaderc_include_result;
// The kinds of include requests.
enum shaderc_include_type {
shaderc_include_type_relative, // E.g. #include "source"
shaderc_include_type_standard // E.g. #include <source>
};
// An includer callback type for mapping an #include request to an include
// result. The user_data parameter specifies the client context. The
// requested_source parameter specifies the name of the source being requested.
// The type parameter specifies the kind of inclusion request being made.
// The requesting_source parameter specifies the name of the source containing
// the #include request. The includer owns the result object and its contents,
// and both must remain valid until the release callback is called on the result
// object.
typedef shaderc_include_result* (*shaderc_include_resolve_fn)(
void* user_data, const char* requested_source, int type,
const char* requesting_source, size_t include_depth);
// An includer callback type for destroying an include result.
typedef void (*shaderc_include_result_release_fn)(
void* user_data, shaderc_include_result* include_result);
// Sets includer callback functions.
SHADERC_EXPORT void shaderc_compile_options_set_include_callbacks(
shaderc_compile_options_t options, shaderc_include_resolve_fn resolver,
shaderc_include_result_release_fn result_releaser, void* user_data);
// Sets the compiler mode to suppress warnings, overriding warnings-as-errors
// mode. When both suppress-warnings and warnings-as-errors modes are
// turned on, warning messages will be inhibited, and will not be emitted
// as error messages.
SHADERC_EXPORT void shaderc_compile_options_set_suppress_warnings(
shaderc_compile_options_t options);
// Sets the target shader environment, affecting which warnings or errors will
// be issued. The version will be for distinguishing between different versions
// of the target environment. The version value should be either 0 or
// a value listed in shaderc_env_version. The 0 value maps to Vulkan 1.0 if
// |target| is Vulkan, and it maps to OpenGL 4.5 if |target| is OpenGL.
SHADERC_EXPORT void shaderc_compile_options_set_target_env(
shaderc_compile_options_t options,
shaderc_target_env target,
uint32_t version);
// Sets the target SPIR-V version. The generated module will use this version
// of SPIR-V. Each target environment determines what versions of SPIR-V
// it can consume. Defaults to the highest version of SPIR-V 1.0 which is
// required to be supported by the target environment. E.g. Default to SPIR-V
// 1.0 for Vulkan 1.0 and SPIR-V 1.3 for Vulkan 1.1.
SHADERC_EXPORT void shaderc_compile_options_set_target_spirv(
shaderc_compile_options_t options, shaderc_spirv_version version);
// Sets the compiler mode to treat all warnings as errors. Note the
// suppress-warnings mode overrides this option, i.e. if both
// warning-as-errors and suppress-warnings modes are set, warnings will not
// be emitted as error messages.
SHADERC_EXPORT void shaderc_compile_options_set_warnings_as_errors(
shaderc_compile_options_t options);
// Sets a resource limit.
SHADERC_EXPORT void shaderc_compile_options_set_limit(
shaderc_compile_options_t options, shaderc_limit limit, int value);
// Sets whether the compiler should automatically assign bindings to uniforms
// that aren't already explicitly bound in the shader source.
SHADERC_EXPORT void shaderc_compile_options_set_auto_bind_uniforms(
shaderc_compile_options_t options, bool auto_bind);
// Sets whether the compiler should automatically remove sampler variables
// and convert image variables to combined image-sampler variables.
SHADERC_EXPORT void shaderc_compile_options_set_auto_combined_image_sampler(
shaderc_compile_options_t options, bool upgrade);
// Sets whether the compiler should use HLSL IO mapping rules for bindings.
// Defaults to false.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_io_mapping(
shaderc_compile_options_t options, bool hlsl_iomap);
// Sets whether the compiler should determine block member offsets using HLSL
// packing rules instead of standard GLSL rules. Defaults to false. Only
// affects GLSL compilation. HLSL rules are always used when compiling HLSL.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_offsets(
shaderc_compile_options_t options, bool hlsl_offsets);
// Sets the base binding number used for for a uniform resource type when
// automatically assigning bindings. For GLSL compilation, sets the lowest
// automatically assigned number. For HLSL compilation, the regsiter number
// assigned to the resource is added to this specified base.
SHADERC_EXPORT void shaderc_compile_options_set_binding_base(
shaderc_compile_options_t options,
shaderc_uniform_kind kind,
uint32_t base);
// Like shaderc_compile_options_set_binding_base, but only takes effect when
// compiling a given shader stage. The stage is assumed to be one of vertex,
// fragment, tessellation evaluation, tesselation control, geometry, or compute.
SHADERC_EXPORT void shaderc_compile_options_set_binding_base_for_stage(
shaderc_compile_options_t options, shaderc_shader_kind shader_kind,
shaderc_uniform_kind kind, uint32_t base);
// Sets whether the compiler should preserve all bindings, even when those
// bindings are not used.
SHADERC_EXPORT void shaderc_compile_options_set_preserve_bindings(
shaderc_compile_options_t options, bool preserve_bindings);
// Sets whether the compiler should automatically assign locations to
// uniform variables that don't have explicit locations in the shader source.
SHADERC_EXPORT void shaderc_compile_options_set_auto_map_locations(
shaderc_compile_options_t options, bool auto_map);
// Sets a descriptor set and binding for an HLSL register in the given stage.
// This method keeps a copy of the string data.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_register_set_and_binding_for_stage(
shaderc_compile_options_t options, shaderc_shader_kind shader_kind,
const char* reg, const char* set, const char* binding);
// Like shaderc_compile_options_set_hlsl_register_set_and_binding_for_stage,
// but affects all shader stages.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_register_set_and_binding(
shaderc_compile_options_t options, const char* reg, const char* set,
const char* binding);
// Sets whether the compiler should enable extension
// SPV_GOOGLE_hlsl_functionality1.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_functionality1(
shaderc_compile_options_t options, bool enable);
// Sets whether 16-bit types are supported in HLSL or not.
SHADERC_EXPORT void shaderc_compile_options_set_hlsl_16bit_types(
shaderc_compile_options_t options, bool enable);
// Sets whether the compiler should invert position.Y output in vertex shader.
SHADERC_EXPORT void shaderc_compile_options_set_invert_y(
shaderc_compile_options_t options, bool enable);
// Sets whether the compiler generates code for max and min builtins which,
// if given a NaN operand, will return the other operand. Similarly, the clamp
// builtin will favour the non-NaN operands, as if clamp were implemented
// as a composition of max and min.
SHADERC_EXPORT void shaderc_compile_options_set_nan_clamp(
shaderc_compile_options_t options, bool enable);
// An opaque handle to the results of a call to any shaderc_compile_into_*()
// function.
typedef struct shaderc_compilation_result* shaderc_compilation_result_t;
// Takes a GLSL source string and the associated shader kind, input file
// name, compiles it according to the given additional_options. If the shader
// kind is not set to a specified kind, but shaderc_glslc_infer_from_source,
// the compiler will try to deduce the shader kind from the source
// string and a failure in deducing will generate an error. Currently only
// #pragma annotation is supported. If the shader kind is set to one of the
// default shader kinds, the compiler will fall back to the default shader
// kind in case it failed to deduce the shader kind from source string.
// The input_file_name is a null-termintated string. It is used as a tag to
// identify the source string in cases like emitting error messages. It
// doesn't have to be a 'file name'.
// The source string will be compiled into SPIR-V binary and a
// shaderc_compilation_result will be returned to hold the results.
// The entry_point_name null-terminated string defines the name of the entry
// point to associate with this GLSL source. If the additional_options
// parameter is not null, then the compilation is modified by any options
// present. May be safely called from multiple threads without explicit
// synchronization. If there was failure in allocating the compiler object,
// null will be returned.
SHADERC_EXPORT shaderc_compilation_result_t shaderc_compile_into_spv(
const shaderc_compiler_t compiler, const char* source_text,
size_t source_text_size, shaderc_shader_kind shader_kind,
const char* input_file_name, const char* entry_point_name,
const shaderc_compile_options_t additional_options);
// Like shaderc_compile_into_spv, but the result contains SPIR-V assembly text
// instead of a SPIR-V binary module. The SPIR-V assembly syntax is as defined
// by the SPIRV-Tools open source project.
SHADERC_EXPORT shaderc_compilation_result_t shaderc_compile_into_spv_assembly(
const shaderc_compiler_t compiler, const char* source_text,
size_t source_text_size, shaderc_shader_kind shader_kind,
const char* input_file_name, const char* entry_point_name,
const shaderc_compile_options_t additional_options);
// Like shaderc_compile_into_spv, but the result contains preprocessed source
// code instead of a SPIR-V binary module
SHADERC_EXPORT shaderc_compilation_result_t shaderc_compile_into_preprocessed_text(
const shaderc_compiler_t compiler, const char* source_text,
size_t source_text_size, shaderc_shader_kind shader_kind,
const char* input_file_name, const char* entry_point_name,
const shaderc_compile_options_t additional_options);
// Takes an assembly string of the format defined in the SPIRV-Tools project
// (https://github.com/KhronosGroup/SPIRV-Tools/blob/master/syntax.md),
// assembles it into SPIR-V binary and a shaderc_compilation_result will be
// returned to hold the results.
// The assembling will pick options suitable for assembling specified in the
// additional_options parameter.
// May be safely called from multiple threads without explicit synchronization.
// If there was failure in allocating the compiler object, null will be
// returned.
SHADERC_EXPORT shaderc_compilation_result_t shaderc_assemble_into_spv(
const shaderc_compiler_t compiler, const char* source_assembly,
size_t source_assembly_size,
const shaderc_compile_options_t additional_options);
// The following functions, operating on shaderc_compilation_result_t objects,
// offer only the basic thread-safety guarantee.
// Releases the resources held by the result object. It is invalid to use the
// result object for any further operations.
SHADERC_EXPORT void shaderc_result_release(shaderc_compilation_result_t result);
// Returns the number of bytes of the compilation output data in a result
// object.
SHADERC_EXPORT size_t shaderc_result_get_length(const shaderc_compilation_result_t result);
// Returns the number of warnings generated during the compilation.
SHADERC_EXPORT size_t shaderc_result_get_num_warnings(
const shaderc_compilation_result_t result);
// Returns the number of errors generated during the compilation.
SHADERC_EXPORT size_t shaderc_result_get_num_errors(const shaderc_compilation_result_t result);
// Returns the compilation status, indicating whether the compilation succeeded,
// or failed due to some reasons, like invalid shader stage or compilation
// errors.
SHADERC_EXPORT shaderc_compilation_status shaderc_result_get_compilation_status(
const shaderc_compilation_result_t);
// Returns a pointer to the start of the compilation output data bytes, either
// SPIR-V binary or char string. When the source string is compiled into SPIR-V
// binary, this is guaranteed to be castable to a uint32_t*. If the result
// contains assembly text or preprocessed source text, the pointer will point to
// the resulting array of characters.
SHADERC_EXPORT const char* shaderc_result_get_bytes(const shaderc_compilation_result_t result);
// Returns a null-terminated string that contains any error messages generated
// during the compilation.
SHADERC_EXPORT const char* shaderc_result_get_error_message(
const shaderc_compilation_result_t result);
// Provides the version & revision of the SPIR-V which will be produced
SHADERC_EXPORT void shaderc_get_spv_version(unsigned int* version, unsigned int* revision);
// Parses the version and profile from a given null-terminated string
// containing both version and profile, like: '450core'. Returns false if
// the string can not be parsed. Returns true when the parsing succeeds. The
// parsed version and profile are returned through arguments.
SHADERC_EXPORT bool shaderc_parse_version_profile(const char* str, int* version,
shaderc_profile* profile);
#ifdef __cplusplus
}
#endif // __cplusplus
#endif // SHADERC_SHADERC_H_

613
include/vulkan/shaderc/shaderc.hpp Normal file
View file

@ -0,0 +1,613 @@
// Copyright 2015 The Shaderc Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef SHADERC_SHADERC_HPP_
#define SHADERC_SHADERC_HPP_
#include <memory>
#include <string>
#include <vector>
#include "shaderc.h"
namespace shaderc {
// A CompilationResult contains the compiler output, compilation status,
// and messages.
//
// The compiler output is stored as an array of elements and accessed
// via random access iterators provided by cbegin() and cend(). The iterators
// are contiguous in the sense of "Contiguous Iterators: A Refinement of
// Random Access Iterators", Nevin Liber, C++ Library Evolution Working
// Group Working Paper N3884.
// http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3884.pdf
//
// Methods begin() and end() are also provided to enable range-based for.
// They are synonyms to cbegin() and cend(), respectively.
template <typename OutputElementType>
class CompilationResult {
public:
typedef OutputElementType element_type;
// The type used to describe the begin and end iterators on the
// compiler output.
typedef const OutputElementType* const_iterator;
// Upon creation, the CompilationResult takes ownership of the
// shaderc_compilation_result instance. During destruction of the
// CompilationResult, the shaderc_compilation_result will be released.
explicit CompilationResult(shaderc_compilation_result_t compilation_result)
: compilation_result_(compilation_result) {}
CompilationResult() : compilation_result_(nullptr) {}
~CompilationResult() { shaderc_result_release(compilation_result_); }
CompilationResult(CompilationResult&& other) : compilation_result_(nullptr) {
*this = std::move(other);
}
CompilationResult& operator=(CompilationResult&& other) {
if (compilation_result_) {
shaderc_result_release(compilation_result_);
}
compilation_result_ = other.compilation_result_;
other.compilation_result_ = nullptr;
return *this;
}
// Returns any error message found during compilation.
std::string GetErrorMessage() const {
if (!compilation_result_) {
return "";
}
return shaderc_result_get_error_message(compilation_result_);
}
// Returns the compilation status, indicating whether the compilation
// succeeded, or failed due to some reasons, like invalid shader stage or
// compilation errors.
shaderc_compilation_status GetCompilationStatus() const {
if (!compilation_result_) {
return shaderc_compilation_status_null_result_object;
}
return shaderc_result_get_compilation_status(compilation_result_);
}
// Returns a random access (contiguous) iterator pointing to the start
// of the compilation output. It is valid for the lifetime of this object.
// If there is no compilation result, then returns nullptr.
const_iterator cbegin() const {
if (!compilation_result_) return nullptr;
return reinterpret_cast<const_iterator>(
shaderc_result_get_bytes(compilation_result_));
}
// Returns a random access (contiguous) iterator pointing to the end of
// the compilation output. It is valid for the lifetime of this object.
// If there is no compilation result, then returns nullptr.
const_iterator cend() const {
if (!compilation_result_) return nullptr;
return cbegin() +
shaderc_result_get_length(compilation_result_) /
sizeof(OutputElementType);
}
// Returns the same iterator as cbegin().
const_iterator begin() const { return cbegin(); }
// Returns the same iterator as cend().
const_iterator end() const { return cend(); }
// Returns the number of warnings generated during the compilation.
size_t GetNumWarnings() const {
if (!compilation_result_) {
return 0;
}
return shaderc_result_get_num_warnings(compilation_result_);
}
// Returns the number of errors generated during the compilation.
size_t GetNumErrors() const {
if (!compilation_result_) {
return 0;
}
return shaderc_result_get_num_errors(compilation_result_);
}
private:
CompilationResult(const CompilationResult& other) = delete;
CompilationResult& operator=(const CompilationResult& other) = delete;
shaderc_compilation_result_t compilation_result_;
};
// A compilation result for a SPIR-V binary module, which is an array
// of uint32_t words.
using SpvCompilationResult = CompilationResult<uint32_t>;
// A compilation result in SPIR-V assembly syntax.
using AssemblyCompilationResult = CompilationResult<char>;
// Preprocessed source text.
using PreprocessedSourceCompilationResult = CompilationResult<char>;
// Contains any options that can have default values for a compilation.
class CompileOptions {
public:
CompileOptions() { options_ = shaderc_compile_options_initialize(); }
~CompileOptions() { shaderc_compile_options_release(options_); }
CompileOptions(const CompileOptions& other) {
options_ = shaderc_compile_options_clone(other.options_);
}
CompileOptions(CompileOptions&& other) {
options_ = other.options_;
other.options_ = nullptr;
}
// Adds a predefined macro to the compilation options. It behaves the same as
// shaderc_compile_options_add_macro_definition in shaderc.h.
void AddMacroDefinition(const char* name, size_t name_length,
const char* value, size_t value_length) {
shaderc_compile_options_add_macro_definition(options_, name, name_length,
value, value_length);
}
// Adds a valueless predefined macro to the compilation options.
void AddMacroDefinition(const std::string& name) {
AddMacroDefinition(name.c_str(), name.size(), nullptr, 0u);
}
// Adds a predefined macro to the compilation options.
void AddMacroDefinition(const std::string& name, const std::string& value) {
AddMacroDefinition(name.c_str(), name.size(), value.c_str(), value.size());
}
// Sets the compiler mode to generate debug information in the output.
void SetGenerateDebugInfo() {
shaderc_compile_options_set_generate_debug_info(options_);
}
// Sets the compiler optimization level to the given level. Only the last one
// takes effect if multiple calls of this function exist.
void SetOptimizationLevel(shaderc_optimization_level level) {
shaderc_compile_options_set_optimization_level(options_, level);
}
// A C++ version of the libshaderc includer interface.
class IncluderInterface {
public:
// Handles shaderc_include_resolver_fn callbacks.
virtual shaderc_include_result* GetInclude(const char* requested_source,
shaderc_include_type type,
const char* requesting_source,
size_t include_depth) = 0;
// Handles shaderc_include_result_release_fn callbacks.
virtual void ReleaseInclude(shaderc_include_result* data) = 0;
virtual ~IncluderInterface() = default;
};
// Sets the includer instance for libshaderc to call during compilation, as
// described in shaderc_compile_options_set_include_callbacks(). Callbacks
// are routed to this includer's methods.
void SetIncluder(std::unique_ptr<IncluderInterface>&& includer) {
includer_ = std::move(includer);
shaderc_compile_options_set_include_callbacks(
options_,
[](void* user_data, const char* requested_source, int type,
const char* requesting_source, size_t include_depth) {
auto* sub_includer = static_cast<IncluderInterface*>(user_data);
return sub_includer->GetInclude(
requested_source, static_cast<shaderc_include_type>(type),
requesting_source, include_depth);
},
[](void* user_data, shaderc_include_result* include_result) {
auto* sub_includer = static_cast<IncluderInterface*>(user_data);
return sub_includer->ReleaseInclude(include_result);
},
includer_.get());
}
// Forces the GLSL language version and profile to a given pair. The version
// number is the same as would appear in the #version annotation in the
// source. Version and profile specified here overrides the #version
// annotation in the source. Use profile: 'shaderc_profile_none' for GLSL
// versions that do not define profiles, e.g. versions below 150.
void SetForcedVersionProfile(int version, shaderc_profile profile) {
shaderc_compile_options_set_forced_version_profile(options_, version,
profile);
}
// Sets the compiler mode to suppress warnings. Note this option overrides
// warnings-as-errors mode. When both suppress-warnings and warnings-as-errors
// modes are turned on, warning messages will be inhibited, and will not be
// emitted as error message.
void SetSuppressWarnings() {
shaderc_compile_options_set_suppress_warnings(options_);
}
// Sets the source language. The default is GLSL.
void SetSourceLanguage(shaderc_source_language lang) {
shaderc_compile_options_set_source_language(options_, lang);
}
// Sets the target shader environment, affecting which warnings or errors will
// be issued. The version will be for distinguishing between different
// versions of the target environment. The version value should be either 0
// or a value listed in shaderc_env_version. The 0 value maps to Vulkan 1.0
// if |target| is Vulkan, and it maps to OpenGL 4.5 if |target| is OpenGL.
void SetTargetEnvironment(shaderc_target_env target, uint32_t version) {
shaderc_compile_options_set_target_env(options_, target, version);
}
// Sets the target SPIR-V version. The generated module will use this version
// of SPIR-V. Each target environment determines what versions of SPIR-V
// it can consume. Defaults to the highest version of SPIR-V 1.0 which is
// required to be supported by the target environment. E.g. Default to SPIR-V
// 1.0 for Vulkan 1.0 and SPIR-V 1.3 for Vulkan 1.1.
void SetTargetSpirv(shaderc_spirv_version version) {
shaderc_compile_options_set_target_spirv(options_, version);
}
// Sets the compiler mode to make all warnings into errors. Note the
// suppress-warnings mode overrides this option, i.e. if both
// warning-as-errors and suppress-warnings modes are set on, warnings will not
// be emitted as error message.
void SetWarningsAsErrors() {
shaderc_compile_options_set_warnings_as_errors(options_);
}
// Sets a resource limit.
void SetLimit(shaderc_limit limit, int value) {
shaderc_compile_options_set_limit(options_, limit, value);
}
// Sets whether the compiler should automatically assign bindings to uniforms
// that aren't already explicitly bound in the shader source.
void SetAutoBindUniforms(bool auto_bind) {
shaderc_compile_options_set_auto_bind_uniforms(options_, auto_bind);
}
// Sets whether the compiler should automatically remove sampler variables
// and convert image variables to combined image sampler variables.
void SetAutoSampledTextures(bool auto_sampled) {
shaderc_compile_options_set_auto_combined_image_sampler(options_,
auto_sampled);
}
// Sets whether the compiler should use HLSL IO mapping rules for bindings.
// Defaults to false.
void SetHlslIoMapping(bool hlsl_iomap) {
shaderc_compile_options_set_hlsl_io_mapping(options_, hlsl_iomap);
}
// Sets whether the compiler should determine block member offsets using HLSL
// packing rules instead of standard GLSL rules. Defaults to false. Only
// affects GLSL compilation. HLSL rules are always used when compiling HLSL.
void SetHlslOffsets(bool hlsl_offsets) {
shaderc_compile_options_set_hlsl_offsets(options_, hlsl_offsets);
}
// Sets the base binding number used for for a uniform resource type when
// automatically assigning bindings. For GLSL compilation, sets the lowest
// automatically assigned number. For HLSL compilation, the regsiter number
// assigned to the resource is added to this specified base.
void SetBindingBase(shaderc_uniform_kind kind, uint32_t base) {
shaderc_compile_options_set_binding_base(options_, kind, base);
}
// Like SetBindingBase, but only takes effect when compiling a given shader
// stage. The stage is assumed to be one of vertex, fragment, tessellation
// evaluation, tesselation control, geometry, or compute.
void SetBindingBaseForStage(shaderc_shader_kind shader_kind,
shaderc_uniform_kind kind, uint32_t base) {
shaderc_compile_options_set_binding_base_for_stage(options_, shader_kind,
kind, base);
}
// Sets whether the compiler should preserve all bindings, even when those
// bindings are not used.
void SetPreserveBindings(bool preserve_bindings) {
shaderc_compile_options_set_preserve_bindings(options_, preserve_bindings);
}
// Sets whether the compiler automatically assigns locations to
// uniform variables that don't have explicit locations.
void SetAutoMapLocations(bool auto_map) {
shaderc_compile_options_set_auto_map_locations(options_, auto_map);
}
// Sets a descriptor set and binding for an HLSL register in the given stage.
// Copies the parameter strings.
void SetHlslRegisterSetAndBindingForStage(shaderc_shader_kind shader_kind,
const std::string& reg,
const std::string& set,
const std::string& binding) {
shaderc_compile_options_set_hlsl_register_set_and_binding_for_stage(
options_, shader_kind, reg.c_str(), set.c_str(), binding.c_str());
}
// Sets a descriptor set and binding for an HLSL register in any stage.
// Copies the parameter strings.
void SetHlslRegisterSetAndBinding(const std::string& reg,
const std::string& set,
const std::string& binding) {
shaderc_compile_options_set_hlsl_register_set_and_binding(
options_, reg.c_str(), set.c_str(), binding.c_str());
}
// Sets whether the compiler should enable extension
// SPV_GOOGLE_hlsl_functionality1.
void SetHlslFunctionality1(bool enable) {
shaderc_compile_options_set_hlsl_functionality1(options_, enable);
}
// Sets whether 16-bit types are supported in HLSL or not.
void SetHlsl16BitTypes(bool enable) {
shaderc_compile_options_set_hlsl_16bit_types(options_, enable);
}
// Sets whether the compiler should invert position.Y output in vertex shader.
void SetInvertY(bool enable) {
shaderc_compile_options_set_invert_y(options_, enable);
}
// Sets whether the compiler should generates code for max an min which,
// if given a NaN operand, will return the other operand. Similarly, the
// clamp builtin will favour the non-NaN operands, as if clamp were
// implemented as a composition of max and min.
void SetNanClamp(bool enable) {
shaderc_compile_options_set_nan_clamp(options_, enable);
}
private:
CompileOptions& operator=(const CompileOptions& other) = delete;
shaderc_compile_options_t options_;
std::unique_ptr<IncluderInterface> includer_;
friend class Compiler;
};
// The compilation context for compiling source to SPIR-V.
class Compiler {
public:
Compiler() : compiler_(shaderc_compiler_initialize()) {}
~Compiler() { shaderc_compiler_release(compiler_); }
Compiler(Compiler&& other) {
compiler_ = other.compiler_;
other.compiler_ = nullptr;
}
bool IsValid() const { return compiler_ != nullptr; }
// Compiles the given source GLSL and returns a SPIR-V binary module
// compilation result.
// The source_text parameter must be a valid pointer.
// The source_text_size parameter must be the length of the source text.
// The shader_kind parameter either forces the compilation to be done with a
// specified shader kind, or hint the compiler how to determine the exact
// shader kind. If the shader kind is set to shaderc_glslc_infer_from_source,
// the compiler will try to deduce the shader kind from the source string and
// a failure in this proess will generate an error. Currently only #pragma
// annotation is supported. If the shader kind is set to one of the default
// shader kinds, the compiler will fall back to the specified default shader
// kind in case it failed to deduce the shader kind from the source string.
// The input_file_name is a null-termintated string. It is used as a tag to
// identify the source string in cases like emitting error messages. It
// doesn't have to be a 'file name'.
// The entry_point_name parameter is a null-terminated string specifying
// the entry point name for HLSL compilation. For GLSL compilation, the
// entry point name is assumed to be "main".
// The compilation is passed any options specified in the CompileOptions
// parameter.
// It is valid for the returned CompilationResult object to outlive this
// compiler object.
// Note when the options_ has disassembly mode or preprocessing only mode set
// on, the returned CompilationResult will hold a text string, instead of a
// SPIR-V binary generated with default options.
SpvCompilationResult CompileGlslToSpv(const char* source_text,
size_t source_text_size,
shaderc_shader_kind shader_kind,
const char* input_file_name,
const char* entry_point_name,
const CompileOptions& options) const {
shaderc_compilation_result_t compilation_result = shaderc_compile_into_spv(
compiler_, source_text, source_text_size, shader_kind, input_file_name,
entry_point_name, options.options_);
return SpvCompilationResult(compilation_result);
}
// Compiles the given source shader and returns a SPIR-V binary module
// compilation result.
// Like the first CompileGlslToSpv method but assumes the entry point name
// is "main".
SpvCompilationResult CompileGlslToSpv(const char* source_text,
size_t source_text_size,
shaderc_shader_kind shader_kind,
const char* input_file_name,
const CompileOptions& options) const {
return CompileGlslToSpv(source_text, source_text_size, shader_kind,
input_file_name, "main", options);
}
// Compiles the given source GLSL and returns a SPIR-V binary module
// compilation result.
// Like the previous CompileGlslToSpv method but uses default options.
SpvCompilationResult CompileGlslToSpv(const char* source_text,
size_t source_text_size,
shaderc_shader_kind shader_kind,
const char* input_file_name) const {
shaderc_compilation_result_t compilation_result =
shaderc_compile_into_spv(compiler_, source_text, source_text_size,
shader_kind, input_file_name, "main", nullptr);
return SpvCompilationResult(compilation_result);
}
// Compiles the given source shader and returns a SPIR-V binary module
// compilation result.
// Like the first CompileGlslToSpv method but the source is provided as
// a std::string, and we assume the entry point is "main".
SpvCompilationResult CompileGlslToSpv(const std::string& source_text,
shaderc_shader_kind shader_kind,
const char* input_file_name,
const CompileOptions& options) const {
return CompileGlslToSpv(source_text.data(), source_text.size(), shader_kind,
input_file_name, options);
}
// Compiles the given source shader and returns a SPIR-V binary module
// compilation result.
// Like the first CompileGlslToSpv method but the source is provided as
// a std::string.
SpvCompilationResult CompileGlslToSpv(const std::string& source_text,
shaderc_shader_kind shader_kind,
const char* input_file_name,
const char* entry_point_name,
const CompileOptions& options) const {
return CompileGlslToSpv(source_text.data(), source_text.size(), shader_kind,
input_file_name, entry_point_name, options);
}
// Compiles the given source GLSL and returns a SPIR-V binary module
// compilation result.
// Like the previous CompileGlslToSpv method but assumes the entry point
// name is "main".
SpvCompilationResult CompileGlslToSpv(const std::string& source_text,
shaderc_shader_kind shader_kind,
const char* input_file_name) const {
return CompileGlslToSpv(source_text.data(), source_text.size(), shader_kind,
input_file_name);
}
// Assembles the given SPIR-V assembly and returns a SPIR-V binary module
// compilation result.
// The assembly should follow the syntax defined in the SPIRV-Tools project
// (https://github.com/KhronosGroup/SPIRV-Tools/blob/master/syntax.md).
// It is valid for the returned CompilationResult object to outlive this
// compiler object.
// The assembling will pick options suitable for assembling specified in the
// CompileOptions parameter.
SpvCompilationResult AssembleToSpv(const char* source_assembly,
size_t source_assembly_size,
const CompileOptions& options) const {
return SpvCompilationResult(shaderc_assemble_into_spv(
compiler_, source_assembly, source_assembly_size, options.options_));
}
// Assembles the given SPIR-V assembly and returns a SPIR-V binary module
// compilation result.
// Like the first AssembleToSpv method but uses the default compiler options.
SpvCompilationResult AssembleToSpv(const char* source_assembly,
size_t source_assembly_size) const {
return SpvCompilationResult(shaderc_assemble_into_spv(
compiler_, source_assembly, source_assembly_size, nullptr));
}
// Assembles the given SPIR-V assembly and returns a SPIR-V binary module
// compilation result.
// Like the first AssembleToSpv method but the source is provided as a
// std::string.
SpvCompilationResult AssembleToSpv(const std::string& source_assembly,
const CompileOptions& options) const {
return SpvCompilationResult(
shaderc_assemble_into_spv(compiler_, source_assembly.data(),
source_assembly.size(), options.options_));
}
// Assembles the given SPIR-V assembly and returns a SPIR-V binary module
// compilation result.
// Like the first AssembleToSpv method but the source is provided as a
// std::string and also uses default compiler options.
SpvCompilationResult AssembleToSpv(const std::string& source_assembly) const {
return SpvCompilationResult(shaderc_assemble_into_spv(
compiler_, source_assembly.data(), source_assembly.size(), nullptr));
}
// Compiles the given source GLSL and returns the SPIR-V assembly text
// compilation result.
// Options are similar to the first CompileToSpv method.
AssemblyCompilationResult CompileGlslToSpvAssembly(
const char* source_text, size_t source_text_size,
shaderc_shader_kind shader_kind, const char* input_file_name,
const char* entry_point_name, const CompileOptions& options) const {
shaderc_compilation_result_t compilation_result =
shaderc_compile_into_spv_assembly(
compiler_, source_text, source_text_size, shader_kind,
input_file_name, entry_point_name, options.options_);
return AssemblyCompilationResult(compilation_result);
}
// Compiles the given source GLSL and returns the SPIR-V assembly text
// compilation result.
// Similare to the previous method, but assumes entry point name is "main".
AssemblyCompilationResult CompileGlslToSpvAssembly(
const char* source_text, size_t source_text_size,
shaderc_shader_kind shader_kind, const char* input_file_name,
const CompileOptions& options) const {
return CompileGlslToSpvAssembly(source_text, source_text_size, shader_kind,
input_file_name, "main", options);
}
// Compiles the given source GLSL and returns the SPIR-V assembly text
// result. Like the first CompileGlslToSpvAssembly method but the source
// is provided as a std::string. Options are otherwise similar to
// the first CompileToSpv method.
AssemblyCompilationResult CompileGlslToSpvAssembly(
const std::string& source_text, shaderc_shader_kind shader_kind,
const char* input_file_name, const char* entry_point_name,
const CompileOptions& options) const {
return CompileGlslToSpvAssembly(source_text.data(), source_text.size(),
shader_kind, input_file_name,
entry_point_name, options);
}
// Compiles the given source GLSL and returns the SPIR-V assembly text
// result. Like the previous CompileGlslToSpvAssembly method but assumes
// the entry point name is "main".
AssemblyCompilationResult CompileGlslToSpvAssembly(
const std::string& source_text, shaderc_shader_kind shader_kind,
const char* input_file_name, const CompileOptions& options) const {
return CompileGlslToSpvAssembly(source_text, shader_kind, input_file_name,
"main", options);
}
// Preprocesses the given source GLSL and returns the preprocessed
// source text as a compilation result.
// Options are similar to the first CompileToSpv method.
PreprocessedSourceCompilationResult PreprocessGlsl(
const char* source_text, size_t source_text_size,
shaderc_shader_kind shader_kind, const char* input_file_name,
const CompileOptions& options) const {
shaderc_compilation_result_t compilation_result =
shaderc_compile_into_preprocessed_text(
compiler_, source_text, source_text_size, shader_kind,
input_file_name, "main", options.options_);
return PreprocessedSourceCompilationResult(compilation_result);
}
// Preprocesses the given source GLSL and returns text result. Like the first
// PreprocessGlsl method but the source is provided as a std::string.
// Options are otherwise similar to the first CompileToSpv method.
PreprocessedSourceCompilationResult PreprocessGlsl(
const std::string& source_text, shaderc_shader_kind shader_kind,
const char* input_file_name, const CompileOptions& options) const {
return PreprocessGlsl(source_text.data(), source_text.size(), shader_kind,
input_file_name, options);
}
private:
Compiler(const Compiler&) = delete;
Compiler& operator=(const Compiler& other) = delete;
shaderc_compiler_t compiler_;
};
} // namespace shaderc
#endif // SHADERC_SHADERC_HPP_

39
include/vulkan/shaderc/status.h Normal file
View file

@ -0,0 +1,39 @@
// Copyright 2018 The Shaderc Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef SHADERC_STATUS_H_
#define SHADERC_STATUS_H_
#ifdef __cplusplus
extern "C" {
#endif
// Indicate the status of a compilation.
typedef enum {
shaderc_compilation_status_success = 0,
shaderc_compilation_status_invalid_stage = 1, // error stage deduction
shaderc_compilation_status_compilation_error = 2,
shaderc_compilation_status_internal_error = 3, // unexpected failure
shaderc_compilation_status_null_result_object = 4,
shaderc_compilation_status_invalid_assembly = 5,
shaderc_compilation_status_validation_error = 6,
shaderc_compilation_status_transformation_error = 7,
shaderc_compilation_status_configuration_error = 8,
} shaderc_compilation_status;
#ifdef __cplusplus
}
#endif // __cplusplus
#endif // SHADERC_STATUS_H_

37
include/vulkan/shaderc/visibility.h Normal file
View file

@ -0,0 +1,37 @@
// Copyright 2018 The Shaderc Authors. All rights reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef SHADERC_VISIBILITY_H_
#define SHADERC_VISIBILITY_H_
// SHADERC_EXPORT tags symbol that will be exposed by the shared libraries.
#if defined(SHADERC_SHAREDLIB)
#if defined(_WIN32)
#if defined(SHADERC_IMPLEMENTATION)
#define SHADERC_EXPORT __declspec(dllexport)
#else
#define SHADERC_EXPORT __declspec(dllimport)
#endif
#else
#if defined(SHADERC_IMPLEMENTATION)
#define SHADERC_EXPORT __attribute__((visibility("default")))
#else
#define SHADERC_EXPORT
#endif
#endif
#else
#define SHADERC_EXPORT
#endif
#endif // SHADERC_VISIBILITY_H_

131
include/vulkan/spirv-headers/GLSL.std.450.h Normal file
View file

@ -0,0 +1,131 @@
/*
** Copyright (c) 2014-2016 The Khronos Group Inc.
**
** Permission is hereby granted, free of charge, to any person obtaining a copy
** of this software and/or associated documentation files (the "Materials"),
** to deal in the Materials without restriction, including without limitation
** the rights to use, copy, modify, merge, publish, distribute, sublicense,
** and/or sell copies of the Materials, and to permit persons to whom the
** Materials are furnished to do so, subject to the following conditions:
**
** The above copyright notice and this permission notice shall be included in
** all copies or substantial portions of the Materials.
**
** MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS KHRONOS
** STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS SPECIFICATIONS AND
** HEADER INFORMATION ARE LOCATED AT https://www.khronos.org/registry/
**
** THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
** OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
** FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
** THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
** LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
** FROM,OUT OF OR IN CONNECTION WITH THE MATERIALS OR THE USE OR OTHER DEALINGS
** IN THE MATERIALS.
*/
#ifndef GLSLstd450_H
#define GLSLstd450_H
static const int GLSLstd450Version = 100;
static const int GLSLstd450Revision = 3;
enum GLSLstd450 {
GLSLstd450Bad = 0, // Don't use
GLSLstd450Round = 1,
GLSLstd450RoundEven = 2,
GLSLstd450Trunc = 3,
GLSLstd450FAbs = 4,
GLSLstd450SAbs = 5,
GLSLstd450FSign = 6,
GLSLstd450SSign = 7,
GLSLstd450Floor = 8,
GLSLstd450Ceil = 9,
GLSLstd450Fract = 10,
GLSLstd450Radians = 11,
GLSLstd450Degrees = 12,
GLSLstd450Sin = 13,
GLSLstd450Cos = 14,
GLSLstd450Tan = 15,
GLSLstd450Asin = 16,
GLSLstd450Acos = 17,
GLSLstd450Atan = 18,
GLSLstd450Sinh = 19,
GLSLstd450Cosh = 20,
GLSLstd450Tanh = 21,
GLSLstd450Asinh = 22,
GLSLstd450Acosh = 23,
GLSLstd450Atanh = 24,
GLSLstd450Atan2 = 25,
GLSLstd450Pow = 26,
GLSLstd450Exp = 27,
GLSLstd450Log = 28,
GLSLstd450Exp2 = 29,
GLSLstd450Log2 = 30,
GLSLstd450Sqrt = 31,
GLSLstd450InverseSqrt = 32,
GLSLstd450Determinant = 33,
GLSLstd450MatrixInverse = 34,
GLSLstd450Modf = 35, // second operand needs an OpVariable to write to
GLSLstd450ModfStruct = 36, // no OpVariable operand
GLSLstd450FMin = 37,
GLSLstd450UMin = 38,
GLSLstd450SMin = 39,
GLSLstd450FMax = 40,
GLSLstd450UMax = 41,
GLSLstd450SMax = 42,
GLSLstd450FClamp = 43,
GLSLstd450UClamp = 44,
GLSLstd450SClamp = 45,
GLSLstd450FMix = 46,
GLSLstd450IMix = 47, // Reserved
GLSLstd450Step = 48,
GLSLstd450SmoothStep = 49,
GLSLstd450Fma = 50,
GLSLstd450Frexp = 51, // second operand needs an OpVariable to write to
GLSLstd450FrexpStruct = 52, // no OpVariable operand
GLSLstd450Ldexp = 53,
GLSLstd450PackSnorm4x8 = 54,
GLSLstd450PackUnorm4x8 = 55,
GLSLstd450PackSnorm2x16 = 56,
GLSLstd450PackUnorm2x16 = 57,
GLSLstd450PackHalf2x16 = 58,
GLSLstd450PackDouble2x32 = 59,
GLSLstd450UnpackSnorm2x16 = 60,
GLSLstd450UnpackUnorm2x16 = 61,
GLSLstd450UnpackHalf2x16 = 62,
GLSLstd450UnpackSnorm4x8 = 63,
GLSLstd450UnpackUnorm4x8 = 64,
GLSLstd450UnpackDouble2x32 = 65,
GLSLstd450Length = 66,
GLSLstd450Distance = 67,
GLSLstd450Cross = 68,
GLSLstd450Normalize = 69,
GLSLstd450FaceForward = 70,
GLSLstd450Reflect = 71,
GLSLstd450Refract = 72,
GLSLstd450FindILsb = 73,
GLSLstd450FindSMsb = 74,
GLSLstd450FindUMsb = 75,
GLSLstd450InterpolateAtCentroid = 76,
GLSLstd450InterpolateAtSample = 77,
GLSLstd450InterpolateAtOffset = 78,
GLSLstd450NMin = 79,
GLSLstd450NMax = 80,
GLSLstd450NClamp = 81,
GLSLstd450Count
};
#endif // #ifndef GLSLstd450_H

2711
include/vulkan/spirv-headers/spirv.h Normal file
View file

File diff suppressed because it is too large Load diff

2752
include/vulkan/spirv-headers/spirv.hpp Normal file
View file

File diff suppressed because it is too large Load diff

2752
include/vulkan/spirv-headers/spirv.hpp11 Normal file
View file

File diff suppressed because it is too large Load diff

1984
include/vulkan/spirv-headers/spirv.json Normal file
View file

File diff suppressed because it is too large Load diff

1943
include/vulkan/spirv-headers/spirv.lua Normal file
View file

File diff suppressed because it is too large Load diff

1943
include/vulkan/spirv-headers/spirv.py Normal file
View file

File diff suppressed because it is too large Load diff

979
include/vulkan/spirv-tools/libspirv.h Normal file
View file

@ -0,0 +1,979 @@
// Copyright (c) 2015-2020 The Khronos Group Inc.
// Modifications Copyright (C) 2020 Advanced Micro Devices, Inc. All rights
// reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef INCLUDE_SPIRV_TOOLS_LIBSPIRV_H_
#define INCLUDE_SPIRV_TOOLS_LIBSPIRV_H_
#ifdef __cplusplus
extern "C" {
#else
#include <stdbool.h>
#endif
#include <stddef.h>
#include <stdint.h>
#if defined(SPIRV_TOOLS_SHAREDLIB)
#if defined(_WIN32)
#if defined(SPIRV_TOOLS_IMPLEMENTATION)
#define SPIRV_TOOLS_EXPORT __declspec(dllexport)
#else
#define SPIRV_TOOLS_EXPORT __declspec(dllimport)
#endif
#else
#if defined(SPIRV_TOOLS_IMPLEMENTATION)
#define SPIRV_TOOLS_EXPORT __attribute__((visibility("default")))
#else
#define SPIRV_TOOLS_EXPORT
#endif
#endif
#else
#define SPIRV_TOOLS_EXPORT
#endif
// Helpers
#define SPV_BIT(shift) (1 << (shift))
#define SPV_FORCE_16_BIT_ENUM(name) SPV_FORCE_16BIT_##name = 0x7fff
#define SPV_FORCE_32_BIT_ENUM(name) SPV_FORCE_32BIT_##name = 0x7fffffff
// Enumerations
typedef enum spv_result_t {
SPV_SUCCESS = 0,
SPV_UNSUPPORTED = 1,
SPV_END_OF_STREAM = 2,
SPV_WARNING = 3,
SPV_FAILED_MATCH = 4,
SPV_REQUESTED_TERMINATION = 5, // Success, but signals early termination.
SPV_ERROR_INTERNAL = -1,
SPV_ERROR_OUT_OF_MEMORY = -2,
SPV_ERROR_INVALID_POINTER = -3,
SPV_ERROR_INVALID_BINARY = -4,
SPV_ERROR_INVALID_TEXT = -5,
SPV_ERROR_INVALID_TABLE = -6,
SPV_ERROR_INVALID_VALUE = -7,
SPV_ERROR_INVALID_DIAGNOSTIC = -8,
SPV_ERROR_INVALID_LOOKUP = -9,
SPV_ERROR_INVALID_ID = -10,
SPV_ERROR_INVALID_CFG = -11,
SPV_ERROR_INVALID_LAYOUT = -12,
SPV_ERROR_INVALID_CAPABILITY = -13,
SPV_ERROR_INVALID_DATA = -14, // Indicates data rules validation failure.
SPV_ERROR_MISSING_EXTENSION = -15,
SPV_ERROR_WRONG_VERSION = -16, // Indicates wrong SPIR-V version
SPV_FORCE_32_BIT_ENUM(spv_result_t)
} spv_result_t;
// Severity levels of messages communicated to the consumer.
typedef enum spv_message_level_t {
SPV_MSG_FATAL, // Unrecoverable error due to environment.
// Will exit the program immediately. E.g.,
// out of memory.
SPV_MSG_INTERNAL_ERROR, // Unrecoverable error due to SPIRV-Tools
// internals.
// Will exit the program immediately. E.g.,
// unimplemented feature.
SPV_MSG_ERROR, // Normal error due to user input.
SPV_MSG_WARNING, // Warning information.
SPV_MSG_INFO, // General information.
SPV_MSG_DEBUG, // Debug information.
} spv_message_level_t;
typedef enum spv_endianness_t {
SPV_ENDIANNESS_LITTLE,
SPV_ENDIANNESS_BIG,
SPV_FORCE_32_BIT_ENUM(spv_endianness_t)
} spv_endianness_t;
// The kinds of operands that an instruction may have.
//
// Some operand types are "concrete". The binary parser uses a concrete
// operand type to describe an operand of a parsed instruction.
//
// The assembler uses all operand types. In addition to determining what
// kind of value an operand may be, non-concrete operand types capture the
// fact that an operand might be optional (may be absent, or present exactly
// once), or might occur zero or more times.
//
// Sometimes we also need to be able to express the fact that an operand
// is a member of an optional tuple of values. In that case the first member
// would be optional, and the subsequent members would be required.
//
// NOTE: Although we don't promise binary compatibility, as a courtesy, please
// add new enum values at the end.
typedef enum spv_operand_type_t {
// A sentinel value.
SPV_OPERAND_TYPE_NONE = 0,
// Set 1: Operands that are IDs.
SPV_OPERAND_TYPE_ID,
SPV_OPERAND_TYPE_TYPE_ID,
SPV_OPERAND_TYPE_RESULT_ID,
SPV_OPERAND_TYPE_MEMORY_SEMANTICS_ID, // SPIR-V Sec 3.25
SPV_OPERAND_TYPE_SCOPE_ID, // SPIR-V Sec 3.27
// Set 2: Operands that are literal numbers.
SPV_OPERAND_TYPE_LITERAL_INTEGER, // Always unsigned 32-bits.
// The Instruction argument to OpExtInst. It's an unsigned 32-bit literal
// number indicating which instruction to use from an extended instruction
// set.
SPV_OPERAND_TYPE_EXTENSION_INSTRUCTION_NUMBER,
// The Opcode argument to OpSpecConstantOp. It determines the operation
// to be performed on constant operands to compute a specialization constant
// result.
SPV_OPERAND_TYPE_SPEC_CONSTANT_OP_NUMBER,
// A literal number whose format and size are determined by a previous operand
// in the same instruction. It's a signed integer, an unsigned integer, or a
// floating point number. It also has a specified bit width. The width
// may be larger than 32, which would require such a typed literal value to
// occupy multiple SPIR-V words.
SPV_OPERAND_TYPE_TYPED_LITERAL_NUMBER,
// Set 3: The literal string operand type.
SPV_OPERAND_TYPE_LITERAL_STRING,
// Set 4: Operands that are a single word enumerated value.
SPV_OPERAND_TYPE_SOURCE_LANGUAGE, // SPIR-V Sec 3.2
SPV_OPERAND_TYPE_EXECUTION_MODEL, // SPIR-V Sec 3.3
SPV_OPERAND_TYPE_ADDRESSING_MODEL, // SPIR-V Sec 3.4
SPV_OPERAND_TYPE_MEMORY_MODEL, // SPIR-V Sec 3.5
SPV_OPERAND_TYPE_EXECUTION_MODE, // SPIR-V Sec 3.6
SPV_OPERAND_TYPE_STORAGE_CLASS, // SPIR-V Sec 3.7
SPV_OPERAND_TYPE_DIMENSIONALITY, // SPIR-V Sec 3.8
SPV_OPERAND_TYPE_SAMPLER_ADDRESSING_MODE, // SPIR-V Sec 3.9
SPV_OPERAND_TYPE_SAMPLER_FILTER_MODE, // SPIR-V Sec 3.10
SPV_OPERAND_TYPE_SAMPLER_IMAGE_FORMAT, // SPIR-V Sec 3.11
SPV_OPERAND_TYPE_IMAGE_CHANNEL_ORDER, // SPIR-V Sec 3.12
SPV_OPERAND_TYPE_IMAGE_CHANNEL_DATA_TYPE, // SPIR-V Sec 3.13
SPV_OPERAND_TYPE_FP_ROUNDING_MODE, // SPIR-V Sec 3.16
SPV_OPERAND_TYPE_LINKAGE_TYPE, // SPIR-V Sec 3.17
SPV_OPERAND_TYPE_ACCESS_QUALIFIER, // SPIR-V Sec 3.18
SPV_OPERAND_TYPE_FUNCTION_PARAMETER_ATTRIBUTE, // SPIR-V Sec 3.19
SPV_OPERAND_TYPE_DECORATION, // SPIR-V Sec 3.20
SPV_OPERAND_TYPE_BUILT_IN, // SPIR-V Sec 3.21
SPV_OPERAND_TYPE_GROUP_OPERATION, // SPIR-V Sec 3.28
SPV_OPERAND_TYPE_KERNEL_ENQ_FLAGS, // SPIR-V Sec 3.29
SPV_OPERAND_TYPE_KERNEL_PROFILING_INFO, // SPIR-V Sec 3.30
SPV_OPERAND_TYPE_CAPABILITY, // SPIR-V Sec 3.31
// NOTE: New concrete enum values should be added at the end.
// Set 5: Operands that are a single word bitmask.
// Sometimes a set bit indicates the instruction requires still more operands.
SPV_OPERAND_TYPE_IMAGE, // SPIR-V Sec 3.14
SPV_OPERAND_TYPE_FP_FAST_MATH_MODE, // SPIR-V Sec 3.15
SPV_OPERAND_TYPE_SELECTION_CONTROL, // SPIR-V Sec 3.22
SPV_OPERAND_TYPE_LOOP_CONTROL, // SPIR-V Sec 3.23
SPV_OPERAND_TYPE_FUNCTION_CONTROL, // SPIR-V Sec 3.24
SPV_OPERAND_TYPE_MEMORY_ACCESS, // SPIR-V Sec 3.26
SPV_OPERAND_TYPE_FRAGMENT_SHADING_RATE, // SPIR-V Sec 3.FSR
// NOTE: New concrete enum values should be added at the end.
// The "optional" and "variable" operand types are only used internally by
// the assembler and the binary parser.
// There are two categories:
// Optional : expands to 0 or 1 operand, like ? in regular expressions.
// Variable : expands to 0, 1 or many operands or pairs of operands.
// This is similar to * in regular expressions.
// NOTE: These FIRST_* and LAST_* enum values are DEPRECATED.
// The concept of "optional" and "variable" operand types are only intended
// for use as an implementation detail of parsing SPIR-V, either in text or
// binary form. Instead of using enum ranges, use characteristic function
// spvOperandIsConcrete.
// The use of enum value ranges in a public API makes it difficult to insert
// new values into a range without also breaking binary compatibility.
//
// Macros for defining bounds on optional and variable operand types.
// Any variable operand type is also optional.
// TODO(dneto): Remove SPV_OPERAND_TYPE_FIRST_* and SPV_OPERAND_TYPE_LAST_*
#define FIRST_OPTIONAL(ENUM) ENUM, SPV_OPERAND_TYPE_FIRST_OPTIONAL_TYPE = ENUM
#define FIRST_VARIABLE(ENUM) ENUM, SPV_OPERAND_TYPE_FIRST_VARIABLE_TYPE = ENUM
#define LAST_VARIABLE(ENUM) \
ENUM, SPV_OPERAND_TYPE_LAST_VARIABLE_TYPE = ENUM, \
SPV_OPERAND_TYPE_LAST_OPTIONAL_TYPE = ENUM
// An optional operand represents zero or one logical operands.
// In an instruction definition, this may only appear at the end of the
// operand types.
FIRST_OPTIONAL(SPV_OPERAND_TYPE_OPTIONAL_ID),
// An optional image operand type.
SPV_OPERAND_TYPE_OPTIONAL_IMAGE,
// An optional memory access type.
SPV_OPERAND_TYPE_OPTIONAL_MEMORY_ACCESS,
// An optional literal integer.
SPV_OPERAND_TYPE_OPTIONAL_LITERAL_INTEGER,
// An optional literal number, which may be either integer or floating point.
SPV_OPERAND_TYPE_OPTIONAL_LITERAL_NUMBER,
// Like SPV_OPERAND_TYPE_TYPED_LITERAL_NUMBER, but optional, and integral.
SPV_OPERAND_TYPE_OPTIONAL_TYPED_LITERAL_INTEGER,
// An optional literal string.
SPV_OPERAND_TYPE_OPTIONAL_LITERAL_STRING,
// An optional access qualifier
SPV_OPERAND_TYPE_OPTIONAL_ACCESS_QUALIFIER,
// An optional context-independent value, or CIV. CIVs are tokens that we can
// assemble regardless of where they occur -- literals, IDs, immediate
// integers, etc.
SPV_OPERAND_TYPE_OPTIONAL_CIV,
// A variable operand represents zero or more logical operands.
// In an instruction definition, this may only appear at the end of the
// operand types.
FIRST_VARIABLE(SPV_OPERAND_TYPE_VARIABLE_ID),
SPV_OPERAND_TYPE_VARIABLE_LITERAL_INTEGER,
// A sequence of zero or more pairs of (typed literal integer, Id).
// Expands to zero or more:
// (SPV_OPERAND_TYPE_TYPED_LITERAL_INTEGER, SPV_OPERAND_TYPE_ID)
// where the literal number must always be an integer of some sort.
SPV_OPERAND_TYPE_VARIABLE_LITERAL_INTEGER_ID,
// A sequence of zero or more pairs of (Id, Literal integer)
LAST_VARIABLE(SPV_OPERAND_TYPE_VARIABLE_ID_LITERAL_INTEGER),
// The following are concrete enum types from the DebugInfo extended
// instruction set.
SPV_OPERAND_TYPE_DEBUG_INFO_FLAGS, // DebugInfo Sec 3.2. A mask.
SPV_OPERAND_TYPE_DEBUG_BASE_TYPE_ATTRIBUTE_ENCODING, // DebugInfo Sec 3.3
SPV_OPERAND_TYPE_DEBUG_COMPOSITE_TYPE, // DebugInfo Sec 3.4
SPV_OPERAND_TYPE_DEBUG_TYPE_QUALIFIER, // DebugInfo Sec 3.5
SPV_OPERAND_TYPE_DEBUG_OPERATION, // DebugInfo Sec 3.6
// The following are concrete enum types from the OpenCL.DebugInfo.100
// extended instruction set.
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_INFO_FLAGS, // Sec 3.2. A Mask
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_BASE_TYPE_ATTRIBUTE_ENCODING, // Sec 3.3
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_COMPOSITE_TYPE, // Sec 3.4
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_TYPE_QUALIFIER, // Sec 3.5
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_OPERATION, // Sec 3.6
SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_IMPORTED_ENTITY, // Sec 3.7
// The following are concrete enum types from SPV_INTEL_float_controls2
// https://github.com/intel/llvm/blob/39fa9b0cbfbae88327118990a05c5b387b56d2ef/sycl/doc/extensions/SPIRV/SPV_INTEL_float_controls2.asciidoc
SPV_OPERAND_TYPE_FPDENORM_MODE, // Sec 3.17 FP Denorm Mode
SPV_OPERAND_TYPE_FPOPERATION_MODE, // Sec 3.18 FP Operation Mode
// A value enum from https://github.com/KhronosGroup/SPIRV-Headers/pull/177
SPV_OPERAND_TYPE_QUANTIZATION_MODES,
// A value enum from https://github.com/KhronosGroup/SPIRV-Headers/pull/177
SPV_OPERAND_TYPE_OVERFLOW_MODES,
// Concrete operand types for the provisional Vulkan ray tracing feature.
SPV_OPERAND_TYPE_RAY_FLAGS, // SPIR-V Sec 3.RF
SPV_OPERAND_TYPE_RAY_QUERY_INTERSECTION, // SPIR-V Sec 3.RQIntersection
SPV_OPERAND_TYPE_RAY_QUERY_COMMITTED_INTERSECTION_TYPE, // SPIR-V Sec
// 3.RQCommitted
SPV_OPERAND_TYPE_RAY_QUERY_CANDIDATE_INTERSECTION_TYPE, // SPIR-V Sec
// 3.RQCandidate
// Concrete operand types for integer dot product.
// Packed vector format
SPV_OPERAND_TYPE_PACKED_VECTOR_FORMAT, // SPIR-V Sec 3.x
// An optional packed vector format
SPV_OPERAND_TYPE_OPTIONAL_PACKED_VECTOR_FORMAT,
// This is a sentinel value, and does not represent an operand type.
// It should come last.
SPV_OPERAND_TYPE_NUM_OPERAND_TYPES,
SPV_FORCE_32_BIT_ENUM(spv_operand_type_t)
} spv_operand_type_t;
// Returns true if the given type is concrete.
bool spvOperandIsConcrete(spv_operand_type_t type);
// Returns true if the given type is concrete and also a mask.
bool spvOperandIsConcreteMask(spv_operand_type_t type);
typedef enum spv_ext_inst_type_t {
SPV_EXT_INST_TYPE_NONE = 0,
SPV_EXT_INST_TYPE_GLSL_STD_450,
SPV_EXT_INST_TYPE_OPENCL_STD,
SPV_EXT_INST_TYPE_SPV_AMD_SHADER_EXPLICIT_VERTEX_PARAMETER,
SPV_EXT_INST_TYPE_SPV_AMD_SHADER_TRINARY_MINMAX,
SPV_EXT_INST_TYPE_SPV_AMD_GCN_SHADER,
SPV_EXT_INST_TYPE_SPV_AMD_SHADER_BALLOT,
SPV_EXT_INST_TYPE_DEBUGINFO,
SPV_EXT_INST_TYPE_OPENCL_DEBUGINFO_100,
SPV_EXT_INST_TYPE_NONSEMANTIC_CLSPVREFLECTION,
SPV_EXT_INST_TYPE_NONSEMANTIC_SHADER_DEBUGINFO_100,
// Multiple distinct extended instruction set types could return this
// value, if they are prefixed with NonSemantic. and are otherwise
// unrecognised
SPV_EXT_INST_TYPE_NONSEMANTIC_UNKNOWN,
SPV_FORCE_32_BIT_ENUM(spv_ext_inst_type_t)
} spv_ext_inst_type_t;
// This determines at a high level the kind of a binary-encoded literal
// number, but not the bit width.
// In principle, these could probably be folded into new entries in
// spv_operand_type_t. But then we'd have some special case differences
// between the assembler and disassembler.
typedef enum spv_number_kind_t {
SPV_NUMBER_NONE = 0, // The default for value initialization.
SPV_NUMBER_UNSIGNED_INT,
SPV_NUMBER_SIGNED_INT,
SPV_NUMBER_FLOATING,
} spv_number_kind_t;
typedef enum spv_text_to_binary_options_t {
SPV_TEXT_TO_BINARY_OPTION_NONE = SPV_BIT(0),
// Numeric IDs in the binary will have the same values as in the source.
// Non-numeric IDs are allocated by filling in the gaps, starting with 1
// and going up.
SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS = SPV_BIT(1),
SPV_FORCE_32_BIT_ENUM(spv_text_to_binary_options_t)
} spv_text_to_binary_options_t;
typedef enum spv_binary_to_text_options_t {
SPV_BINARY_TO_TEXT_OPTION_NONE = SPV_BIT(0),
SPV_BINARY_TO_TEXT_OPTION_PRINT = SPV_BIT(1),
SPV_BINARY_TO_TEXT_OPTION_COLOR = SPV_BIT(2),
SPV_BINARY_TO_TEXT_OPTION_INDENT = SPV_BIT(3),
SPV_BINARY_TO_TEXT_OPTION_SHOW_BYTE_OFFSET = SPV_BIT(4),
// Do not output the module header as leading comments in the assembly.
SPV_BINARY_TO_TEXT_OPTION_NO_HEADER = SPV_BIT(5),
// Use friendly names where possible. The heuristic may expand over
// time, but will use common names for scalar types, and debug names from
// OpName instructions.
SPV_BINARY_TO_TEXT_OPTION_FRIENDLY_NAMES = SPV_BIT(6),
// Add some comments to the generated assembly
SPV_BINARY_TO_TEXT_OPTION_COMMENT = SPV_BIT(7),
SPV_FORCE_32_BIT_ENUM(spv_binary_to_text_options_t)
} spv_binary_to_text_options_t;
// Constants
// The default id bound is to the minimum value for the id limit
// in the spir-v specification under the section "Universal Limits".
const uint32_t kDefaultMaxIdBound = 0x3FFFFF;
// Structures
// Information about an operand parsed from a binary SPIR-V module.
// Note that the values are not included. You still need access to the binary
// to extract the values.
typedef struct spv_parsed_operand_t {
// Location of the operand, in words from the start of the instruction.
uint16_t offset;
// Number of words occupied by this operand.
uint16_t num_words;
// The "concrete" operand type. See the definition of spv_operand_type_t
// for details.
spv_operand_type_t type;
// If type is a literal number type, then number_kind says whether it's
// a signed integer, an unsigned integer, or a floating point number.
spv_number_kind_t number_kind;
// The number of bits for a literal number type.
uint32_t number_bit_width;
} spv_parsed_operand_t;
// An instruction parsed from a binary SPIR-V module.
typedef struct spv_parsed_instruction_t {
// An array of words for this instruction, in native endianness.
const uint32_t* words;
// The number of words in this instruction.
uint16_t num_words;
uint16_t opcode;
// The extended instruction type, if opcode is OpExtInst. Otherwise
// this is the "none" value.
spv_ext_inst_type_t ext_inst_type;
// The type id, or 0 if this instruction doesn't have one.
uint32_t type_id;
// The result id, or 0 if this instruction doesn't have one.
uint32_t result_id;
// The array of parsed operands.
const spv_parsed_operand_t* operands;
uint16_t num_operands;
} spv_parsed_instruction_t;
typedef struct spv_parsed_header_t {
// The magic number of the SPIR-V module.
uint32_t magic;
// Version number.
uint32_t version;
// Generator's magic number.
uint32_t generator;
// IDs bound for this module (0 < id < bound).
uint32_t bound;
// reserved.
uint32_t reserved;
} spv_parsed_header_t;
typedef struct spv_const_binary_t {
const uint32_t* code;
const size_t wordCount;
} spv_const_binary_t;
typedef struct spv_binary_t {
uint32_t* code;
size_t wordCount;
} spv_binary_t;
typedef struct spv_text_t {
const char* str;
size_t length;
} spv_text_t;
typedef struct spv_position_t {
size_t line;
size_t column;
size_t index;
} spv_position_t;
typedef struct spv_diagnostic_t {
spv_position_t position;
char* error;
bool isTextSource;
} spv_diagnostic_t;
// Opaque struct containing the context used to operate on a SPIR-V module.
// Its object is used by various translation API functions.
typedef struct spv_context_t spv_context_t;
typedef struct spv_validator_options_t spv_validator_options_t;
typedef struct spv_optimizer_options_t spv_optimizer_options_t;
typedef struct spv_reducer_options_t spv_reducer_options_t;
typedef struct spv_fuzzer_options_t spv_fuzzer_options_t;
typedef struct spv_optimizer_t spv_optimizer_t;
// Type Definitions
typedef spv_const_binary_t* spv_const_binary;
typedef spv_binary_t* spv_binary;
typedef spv_text_t* spv_text;
typedef spv_position_t* spv_position;
typedef spv_diagnostic_t* spv_diagnostic;
typedef const spv_context_t* spv_const_context;
typedef spv_context_t* spv_context;
typedef spv_validator_options_t* spv_validator_options;
typedef const spv_validator_options_t* spv_const_validator_options;
typedef spv_optimizer_options_t* spv_optimizer_options;
typedef const spv_optimizer_options_t* spv_const_optimizer_options;
typedef spv_reducer_options_t* spv_reducer_options;
typedef const spv_reducer_options_t* spv_const_reducer_options;
typedef spv_fuzzer_options_t* spv_fuzzer_options;
typedef const spv_fuzzer_options_t* spv_const_fuzzer_options;
// Platform API
// Returns the SPIRV-Tools software version as a null-terminated string.
// The contents of the underlying storage is valid for the remainder of
// the process.
SPIRV_TOOLS_EXPORT const char* spvSoftwareVersionString(void);
// Returns a null-terminated string containing the name of the project,
// the software version string, and commit details.
// The contents of the underlying storage is valid for the remainder of
// the process.
SPIRV_TOOLS_EXPORT const char* spvSoftwareVersionDetailsString(void);
// Certain target environments impose additional restrictions on SPIR-V, so it's
// often necessary to specify which one applies. SPV_ENV_UNIVERSAL_* implies an
// environment-agnostic SPIR-V.
//
// When an API method needs to derive a SPIR-V version from a target environment
// (from the spv_context object), the method will choose the highest version of
// SPIR-V supported by the target environment. Examples:
// SPV_ENV_VULKAN_1_0 -> SPIR-V 1.0
// SPV_ENV_VULKAN_1_1 -> SPIR-V 1.3
// SPV_ENV_VULKAN_1_1_SPIRV_1_4 -> SPIR-V 1.4
// SPV_ENV_VULKAN_1_2 -> SPIR-V 1.5
// SPV_ENV_VULKAN_1_3 -> SPIR-V 1.6
// Consult the description of API entry points for specific rules.
typedef enum {
SPV_ENV_UNIVERSAL_1_0, // SPIR-V 1.0 latest revision, no other restrictions.
SPV_ENV_VULKAN_1_0, // Vulkan 1.0 latest revision.
SPV_ENV_UNIVERSAL_1_1, // SPIR-V 1.1 latest revision, no other restrictions.
SPV_ENV_OPENCL_2_1, // OpenCL Full Profile 2.1 latest revision.
SPV_ENV_OPENCL_2_2, // OpenCL Full Profile 2.2 latest revision.
SPV_ENV_OPENGL_4_0, // OpenGL 4.0 plus GL_ARB_gl_spirv, latest revisions.
SPV_ENV_OPENGL_4_1, // OpenGL 4.1 plus GL_ARB_gl_spirv, latest revisions.
SPV_ENV_OPENGL_4_2, // OpenGL 4.2 plus GL_ARB_gl_spirv, latest revisions.
SPV_ENV_OPENGL_4_3, // OpenGL 4.3 plus GL_ARB_gl_spirv, latest revisions.
// There is no variant for OpenGL 4.4.
SPV_ENV_OPENGL_4_5, // OpenGL 4.5 plus GL_ARB_gl_spirv, latest revisions.
SPV_ENV_UNIVERSAL_1_2, // SPIR-V 1.2, latest revision, no other restrictions.
SPV_ENV_OPENCL_1_2, // OpenCL Full Profile 1.2 plus cl_khr_il_program,
// latest revision.
SPV_ENV_OPENCL_EMBEDDED_1_2, // OpenCL Embedded Profile 1.2 plus
// cl_khr_il_program, latest revision.
SPV_ENV_OPENCL_2_0, // OpenCL Full Profile 2.0 plus cl_khr_il_program,
// latest revision.
SPV_ENV_OPENCL_EMBEDDED_2_0, // OpenCL Embedded Profile 2.0 plus
// cl_khr_il_program, latest revision.
SPV_ENV_OPENCL_EMBEDDED_2_1, // OpenCL Embedded Profile 2.1 latest revision.
SPV_ENV_OPENCL_EMBEDDED_2_2, // OpenCL Embedded Profile 2.2 latest revision.
SPV_ENV_UNIVERSAL_1_3, // SPIR-V 1.3 latest revision, no other restrictions.
SPV_ENV_VULKAN_1_1, // Vulkan 1.1 latest revision.
SPV_ENV_WEBGPU_0, // DEPRECATED, may be removed in the future.
SPV_ENV_UNIVERSAL_1_4, // SPIR-V 1.4 latest revision, no other restrictions.
// Vulkan 1.1 with VK_KHR_spirv_1_4, i.e. SPIR-V 1.4 binary.
SPV_ENV_VULKAN_1_1_SPIRV_1_4,
SPV_ENV_UNIVERSAL_1_5, // SPIR-V 1.5 latest revision, no other restrictions.
SPV_ENV_VULKAN_1_2, // Vulkan 1.2 latest revision.
SPV_ENV_UNIVERSAL_1_6, // SPIR-V 1.6 latest revision, no other restrictions.
SPV_ENV_VULKAN_1_3, // Vulkan 1.3 latest revision.
SPV_ENV_MAX // Keep this as the last enum value.
} spv_target_env;
// SPIR-V Validator can be parameterized with the following Universal Limits.
typedef enum {
spv_validator_limit_max_struct_members,
spv_validator_limit_max_struct_depth,
spv_validator_limit_max_local_variables,
spv_validator_limit_max_global_variables,
spv_validator_limit_max_switch_branches,
spv_validator_limit_max_function_args,
spv_validator_limit_max_control_flow_nesting_depth,
spv_validator_limit_max_access_chain_indexes,
spv_validator_limit_max_id_bound,
} spv_validator_limit;
// Returns a string describing the given SPIR-V target environment.
SPIRV_TOOLS_EXPORT const char* spvTargetEnvDescription(spv_target_env env);
// Parses s into *env and returns true if successful. If unparsable, returns
// false and sets *env to SPV_ENV_UNIVERSAL_1_0.
SPIRV_TOOLS_EXPORT bool spvParseTargetEnv(const char* s, spv_target_env* env);
// Determines the target env value with the least features but which enables
// the given Vulkan and SPIR-V versions. If such a target is supported, returns
// true and writes the value to |env|, otherwise returns false.
//
// The Vulkan version is given as an unsigned 32-bit number as specified in
// Vulkan section "29.2.1 Version Numbers": the major version number appears
// in bits 22 to 21, and the minor version is in bits 12 to 21. The SPIR-V
// version is given in the SPIR-V version header word: major version in bits
// 16 to 23, and minor version in bits 8 to 15.
SPIRV_TOOLS_EXPORT bool spvParseVulkanEnv(uint32_t vulkan_ver,
uint32_t spirv_ver,
spv_target_env* env);
// Creates a context object for most of the SPIRV-Tools API.
// Returns null if env is invalid.
//
// See specific API calls for how the target environment is interpreted
// (particularly assembly and validation).
SPIRV_TOOLS_EXPORT spv_context spvContextCreate(spv_target_env env);
// Destroys the given context object.
SPIRV_TOOLS_EXPORT void spvContextDestroy(spv_context context);
// Creates a Validator options object with default options. Returns a valid
// options object. The object remains valid until it is passed into
// spvValidatorOptionsDestroy.
SPIRV_TOOLS_EXPORT spv_validator_options spvValidatorOptionsCreate(void);
// Destroys the given Validator options object.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsDestroy(
spv_validator_options options);
// Records the maximum Universal Limit that is considered valid in the given
// Validator options object. <options> argument must be a valid options object.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetUniversalLimit(
spv_validator_options options, spv_validator_limit limit_type,
uint32_t limit);
// Record whether or not the validator should relax the rules on types for
// stores to structs. When relaxed, it will allow a type mismatch as long as
// the types are structs with the same layout. Two structs have the same layout
// if
//
// 1) the members of the structs are either the same type or are structs with
// same layout, and
//
// 2) the decorations that affect the memory layout are identical for both
// types. Other decorations are not relevant.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetRelaxStoreStruct(
spv_validator_options options, bool val);
// Records whether or not the validator should relax the rules on pointer usage
// in logical addressing mode.
//
// When relaxed, it will allow the following usage cases of pointers:
// 1) OpVariable allocating an object whose type is a pointer type
// 2) OpReturnValue returning a pointer value
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetRelaxLogicalPointer(
spv_validator_options options, bool val);
// Records whether or not the validator should relax the rules because it is
// expected that the optimizations will make the code legal.
//
// When relaxed, it will allow the following:
// 1) It will allow relaxed logical pointers. Setting this option will also
// set that option.
// 2) Pointers that are pass as parameters to function calls do not have to
// match the storage class of the formal parameter.
// 3) Pointers that are actual parameters on function calls do not have to point
// to the same type pointed as the formal parameter. The types just need to
// logically match.
// 4) GLSLstd450 Interpolate* instructions can have a load of an interpolant
// for a first argument.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetBeforeHlslLegalization(
spv_validator_options options, bool val);
// Records whether the validator should use "relaxed" block layout rules.
// Relaxed layout rules are described by Vulkan extension
// VK_KHR_relaxed_block_layout, and they affect uniform blocks, storage blocks,
// and push constants.
//
// This is enabled by default when targeting Vulkan 1.1 or later.
// Relaxed layout is more permissive than the default rules in Vulkan 1.0.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetRelaxBlockLayout(
spv_validator_options options, bool val);
// Records whether the validator should use standard block layout rules for
// uniform blocks.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetUniformBufferStandardLayout(
spv_validator_options options, bool val);
// Records whether the validator should use "scalar" block layout rules.
// Scalar layout rules are more permissive than relaxed block layout.
//
// See Vulkan extension VK_EXT_scalar_block_layout. The scalar alignment is
// defined as follows:
// - scalar alignment of a scalar is the scalar size
// - scalar alignment of a vector is the scalar alignment of its component
// - scalar alignment of a matrix is the scalar alignment of its component
// - scalar alignment of an array is the scalar alignment of its element
// - scalar alignment of a struct is the max scalar alignment among its
// members
//
// For a struct in Uniform, StorageClass, or PushConstant:
// - a member Offset must be a multiple of the member's scalar alignment
// - ArrayStride or MatrixStride must be a multiple of the array or matrix
// scalar alignment
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetScalarBlockLayout(
spv_validator_options options, bool val);
// Records whether the validator should use "scalar" block layout
// rules (as defined above) for Workgroup blocks. See Vulkan
// extension VK_KHR_workgroup_memory_explicit_layout.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetWorkgroupScalarBlockLayout(
spv_validator_options options, bool val);
// Records whether or not the validator should skip validating standard
// uniform/storage block layout.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetSkipBlockLayout(
spv_validator_options options, bool val);
// Records whether or not the validator should allow the LocalSizeId
// decoration where the environment otherwise would not allow it.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetAllowLocalSizeId(
spv_validator_options options, bool val);
// Whether friendly names should be used in validation error messages.
SPIRV_TOOLS_EXPORT void spvValidatorOptionsSetFriendlyNames(
spv_validator_options options, bool val);
// Creates an optimizer options object with default options. Returns a valid
// options object. The object remains valid until it is passed into
// |spvOptimizerOptionsDestroy|.
SPIRV_TOOLS_EXPORT spv_optimizer_options spvOptimizerOptionsCreate(void);
// Destroys the given optimizer options object.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsDestroy(
spv_optimizer_options options);
// Records whether or not the optimizer should run the validator before
// optimizing. If |val| is true, the validator will be run.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsSetRunValidator(
spv_optimizer_options options, bool val);
// Records the validator options that should be passed to the validator if it is
// run.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsSetValidatorOptions(
spv_optimizer_options options, spv_validator_options val);
// Records the maximum possible value for the id bound.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsSetMaxIdBound(
spv_optimizer_options options, uint32_t val);
// Records whether all bindings within the module should be preserved.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsSetPreserveBindings(
spv_optimizer_options options, bool val);
// Records whether all specialization constants within the module
// should be preserved.
SPIRV_TOOLS_EXPORT void spvOptimizerOptionsSetPreserveSpecConstants(
spv_optimizer_options options, bool val);
// Creates a reducer options object with default options. Returns a valid
// options object. The object remains valid until it is passed into
// |spvReducerOptionsDestroy|.
SPIRV_TOOLS_EXPORT spv_reducer_options spvReducerOptionsCreate(void);
// Destroys the given reducer options object.
SPIRV_TOOLS_EXPORT void spvReducerOptionsDestroy(spv_reducer_options options);
// Sets the maximum number of reduction steps that should run before the reducer
// gives up.
SPIRV_TOOLS_EXPORT void spvReducerOptionsSetStepLimit(
spv_reducer_options options, uint32_t step_limit);
// Sets the fail-on-validation-error option; if true, the reducer will return
// kStateInvalid if a reduction step yields a state that fails SPIR-V
// validation. Otherwise, an invalid state is treated as uninteresting and the
// reduction backtracks and continues.
SPIRV_TOOLS_EXPORT void spvReducerOptionsSetFailOnValidationError(
spv_reducer_options options, bool fail_on_validation_error);
// Sets the function that the reducer should target. If set to zero the reducer
// will target all functions as well as parts of the module that lie outside
// functions. Otherwise the reducer will restrict reduction to the function
// with result id |target_function|, which is required to exist.
SPIRV_TOOLS_EXPORT void spvReducerOptionsSetTargetFunction(
spv_reducer_options options, uint32_t target_function);
// Creates a fuzzer options object with default options. Returns a valid
// options object. The object remains valid until it is passed into
// |spvFuzzerOptionsDestroy|.
SPIRV_TOOLS_EXPORT spv_fuzzer_options spvFuzzerOptionsCreate(void);
// Destroys the given fuzzer options object.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsDestroy(spv_fuzzer_options options);
// Enables running the validator after every transformation is applied during
// a replay.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsEnableReplayValidation(
spv_fuzzer_options options);
// Sets the seed with which the random number generator used by the fuzzer
// should be initialized.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsSetRandomSeed(
spv_fuzzer_options options, uint32_t seed);
// Sets the range of transformations that should be applied during replay: 0
// means all transformations, +N means the first N transformations, -N means all
// except the final N transformations.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsSetReplayRange(
spv_fuzzer_options options, int32_t replay_range);
// Sets the maximum number of steps that the shrinker should take before giving
// up.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsSetShrinkerStepLimit(
spv_fuzzer_options options, uint32_t shrinker_step_limit);
// Enables running the validator after every pass is applied during a fuzzing
// run.
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsEnableFuzzerPassValidation(
spv_fuzzer_options options);
// Enables all fuzzer passes during a fuzzing run (instead of a random subset
// of passes).
SPIRV_TOOLS_EXPORT void spvFuzzerOptionsEnableAllPasses(
spv_fuzzer_options options);
// Encodes the given SPIR-V assembly text to its binary representation. The
// length parameter specifies the number of bytes for text. Encoded binary will
// be stored into *binary. Any error will be written into *diagnostic if
// diagnostic is non-null, otherwise the context's message consumer will be
// used. The generated binary is independent of the context and may outlive it.
// The SPIR-V binary version is set to the highest version of SPIR-V supported
// by the context's target environment.
SPIRV_TOOLS_EXPORT spv_result_t spvTextToBinary(const spv_const_context context,
const char* text,
const size_t length,
spv_binary* binary,
spv_diagnostic* diagnostic);
// Encodes the given SPIR-V assembly text to its binary representation. Same as
// spvTextToBinary but with options. The options parameter is a bit field of
// spv_text_to_binary_options_t.
SPIRV_TOOLS_EXPORT spv_result_t spvTextToBinaryWithOptions(
const spv_const_context context, const char* text, const size_t length,
const uint32_t options, spv_binary* binary, spv_diagnostic* diagnostic);
// Frees an allocated text stream. This is a no-op if the text parameter
// is a null pointer.
SPIRV_TOOLS_EXPORT void spvTextDestroy(spv_text text);
// Decodes the given SPIR-V binary representation to its assembly text. The
// word_count parameter specifies the number of words for binary. The options
// parameter is a bit field of spv_binary_to_text_options_t. Decoded text will
// be stored into *text. Any error will be written into *diagnostic if
// diagnostic is non-null, otherwise the context's message consumer will be
// used.
SPIRV_TOOLS_EXPORT spv_result_t spvBinaryToText(const spv_const_context context,
const uint32_t* binary,
const size_t word_count,
const uint32_t options,
spv_text* text,
spv_diagnostic* diagnostic);
// Frees a binary stream from memory. This is a no-op if binary is a null
// pointer.
SPIRV_TOOLS_EXPORT void spvBinaryDestroy(spv_binary binary);
// Validates a SPIR-V binary for correctness. Any errors will be written into
// *diagnostic if diagnostic is non-null, otherwise the context's message
// consumer will be used.
//
// Validate for SPIR-V spec rules for the SPIR-V version named in the
// binary's header (at word offset 1). Additionally, if the context target
// environment is a client API (such as Vulkan 1.1), then validate for that
// client API version, to the extent that it is verifiable from data in the
// binary itself.
SPIRV_TOOLS_EXPORT spv_result_t spvValidate(const spv_const_context context,
const spv_const_binary binary,
spv_diagnostic* diagnostic);
// Validates a SPIR-V binary for correctness. Uses the provided Validator
// options. Any errors will be written into *diagnostic if diagnostic is
// non-null, otherwise the context's message consumer will be used.
//
// Validate for SPIR-V spec rules for the SPIR-V version named in the
// binary's header (at word offset 1). Additionally, if the context target
// environment is a client API (such as Vulkan 1.1), then validate for that
// client API version, to the extent that it is verifiable from data in the
// binary itself, or in the validator options.
SPIRV_TOOLS_EXPORT spv_result_t spvValidateWithOptions(
const spv_const_context context, const spv_const_validator_options options,
const spv_const_binary binary, spv_diagnostic* diagnostic);
// Validates a raw SPIR-V binary for correctness. Any errors will be written
// into *diagnostic if diagnostic is non-null, otherwise the context's message
// consumer will be used.
SPIRV_TOOLS_EXPORT spv_result_t
spvValidateBinary(const spv_const_context context, const uint32_t* words,
const size_t num_words, spv_diagnostic* diagnostic);
// Creates a diagnostic object. The position parameter specifies the location in
// the text/binary stream. The message parameter, copied into the diagnostic
// object, contains the error message to display.
SPIRV_TOOLS_EXPORT spv_diagnostic
spvDiagnosticCreate(const spv_position position, const char* message);
// Destroys a diagnostic object. This is a no-op if diagnostic is a null
// pointer.
SPIRV_TOOLS_EXPORT void spvDiagnosticDestroy(spv_diagnostic diagnostic);
// Prints the diagnostic to stderr.
SPIRV_TOOLS_EXPORT spv_result_t
spvDiagnosticPrint(const spv_diagnostic diagnostic);
// Gets the name of an instruction, without the "Op" prefix.
SPIRV_TOOLS_EXPORT const char* spvOpcodeString(const uint32_t opcode);
// The binary parser interface.
// A pointer to a function that accepts a parsed SPIR-V header.
// The integer arguments are the 32-bit words from the header, as specified
// in SPIR-V 1.0 Section 2.3 Table 1.
// The function should return SPV_SUCCESS if parsing should continue.
typedef spv_result_t (*spv_parsed_header_fn_t)(
void* user_data, spv_endianness_t endian, uint32_t magic, uint32_t version,
uint32_t generator, uint32_t id_bound, uint32_t reserved);
// A pointer to a function that accepts a parsed SPIR-V instruction.
// The parsed_instruction value is transient: it may be overwritten
// or released immediately after the function has returned. That also
// applies to the words array member of the parsed instruction. The
// function should return SPV_SUCCESS if and only if parsing should
// continue.
typedef spv_result_t (*spv_parsed_instruction_fn_t)(
void* user_data, const spv_parsed_instruction_t* parsed_instruction);
// Parses a SPIR-V binary, specified as counted sequence of 32-bit words.
// Parsing feedback is provided via two callbacks provided as function
// pointers. Each callback function pointer can be a null pointer, in
// which case it is never called. Otherwise, in a valid parse the
// parsed-header callback is called once, and then the parsed-instruction
// callback once for each instruction in the stream. The user_data parameter
// is supplied as context to the callbacks. Returns SPV_SUCCESS on successful
// parse where the callbacks always return SPV_SUCCESS. For an invalid parse,
// returns a status code other than SPV_SUCCESS, and if diagnostic is non-null
// also emits a diagnostic. If diagnostic is null the context's message consumer
// will be used to emit any errors. If a callback returns anything other than
// SPV_SUCCESS, then that status code is returned, no further callbacks are
// issued, and no additional diagnostics are emitted.
SPIRV_TOOLS_EXPORT spv_result_t spvBinaryParse(
const spv_const_context context, void* user_data, const uint32_t* words,
const size_t num_words, spv_parsed_header_fn_t parse_header,
spv_parsed_instruction_fn_t parse_instruction, spv_diagnostic* diagnostic);
// The optimizer interface.
// A pointer to a function that accepts a log message from an optimizer.
typedef void (*spv_message_consumer)(
spv_message_level_t, const char*, const spv_position_t*, const char*);
// Creates and returns an optimizer object. This object must be passed to
// optimizer APIs below and is valid until passed to spvOptimizerDestroy.
SPIRV_TOOLS_EXPORT spv_optimizer_t* spvOptimizerCreate(spv_target_env env);
// Destroys the given optimizer object.
SPIRV_TOOLS_EXPORT void spvOptimizerDestroy(spv_optimizer_t* optimizer);
// Sets an spv_message_consumer on an optimizer object.
SPIRV_TOOLS_EXPORT void spvOptimizerSetMessageConsumer(
spv_optimizer_t* optimizer, spv_message_consumer consumer);
// Registers passes that attempt to legalize the generated code.
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterLegalizationPasses(
spv_optimizer_t* optimizer);
// Registers passes that attempt to improve performance of generated code.
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterPerformancePasses(
spv_optimizer_t* optimizer);
// Registers passes that attempt to improve the size of generated code.
SPIRV_TOOLS_EXPORT void spvOptimizerRegisterSizePasses(
spv_optimizer_t* optimizer);
// Registers a pass specified by a flag in an optimizer object.
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassFromFlag(
spv_optimizer_t* optimizer, const char* flag);
// Registers passes specified by length number of flags in an optimizer object.
SPIRV_TOOLS_EXPORT bool spvOptimizerRegisterPassesFromFlags(
spv_optimizer_t* optimizer, const char** flags, const size_t flag_count);
// Optimizes the SPIR-V code of size |word_count| pointed to by |binary| and
// returns an optimized spv_binary in |optimized_binary|.
//
// Returns SPV_SUCCESS on successful optimization, whether or not the module is
// modified. Returns an SPV_ERROR_* if the module fails to validate or if
// errors occur when processing using any of the registered passes. In that
// case, no further passes are executed and the |optimized_binary| contents may
// be invalid.
//
// By default, the binary is validated before any transforms are performed,
// and optionally after each transform. Validation uses SPIR-V spec rules
// for the SPIR-V version named in the binary's header (at word offset 1).
// Additionally, if the target environment is a client API (such as
// Vulkan 1.1), then validate for that client API version, to the extent
// that it is verifiable from data in the binary itself, or from the
// validator options set on the optimizer options.
SPIRV_TOOLS_EXPORT spv_result_t spvOptimizerRun(
spv_optimizer_t* optimizer, const uint32_t* binary, const size_t word_count,
spv_binary* optimized_binary, const spv_optimizer_options options);
#ifdef __cplusplus
}
#endif
#endif // INCLUDE_SPIRV_TOOLS_LIBSPIRV_H_

397
include/vulkan/spirv-tools/libspirv.hpp Normal file
View file

@ -0,0 +1,397 @@
// Copyright (c) 2016 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef INCLUDE_SPIRV_TOOLS_LIBSPIRV_HPP_
#define INCLUDE_SPIRV_TOOLS_LIBSPIRV_HPP_
#include <functional>
#include <memory>
#include <string>
#include <vector>
#include "spirv-tools/libspirv.h"
namespace spvtools {
// Message consumer. The C strings for source and message are only alive for the
// specific invocation.
using MessageConsumer = std::function<void(
spv_message_level_t /* level */, const char* /* source */,
const spv_position_t& /* position */, const char* /* message */
)>;
using HeaderParser = std::function<spv_result_t(
const spv_endianness_t endianess, const spv_parsed_header_t& instruction)>;
using InstructionParser =
std::function<spv_result_t(const spv_parsed_instruction_t& instruction)>;
// C++ RAII wrapper around the C context object spv_context.
class Context {
public:
// Constructs a context targeting the given environment |env|.
//
// See specific API calls for how the target environment is interpreted
// (particularly assembly and validation).
//
// The constructed instance will have an empty message consumer, which just
// ignores all messages from the library. Use SetMessageConsumer() to supply
// one if messages are of concern.
explicit Context(spv_target_env env);
// Enables move constructor/assignment operations.
Context(Context&& other);
Context& operator=(Context&& other);
// Disables copy constructor/assignment operations.
Context(const Context&) = delete;
Context& operator=(const Context&) = delete;
// Destructs this instance.
~Context();
// Sets the message consumer to the given |consumer|. The |consumer| will be
// invoked once for each message communicated from the library.
void SetMessageConsumer(MessageConsumer consumer);
// Returns the underlying spv_context.
spv_context& CContext();
const spv_context& CContext() const;
private:
spv_context context_;
};
// A RAII wrapper around a validator options object.
class ValidatorOptions {
public:
ValidatorOptions() : options_(spvValidatorOptionsCreate()) {}
~ValidatorOptions() { spvValidatorOptionsDestroy(options_); }
// Allow implicit conversion to the underlying object.
operator spv_validator_options() const { return options_; }
// Sets a limit.
void SetUniversalLimit(spv_validator_limit limit_type, uint32_t limit) {
spvValidatorOptionsSetUniversalLimit(options_, limit_type, limit);
}
void SetRelaxStructStore(bool val) {
spvValidatorOptionsSetRelaxStoreStruct(options_, val);
}
// Enables VK_KHR_relaxed_block_layout when validating standard
// uniform/storage buffer/push-constant layout. If true, disables
// scalar block layout rules.
void SetRelaxBlockLayout(bool val) {
spvValidatorOptionsSetRelaxBlockLayout(options_, val);
}
// Enables VK_KHR_uniform_buffer_standard_layout when validating standard
// uniform layout. If true, disables scalar block layout rules.
void SetUniformBufferStandardLayout(bool val) {
spvValidatorOptionsSetUniformBufferStandardLayout(options_, val);
}
// Enables VK_EXT_scalar_block_layout when validating standard
// uniform/storage buffer/push-constant layout. If true, disables
// relaxed block layout rules.
void SetScalarBlockLayout(bool val) {
spvValidatorOptionsSetScalarBlockLayout(options_, val);
}
// Enables scalar layout when validating Workgroup blocks. See
// VK_KHR_workgroup_memory_explicit_layout.
void SetWorkgroupScalarBlockLayout(bool val) {
spvValidatorOptionsSetWorkgroupScalarBlockLayout(options_, val);
}
// Skips validating standard uniform/storage buffer/push-constant layout.
void SetSkipBlockLayout(bool val) {
spvValidatorOptionsSetSkipBlockLayout(options_, val);
}
// Enables LocalSizeId decorations where the environment would not otherwise
// allow them.
void SetAllowLocalSizeId(bool val) {
spvValidatorOptionsSetAllowLocalSizeId(options_, val);
}
// Records whether or not the validator should relax the rules on pointer
// usage in logical addressing mode.
//
// When relaxed, it will allow the following usage cases of pointers:
// 1) OpVariable allocating an object whose type is a pointer type
// 2) OpReturnValue returning a pointer value
void SetRelaxLogicalPointer(bool val) {
spvValidatorOptionsSetRelaxLogicalPointer(options_, val);
}
// Records whether or not the validator should relax the rules because it is
// expected that the optimizations will make the code legal.
//
// When relaxed, it will allow the following:
// 1) It will allow relaxed logical pointers. Setting this option will also
// set that option.
// 2) Pointers that are pass as parameters to function calls do not have to
// match the storage class of the formal parameter.
// 3) Pointers that are actual parameters on function calls do not have to
// point to the same type pointed as the formal parameter. The types just
// need to logically match.
// 4) GLSLstd450 Interpolate* instructions can have a load of an interpolant
// for a first argument.
void SetBeforeHlslLegalization(bool val) {
spvValidatorOptionsSetBeforeHlslLegalization(options_, val);
}
// Whether friendly names should be used in validation error messages.
void SetFriendlyNames(bool val) {
spvValidatorOptionsSetFriendlyNames(options_, val);
}
private:
spv_validator_options options_;
};
// A C++ wrapper around an optimization options object.
class OptimizerOptions {
public:
OptimizerOptions() : options_(spvOptimizerOptionsCreate()) {}
~OptimizerOptions() { spvOptimizerOptionsDestroy(options_); }
// Allow implicit conversion to the underlying object.
operator spv_optimizer_options() const { return options_; }
// Records whether or not the optimizer should run the validator before
// optimizing. If |run| is true, the validator will be run.
void set_run_validator(bool run) {
spvOptimizerOptionsSetRunValidator(options_, run);
}
// Records the validator options that should be passed to the validator if it
// is run.
void set_validator_options(const ValidatorOptions& val_options) {
spvOptimizerOptionsSetValidatorOptions(options_, val_options);
}
// Records the maximum possible value for the id bound.
void set_max_id_bound(uint32_t new_bound) {
spvOptimizerOptionsSetMaxIdBound(options_, new_bound);
}
// Records whether all bindings within the module should be preserved.
void set_preserve_bindings(bool preserve_bindings) {
spvOptimizerOptionsSetPreserveBindings(options_, preserve_bindings);
}
// Records whether all specialization constants within the module
// should be preserved.
void set_preserve_spec_constants(bool preserve_spec_constants) {
spvOptimizerOptionsSetPreserveSpecConstants(options_,
preserve_spec_constants);
}
private:
spv_optimizer_options options_;
};
// A C++ wrapper around a reducer options object.
class ReducerOptions {
public:
ReducerOptions() : options_(spvReducerOptionsCreate()) {}
~ReducerOptions() { spvReducerOptionsDestroy(options_); }
// Allow implicit conversion to the underlying object.
operator spv_reducer_options() const { // NOLINT(google-explicit-constructor)
return options_;
}
// See spvReducerOptionsSetStepLimit.
void set_step_limit(uint32_t step_limit) {
spvReducerOptionsSetStepLimit(options_, step_limit);
}
// See spvReducerOptionsSetFailOnValidationError.
void set_fail_on_validation_error(bool fail_on_validation_error) {
spvReducerOptionsSetFailOnValidationError(options_,
fail_on_validation_error);
}
// See spvReducerOptionsSetTargetFunction.
void set_target_function(uint32_t target_function) {
spvReducerOptionsSetTargetFunction(options_, target_function);
}
private:
spv_reducer_options options_;
};
// A C++ wrapper around a fuzzer options object.
class FuzzerOptions {
public:
FuzzerOptions() : options_(spvFuzzerOptionsCreate()) {}
~FuzzerOptions() { spvFuzzerOptionsDestroy(options_); }
// Allow implicit conversion to the underlying object.
operator spv_fuzzer_options() const { // NOLINT(google-explicit-constructor)
return options_;
}
// See spvFuzzerOptionsEnableReplayValidation.
void enable_replay_validation() {
spvFuzzerOptionsEnableReplayValidation(options_);
}
// See spvFuzzerOptionsSetRandomSeed.
void set_random_seed(uint32_t seed) {
spvFuzzerOptionsSetRandomSeed(options_, seed);
}
// See spvFuzzerOptionsSetReplayRange.
void set_replay_range(int32_t replay_range) {
spvFuzzerOptionsSetReplayRange(options_, replay_range);
}
// See spvFuzzerOptionsSetShrinkerStepLimit.
void set_shrinker_step_limit(uint32_t shrinker_step_limit) {
spvFuzzerOptionsSetShrinkerStepLimit(options_, shrinker_step_limit);
}
// See spvFuzzerOptionsEnableFuzzerPassValidation.
void enable_fuzzer_pass_validation() {
spvFuzzerOptionsEnableFuzzerPassValidation(options_);
}
// See spvFuzzerOptionsEnableAllPasses.
void enable_all_passes() { spvFuzzerOptionsEnableAllPasses(options_); }
private:
spv_fuzzer_options options_;
};
// C++ interface for SPIRV-Tools functionalities. It wraps the context
// (including target environment and the corresponding SPIR-V grammar) and
// provides methods for assembling, disassembling, and validating.
//
// Instances of this class provide basic thread-safety guarantee.
class SpirvTools {
public:
enum {
// Default assembling option used by assemble():
kDefaultAssembleOption = SPV_TEXT_TO_BINARY_OPTION_NONE,
// Default disassembling option used by Disassemble():
// * Avoid prefix comments from decoding the SPIR-V module header, and
// * Use friendly names for variables.
kDefaultDisassembleOption = SPV_BINARY_TO_TEXT_OPTION_NO_HEADER |
SPV_BINARY_TO_TEXT_OPTION_FRIENDLY_NAMES
};
// Constructs an instance targeting the given environment |env|.
//
// The constructed instance will have an empty message consumer, which just
// ignores all messages from the library. Use SetMessageConsumer() to supply
// one if messages are of concern.
explicit SpirvTools(spv_target_env env);
// Disables copy/move constructor/assignment operations.
SpirvTools(const SpirvTools&) = delete;
SpirvTools(SpirvTools&&) = delete;
SpirvTools& operator=(const SpirvTools&) = delete;
SpirvTools& operator=(SpirvTools&&) = delete;
// Destructs this instance.
~SpirvTools();
// Sets the message consumer to the given |consumer|. The |consumer| will be
// invoked once for each message communicated from the library.
void SetMessageConsumer(MessageConsumer consumer);
// Assembles the given assembly |text| and writes the result to |binary|.
// Returns true on successful assembling. |binary| will be kept untouched if
// assembling is unsuccessful.
// The SPIR-V binary version is set to the highest version of SPIR-V supported
// by the target environment with which this SpirvTools object was created.
bool Assemble(const std::string& text, std::vector<uint32_t>* binary,
uint32_t options = kDefaultAssembleOption) const;
// |text_size| specifies the number of bytes in |text|. A terminating null
// character is not required to present in |text| as long as |text| is valid.
// The SPIR-V binary version is set to the highest version of SPIR-V supported
// by the target environment with which this SpirvTools object was created.
bool Assemble(const char* text, size_t text_size,
std::vector<uint32_t>* binary,
uint32_t options = kDefaultAssembleOption) const;
// Disassembles the given SPIR-V |binary| with the given |options| and writes
// the assembly to |text|. Returns true on successful disassembling. |text|
// will be kept untouched if diassembling is unsuccessful.
bool Disassemble(const std::vector<uint32_t>& binary, std::string* text,
uint32_t options = kDefaultDisassembleOption) const;
// |binary_size| specifies the number of words in |binary|.
bool Disassemble(const uint32_t* binary, size_t binary_size,
std::string* text,
uint32_t options = kDefaultDisassembleOption) const;
// Parses a SPIR-V binary, specified as counted sequence of 32-bit words.
// Parsing feedback is provided via two callbacks provided as std::function.
// In a valid parse the parsed-header callback is called once, and
// then the parsed-instruction callback is called once for each instruction
// in the stream.
// Returns true on successful parsing.
// If diagnostic is non-null, a diagnostic is emitted on failed parsing.
// If diagnostic is null the context's message consumer
// will be used to emit any errors. If a callback returns anything other than
// SPV_SUCCESS, then that status code is returned, no further callbacks are
// issued, and no additional diagnostics are emitted.
// This is a wrapper around the C API spvBinaryParse.
bool Parse(const std::vector<uint32_t>& binary,
const HeaderParser& header_parser,
const InstructionParser& instruction_parser,
spv_diagnostic* diagnostic = nullptr);
// Validates the given SPIR-V |binary|. Returns true if no issues are found.
// Otherwise, returns false and communicates issues via the message consumer
// registered.
// Validates for SPIR-V spec rules for the SPIR-V version named in the
// binary's header (at word offset 1). Additionally, if the target
// environment is a client API (such as Vulkan 1.1), then validate for that
// client API version, to the extent that it is verifiable from data in the
// binary itself.
bool Validate(const std::vector<uint32_t>& binary) const;
// Like the previous overload, but provides the binary as a pointer and size:
// |binary_size| specifies the number of words in |binary|.
// Validates for SPIR-V spec rules for the SPIR-V version named in the
// binary's header (at word offset 1). Additionally, if the target
// environment is a client API (such as Vulkan 1.1), then validate for that
// client API version, to the extent that it is verifiable from data in the
// binary itself.
bool Validate(const uint32_t* binary, size_t binary_size) const;
// Like the previous overload, but takes an options object.
// Validates for SPIR-V spec rules for the SPIR-V version named in the
// binary's header (at word offset 1). Additionally, if the target
// environment is a client API (such as Vulkan 1.1), then validate for that
// client API version, to the extent that it is verifiable from data in the
// binary itself, or in the validator options.
bool Validate(const uint32_t* binary, size_t binary_size,
spv_validator_options options) const;
// Was this object successfully constructed.
bool IsValid() const;
private:
struct Impl; // Opaque struct for holding the data fields used by this class.
std::unique_ptr<Impl> impl_; // Unique pointer to implementation data.
};
} // namespace spvtools
#endif // INCLUDE_SPIRV_TOOLS_LIBSPIRV_HPP_

97
include/vulkan/spirv-tools/linker.hpp Normal file
View file

@ -0,0 +1,97 @@
// Copyright (c) 2017 Pierre Moreau
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef INCLUDE_SPIRV_TOOLS_LINKER_HPP_
#define INCLUDE_SPIRV_TOOLS_LINKER_HPP_
#include <cstdint>
#include <memory>
#include <vector>
#include "libspirv.hpp"
namespace spvtools {
class LinkerOptions {
public:
LinkerOptions()
: create_library_(false),
verify_ids_(false),
allow_partial_linkage_(false) {}
// Returns whether a library or an executable should be produced by the
// linking phase.
//
// All exported symbols are kept when creating a library, whereas they will
// be removed when creating an executable.
// The returned value will be true if creating a library, and false if
// creating an executable.
bool GetCreateLibrary() const { return create_library_; }
// Sets whether a library or an executable should be produced.
void SetCreateLibrary(bool create_library) {
create_library_ = create_library;
}
// Returns whether to verify the uniqueness of the unique ids in the merged
// context.
bool GetVerifyIds() const { return verify_ids_; }
// Sets whether to verify the uniqueness of the unique ids in the merged
// context.
void SetVerifyIds(bool verify_ids) { verify_ids_ = verify_ids; }
// Returns whether to allow for imported symbols to have no corresponding
// exported symbols
bool GetAllowPartialLinkage() const { return allow_partial_linkage_; }
// Sets whether to allow for imported symbols to have no corresponding
// exported symbols
void SetAllowPartialLinkage(bool allow_partial_linkage) {
allow_partial_linkage_ = allow_partial_linkage;
}
private:
bool create_library_;
bool verify_ids_;
bool allow_partial_linkage_;
};
// Links one or more SPIR-V modules into a new SPIR-V module. That is, combine
// several SPIR-V modules into one, resolving link dependencies between them.
//
// At least one binary has to be provided in |binaries|. Those binaries do not
// have to be valid, but they should be at least parseable.
// The functions can fail due to the following:
// * The given context was not initialised using `spvContextCreate()`;
// * No input modules were given;
// * One or more of those modules were not parseable;
// * The input modules used different addressing or memory models;
// * The ID or global variable number limit were exceeded;
// * Some entry points were defined multiple times;
// * Some imported symbols did not have an exported counterpart;
// * Possibly other reasons.
spv_result_t Link(const Context& context,
const std::vector<std::vector<uint32_t>>& binaries,
std::vector<uint32_t>* linked_binary,
const LinkerOptions& options = LinkerOptions());
spv_result_t Link(const Context& context, const uint32_t* const* binaries,
const size_t* binary_sizes, size_t num_binaries,
std::vector<uint32_t>* linked_binary,
const LinkerOptions& options = LinkerOptions());
} // namespace spvtools
#endif // INCLUDE_SPIRV_TOOLS_LINKER_HPP_

976
include/vulkan/spirv-tools/optimizer.hpp Normal file
View file

@ -0,0 +1,976 @@
// Copyright (c) 2016 Google Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef INCLUDE_SPIRV_TOOLS_OPTIMIZER_HPP_
#define INCLUDE_SPIRV_TOOLS_OPTIMIZER_HPP_
#include <memory>
#include <ostream>
#include <string>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#include "libspirv.hpp"
namespace spvtools {
namespace opt {
class Pass;
struct DescriptorSetAndBinding;
} // namespace opt
// C++ interface for SPIR-V optimization functionalities. It wraps the context
// (including target environment and the corresponding SPIR-V grammar) and
// provides methods for registering optimization passes and optimizing.
//
// Instances of this class provides basic thread-safety guarantee.
class Optimizer {
public:
// The token for an optimization pass. It is returned via one of the
// Create*Pass() standalone functions at the end of this header file and
// consumed by the RegisterPass() method. Tokens are one-time objects that
// only support move; copying is not allowed.
struct PassToken {
struct Impl; // Opaque struct for holding internal data.
PassToken(std::unique_ptr<Impl>);
// Tokens for built-in passes should be created using Create*Pass functions
// below; for out-of-tree passes, use this constructor instead.
// Note that this API isn't guaranteed to be stable and may change without
// preserving source or binary compatibility in the future.
PassToken(std::unique_ptr<opt::Pass>&& pass);
// Tokens can only be moved. Copying is disabled.
PassToken(const PassToken&) = delete;
PassToken(PassToken&&);
PassToken& operator=(const PassToken&) = delete;
PassToken& operator=(PassToken&&);
~PassToken();
std::unique_ptr<Impl> impl_; // Unique pointer to internal data.
};
// Constructs an instance with the given target |env|, which is used to decode
// the binaries to be optimized later.
//
// The instance will have an empty message consumer, which ignores all
// messages from the library. Use SetMessageConsumer() to supply a consumer
// if messages are of concern.
explicit Optimizer(spv_target_env env);
// Disables copy/move constructor/assignment operations.
Optimizer(const Optimizer&) = delete;
Optimizer(Optimizer&&) = delete;
Optimizer& operator=(const Optimizer&) = delete;
Optimizer& operator=(Optimizer&&) = delete;
// Destructs this instance.
~Optimizer();
// Sets the message consumer to the given |consumer|. The |consumer| will be
// invoked once for each message communicated from the library.
void SetMessageConsumer(MessageConsumer consumer);
// Returns a reference to the registered message consumer.
const MessageConsumer& consumer() const;
// Registers the given |pass| to this optimizer. Passes will be run in the
// exact order of registration. The token passed in will be consumed by this
// method.
Optimizer& RegisterPass(PassToken&& pass);
// Registers passes that attempt to improve performance of generated code.
// This sequence of passes is subject to constant review and will change
// from time to time.
Optimizer& RegisterPerformancePasses();
// Registers passes that attempt to improve the size of generated code.
// This sequence of passes is subject to constant review and will change
// from time to time.
Optimizer& RegisterSizePasses();
// Registers passes that attempt to legalize the generated code.
//
// Note: this recipe is specially designed for legalizing SPIR-V. It should be
// used by compilers after translating HLSL source code literally. It should
// *not* be used by general workloads for performance or size improvement.
//
// This sequence of passes is subject to constant review and will change
// from time to time.
Optimizer& RegisterLegalizationPasses();
// Register passes specified in the list of |flags|. Each flag must be a
// string of a form accepted by Optimizer::FlagHasValidForm().
//
// If the list of flags contains an invalid entry, it returns false and an
// error message is emitted to the MessageConsumer object (use
// Optimizer::SetMessageConsumer to define a message consumer, if needed).
//
// If all the passes are registered successfully, it returns true.
bool RegisterPassesFromFlags(const std::vector<std::string>& flags);
// Registers the optimization pass associated with |flag|. This only accepts
// |flag| values of the form "--pass_name[=pass_args]". If no such pass
// exists, it returns false. Otherwise, the pass is registered and it returns
// true.
//
// The following flags have special meaning:
//
// -O: Registers all performance optimization passes
// (Optimizer::RegisterPerformancePasses)
//
// -Os: Registers all size optimization passes
// (Optimizer::RegisterSizePasses).
//
// --legalize-hlsl: Registers all passes that legalize SPIR-V generated by an
// HLSL front-end.
bool RegisterPassFromFlag(const std::string& flag);
// Validates that |flag| has a valid format. Strings accepted:
//
// --pass_name[=pass_args]
// -O
// -Os
//
// If |flag| takes one of the forms above, it returns true. Otherwise, it
// returns false.
bool FlagHasValidForm(const std::string& flag) const;
// Allows changing, after creation time, the target environment to be
// optimized for and validated. Should be called before calling Run().
void SetTargetEnv(const spv_target_env env);
// Optimizes the given SPIR-V module |original_binary| and writes the
// optimized binary into |optimized_binary|. The optimized binary uses
// the same SPIR-V version as the original binary.
//
// Returns true on successful optimization, whether or not the module is
// modified. Returns false if |original_binary| fails to validate or if errors
// occur when processing |original_binary| using any of the registered passes.
// In that case, no further passes are executed and the contents in
// |optimized_binary| may be invalid.
//
// By default, the binary is validated before any transforms are performed,
// and optionally after each transform. Validation uses SPIR-V spec rules
// for the SPIR-V version named in the binary's header (at word offset 1).
// Additionally, if the target environment is a client API (such as
// Vulkan 1.1), then validate for that client API version, to the extent
// that it is verifiable from data in the binary itself.
//
// It's allowed to alias |original_binary| to the start of |optimized_binary|.
bool Run(const uint32_t* original_binary, size_t original_binary_size,
std::vector<uint32_t>* optimized_binary) const;
// DEPRECATED: Same as above, except passes |options| to the validator when
// trying to validate the binary. If |skip_validation| is true, then the
// caller is guaranteeing that |original_binary| is valid, and the validator
// will not be run. The |max_id_bound| is the limit on the max id in the
// module.
bool Run(const uint32_t* original_binary, const size_t original_binary_size,
std::vector<uint32_t>* optimized_binary,
const ValidatorOptions& options, bool skip_validation) const;
// Same as above, except it takes an options object. See the documentation
// for |OptimizerOptions| to see which options can be set.
//
// By default, the binary is validated before any transforms are performed,
// and optionally after each transform. Validation uses SPIR-V spec rules
// for the SPIR-V version named in the binary's header (at word offset 1).
// Additionally, if the target environment is a client API (such as
// Vulkan 1.1), then validate for that client API version, to the extent
// that it is verifiable from data in the binary itself, or from the
// validator options set on the optimizer options.
bool Run(const uint32_t* original_binary, const size_t original_binary_size,
std::vector<uint32_t>* optimized_binary,
const spv_optimizer_options opt_options) const;
// Returns a vector of strings with all the pass names added to this
// optimizer's pass manager. These strings are valid until the associated
// pass manager is destroyed.
std::vector<const char*> GetPassNames() const;
// Sets the option to print the disassembly before each pass and after the
// last pass. If |out| is null, then no output is generated. Otherwise,
// output is sent to the |out| output stream.
Optimizer& SetPrintAll(std::ostream* out);
// Sets the option to print the resource utilization of each pass. If |out|
// is null, then no output is generated. Otherwise, output is sent to the
// |out| output stream.
Optimizer& SetTimeReport(std::ostream* out);
// Sets the option to validate the module after each pass.
Optimizer& SetValidateAfterAll(bool validate);
private:
struct Impl; // Opaque struct for holding internal data.
std::unique_ptr<Impl> impl_; // Unique pointer to internal data.
};
// Creates a null pass.
// A null pass does nothing to the SPIR-V module to be optimized.
Optimizer::PassToken CreateNullPass();
// Creates a strip-debug-info pass.
// A strip-debug-info pass removes all debug instructions (as documented in
// Section 3.42.2 of the SPIR-V spec) of the SPIR-V module to be optimized.
Optimizer::PassToken CreateStripDebugInfoPass();
// [Deprecated] This will create a strip-nonsemantic-info pass. See below.
Optimizer::PassToken CreateStripReflectInfoPass();
// Creates a strip-nonsemantic-info pass.
// A strip-nonsemantic-info pass removes all reflections and explicitly
// non-semantic instructions.
Optimizer::PassToken CreateStripNonSemanticInfoPass();
// Creates an eliminate-dead-functions pass.
// An eliminate-dead-functions pass will remove all functions that are not in
// the call trees rooted at entry points and exported functions. These
// functions are not needed because they will never be called.
Optimizer::PassToken CreateEliminateDeadFunctionsPass();
// Creates an eliminate-dead-members pass.
// An eliminate-dead-members pass will remove all unused members of structures.
// This will not affect the data layout of the remaining members.
Optimizer::PassToken CreateEliminateDeadMembersPass();
// Creates a set-spec-constant-default-value pass from a mapping from spec-ids
// to the default values in the form of string.
// A set-spec-constant-default-value pass sets the default values for the
// spec constants that have SpecId decorations (i.e., those defined by
// OpSpecConstant{|True|False} instructions).
Optimizer::PassToken CreateSetSpecConstantDefaultValuePass(
const std::unordered_map<uint32_t, std::string>& id_value_map);
// Creates a set-spec-constant-default-value pass from a mapping from spec-ids
// to the default values in the form of bit pattern.
// A set-spec-constant-default-value pass sets the default values for the
// spec constants that have SpecId decorations (i.e., those defined by
// OpSpecConstant{|True|False} instructions).
Optimizer::PassToken CreateSetSpecConstantDefaultValuePass(
const std::unordered_map<uint32_t, std::vector<uint32_t>>& id_value_map);
// Creates a flatten-decoration pass.
// A flatten-decoration pass replaces grouped decorations with equivalent
// ungrouped decorations. That is, it replaces each OpDecorationGroup
// instruction and associated OpGroupDecorate and OpGroupMemberDecorate
// instructions with equivalent OpDecorate and OpMemberDecorate instructions.
// The pass does not attempt to preserve debug information for instructions
// it removes.
Optimizer::PassToken CreateFlattenDecorationPass();
// Creates a freeze-spec-constant-value pass.
// A freeze-spec-constant pass specializes the value of spec constants to
// their default values. This pass only processes the spec constants that have
// SpecId decorations (defined by OpSpecConstant, OpSpecConstantTrue, or
// OpSpecConstantFalse instructions) and replaces them with their normal
// counterparts (OpConstant, OpConstantTrue, or OpConstantFalse). The
// corresponding SpecId annotation instructions will also be removed. This
// pass does not fold the newly added normal constants and does not process
// other spec constants defined by OpSpecConstantComposite or
// OpSpecConstantOp.
Optimizer::PassToken CreateFreezeSpecConstantValuePass();
// Creates a fold-spec-constant-op-and-composite pass.
// A fold-spec-constant-op-and-composite pass folds spec constants defined by
// OpSpecConstantOp or OpSpecConstantComposite instruction, to normal Constants
// defined by OpConstantTrue, OpConstantFalse, OpConstant, OpConstantNull, or
// OpConstantComposite instructions. Note that spec constants defined with
// OpSpecConstant, OpSpecConstantTrue, or OpSpecConstantFalse instructions are
// not handled, as these instructions indicate their value are not determined
// and can be changed in future. A spec constant is foldable if all of its
// value(s) can be determined from the module. E.g., an integer spec constant
// defined with OpSpecConstantOp instruction can be folded if its value won't
// change later. This pass will replace the original OpSpecConstantOp
// instruction with an OpConstant instruction. When folding composite spec
// constants, new instructions may be inserted to define the components of the
// composite constant first, then the original spec constants will be replaced
// by OpConstantComposite instructions.
//
// There are some operations not supported yet:
// OpSConvert, OpFConvert, OpQuantizeToF16 and
// all the operations under Kernel capability.
// TODO(qining): Add support for the operations listed above.
Optimizer::PassToken CreateFoldSpecConstantOpAndCompositePass();
// Creates a unify-constant pass.
// A unify-constant pass de-duplicates the constants. Constants with the exact
// same value and identical form will be unified and only one constant will
// be kept for each unique pair of type and value.
// There are several cases not handled by this pass:
// 1) Constants defined by OpConstantNull instructions (null constants) and
// constants defined by OpConstantFalse, OpConstant or OpConstantComposite
// with value 0 (zero-valued normal constants) are not considered equivalent.
// So null constants won't be used to replace zero-valued normal constants,
// vice versa.
// 2) Whenever there are decorations to the constant's result id id, the
// constant won't be handled, which means, it won't be used to replace any
// other constants, neither can other constants replace it.
// 3) NaN in float point format with different bit patterns are not unified.
Optimizer::PassToken CreateUnifyConstantPass();
// Creates a eliminate-dead-constant pass.
// A eliminate-dead-constant pass removes dead constants, including normal
// constants defined by OpConstant, OpConstantComposite, OpConstantTrue, or
// OpConstantFalse and spec constants defined by OpSpecConstant,
// OpSpecConstantComposite, OpSpecConstantTrue, OpSpecConstantFalse or
// OpSpecConstantOp.
Optimizer::PassToken CreateEliminateDeadConstantPass();
// Creates a strength-reduction pass.
// A strength-reduction pass will look for opportunities to replace an
// instruction with an equivalent and less expensive one. For example,
// multiplying by a power of 2 can be replaced by a bit shift.
Optimizer::PassToken CreateStrengthReductionPass();
// Creates a block merge pass.
// This pass searches for blocks with a single Branch to a block with no
// other predecessors and merges the blocks into a single block. Continue
// blocks and Merge blocks are not candidates for the second block.
//
// The pass is most useful after Dead Branch Elimination, which can leave
// such sequences of blocks. Merging them makes subsequent passes more
// effective, such as single block local store-load elimination.
//
// While this pass reduces the number of occurrences of this sequence, at
// this time it does not guarantee all such sequences are eliminated.
//
// Presence of phi instructions can inhibit this optimization. Handling
// these is left for future improvements.
Optimizer::PassToken CreateBlockMergePass();
// Creates an exhaustive inline pass.
// An exhaustive inline pass attempts to exhaustively inline all function
// calls in all functions in an entry point call tree. The intent is to enable,
// albeit through brute force, analysis and optimization across function
// calls by subsequent optimization passes. As the inlining is exhaustive,
// there is no attempt to optimize for size or runtime performance. Functions
// that are not in the call tree of an entry point are not changed.
Optimizer::PassToken CreateInlineExhaustivePass();
// Creates an opaque inline pass.
// An opaque inline pass inlines all function calls in all functions in all
// entry point call trees where the called function contains an opaque type
// in either its parameter types or return type. An opaque type is currently
// defined as Image, Sampler or SampledImage. The intent is to enable, albeit
// through brute force, analysis and optimization across these function calls
// by subsequent passes in order to remove the storing of opaque types which is
// not legal in Vulkan. Functions that are not in the call tree of an entry
// point are not changed.
Optimizer::PassToken CreateInlineOpaquePass();
// Creates a single-block local variable load/store elimination pass.
// For every entry point function, do single block memory optimization of
// function variables referenced only with non-access-chain loads and stores.
// For each targeted variable load, if previous store to that variable in the
// block, replace the load's result id with the value id of the store.
// If previous load within the block, replace the current load's result id
// with the previous load's result id. In either case, delete the current
// load. Finally, check if any remaining stores are useless, and delete store
// and variable if possible.
//
// The presence of access chain references and function calls can inhibit
// the above optimization.
//
// Only modules with relaxed logical addressing (see opt/instruction.h) are
// currently processed.
//
// This pass is most effective if preceded by Inlining and
// LocalAccessChainConvert. This pass will reduce the work needed to be done
// by LocalSingleStoreElim and LocalMultiStoreElim.
//
// Only functions in the call tree of an entry point are processed.
Optimizer::PassToken CreateLocalSingleBlockLoadStoreElimPass();
// Create dead branch elimination pass.
// For each entry point function, this pass will look for SelectionMerge
// BranchConditionals with constant condition and convert to a Branch to
// the indicated label. It will delete resulting dead blocks.
//
// For all phi functions in merge block, replace all uses with the id
// corresponding to the living predecessor.
//
// Note that some branches and blocks may be left to avoid creating invalid
// control flow. Improving this is left to future work.
//
// This pass is most effective when preceded by passes which eliminate
// local loads and stores, effectively propagating constant values where
// possible.
Optimizer::PassToken CreateDeadBranchElimPass();
// Creates an SSA local variable load/store elimination pass.
// For every entry point function, eliminate all loads and stores of function
// scope variables only referenced with non-access-chain loads and stores.
// Eliminate the variables as well.
//
// The presence of access chain references and function calls can inhibit
// the above optimization.
//
// Only shader modules with relaxed logical addressing (see opt/instruction.h)
// are currently processed. Currently modules with any extensions enabled are
// not processed. This is left for future work.
//
// This pass is most effective if preceded by Inlining and
// LocalAccessChainConvert. LocalSingleStoreElim and LocalSingleBlockElim
// will reduce the work that this pass has to do.
Optimizer::PassToken CreateLocalMultiStoreElimPass();
// Creates a local access chain conversion pass.
// A local access chain conversion pass identifies all function scope
// variables which are accessed only with loads, stores and access chains
// with constant indices. It then converts all loads and stores of such
// variables into equivalent sequences of loads, stores, extracts and inserts.
//
// This pass only processes entry point functions. It currently only converts
// non-nested, non-ptr access chains. It does not process modules with
// non-32-bit integer types present. Optional memory access options on loads
// and stores are ignored as we are only processing function scope variables.
//
// This pass unifies access to these variables to a single mode and simplifies
// subsequent analysis and elimination of these variables along with their
// loads and stores allowing values to propagate to their points of use where
// possible.
Optimizer::PassToken CreateLocalAccessChainConvertPass();
// Creates a local single store elimination pass.
// For each entry point function, this pass eliminates loads and stores for
// function scope variable that are stored to only once, where possible. Only
// whole variable loads and stores are eliminated; access-chain references are
// not optimized. Replace all loads of such variables with the value that is
// stored and eliminate any resulting dead code.
//
// Currently, the presence of access chains and function calls can inhibit this
// pass, however the Inlining and LocalAccessChainConvert passes can make it
// more effective. In additional, many non-load/store memory operations are
// not supported and will prohibit optimization of a function. Support of
// these operations are future work.
//
// Only shader modules with relaxed logical addressing (see opt/instruction.h)
// are currently processed.
//
// This pass will reduce the work needed to be done by LocalSingleBlockElim
// and LocalMultiStoreElim and can improve the effectiveness of other passes
// such as DeadBranchElimination which depend on values for their analysis.
Optimizer::PassToken CreateLocalSingleStoreElimPass();
// Creates an insert/extract elimination pass.
// This pass processes each entry point function in the module, searching for
// extracts on a sequence of inserts. It further searches the sequence for an
// insert with indices identical to the extract. If such an insert can be
// found before hitting a conflicting insert, the extract's result id is
// replaced with the id of the values from the insert.
//
// Besides removing extracts this pass enables subsequent dead code elimination
// passes to delete the inserts. This pass performs best after access chains are
// converted to inserts and extracts and local loads and stores are eliminated.
Optimizer::PassToken CreateInsertExtractElimPass();
// Creates a dead insert elimination pass.
// This pass processes each entry point function in the module, searching for
// unreferenced inserts into composite types. These are most often unused
// stores to vector components. They are unused because they are never
// referenced, or because there is another insert to the same component between
// the insert and the reference. After removing the inserts, dead code
// elimination is attempted on the inserted values.
//
// This pass performs best after access chains are converted to inserts and
// extracts and local loads and stores are eliminated. While executing this
// pass can be advantageous on its own, it is also advantageous to execute
// this pass after CreateInsertExtractPass() as it will remove any unused
// inserts created by that pass.
Optimizer::PassToken CreateDeadInsertElimPass();
// Create aggressive dead code elimination pass
// This pass eliminates unused code from the module. In addition,
// it detects and eliminates code which may have spurious uses but which do
// not contribute to the output of the function. The most common cause of
// such code sequences is summations in loops whose result is no longer used
// due to dead code elimination. This optimization has additional compile
// time cost over standard dead code elimination.
//
// This pass only processes entry point functions. It also only processes
// shaders with relaxed logical addressing (see opt/instruction.h). It
// currently will not process functions with function calls. Unreachable
// functions are deleted.
//
// This pass will be made more effective by first running passes that remove
// dead control flow and inlines function calls.
//
// This pass can be especially useful after running Local Access Chain
// Conversion, which tends to cause cycles of dead code to be left after
// Store/Load elimination passes are completed. These cycles cannot be
// eliminated with standard dead code elimination.
//
// If |preserve_interface| is true, all non-io variables in the entry point
// interface are considered live and are not eliminated. This mode is needed
// by GPU-Assisted validation instrumentation, where a change in the interface
// is not allowed.
//
// If |remove_outputs| is true, allow outputs to be removed from the interface.
// This is only safe if the caller knows that there is no corresponding input
// variable in the following shader. It is false by default.
Optimizer::PassToken CreateAggressiveDCEPass();
Optimizer::PassToken CreateAggressiveDCEPass(bool preserve_interface);
Optimizer::PassToken CreateAggressiveDCEPass(bool preserve_interface,
bool remove_outputs);
// Creates a remove-unused-interface-variables pass.
// Removes variables referenced on the |OpEntryPoint| instruction that are not
// referenced in the entry point function or any function in its call tree. Note
// that this could cause the shader interface to no longer match other shader
// stages.
Optimizer::PassToken CreateRemoveUnusedInterfaceVariablesPass();
// Creates an empty pass.
// This is deprecated and will be removed.
// TODO(jaebaek): remove this pass after handling glslang's broken unit tests.
// https://github.com/KhronosGroup/glslang/pull/2440
Optimizer::PassToken CreatePropagateLineInfoPass();
// Creates an empty pass.
// This is deprecated and will be removed.
// TODO(jaebaek): remove this pass after handling glslang's broken unit tests.
// https://github.com/KhronosGroup/glslang/pull/2440
Optimizer::PassToken CreateRedundantLineInfoElimPass();
// Creates a compact ids pass.
// The pass remaps result ids to a compact and gapless range starting from %1.
Optimizer::PassToken CreateCompactIdsPass();
// Creates a remove duplicate pass.
// This pass removes various duplicates:
// * duplicate capabilities;
// * duplicate extended instruction imports;
// * duplicate types;
// * duplicate decorations.
Optimizer::PassToken CreateRemoveDuplicatesPass();
// Creates a CFG cleanup pass.
// This pass removes cruft from the control flow graph of functions that are
// reachable from entry points and exported functions. It currently includes the
// following functionality:
//
// - Removal of unreachable basic blocks.
Optimizer::PassToken CreateCFGCleanupPass();
// Create dead variable elimination pass.
// This pass will delete module scope variables, along with their decorations,
// that are not referenced.
Optimizer::PassToken CreateDeadVariableEliminationPass();
// create merge return pass.
// changes functions that have multiple return statements so they have a single
// return statement.
//
// for structured control flow it is assumed that the only unreachable blocks in
// the function are trivial merge and continue blocks.
//
// a trivial merge block contains the label and an opunreachable instructions,
// nothing else. a trivial continue block contain a label and an opbranch to
// the header, nothing else.
//
// these conditions are guaranteed to be met after running dead-branch
// elimination.
Optimizer::PassToken CreateMergeReturnPass();
// Create value numbering pass.
// This pass will look for instructions in the same basic block that compute the
// same value, and remove the redundant ones.
Optimizer::PassToken CreateLocalRedundancyEliminationPass();
// Create LICM pass.
// This pass will look for invariant instructions inside loops and hoist them to
// the loops preheader.
Optimizer::PassToken CreateLoopInvariantCodeMotionPass();
// Creates a loop fission pass.
// This pass will split all top level loops whose register pressure exceedes the
// given |threshold|.
Optimizer::PassToken CreateLoopFissionPass(size_t threshold);
// Creates a loop fusion pass.
// This pass will look for adjacent loops that are compatible and legal to be
// fused. The fuse all such loops as long as the register usage for the fused
// loop stays under the threshold defined by |max_registers_per_loop|.
Optimizer::PassToken CreateLoopFusionPass(size_t max_registers_per_loop);
// Creates a loop peeling pass.
// This pass will look for conditions inside a loop that are true or false only
// for the N first or last iteration. For loop with such condition, those N
// iterations of the loop will be executed outside of the main loop.
// To limit code size explosion, the loop peeling can only happen if the code
// size growth for each loop is under |code_growth_threshold|.
Optimizer::PassToken CreateLoopPeelingPass();
// Creates a loop unswitch pass.
// This pass will look for loop independent branch conditions and move the
// condition out of the loop and version the loop based on the taken branch.
// Works best after LICM and local multi store elimination pass.
Optimizer::PassToken CreateLoopUnswitchPass();
// Create global value numbering pass.
// This pass will look for instructions where the same value is computed on all
// paths leading to the instruction. Those instructions are deleted.
Optimizer::PassToken CreateRedundancyEliminationPass();
// Create scalar replacement pass.
// This pass replaces composite function scope variables with variables for each
// element if those elements are accessed individually. The parameter is a
// limit on the number of members in the composite variable that the pass will
// consider replacing.
Optimizer::PassToken CreateScalarReplacementPass(uint32_t size_limit = 100);
// Create a private to local pass.
// This pass looks for variables declared in the private storage class that are
// used in only one function. Those variables are moved to the function storage
// class in the function that they are used.
Optimizer::PassToken CreatePrivateToLocalPass();
// Creates a conditional constant propagation (CCP) pass.
// This pass implements the SSA-CCP algorithm in
//
// Constant propagation with conditional branches,
// Wegman and Zadeck, ACM TOPLAS 13(2):181-210.
//
// Constant values in expressions and conditional jumps are folded and
// simplified. This may reduce code size by removing never executed jump targets
// and computations with constant operands.
Optimizer::PassToken CreateCCPPass();
// Creates a workaround driver bugs pass. This pass attempts to work around
// a known driver bug (issue #1209) by identifying the bad code sequences and
// rewriting them.
//
// Current workaround: Avoid OpUnreachable instructions in loops.
Optimizer::PassToken CreateWorkaround1209Pass();
// Creates a pass that converts if-then-else like assignments into OpSelect.
Optimizer::PassToken CreateIfConversionPass();
// Creates a pass that will replace instructions that are not valid for the
// current shader stage by constants. Has no effect on non-shader modules.
Optimizer::PassToken CreateReplaceInvalidOpcodePass();
// Creates a pass that simplifies instructions using the instruction folder.
Optimizer::PassToken CreateSimplificationPass();
// Create loop unroller pass.
// Creates a pass to unroll loops which have the "Unroll" loop control
// mask set. The loops must meet a specific criteria in order to be unrolled
// safely this criteria is checked before doing the unroll by the
// LoopUtils::CanPerformUnroll method. Any loop that does not meet the criteria
// won't be unrolled. See CanPerformUnroll LoopUtils.h for more information.
Optimizer::PassToken CreateLoopUnrollPass(bool fully_unroll, int factor = 0);
// Create the SSA rewrite pass.
// This pass converts load/store operations on function local variables into
// operations on SSA IDs. This allows SSA optimizers to act on these variables.
// Only variables that are local to the function and of supported types are
// processed (see IsSSATargetVar for details).
Optimizer::PassToken CreateSSARewritePass();
// Create pass to convert relaxed precision instructions to half precision.
// This pass converts as many relaxed float32 arithmetic operations to half as
// possible. It converts any float32 operands to half if needed. It converts
// any resulting half precision values back to float32 as needed. No variables
// are changed. No image operations are changed.
//
// Best if run after function scope store/load and composite operation
// eliminations are run. Also best if followed by instruction simplification,
// redundancy elimination and DCE.
Optimizer::PassToken CreateConvertRelaxedToHalfPass();
// Create relax float ops pass.
// This pass decorates all float32 result instructions with RelaxedPrecision
// if not already so decorated.
Optimizer::PassToken CreateRelaxFloatOpsPass();
// Create copy propagate arrays pass.
// This pass looks to copy propagate memory references for arrays. It looks
// for specific code patterns to recognize array copies.
Optimizer::PassToken CreateCopyPropagateArraysPass();
// Create a vector dce pass.
// This pass looks for components of vectors that are unused, and removes them
// from the vector. Note this would still leave around lots of dead code that
// a pass of ADCE will be able to remove.
Optimizer::PassToken CreateVectorDCEPass();
// Create a pass to reduce the size of loads.
// This pass looks for loads of structures where only a few of its members are
// used. It replaces the loads feeding an OpExtract with an OpAccessChain and
// a load of the specific elements. The parameter is a threshold to determine
// whether we have to replace the load or not. If the ratio of the used
// components of the load is less than the threshold, we replace the load.
Optimizer::PassToken CreateReduceLoadSizePass(
double load_replacement_threshold = 0.9);
// Create a pass to combine chained access chains.
// This pass looks for access chains fed by other access chains and combines
// them into a single instruction where possible.
Optimizer::PassToken CreateCombineAccessChainsPass();
// Create a pass to instrument bindless descriptor checking
// This pass instruments all bindless references to check that descriptor
// array indices are inbounds, and if the descriptor indexing extension is
// enabled, that the descriptor has been initialized. If the reference is
// invalid, a record is written to the debug output buffer (if space allows)
// and a null value is returned. This pass is designed to support bindless
// validation in the Vulkan validation layers.
//
// TODO(greg-lunarg): Add support for buffer references. Currently only does
// checking for image references.
//
// Dead code elimination should be run after this pass as the original,
// potentially invalid code is not removed and could cause undefined behavior,
// including crashes. It may also be beneficial to run Simplification
// (ie Constant Propagation), DeadBranchElim and BlockMerge after this pass to
// optimize instrument code involving the testing of compile-time constants.
// It is also generally recommended that this pass (and all
// instrumentation passes) be run after any legalization and optimization
// passes. This will give better analysis for the instrumentation and avoid
// potentially de-optimizing the instrument code, for example, inlining
// the debug record output function throughout the module.
//
// The instrumentation will read and write buffers in debug
// descriptor set |desc_set|. It will write |shader_id| in each output record
// to identify the shader module which generated the record.
// |desc_length_enable| controls instrumentation of runtime descriptor array
// references, |desc_init_enable| controls instrumentation of descriptor
// initialization checking, and |buff_oob_enable| controls instrumentation
// of storage and uniform buffer bounds checking, all of which require input
// buffer support. |texbuff_oob_enable| controls instrumentation of texel
// buffers, which does not require input buffer support.
Optimizer::PassToken CreateInstBindlessCheckPass(
uint32_t desc_set, uint32_t shader_id, bool desc_length_enable = false,
bool desc_init_enable = false, bool buff_oob_enable = false,
bool texbuff_oob_enable = false);
// Create a pass to instrument physical buffer address checking
// This pass instruments all physical buffer address references to check that
// all referenced bytes fall in a valid buffer. If the reference is
// invalid, a record is written to the debug output buffer (if space allows)
// and a null value is returned. This pass is designed to support buffer
// address validation in the Vulkan validation layers.
//
// Dead code elimination should be run after this pass as the original,
// potentially invalid code is not removed and could cause undefined behavior,
// including crashes. Instruction simplification would likely also be
// beneficial. It is also generally recommended that this pass (and all
// instrumentation passes) be run after any legalization and optimization
// passes. This will give better analysis for the instrumentation and avoid
// potentially de-optimizing the instrument code, for example, inlining
// the debug record output function throughout the module.
//
// The instrumentation will read and write buffers in debug
// descriptor set |desc_set|. It will write |shader_id| in each output record
// to identify the shader module which generated the record.
Optimizer::PassToken CreateInstBuffAddrCheckPass(uint32_t desc_set,
uint32_t shader_id);
// Create a pass to instrument OpDebugPrintf instructions.
// This pass replaces all OpDebugPrintf instructions with instructions to write
// a record containing the string id and the all specified values into a special
// printf output buffer (if space allows). This pass is designed to support
// the printf validation in the Vulkan validation layers.
//
// The instrumentation will write buffers in debug descriptor set |desc_set|.
// It will write |shader_id| in each output record to identify the shader
// module which generated the record.
Optimizer::PassToken CreateInstDebugPrintfPass(uint32_t desc_set,
uint32_t shader_id);
// Create a pass to upgrade to the VulkanKHR memory model.
// This pass upgrades the Logical GLSL450 memory model to Logical VulkanKHR.
// Additionally, it modifies memory, image, atomic and barrier operations to
// conform to that model's requirements.
Optimizer::PassToken CreateUpgradeMemoryModelPass();
// Create a pass to do code sinking. Code sinking is a transformation
// where an instruction is moved into a more deeply nested construct.
Optimizer::PassToken CreateCodeSinkingPass();
// Create a pass to fix incorrect storage classes. In order to make code
// generation simpler, DXC may generate code where the storage classes do not
// match up correctly. This pass will fix the errors that it can.
Optimizer::PassToken CreateFixStorageClassPass();
// Creates a graphics robust access pass.
//
// This pass injects code to clamp indexed accesses to buffers and internal
// arrays, providing guarantees satisfying Vulkan's robustBufferAccess rules.
//
// TODO(dneto): Clamps coordinates and sample index for pointer calculations
// into storage images (OpImageTexelPointer). For an cube array image, it
// assumes the maximum layer count times 6 is at most 0xffffffff.
//
// NOTE: This pass will fail with a message if:
// - The module is not a Shader module.
// - The module declares VariablePointers, VariablePointersStorageBuffer, or
// RuntimeDescriptorArrayEXT capabilities.
// - The module uses an addressing model other than Logical
// - Access chain indices are wider than 64 bits.
// - Access chain index for a struct is not an OpConstant integer or is out
// of range. (The module is already invalid if that is the case.)
// - TODO(dneto): The OpImageTexelPointer coordinate component is not 32-bits
// wide.
//
// NOTE: Access chain indices are always treated as signed integers. So
// if an array has a fixed size of more than 2^31 elements, then elements
// from 2^31 and above are never accessible with a 32-bit index,
// signed or unsigned. For this case, this pass will clamp the index
// between 0 and at 2^31-1, inclusive.
// Similarly, if an array has more then 2^15 element and is accessed with
// a 16-bit index, then elements from 2^15 and above are not accessible.
// In this case, the pass will clamp the index between 0 and 2^15-1
// inclusive.
Optimizer::PassToken CreateGraphicsRobustAccessPass();
// Create a pass to spread Volatile semantics to variables with SMIDNV,
// WarpIDNV, SubgroupSize, SubgroupLocalInvocationId, SubgroupEqMask,
// SubgroupGeMask, SubgroupGtMask, SubgroupLeMask, or SubgroupLtMask BuiltIn
// decorations or OpLoad for them when the shader model is the ray generation,
// closest hit, miss, intersection, or callable. This pass can be used for
// VUID-StandaloneSpirv-VulkanMemoryModel-04678 and
// VUID-StandaloneSpirv-VulkanMemoryModel-04679 (See "Standalone SPIR-V
// Validation" section of Vulkan spec "Appendix A: Vulkan Environment for
// SPIR-V"). When the SPIR-V version is 1.6 or above, the pass also spreads
// the Volatile semantics to a variable with HelperInvocation BuiltIn decoration
// in the fragement shader.
Optimizer::PassToken CreateSpreadVolatileSemanticsPass();
// Create a pass to replace a descriptor access using variable index.
// This pass replaces every access using a variable index to array variable
// |desc| that has a DescriptorSet and Binding decorations with a constant
// element of the array. In order to replace the access using a variable index
// with the constant element, it uses a switch statement.
Optimizer::PassToken CreateReplaceDescArrayAccessUsingVarIndexPass();
// Create descriptor scalar replacement pass.
// This pass replaces every array variable |desc| that has a DescriptorSet and
// Binding decorations with a new variable for each element of the array.
// Suppose |desc| was bound at binding |b|. Then the variable corresponding to
// |desc[i]| will have binding |b+i|. The descriptor set will be the same. It
// is assumed that no other variable already has a binding that will used by one
// of the new variables. If not, the pass will generate invalid Spir-V. All
// accesses to |desc| must be OpAccessChain instructions with a literal index
// for the first index.
Optimizer::PassToken CreateDescriptorScalarReplacementPass();
// Create a pass to replace each OpKill instruction with a function call to a
// function that has a single OpKill. Also replace each OpTerminateInvocation
// instruction with a function call to a function that has a single
// OpTerminateInvocation. This allows more code to be inlined.
Optimizer::PassToken CreateWrapOpKillPass();
// Replaces the extensions VK_AMD_shader_ballot,VK_AMD_gcn_shader, and
// VK_AMD_shader_trinary_minmax with equivalent code using core instructions and
// capabilities.
Optimizer::PassToken CreateAmdExtToKhrPass();
// Replaces the internal version of GLSLstd450 InterpolateAt* extended
// instructions with the externally valid version. The internal version allows
// an OpLoad of the interpolant for the first argument. This pass removes the
// OpLoad and replaces it with its pointer. glslang and possibly other
// frontends will create the internal version for HLSL. This pass will be part
// of HLSL legalization and should be called after interpolants have been
// propagated into their final positions.
Optimizer::PassToken CreateInterpolateFixupPass();
// Removes unused components from composite input variables. Current
// implementation just removes trailing unused components from input arrays
// and structs. The pass performs best after maximizing dead code removal.
// A subsequent dead code elimination pass would be beneficial in removing
// newly unused component types.
//
// WARNING: This pass can only be safely applied standalone to vertex shaders
// as it can otherwise cause interface incompatibilities with the preceding
// shader in the pipeline. If applied to non-vertex shaders, the user should
// follow by applying EliminateDeadOutputStores and
// EliminateDeadOutputComponents to the preceding shader.
Optimizer::PassToken CreateEliminateDeadInputComponentsPass();
// Removes unused components from composite output variables. Current
// implementation just removes trailing unused components from output arrays
// and structs. The pass performs best after eliminating dead output stores.
// A subsequent dead code elimination pass would be beneficial in removing
// newly unused component types. Currently only supports vertex and fragment
// shaders.
//
// WARNING: This pass cannot be safely applied standalone as it can cause
// interface incompatibility with the following shader in the pipeline. The
// user should first apply EliminateDeadInputComponents to the following
// shader, then apply EliminateDeadOutputStores to this shader.
Optimizer::PassToken CreateEliminateDeadOutputComponentsPass();
// Removes unused components from composite input variables. This safe
// version will not cause interface incompatibilities since it only changes
// vertex shaders. The current implementation just removes trailing unused
// components from input structs and input arrays. The pass performs best
// after maximizing dead code removal. A subsequent dead code elimination
// pass would be beneficial in removing newly unused component types.
Optimizer::PassToken CreateEliminateDeadInputComponentsSafePass();
// Analyzes shader and populates |live_locs| and |live_builtins|. Best results
// will be obtained if shader has all dead code eliminated first. |live_locs|
// and |live_builtins| are subsequently used when calling
// CreateEliminateDeadOutputStoresPass on the preceding shader. Currently only
// supports tesc, tese, geom, and frag shaders.
Optimizer::PassToken CreateAnalyzeLiveInputPass(
std::unordered_set<uint32_t>* live_locs,
std::unordered_set<uint32_t>* live_builtins);
// Removes stores to output locations not listed in |live_locs| or
// |live_builtins|. Best results are obtained if constant propagation is
// performed first. A subsequent call to ADCE will eliminate any dead code
// created by the removal of the stores. A subsequent call to
// CreateEliminateDeadOutputComponentsPass will eliminate any dead output
// components created by the elimination of the stores. Currently only supports
// vert, tesc, tese, and geom shaders.
Optimizer::PassToken CreateEliminateDeadOutputStoresPass(
std::unordered_set<uint32_t>* live_locs,
std::unordered_set<uint32_t>* live_builtins);
// Creates a convert-to-sampled-image pass to convert images and/or
// samplers with given pairs of descriptor set and binding to sampled image.
// If a pair of an image and a sampler have the same pair of descriptor set and
// binding that is one of the given pairs, they will be converted to a sampled
// image. In addition, if only an image has the descriptor set and binding that
// is one of the given pairs, it will be converted to a sampled image as well.
Optimizer::PassToken CreateConvertToSampledImagePass(
const std::vector<opt::DescriptorSetAndBinding>&
descriptor_set_binding_pairs);
// Create an interface-variable-scalar-replacement pass that replaces array or
// matrix interface variables with a series of scalar or vector interface
// variables. For example, it replaces `float3 foo[2]` with `float3 foo0, foo1`.
Optimizer::PassToken CreateInterfaceVariableScalarReplacementPass();
// Creates a remove-dont-inline pass to remove the |DontInline| function control
// from every function in the module. This is useful if you want the inliner to
// inline these functions some reason.
Optimizer::PassToken CreateRemoveDontInlinePass();
// Create a fix-func-call-param pass to fix non memory argument for the function
// call, as spirv-validation requires function parameters to be an memory
// object, currently the pass would remove accesschain pointer argument passed
// to the function
Optimizer::PassToken CreateFixFuncCallArgumentsPass();
} // namespace spvtools
#endif // INCLUDE_SPIRV_TOOLS_OPTIMIZER_HPP_

114
include/vulkan/spirv_cross/GLSL.std.450.h Normal file
View file

@ -0,0 +1,114 @@
/*
* Copyright 2014-2016,2021 The Khronos Group, Inc.
* SPDX-License-Identifier: MIT
*
* MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS KHRONOS
* STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS SPECIFICATIONS AND
* HEADER INFORMATION ARE LOCATED AT https://www.khronos.org/registry/
*/
#ifndef GLSLstd450_H
#define GLSLstd450_H
static const int GLSLstd450Version = 100;
static const int GLSLstd450Revision = 3;
enum GLSLstd450 {
GLSLstd450Bad = 0, // Don't use
GLSLstd450Round = 1,
GLSLstd450RoundEven = 2,
GLSLstd450Trunc = 3,
GLSLstd450FAbs = 4,
GLSLstd450SAbs = 5,
GLSLstd450FSign = 6,
GLSLstd450SSign = 7,
GLSLstd450Floor = 8,
GLSLstd450Ceil = 9,
GLSLstd450Fract = 10,
GLSLstd450Radians = 11,
GLSLstd450Degrees = 12,
GLSLstd450Sin = 13,
GLSLstd450Cos = 14,
GLSLstd450Tan = 15,
GLSLstd450Asin = 16,
GLSLstd450Acos = 17,
GLSLstd450Atan = 18,
GLSLstd450Sinh = 19,
GLSLstd450Cosh = 20,
GLSLstd450Tanh = 21,
GLSLstd450Asinh = 22,
GLSLstd450Acosh = 23,
GLSLstd450Atanh = 24,
GLSLstd450Atan2 = 25,
GLSLstd450Pow = 26,
GLSLstd450Exp = 27,
GLSLstd450Log = 28,
GLSLstd450Exp2 = 29,
GLSLstd450Log2 = 30,
GLSLstd450Sqrt = 31,
GLSLstd450InverseSqrt = 32,
GLSLstd450Determinant = 33,
GLSLstd450MatrixInverse = 34,
GLSLstd450Modf = 35, // second operand needs an OpVariable to write to
GLSLstd450ModfStruct = 36, // no OpVariable operand
GLSLstd450FMin = 37,
GLSLstd450UMin = 38,
GLSLstd450SMin = 39,
GLSLstd450FMax = 40,
GLSLstd450UMax = 41,
GLSLstd450SMax = 42,
GLSLstd450FClamp = 43,
GLSLstd450UClamp = 44,
GLSLstd450SClamp = 45,
GLSLstd450FMix = 46,
GLSLstd450IMix = 47, // Reserved
GLSLstd450Step = 48,
GLSLstd450SmoothStep = 49,
GLSLstd450Fma = 50,
GLSLstd450Frexp = 51, // second operand needs an OpVariable to write to
GLSLstd450FrexpStruct = 52, // no OpVariable operand
GLSLstd450Ldexp = 53,
GLSLstd450PackSnorm4x8 = 54,
GLSLstd450PackUnorm4x8 = 55,
GLSLstd450PackSnorm2x16 = 56,
GLSLstd450PackUnorm2x16 = 57,
GLSLstd450PackHalf2x16 = 58,
GLSLstd450PackDouble2x32 = 59,
GLSLstd450UnpackSnorm2x16 = 60,
GLSLstd450UnpackUnorm2x16 = 61,
GLSLstd450UnpackHalf2x16 = 62,
GLSLstd450UnpackSnorm4x8 = 63,
GLSLstd450UnpackUnorm4x8 = 64,
GLSLstd450UnpackDouble2x32 = 65,
GLSLstd450Length = 66,
GLSLstd450Distance = 67,
GLSLstd450Cross = 68,
GLSLstd450Normalize = 69,
GLSLstd450FaceForward = 70,
GLSLstd450Reflect = 71,
GLSLstd450Refract = 72,
GLSLstd450FindILsb = 73,
GLSLstd450FindSMsb = 74,
GLSLstd450FindUMsb = 75,
GLSLstd450InterpolateAtCentroid = 76,
GLSLstd450InterpolateAtSample = 77,
GLSLstd450InterpolateAtOffset = 78,
GLSLstd450NMin = 79,
GLSLstd450NMax = 80,
GLSLstd450NClamp = 81,
GLSLstd450Count
};
#endif // #ifndef GLSLstd450_H

2568
include/vulkan/spirv_cross/spirv.h Normal file
View file

File diff suppressed because it is too large Load diff

2579
include/vulkan/spirv_cross/spirv.hpp Normal file
View file

File diff suppressed because it is too large Load diff

168
include/vulkan/spirv_cross/spirv_cfg.hpp Normal file
View file

@ -0,0 +1,168 @@
/*
* Copyright 2016-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_CFG_HPP
#define SPIRV_CROSS_CFG_HPP
#include "spirv_common.hpp"
#include <assert.h>
namespace SPIRV_CROSS_NAMESPACE
{
class Compiler;
class CFG
{
public:
CFG(Compiler &compiler, const SPIRFunction &function);
Compiler &get_compiler()
{
return compiler;
}
const Compiler &get_compiler() const
{
return compiler;
}
const SPIRFunction &get_function() const
{
return func;
}
uint32_t get_immediate_dominator(uint32_t block) const
{
auto itr = immediate_dominators.find(block);
if (itr != std::end(immediate_dominators))
return itr->second;
else
return 0;
}
bool is_reachable(uint32_t block) const
{
return visit_order.count(block) != 0;
}
uint32_t get_visit_order(uint32_t block) const
{
auto itr = visit_order.find(block);
assert(itr != std::end(visit_order));
int v = itr->second.get();
assert(v > 0);
return uint32_t(v);
}
uint32_t find_common_dominator(uint32_t a, uint32_t b) const;
const SmallVector<uint32_t> &get_preceding_edges(uint32_t block) const
{
auto itr = preceding_edges.find(block);
if (itr != std::end(preceding_edges))
return itr->second;
else
return empty_vector;
}
const SmallVector<uint32_t> &get_succeeding_edges(uint32_t block) const
{
auto itr = succeeding_edges.find(block);
if (itr != std::end(succeeding_edges))
return itr->second;
else
return empty_vector;
}
template <typename Op>
void walk_from(std::unordered_set<uint32_t> &seen_blocks, uint32_t block, const Op &op) const
{
if (seen_blocks.count(block))
return;
seen_blocks.insert(block);
if (op(block))
{
for (auto b : get_succeeding_edges(block))
walk_from(seen_blocks, b, op);
}
}
uint32_t find_loop_dominator(uint32_t block) const;
bool node_terminates_control_flow_in_sub_graph(BlockID from, BlockID to) const;
private:
struct VisitOrder
{
int &get()
{
return v;
}
const int &get() const
{
return v;
}
int v = -1;
};
Compiler &compiler;
const SPIRFunction &func;
std::unordered_map<uint32_t, SmallVector<uint32_t>> preceding_edges;
std::unordered_map<uint32_t, SmallVector<uint32_t>> succeeding_edges;
std::unordered_map<uint32_t, uint32_t> immediate_dominators;
std::unordered_map<uint32_t, VisitOrder> visit_order;
SmallVector<uint32_t> post_order;
SmallVector<uint32_t> empty_vector;
void add_branch(uint32_t from, uint32_t to);
void build_post_order_visit_order();
void build_immediate_dominators();
bool post_order_visit(uint32_t block);
uint32_t visit_count = 0;
bool is_back_edge(uint32_t to) const;
bool has_visited_forward_edge(uint32_t to) const;
};
class DominatorBuilder
{
public:
DominatorBuilder(const CFG &cfg);
void add_block(uint32_t block);
uint32_t get_dominator() const
{
return dominator;
}
void lift_continue_block_dominator();
private:
const CFG &cfg;
uint32_t dominator = 0;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

1921
include/vulkan/spirv_cross/spirv_common.hpp Normal file
View file

File diff suppressed because it is too large Load diff

1175
include/vulkan/spirv_cross/spirv_cross.hpp Normal file
View file

File diff suppressed because it is too large Load diff

1092
include/vulkan/spirv_cross/spirv_cross_c.h Normal file
View file

File diff suppressed because it is too large Load diff

755
include/vulkan/spirv_cross/spirv_cross_containers.hpp Normal file
View file

@ -0,0 +1,755 @@
/*
* Copyright 2019-2021 Hans-Kristian Arntzen
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_CONTAINERS_HPP
#define SPIRV_CROSS_CONTAINERS_HPP
#include "spirv_cross_error_handling.hpp"
#include <algorithm>
#include <exception>
#include <functional>
#include <iterator>
#include <limits>
#include <memory>
#include <stack>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <type_traits>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
#ifdef SPIRV_CROSS_NAMESPACE_OVERRIDE
#define SPIRV_CROSS_NAMESPACE SPIRV_CROSS_NAMESPACE_OVERRIDE
#else
#define SPIRV_CROSS_NAMESPACE spirv_cross
#endif
namespace SPIRV_CROSS_NAMESPACE
{
#ifndef SPIRV_CROSS_FORCE_STL_TYPES
// std::aligned_storage does not support size == 0, so roll our own.
template <typename T, size_t N>
class AlignedBuffer
{
public:
T *data()
{
#if defined(_MSC_VER) && _MSC_VER < 1900
// MSVC 2013 workarounds, sigh ...
// Only use this workaround on MSVC 2013 due to some confusion around default initialized unions.
// Spec seems to suggest the memory will be zero-initialized, which is *not* what we want.
return reinterpret_cast<T *>(u.aligned_char);
#else
return reinterpret_cast<T *>(aligned_char);
#endif
}
private:
#if defined(_MSC_VER) && _MSC_VER < 1900
// MSVC 2013 workarounds, sigh ...
union
{
char aligned_char[sizeof(T) * N];
double dummy_aligner;
} u;
#else
alignas(T) char aligned_char[sizeof(T) * N];
#endif
};
template <typename T>
class AlignedBuffer<T, 0>
{
public:
T *data()
{
return nullptr;
}
};
// An immutable version of SmallVector which erases type information about storage.
template <typename T>
class VectorView
{
public:
T &operator[](size_t i) SPIRV_CROSS_NOEXCEPT
{
return ptr[i];
}
const T &operator[](size_t i) const SPIRV_CROSS_NOEXCEPT
{
return ptr[i];
}
bool empty() const SPIRV_CROSS_NOEXCEPT
{
return buffer_size == 0;
}
size_t size() const SPIRV_CROSS_NOEXCEPT
{
return buffer_size;
}
T *data() SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
const T *data() const SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
T *begin() SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
T *end() SPIRV_CROSS_NOEXCEPT
{
return ptr + buffer_size;
}
const T *begin() const SPIRV_CROSS_NOEXCEPT
{
return ptr;
}
const T *end() const SPIRV_CROSS_NOEXCEPT
{
return ptr + buffer_size;
}
T &front() SPIRV_CROSS_NOEXCEPT
{
return ptr[0];
}
const T &front() const SPIRV_CROSS_NOEXCEPT
{
return ptr[0];
}
T &back() SPIRV_CROSS_NOEXCEPT
{
return ptr[buffer_size - 1];
}
const T &back() const SPIRV_CROSS_NOEXCEPT
{
return ptr[buffer_size - 1];
}
// Makes it easier to consume SmallVector.
#if defined(_MSC_VER) && _MSC_VER < 1900
explicit operator std::vector<T>() const
{
// Another MSVC 2013 workaround. It does not understand lvalue/rvalue qualified operations.
return std::vector<T>(ptr, ptr + buffer_size);
}
#else
// Makes it easier to consume SmallVector.
explicit operator std::vector<T>() const &
{
return std::vector<T>(ptr, ptr + buffer_size);
}
// If we are converting as an r-value, we can pilfer our elements.
explicit operator std::vector<T>() &&
{
return std::vector<T>(std::make_move_iterator(ptr), std::make_move_iterator(ptr + buffer_size));
}
#endif
// Avoid sliced copies. Base class should only be read as a reference.
VectorView(const VectorView &) = delete;
void operator=(const VectorView &) = delete;
protected:
VectorView() = default;
T *ptr = nullptr;
size_t buffer_size = 0;
};
// Simple vector which supports up to N elements inline, without malloc/free.
// We use a lot of throwaway vectors all over the place which triggers allocations.
// This class only implements the subset of std::vector we need in SPIRV-Cross.
// It is *NOT* a drop-in replacement in general projects.
template <typename T, size_t N = 8>
class SmallVector : public VectorView<T>
{
public:
SmallVector() SPIRV_CROSS_NOEXCEPT
{
this->ptr = stack_storage.data();
buffer_capacity = N;
}
template <typename U>
SmallVector(const U *arg_list_begin, const U *arg_list_end) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
auto count = size_t(arg_list_end - arg_list_begin);
reserve(count);
for (size_t i = 0; i < count; i++, arg_list_begin++)
new (&this->ptr[i]) T(*arg_list_begin);
this->buffer_size = count;
}
template <typename U>
SmallVector(std::initializer_list<U> init) SPIRV_CROSS_NOEXCEPT : SmallVector(init.begin(), init.end())
{
}
template <typename U, size_t M>
explicit SmallVector(const U (&init)[M]) SPIRV_CROSS_NOEXCEPT : SmallVector(init, init + M)
{
}
SmallVector(SmallVector &&other) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
*this = std::move(other);
}
SmallVector &operator=(SmallVector &&other) SPIRV_CROSS_NOEXCEPT
{
clear();
if (other.ptr != other.stack_storage.data())
{
// Pilfer allocated pointer.
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = other.ptr;
this->buffer_size = other.buffer_size;
buffer_capacity = other.buffer_capacity;
other.ptr = nullptr;
other.buffer_size = 0;
other.buffer_capacity = 0;
}
else
{
// Need to move the stack contents individually.
reserve(other.buffer_size);
for (size_t i = 0; i < other.buffer_size; i++)
{
new (&this->ptr[i]) T(std::move(other.ptr[i]));
other.ptr[i].~T();
}
this->buffer_size = other.buffer_size;
other.buffer_size = 0;
}
return *this;
}
SmallVector(const SmallVector &other) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
*this = other;
}
SmallVector &operator=(const SmallVector &other) SPIRV_CROSS_NOEXCEPT
{
if (this == &other)
return *this;
clear();
reserve(other.buffer_size);
for (size_t i = 0; i < other.buffer_size; i++)
new (&this->ptr[i]) T(other.ptr[i]);
this->buffer_size = other.buffer_size;
return *this;
}
explicit SmallVector(size_t count) SPIRV_CROSS_NOEXCEPT : SmallVector()
{
resize(count);
}
~SmallVector()
{
clear();
if (this->ptr != stack_storage.data())
free(this->ptr);
}
void clear() SPIRV_CROSS_NOEXCEPT
{
for (size_t i = 0; i < this->buffer_size; i++)
this->ptr[i].~T();
this->buffer_size = 0;
}
void push_back(const T &t) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(t);
this->buffer_size++;
}
void push_back(T &&t) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(std::move(t));
this->buffer_size++;
}
void pop_back() SPIRV_CROSS_NOEXCEPT
{
// Work around false positive warning on GCC 8.3.
// Calling pop_back on empty vector is undefined.
if (!this->empty())
resize(this->buffer_size - 1);
}
template <typename... Ts>
void emplace_back(Ts &&... ts) SPIRV_CROSS_NOEXCEPT
{
reserve(this->buffer_size + 1);
new (&this->ptr[this->buffer_size]) T(std::forward<Ts>(ts)...);
this->buffer_size++;
}
void reserve(size_t count) SPIRV_CROSS_NOEXCEPT
{
if ((count > (std::numeric_limits<size_t>::max)() / sizeof(T)) ||
(count > (std::numeric_limits<size_t>::max)() / 2))
{
// Only way this should ever happen is with garbage input, terminate.
std::terminate();
}
if (count > buffer_capacity)
{
size_t target_capacity = buffer_capacity;
if (target_capacity == 0)
target_capacity = 1;
// Weird parens works around macro issues on Windows if NOMINMAX is not used.
target_capacity = (std::max)(target_capacity, N);
// Need to ensure there is a POT value of target capacity which is larger than count,
// otherwise this will overflow.
while (target_capacity < count)
target_capacity <<= 1u;
T *new_buffer =
target_capacity > N ? static_cast<T *>(malloc(target_capacity * sizeof(T))) : stack_storage.data();
// If we actually fail this malloc, we are hosed anyways, there is no reason to attempt recovery.
if (!new_buffer)
std::terminate();
// In case for some reason two allocations both come from same stack.
if (new_buffer != this->ptr)
{
// We don't deal with types which can throw in move constructor.
for (size_t i = 0; i < this->buffer_size; i++)
{
new (&new_buffer[i]) T(std::move(this->ptr[i]));
this->ptr[i].~T();
}
}
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = new_buffer;
buffer_capacity = target_capacity;
}
}
void insert(T *itr, const T *insert_begin, const T *insert_end) SPIRV_CROSS_NOEXCEPT
{
auto count = size_t(insert_end - insert_begin);
if (itr == this->end())
{
reserve(this->buffer_size + count);
for (size_t i = 0; i < count; i++, insert_begin++)
new (&this->ptr[this->buffer_size + i]) T(*insert_begin);
this->buffer_size += count;
}
else
{
if (this->buffer_size + count > buffer_capacity)
{
auto target_capacity = this->buffer_size + count;
if (target_capacity == 0)
target_capacity = 1;
if (target_capacity < N)
target_capacity = N;
while (target_capacity < count)
target_capacity <<= 1u;
// Need to allocate new buffer. Move everything to a new buffer.
T *new_buffer =
target_capacity > N ? static_cast<T *>(malloc(target_capacity * sizeof(T))) : stack_storage.data();
// If we actually fail this malloc, we are hosed anyways, there is no reason to attempt recovery.
if (!new_buffer)
std::terminate();
// First, move elements from source buffer to new buffer.
// We don't deal with types which can throw in move constructor.
auto *target_itr = new_buffer;
auto *original_source_itr = this->begin();
if (new_buffer != this->ptr)
{
while (original_source_itr != itr)
{
new (target_itr) T(std::move(*original_source_itr));
original_source_itr->~T();
++original_source_itr;
++target_itr;
}
}
// Copy-construct new elements.
for (auto *source_itr = insert_begin; source_itr != insert_end; ++source_itr, ++target_itr)
new (target_itr) T(*source_itr);
// Move over the other half.
if (new_buffer != this->ptr || insert_begin != insert_end)
{
while (original_source_itr != this->end())
{
new (target_itr) T(std::move(*original_source_itr));
original_source_itr->~T();
++original_source_itr;
++target_itr;
}
}
if (this->ptr != stack_storage.data())
free(this->ptr);
this->ptr = new_buffer;
buffer_capacity = target_capacity;
}
else
{
// Move in place, need to be a bit careful about which elements are constructed and which are not.
// Move the end and construct the new elements.
auto *target_itr = this->end() + count;
auto *source_itr = this->end();
while (target_itr != this->end() && source_itr != itr)
{
--target_itr;
--source_itr;
new (target_itr) T(std::move(*source_itr));
}
// For already constructed elements we can move-assign.
std::move_backward(itr, source_itr, target_itr);
// For the inserts which go to already constructed elements, we can do a plain copy.
while (itr != this->end() && insert_begin != insert_end)
*itr++ = *insert_begin++;
// For inserts into newly allocated memory, we must copy-construct instead.
while (insert_begin != insert_end)
{
new (itr) T(*insert_begin);
++itr;
++insert_begin;
}
}
this->buffer_size += count;
}
}
void insert(T *itr, const T &value) SPIRV_CROSS_NOEXCEPT
{
insert(itr, &value, &value + 1);
}
T *erase(T *itr) SPIRV_CROSS_NOEXCEPT
{
std::move(itr + 1, this->end(), itr);
this->ptr[--this->buffer_size].~T();
return itr;
}
void erase(T *start_erase, T *end_erase) SPIRV_CROSS_NOEXCEPT
{
if (end_erase == this->end())
{
resize(size_t(start_erase - this->begin()));
}
else
{
auto new_size = this->buffer_size - (end_erase - start_erase);
std::move(end_erase, this->end(), start_erase);
resize(new_size);
}
}
void resize(size_t new_size) SPIRV_CROSS_NOEXCEPT
{
if (new_size < this->buffer_size)
{
for (size_t i = new_size; i < this->buffer_size; i++)
this->ptr[i].~T();
}
else if (new_size > this->buffer_size)
{
reserve(new_size);
for (size_t i = this->buffer_size; i < new_size; i++)
new (&this->ptr[i]) T();
}
this->buffer_size = new_size;
}
private:
size_t buffer_capacity = 0;
AlignedBuffer<T, N> stack_storage;
};
// A vector without stack storage.
// Could also be a typedef-ed to std::vector,
// but might as well use the one we have.
template <typename T>
using Vector = SmallVector<T, 0>;
#else // SPIRV_CROSS_FORCE_STL_TYPES
template <typename T, size_t N = 8>
using SmallVector = std::vector<T>;
template <typename T>
using Vector = std::vector<T>;
template <typename T>
using VectorView = std::vector<T>;
#endif // SPIRV_CROSS_FORCE_STL_TYPES
// An object pool which we use for allocating IVariant-derived objects.
// We know we are going to allocate a bunch of objects of each type,
// so amortize the mallocs.
class ObjectPoolBase
{
public:
virtual ~ObjectPoolBase() = default;
virtual void deallocate_opaque(void *ptr) = 0;
};
template <typename T>
class ObjectPool : public ObjectPoolBase
{
public:
explicit ObjectPool(unsigned start_object_count_ = 16)
: start_object_count(start_object_count_)
{
}
template <typename... P>
T *allocate(P &&... p)
{
if (vacants.empty())
{
unsigned num_objects = start_object_count << memory.size();
T *ptr = static_cast<T *>(malloc(num_objects * sizeof(T)));
if (!ptr)
return nullptr;
for (unsigned i = 0; i < num_objects; i++)
vacants.push_back(&ptr[i]);
memory.emplace_back(ptr);
}
T *ptr = vacants.back();
vacants.pop_back();
new (ptr) T(std::forward<P>(p)...);
return ptr;
}
void deallocate(T *ptr)
{
ptr->~T();
vacants.push_back(ptr);
}
void deallocate_opaque(void *ptr) override
{
deallocate(static_cast<T *>(ptr));
}
void clear()
{
vacants.clear();
memory.clear();
}
protected:
Vector<T *> vacants;
struct MallocDeleter
{
void operator()(T *ptr)
{
::free(ptr);
}
};
SmallVector<std::unique_ptr<T, MallocDeleter>> memory;
unsigned start_object_count;
};
template <size_t StackSize = 4096, size_t BlockSize = 4096>
class StringStream
{
public:
StringStream()
{
reset();
}
~StringStream()
{
reset();
}
// Disable copies and moves. Makes it easier to implement, and we don't need it.
StringStream(const StringStream &) = delete;
void operator=(const StringStream &) = delete;
template <typename T, typename std::enable_if<!std::is_floating_point<T>::value, int>::type = 0>
StringStream &operator<<(const T &t)
{
auto s = std::to_string(t);
append(s.data(), s.size());
return *this;
}
// Only overload this to make float/double conversions ambiguous.
StringStream &operator<<(uint32_t v)
{
auto s = std::to_string(v);
append(s.data(), s.size());
return *this;
}
StringStream &operator<<(char c)
{
append(&c, 1);
return *this;
}
StringStream &operator<<(const std::string &s)
{
append(s.data(), s.size());
return *this;
}
StringStream &operator<<(const char *s)
{
append(s, strlen(s));
return *this;
}
template <size_t N>
StringStream &operator<<(const char (&s)[N])
{
append(s, strlen(s));
return *this;
}
std::string str() const
{
std::string ret;
size_t target_size = 0;
for (auto &saved : saved_buffers)
target_size += saved.offset;
target_size += current_buffer.offset;
ret.reserve(target_size);
for (auto &saved : saved_buffers)
ret.insert(ret.end(), saved.buffer, saved.buffer + saved.offset);
ret.insert(ret.end(), current_buffer.buffer, current_buffer.buffer + current_buffer.offset);
return ret;
}
void reset()
{
for (auto &saved : saved_buffers)
if (saved.buffer != stack_buffer)
free(saved.buffer);
if (current_buffer.buffer != stack_buffer)
free(current_buffer.buffer);
saved_buffers.clear();
current_buffer.buffer = stack_buffer;
current_buffer.offset = 0;
current_buffer.size = sizeof(stack_buffer);
}
private:
struct Buffer
{
char *buffer = nullptr;
size_t offset = 0;
size_t size = 0;
};
Buffer current_buffer;
char stack_buffer[StackSize];
SmallVector<Buffer> saved_buffers;
void append(const char *s, size_t len)
{
size_t avail = current_buffer.size - current_buffer.offset;
if (avail < len)
{
if (avail > 0)
{
memcpy(current_buffer.buffer + current_buffer.offset, s, avail);
s += avail;
len -= avail;
current_buffer.offset += avail;
}
saved_buffers.push_back(current_buffer);
size_t target_size = len > BlockSize ? len : BlockSize;
current_buffer.buffer = static_cast<char *>(malloc(target_size));
if (!current_buffer.buffer)
SPIRV_CROSS_THROW("Out of memory.");
memcpy(current_buffer.buffer, s, len);
current_buffer.offset = len;
current_buffer.size = target_size;
}
else
{
memcpy(current_buffer.buffer + current_buffer.offset, s, len);
current_buffer.offset += len;
}
}
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

94
include/vulkan/spirv_cross/spirv_cross_error_handling.hpp Normal file
View file

@ -0,0 +1,94 @@
/*
* Copyright 2015-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_ERROR_HANDLING
#define SPIRV_CROSS_ERROR_HANDLING
#include <stdio.h>
#include <stdlib.h>
#include <string>
#ifndef SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS
#include <stdexcept>
#endif
#ifdef SPIRV_CROSS_NAMESPACE_OVERRIDE
#define SPIRV_CROSS_NAMESPACE SPIRV_CROSS_NAMESPACE_OVERRIDE
#else
#define SPIRV_CROSS_NAMESPACE spirv_cross
#endif
namespace SPIRV_CROSS_NAMESPACE
{
#ifdef SPIRV_CROSS_EXCEPTIONS_TO_ASSERTIONS
#if !defined(_MSC_VER) || defined(__clang__)
[[noreturn]]
#elif defined(_MSC_VER)
__declspec(noreturn)
#endif
inline void
report_and_abort(const std::string &msg)
{
#ifdef NDEBUG
(void)msg;
#else
fprintf(stderr, "There was a compiler error: %s\n", msg.c_str());
#endif
fflush(stderr);
abort();
}
#define SPIRV_CROSS_THROW(x) report_and_abort(x)
#else
class CompilerError : public std::runtime_error
{
public:
explicit CompilerError(const std::string &str)
: std::runtime_error(str)
{
}
};
#define SPIRV_CROSS_THROW(x) throw CompilerError(x)
#endif
// MSVC 2013 does not have noexcept. We need this for Variant to get move constructor to work correctly
// instead of copy constructor.
// MSVC 2013 ignores that move constructors cannot throw in std::vector, so just don't define it.
#if defined(_MSC_VER) && _MSC_VER < 1900
#define SPIRV_CROSS_NOEXCEPT
#else
#define SPIRV_CROSS_NOEXCEPT noexcept
#endif
#if __cplusplus >= 201402l
#define SPIRV_CROSS_DEPRECATED(reason) [[deprecated(reason)]]
#elif defined(__GNUC__)
#define SPIRV_CROSS_DEPRECATED(reason) __attribute__((deprecated))
#elif defined(_MSC_VER)
#define SPIRV_CROSS_DEPRECATED(reason) __declspec(deprecated(reason))
#else
#define SPIRV_CROSS_DEPRECATED(reason)
#endif
} // namespace SPIRV_CROSS_NAMESPACE
#endif

256
include/vulkan/spirv_cross/spirv_cross_parsed_ir.hpp Normal file
View file

@ -0,0 +1,256 @@
/*
* Copyright 2018-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_PARSED_IR_HPP
#define SPIRV_CROSS_PARSED_IR_HPP
#include "spirv_common.hpp"
#include <stdint.h>
#include <unordered_map>
namespace SPIRV_CROSS_NAMESPACE
{
// This data structure holds all information needed to perform cross-compilation and reflection.
// It is the output of the Parser, but any implementation could create this structure.
// It is intentionally very "open" and struct-like with some helper functions to deal with decorations.
// Parser is the reference implementation of how this data structure should be filled in.
class ParsedIR
{
private:
// This must be destroyed after the "ids" vector.
std::unique_ptr<ObjectPoolGroup> pool_group;
public:
ParsedIR();
// Due to custom allocations from object pools, we cannot use a default copy constructor.
ParsedIR(const ParsedIR &other);
ParsedIR &operator=(const ParsedIR &other);
// Moves are unproblematic, but we need to implement it anyways, since MSVC 2013 does not understand
// how to default-implement these.
ParsedIR(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT;
ParsedIR &operator=(ParsedIR &&other) SPIRV_CROSS_NOEXCEPT;
// Resizes ids, meta and block_meta.
void set_id_bounds(uint32_t bounds);
// The raw SPIR-V, instructions and opcodes refer to this by offset + count.
std::vector<uint32_t> spirv;
// Holds various data structures which inherit from IVariant.
SmallVector<Variant> ids;
// Various meta data for IDs, decorations, names, etc.
std::unordered_map<ID, Meta> meta;
// Holds all IDs which have a certain type.
// This is needed so we can iterate through a specific kind of resource quickly,
// and in-order of module declaration.
SmallVector<ID> ids_for_type[TypeCount];
// Special purpose lists which contain a union of types.
// This is needed so we can declare specialization constants and structs in an interleaved fashion,
// among other things.
// Constants can be undef or of struct type, and struct array sizes can use specialization constants.
SmallVector<ID> ids_for_constant_undef_or_type;
SmallVector<ID> ids_for_constant_or_variable;
// We need to keep track of the width the Ops that contains a type for the
// OpSwitch instruction, since this one doesn't contains the type in the
// instruction itself. And in some case we need to cast the condition to
// wider types. We only need the width to do the branch fixup since the
// type check itself can be done at runtime
std::unordered_map<ID, uint32_t> load_type_width;
// Declared capabilities and extensions in the SPIR-V module.
// Not really used except for reflection at the moment.
SmallVector<spv::Capability> declared_capabilities;
SmallVector<std::string> declared_extensions;
// Meta data about blocks. The cross-compiler needs to query if a block is either of these types.
// It is a bitset as there can be more than one tag per block.
enum BlockMetaFlagBits
{
BLOCK_META_LOOP_HEADER_BIT = 1 << 0,
BLOCK_META_CONTINUE_BIT = 1 << 1,
BLOCK_META_LOOP_MERGE_BIT = 1 << 2,
BLOCK_META_SELECTION_MERGE_BIT = 1 << 3,
BLOCK_META_MULTISELECT_MERGE_BIT = 1 << 4
};
using BlockMetaFlags = uint8_t;
SmallVector<BlockMetaFlags> block_meta;
std::unordered_map<BlockID, BlockID> continue_block_to_loop_header;
// Normally, we'd stick SPIREntryPoint in ids array, but it conflicts with SPIRFunction.
// Entry points can therefore be seen as some sort of meta structure.
std::unordered_map<FunctionID, SPIREntryPoint> entry_points;
FunctionID default_entry_point = 0;
struct Source
{
uint32_t version = 0;
bool es = false;
bool known = false;
bool hlsl = false;
Source() = default;
};
Source source;
spv::AddressingModel addressing_model = spv::AddressingModelMax;
spv::MemoryModel memory_model = spv::MemoryModelMax;
// Decoration handling methods.
// Can be useful for simple "raw" reflection.
// However, most members are here because the Parser needs most of these,
// and might as well just have the whole suite of decoration/name handling in one place.
void set_name(ID id, const std::string &name);
const std::string &get_name(ID id) const;
void set_decoration(ID id, spv::Decoration decoration, uint32_t argument = 0);
void set_decoration_string(ID id, spv::Decoration decoration, const std::string &argument);
bool has_decoration(ID id, spv::Decoration decoration) const;
uint32_t get_decoration(ID id, spv::Decoration decoration) const;
const std::string &get_decoration_string(ID id, spv::Decoration decoration) const;
const Bitset &get_decoration_bitset(ID id) const;
void unset_decoration(ID id, spv::Decoration decoration);
// Decoration handling methods (for members of a struct).
void set_member_name(TypeID id, uint32_t index, const std::string &name);
const std::string &get_member_name(TypeID id, uint32_t index) const;
void set_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration, uint32_t argument = 0);
void set_member_decoration_string(TypeID id, uint32_t index, spv::Decoration decoration,
const std::string &argument);
uint32_t get_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration) const;
const std::string &get_member_decoration_string(TypeID id, uint32_t index, spv::Decoration decoration) const;
bool has_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration) const;
const Bitset &get_member_decoration_bitset(TypeID id, uint32_t index) const;
void unset_member_decoration(TypeID id, uint32_t index, spv::Decoration decoration);
void mark_used_as_array_length(ID id);
uint32_t increase_bound_by(uint32_t count);
Bitset get_buffer_block_flags(const SPIRVariable &var) const;
Bitset get_buffer_block_type_flags(const SPIRType &type) const;
void add_typed_id(Types type, ID id);
void remove_typed_id(Types type, ID id);
class LoopLock
{
public:
explicit LoopLock(uint32_t *counter);
LoopLock(const LoopLock &) = delete;
void operator=(const LoopLock &) = delete;
LoopLock(LoopLock &&other) SPIRV_CROSS_NOEXCEPT;
LoopLock &operator=(LoopLock &&other) SPIRV_CROSS_NOEXCEPT;
~LoopLock();
private:
uint32_t *lock;
};
// This must be held while iterating over a type ID array.
// It is undefined if someone calls set<>() while we're iterating over a data structure, so we must
// make sure that this case is avoided.
// If we have a hard lock, it is an error to call set<>(), and an exception is thrown.
// If we have a soft lock, we silently ignore any additions to the typed arrays.
// This should only be used for physical ID remapping where we need to create an ID, but we will never
// care about iterating over them.
LoopLock create_loop_hard_lock() const;
LoopLock create_loop_soft_lock() const;
template <typename T, typename Op>
void for_each_typed_id(const Op &op)
{
auto loop_lock = create_loop_hard_lock();
for (auto &id : ids_for_type[T::type])
{
if (ids[id].get_type() == static_cast<Types>(T::type))
op(id, get<T>(id));
}
}
template <typename T, typename Op>
void for_each_typed_id(const Op &op) const
{
auto loop_lock = create_loop_hard_lock();
for (auto &id : ids_for_type[T::type])
{
if (ids[id].get_type() == static_cast<Types>(T::type))
op(id, get<T>(id));
}
}
template <typename T>
void reset_all_of_type()
{
reset_all_of_type(static_cast<Types>(T::type));
}
void reset_all_of_type(Types type);
Meta *find_meta(ID id);
const Meta *find_meta(ID id) const;
const std::string &get_empty_string() const
{
return empty_string;
}
void make_constant_null(uint32_t id, uint32_t type, bool add_to_typed_id_set);
void fixup_reserved_names();
static void sanitize_underscores(std::string &str);
static void sanitize_identifier(std::string &str, bool member, bool allow_reserved_prefixes);
static bool is_globally_reserved_identifier(std::string &str, bool allow_reserved_prefixes);
uint32_t get_spirv_version() const;
private:
template <typename T>
T &get(uint32_t id)
{
return variant_get<T>(ids[id]);
}
template <typename T>
const T &get(uint32_t id) const
{
return variant_get<T>(ids[id]);
}
mutable uint32_t loop_iteration_depth_hard = 0;
mutable uint32_t loop_iteration_depth_soft = 0;
std::string empty_string;
Bitset cleared_bitset;
std::unordered_set<uint32_t> meta_needing_name_fixup;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

1033
include/vulkan/spirv_cross/spirv_glsl.hpp Normal file
View file

File diff suppressed because it is too large Load diff

407
include/vulkan/spirv_cross/spirv_hlsl.hpp Normal file
View file

@ -0,0 +1,407 @@
/*
* Copyright 2016-2021 Robert Konrad
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_HLSL_HPP
#define SPIRV_HLSL_HPP
#include "spirv_glsl.hpp"
#include <utility>
namespace SPIRV_CROSS_NAMESPACE
{
// Interface which remaps vertex inputs to a fixed semantic name to make linking easier.
struct HLSLVertexAttributeRemap
{
uint32_t location;
std::string semantic;
};
// Specifying a root constant (d3d12) or push constant range (vulkan).
//
// `start` and `end` denotes the range of the root constant in bytes.
// Both values need to be multiple of 4.
struct RootConstants
{
uint32_t start;
uint32_t end;
uint32_t binding;
uint32_t space;
};
// For finer control, decorations may be removed from specific resources instead with unset_decoration().
enum HLSLBindingFlagBits
{
HLSL_BINDING_AUTO_NONE_BIT = 0,
// Push constant (root constant) resources will be declared as CBVs (b-space) without a register() declaration.
// A register will be automatically assigned by the D3D compiler, but must therefore be reflected in D3D-land.
// Push constants do not normally have a DecorationBinding set, but if they do, this can be used to ignore it.
HLSL_BINDING_AUTO_PUSH_CONSTANT_BIT = 1 << 0,
// cbuffer resources will be declared as CBVs (b-space) without a register() declaration.
// A register will be automatically assigned, but must be reflected in D3D-land.
HLSL_BINDING_AUTO_CBV_BIT = 1 << 1,
// All SRVs (t-space) will be declared without a register() declaration.
HLSL_BINDING_AUTO_SRV_BIT = 1 << 2,
// All UAVs (u-space) will be declared without a register() declaration.
HLSL_BINDING_AUTO_UAV_BIT = 1 << 3,
// All samplers (s-space) will be declared without a register() declaration.
HLSL_BINDING_AUTO_SAMPLER_BIT = 1 << 4,
// No resources will be declared with register().
HLSL_BINDING_AUTO_ALL = 0x7fffffff
};
using HLSLBindingFlags = uint32_t;
// By matching stage, desc_set and binding for a SPIR-V resource,
// register bindings are set based on whether the HLSL resource is a
// CBV, UAV, SRV or Sampler. A single binding in SPIR-V might contain multiple
// resource types, e.g. COMBINED_IMAGE_SAMPLER, and SRV/Sampler bindings will be used respectively.
// On SM 5.0 and lower, register_space is ignored.
//
// To remap a push constant block which does not have any desc_set/binding associated with it,
// use ResourceBindingPushConstant{DescriptorSet,Binding} as values for desc_set/binding.
// For deeper control of push constants, set_root_constant_layouts() can be used instead.
struct HLSLResourceBinding
{
spv::ExecutionModel stage = spv::ExecutionModelMax;
uint32_t desc_set = 0;
uint32_t binding = 0;
struct Binding
{
uint32_t register_space = 0;
uint32_t register_binding = 0;
} cbv, uav, srv, sampler;
};
enum HLSLAuxBinding
{
HLSL_AUX_BINDING_BASE_VERTEX_INSTANCE = 0
};
class CompilerHLSL : public CompilerGLSL
{
public:
struct Options
{
uint32_t shader_model = 30; // TODO: map ps_4_0_level_9_0,... somehow
// Allows the PointSize builtin in SM 4.0+, and ignores it, as PointSize is not supported in SM 4+.
bool point_size_compat = false;
// Allows the PointCoord builtin, returns float2(0.5, 0.5), as PointCoord is not supported in HLSL.
bool point_coord_compat = false;
// If true, the backend will assume that VertexIndex and InstanceIndex will need to apply
// a base offset, and you will need to fill in a cbuffer with offsets.
// Set to false if you know you will never use base instance or base vertex
// functionality as it might remove an internal cbuffer.
bool support_nonzero_base_vertex_base_instance = false;
// Forces a storage buffer to always be declared as UAV, even if the readonly decoration is used.
// By default, a readonly storage buffer will be declared as ByteAddressBuffer (SRV) instead.
// Alternatively, use set_hlsl_force_storage_buffer_as_uav to specify individually.
bool force_storage_buffer_as_uav = false;
// Forces any storage image type marked as NonWritable to be considered an SRV instead.
// For this to work with function call parameters, NonWritable must be considered to be part of the type system
// so that NonWritable image arguments are also translated to Texture rather than RWTexture.
bool nonwritable_uav_texture_as_srv = false;
// Enables native 16-bit types. Needs SM 6.2.
// Uses half/int16_t/uint16_t instead of min16* types.
// Also adds support for 16-bit load-store from (RW)ByteAddressBuffer.
bool enable_16bit_types = false;
// If matrices are used as IO variables, flatten the attribute declaration to use
// TEXCOORD{N,N+1,N+2,...} rather than TEXCOORDN_{0,1,2,3}.
// If add_vertex_attribute_remap is used and this feature is used,
// the semantic name will be queried once per active location.
bool flatten_matrix_vertex_input_semantics = false;
// Rather than emitting main() for the entry point, use the name in SPIR-V.
bool use_entry_point_name = false;
};
explicit CompilerHLSL(std::vector<uint32_t> spirv_)
: CompilerGLSL(std::move(spirv_))
{
}
CompilerHLSL(const uint32_t *ir_, size_t size)
: CompilerGLSL(ir_, size)
{
}
explicit CompilerHLSL(const ParsedIR &ir_)
: CompilerGLSL(ir_)
{
}
explicit CompilerHLSL(ParsedIR &&ir_)
: CompilerGLSL(std::move(ir_))
{
}
const Options &get_hlsl_options() const
{
return hlsl_options;
}
void set_hlsl_options(const Options &opts)
{
hlsl_options = opts;
}
// Optionally specify a custom root constant layout.
//
// Push constants ranges will be split up according to the
// layout specified.
void set_root_constant_layouts(std::vector<RootConstants> layout);
// Compiles and remaps vertex attributes at specific locations to a fixed semantic.
// The default is TEXCOORD# where # denotes location.
// Matrices are unrolled to vectors with notation ${SEMANTIC}_#, where # denotes row.
// $SEMANTIC is either TEXCOORD# or a semantic name specified here.
void add_vertex_attribute_remap(const HLSLVertexAttributeRemap &vertex_attributes);
std::string compile() override;
// This is a special HLSL workaround for the NumWorkGroups builtin.
// This does not exist in HLSL, so the calling application must create a dummy cbuffer in
// which the application will store this builtin.
// The cbuffer layout will be:
// cbuffer SPIRV_Cross_NumWorkgroups : register(b#, space#) { uint3 SPIRV_Cross_NumWorkgroups_count; };
// This must be called before compile().
// The function returns 0 if NumWorkGroups builtin is not statically used in the shader from the current entry point.
// If non-zero, this returns the variable ID of a cbuffer which corresponds to
// the cbuffer declared above. By default, no binding or descriptor set decoration is set,
// so the calling application should declare explicit bindings on this ID before calling compile().
VariableID remap_num_workgroups_builtin();
// Controls how resource bindings are declared in the output HLSL.
void set_resource_binding_flags(HLSLBindingFlags flags);
// resource is a resource binding to indicate the HLSL CBV, SRV, UAV or sampler binding
// to use for a particular SPIR-V description set
// and binding. If resource bindings are provided,
// is_hlsl_resource_binding_used() will return true after calling ::compile() if
// the set/binding combination was used by the HLSL code.
void add_hlsl_resource_binding(const HLSLResourceBinding &resource);
bool is_hlsl_resource_binding_used(spv::ExecutionModel model, uint32_t set, uint32_t binding) const;
// Controls which storage buffer bindings will be forced to be declared as UAVs.
void set_hlsl_force_storage_buffer_as_uav(uint32_t desc_set, uint32_t binding);
// By default, these magic buffers are not assigned a specific binding.
void set_hlsl_aux_buffer_binding(HLSLAuxBinding binding, uint32_t register_index, uint32_t register_space);
void unset_hlsl_aux_buffer_binding(HLSLAuxBinding binding);
bool is_hlsl_aux_buffer_binding_used(HLSLAuxBinding binding) const;
private:
std::string type_to_glsl(const SPIRType &type, uint32_t id = 0) override;
std::string image_type_hlsl(const SPIRType &type, uint32_t id);
std::string image_type_hlsl_modern(const SPIRType &type, uint32_t id);
std::string image_type_hlsl_legacy(const SPIRType &type, uint32_t id);
void emit_function_prototype(SPIRFunction &func, const Bitset &return_flags) override;
void emit_hlsl_entry_point();
void emit_header() override;
void emit_resources();
void emit_interface_block_globally(const SPIRVariable &type);
void emit_interface_block_in_struct(const SPIRVariable &var, std::unordered_set<uint32_t> &active_locations);
void emit_interface_block_member_in_struct(const SPIRVariable &var, uint32_t member_index, uint32_t location,
std::unordered_set<uint32_t> &active_locations);
void emit_builtin_inputs_in_struct();
void emit_builtin_outputs_in_struct();
void emit_builtin_primitive_outputs_in_struct();
void emit_texture_op(const Instruction &i, bool sparse) override;
void emit_instruction(const Instruction &instruction) override;
void emit_glsl_op(uint32_t result_type, uint32_t result_id, uint32_t op, const uint32_t *args,
uint32_t count) override;
void emit_buffer_block(const SPIRVariable &type) override;
void emit_push_constant_block(const SPIRVariable &var) override;
void emit_uniform(const SPIRVariable &var) override;
void emit_modern_uniform(const SPIRVariable &var);
void emit_legacy_uniform(const SPIRVariable &var);
void emit_specialization_constants_and_structs();
void emit_composite_constants();
void emit_fixup() override;
std::string builtin_to_glsl(spv::BuiltIn builtin, spv::StorageClass storage) override;
std::string layout_for_member(const SPIRType &type, uint32_t index) override;
std::string to_interpolation_qualifiers(const Bitset &flags) override;
std::string bitcast_glsl_op(const SPIRType &result_type, const SPIRType &argument_type) override;
bool emit_complex_bitcast(uint32_t result_type, uint32_t id, uint32_t op0) override;
std::string to_func_call_arg(const SPIRFunction::Parameter &arg, uint32_t id) override;
std::string to_sampler_expression(uint32_t id);
std::string to_resource_binding(const SPIRVariable &var);
std::string to_resource_binding_sampler(const SPIRVariable &var);
std::string to_resource_register(HLSLBindingFlagBits flag, char space, uint32_t binding, uint32_t set);
std::string to_initializer_expression(const SPIRVariable &var) override;
void emit_sampled_image_op(uint32_t result_type, uint32_t result_id, uint32_t image_id, uint32_t samp_id) override;
void emit_access_chain(const Instruction &instruction);
void emit_load(const Instruction &instruction);
void read_access_chain(std::string *expr, const std::string &lhs, const SPIRAccessChain &chain);
void read_access_chain_struct(const std::string &lhs, const SPIRAccessChain &chain);
void read_access_chain_array(const std::string &lhs, const SPIRAccessChain &chain);
void write_access_chain(const SPIRAccessChain &chain, uint32_t value, const SmallVector<uint32_t> &composite_chain);
void write_access_chain_struct(const SPIRAccessChain &chain, uint32_t value,
const SmallVector<uint32_t> &composite_chain);
void write_access_chain_array(const SPIRAccessChain &chain, uint32_t value,
const SmallVector<uint32_t> &composite_chain);
std::string write_access_chain_value(uint32_t value, const SmallVector<uint32_t> &composite_chain, bool enclose);
void emit_store(const Instruction &instruction);
void emit_atomic(const uint32_t *ops, uint32_t length, spv::Op op);
void emit_subgroup_op(const Instruction &i) override;
void emit_block_hints(const SPIRBlock &block) override;
void emit_struct_member(const SPIRType &type, uint32_t member_type_id, uint32_t index, const std::string &qualifier,
uint32_t base_offset = 0) override;
void emit_rayquery_function(const char *commited, const char *candidate, const uint32_t *ops);
void emit_mesh_tasks(SPIRBlock &block) override;
const char *to_storage_qualifiers_glsl(const SPIRVariable &var) override;
void replace_illegal_names() override;
bool is_hlsl_force_storage_buffer_as_uav(ID id) const;
Options hlsl_options;
// TODO: Refactor this to be more similar to MSL, maybe have some common system in place?
bool requires_op_fmod = false;
bool requires_fp16_packing = false;
bool requires_uint2_packing = false;
bool requires_explicit_fp16_packing = false;
bool requires_unorm8_packing = false;
bool requires_snorm8_packing = false;
bool requires_unorm16_packing = false;
bool requires_snorm16_packing = false;
bool requires_bitfield_insert = false;
bool requires_bitfield_extract = false;
bool requires_inverse_2x2 = false;
bool requires_inverse_3x3 = false;
bool requires_inverse_4x4 = false;
bool requires_scalar_reflect = false;
bool requires_scalar_refract = false;
bool requires_scalar_faceforward = false;
struct TextureSizeVariants
{
// MSVC 2013 workaround.
TextureSizeVariants()
{
srv = 0;
for (auto &unorm : uav)
for (auto &u : unorm)
u = 0;
}
uint64_t srv;
uint64_t uav[3][4];
} required_texture_size_variants;
void require_texture_query_variant(uint32_t var_id);
void emit_texture_size_variants(uint64_t variant_mask, const char *vecsize_qualifier, bool uav,
const char *type_qualifier);
enum TextureQueryVariantDim
{
Query1D = 0,
Query1DArray,
Query2D,
Query2DArray,
Query3D,
QueryBuffer,
QueryCube,
QueryCubeArray,
Query2DMS,
Query2DMSArray,
QueryDimCount
};
enum TextureQueryVariantType
{
QueryTypeFloat = 0,
QueryTypeInt = 16,
QueryTypeUInt = 32,
QueryTypeCount = 3
};
enum BitcastType
{
TypeNormal,
TypePackUint2x32,
TypeUnpackUint64
};
void analyze_meshlet_writes();
void analyze_meshlet_writes(uint32_t func_id, uint32_t id_per_vertex, uint32_t id_per_primitive,
std::unordered_set<uint32_t> &processed_func_ids);
BitcastType get_bitcast_type(uint32_t result_type, uint32_t op0);
void emit_builtin_variables();
bool require_output = false;
bool require_input = false;
SmallVector<HLSLVertexAttributeRemap> remap_vertex_attributes;
uint32_t type_to_consumed_locations(const SPIRType &type) const;
std::string to_semantic(uint32_t location, spv::ExecutionModel em, spv::StorageClass sc);
uint32_t num_workgroups_builtin = 0;
HLSLBindingFlags resource_binding_flags = 0;
// Custom root constant layout, which should be emitted
// when translating push constant ranges.
std::vector<RootConstants> root_constants_layout;
void validate_shader_model();
std::string get_unique_identifier();
uint32_t unique_identifier_count = 0;
std::unordered_map<StageSetBinding, std::pair<HLSLResourceBinding, bool>, InternalHasher> resource_bindings;
void remap_hlsl_resource_binding(HLSLBindingFlagBits type, uint32_t &desc_set, uint32_t &binding);
std::unordered_set<SetBindingPair, InternalHasher> force_uav_buffer_bindings;
struct
{
uint32_t register_index = 0;
uint32_t register_space = 0;
bool explicit_binding = false;
bool used = false;
} base_vertex_info;
// Returns true for BuiltInSampleMask because gl_SampleMask[] is an array in SPIR-V, but SV_Coverage is a scalar in HLSL.
bool builtin_translates_to_nonarray(spv::BuiltIn builtin) const override;
std::vector<TypeID> composite_selection_workaround_types;
std::string get_inner_entry_point_name() const;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

1297
include/vulkan/spirv_cross/spirv_msl.hpp Normal file
View file

File diff suppressed because it is too large Load diff

103
include/vulkan/spirv_cross/spirv_parser.hpp Normal file
View file

@ -0,0 +1,103 @@
/*
* Copyright 2018-2021 Arm Limited
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_PARSER_HPP
#define SPIRV_CROSS_PARSER_HPP
#include "spirv_cross_parsed_ir.hpp"
#include <stdint.h>
namespace SPIRV_CROSS_NAMESPACE
{
class Parser
{
public:
Parser(const uint32_t *spirv_data, size_t word_count);
Parser(std::vector<uint32_t> spirv);
void parse();
ParsedIR &get_parsed_ir()
{
return ir;
}
private:
ParsedIR ir;
SPIRFunction *current_function = nullptr;
SPIRBlock *current_block = nullptr;
// For workarounds.
bool ignore_trailing_block_opcodes = false;
void parse(const Instruction &instr);
const uint32_t *stream(const Instruction &instr) const;
template <typename T, typename... P>
T &set(uint32_t id, P &&... args)
{
ir.add_typed_id(static_cast<Types>(T::type), id);
auto &var = variant_set<T>(ir.ids[id], std::forward<P>(args)...);
var.self = id;
return var;
}
template <typename T>
T &get(uint32_t id)
{
return variant_get<T>(ir.ids[id]);
}
template <typename T>
T *maybe_get(uint32_t id)
{
if (ir.ids[id].get_type() == static_cast<Types>(T::type))
return &get<T>(id);
else
return nullptr;
}
template <typename T>
const T &get(uint32_t id) const
{
return variant_get<T>(ir.ids[id]);
}
template <typename T>
const T *maybe_get(uint32_t id) const
{
if (ir.ids[id].get_type() == T::type)
return &get<T>(id);
else
return nullptr;
}
// This must be an ordered data structure so we always pick the same type aliases.
SmallVector<uint32_t> global_struct_cache;
SmallVector<std::pair<uint32_t, uint32_t>> forward_pointer_fixups;
bool types_are_logically_equivalent(const SPIRType &a, const SPIRType &b) const;
bool variable_storage_is_aliased(const SPIRVariable &v) const;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

91
include/vulkan/spirv_cross/spirv_reflect.hpp Normal file
View file

@ -0,0 +1,91 @@
/*
* Copyright 2018-2021 Bradley Austin Davis
* SPDX-License-Identifier: Apache-2.0 OR MIT
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
/*
* At your option, you may choose to accept this material under either:
* 1. The Apache License, Version 2.0, found at <http://www.apache.org/licenses/LICENSE-2.0>, or
* 2. The MIT License, found at <http://opensource.org/licenses/MIT>.
*/
#ifndef SPIRV_CROSS_REFLECT_HPP
#define SPIRV_CROSS_REFLECT_HPP
#include "spirv_glsl.hpp"
#include <utility>
namespace simple_json
{
class Stream;
}
namespace SPIRV_CROSS_NAMESPACE
{
class CompilerReflection : public CompilerGLSL
{
using Parent = CompilerGLSL;
public:
explicit CompilerReflection(std::vector<uint32_t> spirv_)
: Parent(std::move(spirv_))
{
options.vulkan_semantics = true;
}
CompilerReflection(const uint32_t *ir_, size_t word_count)
: Parent(ir_, word_count)
{
options.vulkan_semantics = true;
}
explicit CompilerReflection(const ParsedIR &ir_)
: CompilerGLSL(ir_)
{
options.vulkan_semantics = true;
}
explicit CompilerReflection(ParsedIR &&ir_)
: CompilerGLSL(std::move(ir_))
{
options.vulkan_semantics = true;
}
void set_format(const std::string &format);
std::string compile() override;
private:
static std::string execution_model_to_str(spv::ExecutionModel model);
void emit_entry_points();
void emit_types();
void emit_resources();
void emit_specialization_constants();
void emit_type(uint32_t type_id, bool &emitted_open_tag);
void emit_type_member(const SPIRType &type, uint32_t index);
void emit_type_member_qualifiers(const SPIRType &type, uint32_t index);
void emit_type_array(const SPIRType &type);
void emit_resources(const char *tag, const SmallVector<Resource> &resources);
bool type_is_reference(const SPIRType &type) const;
std::string to_member_name(const SPIRType &type, uint32_t index) const;
std::shared_ptr<simple_json::Stream> json_stream;
};
} // namespace SPIRV_CROSS_NAMESPACE
#endif

310
include/vulkan/vk_video/vulkan_video_codec_h264std.h Normal file
View file

@ -0,0 +1,310 @@
#ifndef VULKAN_VIDEO_CODEC_H264STD_H_
#define VULKAN_VIDEO_CODEC_H264STD_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h264std 1
#include <stdint.h>
#define STD_VIDEO_H264_CPB_CNT_LIST_SIZE 32
#define STD_VIDEO_H264_SCALING_LIST_4X4_NUM_LISTS 6
#define STD_VIDEO_H264_SCALING_LIST_4X4_NUM_ELEMENTS 16
#define STD_VIDEO_H264_SCALING_LIST_8X8_NUM_LISTS 6
#define STD_VIDEO_H264_SCALING_LIST_8X8_NUM_ELEMENTS 64
#define STD_VIDEO_H264_MAX_NUM_LIST_REF 32
#define STD_VIDEO_H264_MAX_CHROMA_PLANES 2
typedef enum StdVideoH264ChromaFormatIdc {
STD_VIDEO_H264_CHROMA_FORMAT_IDC_MONOCHROME = 0,
STD_VIDEO_H264_CHROMA_FORMAT_IDC_420 = 1,
STD_VIDEO_H264_CHROMA_FORMAT_IDC_422 = 2,
STD_VIDEO_H264_CHROMA_FORMAT_IDC_444 = 3,
STD_VIDEO_H264_CHROMA_FORMAT_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_CHROMA_FORMAT_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264ChromaFormatIdc;
typedef enum StdVideoH264ProfileIdc {
STD_VIDEO_H264_PROFILE_IDC_BASELINE = 66,
STD_VIDEO_H264_PROFILE_IDC_MAIN = 77,
STD_VIDEO_H264_PROFILE_IDC_HIGH = 100,
STD_VIDEO_H264_PROFILE_IDC_HIGH_444_PREDICTIVE = 244,
STD_VIDEO_H264_PROFILE_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_PROFILE_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264ProfileIdc;
typedef enum StdVideoH264LevelIdc {
STD_VIDEO_H264_LEVEL_IDC_1_0 = 0,
STD_VIDEO_H264_LEVEL_IDC_1_1 = 1,
STD_VIDEO_H264_LEVEL_IDC_1_2 = 2,
STD_VIDEO_H264_LEVEL_IDC_1_3 = 3,
STD_VIDEO_H264_LEVEL_IDC_2_0 = 4,
STD_VIDEO_H264_LEVEL_IDC_2_1 = 5,
STD_VIDEO_H264_LEVEL_IDC_2_2 = 6,
STD_VIDEO_H264_LEVEL_IDC_3_0 = 7,
STD_VIDEO_H264_LEVEL_IDC_3_1 = 8,
STD_VIDEO_H264_LEVEL_IDC_3_2 = 9,
STD_VIDEO_H264_LEVEL_IDC_4_0 = 10,
STD_VIDEO_H264_LEVEL_IDC_4_1 = 11,
STD_VIDEO_H264_LEVEL_IDC_4_2 = 12,
STD_VIDEO_H264_LEVEL_IDC_5_0 = 13,
STD_VIDEO_H264_LEVEL_IDC_5_1 = 14,
STD_VIDEO_H264_LEVEL_IDC_5_2 = 15,
STD_VIDEO_H264_LEVEL_IDC_6_0 = 16,
STD_VIDEO_H264_LEVEL_IDC_6_1 = 17,
STD_VIDEO_H264_LEVEL_IDC_6_2 = 18,
STD_VIDEO_H264_LEVEL_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_LEVEL_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264LevelIdc;
typedef enum StdVideoH264PocType {
STD_VIDEO_H264_POC_TYPE_0 = 0,
STD_VIDEO_H264_POC_TYPE_1 = 1,
STD_VIDEO_H264_POC_TYPE_2 = 2,
STD_VIDEO_H264_POC_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_POC_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264PocType;
typedef enum StdVideoH264AspectRatioIdc {
STD_VIDEO_H264_ASPECT_RATIO_IDC_UNSPECIFIED = 0,
STD_VIDEO_H264_ASPECT_RATIO_IDC_SQUARE = 1,
STD_VIDEO_H264_ASPECT_RATIO_IDC_12_11 = 2,
STD_VIDEO_H264_ASPECT_RATIO_IDC_10_11 = 3,
STD_VIDEO_H264_ASPECT_RATIO_IDC_16_11 = 4,
STD_VIDEO_H264_ASPECT_RATIO_IDC_40_33 = 5,
STD_VIDEO_H264_ASPECT_RATIO_IDC_24_11 = 6,
STD_VIDEO_H264_ASPECT_RATIO_IDC_20_11 = 7,
STD_VIDEO_H264_ASPECT_RATIO_IDC_32_11 = 8,
STD_VIDEO_H264_ASPECT_RATIO_IDC_80_33 = 9,
STD_VIDEO_H264_ASPECT_RATIO_IDC_18_11 = 10,
STD_VIDEO_H264_ASPECT_RATIO_IDC_15_11 = 11,
STD_VIDEO_H264_ASPECT_RATIO_IDC_64_33 = 12,
STD_VIDEO_H264_ASPECT_RATIO_IDC_160_99 = 13,
STD_VIDEO_H264_ASPECT_RATIO_IDC_4_3 = 14,
STD_VIDEO_H264_ASPECT_RATIO_IDC_3_2 = 15,
STD_VIDEO_H264_ASPECT_RATIO_IDC_2_1 = 16,
STD_VIDEO_H264_ASPECT_RATIO_IDC_EXTENDED_SAR = 255,
STD_VIDEO_H264_ASPECT_RATIO_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_ASPECT_RATIO_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264AspectRatioIdc;
typedef enum StdVideoH264WeightedBipredIdc {
STD_VIDEO_H264_WEIGHTED_BIPRED_IDC_DEFAULT = 0,
STD_VIDEO_H264_WEIGHTED_BIPRED_IDC_EXPLICIT = 1,
STD_VIDEO_H264_WEIGHTED_BIPRED_IDC_IMPLICIT = 2,
STD_VIDEO_H264_WEIGHTED_BIPRED_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_WEIGHTED_BIPRED_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264WeightedBipredIdc;
typedef enum StdVideoH264ModificationOfPicNumsIdc {
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_SHORT_TERM_SUBTRACT = 0,
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_SHORT_TERM_ADD = 1,
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_LONG_TERM = 2,
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_END = 3,
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_MODIFICATION_OF_PIC_NUMS_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264ModificationOfPicNumsIdc;
typedef enum StdVideoH264MemMgmtControlOp {
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_END = 0,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_UNMARK_SHORT_TERM = 1,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_UNMARK_LONG_TERM = 2,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_MARK_LONG_TERM = 3,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_SET_MAX_LONG_TERM_INDEX = 4,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_UNMARK_ALL = 5,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_MARK_CURRENT_AS_LONG_TERM = 6,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_MEM_MGMT_CONTROL_OP_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264MemMgmtControlOp;
typedef enum StdVideoH264CabacInitIdc {
STD_VIDEO_H264_CABAC_INIT_IDC_0 = 0,
STD_VIDEO_H264_CABAC_INIT_IDC_1 = 1,
STD_VIDEO_H264_CABAC_INIT_IDC_2 = 2,
STD_VIDEO_H264_CABAC_INIT_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_CABAC_INIT_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264CabacInitIdc;
typedef enum StdVideoH264DisableDeblockingFilterIdc {
STD_VIDEO_H264_DISABLE_DEBLOCKING_FILTER_IDC_DISABLED = 0,
STD_VIDEO_H264_DISABLE_DEBLOCKING_FILTER_IDC_ENABLED = 1,
STD_VIDEO_H264_DISABLE_DEBLOCKING_FILTER_IDC_PARTIAL = 2,
STD_VIDEO_H264_DISABLE_DEBLOCKING_FILTER_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_DISABLE_DEBLOCKING_FILTER_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264DisableDeblockingFilterIdc;
typedef enum StdVideoH264SliceType {
STD_VIDEO_H264_SLICE_TYPE_P = 0,
STD_VIDEO_H264_SLICE_TYPE_B = 1,
STD_VIDEO_H264_SLICE_TYPE_I = 2,
STD_VIDEO_H264_SLICE_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_SLICE_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264SliceType;
typedef enum StdVideoH264PictureType {
STD_VIDEO_H264_PICTURE_TYPE_P = 0,
STD_VIDEO_H264_PICTURE_TYPE_B = 1,
STD_VIDEO_H264_PICTURE_TYPE_I = 2,
STD_VIDEO_H264_PICTURE_TYPE_IDR = 5,
STD_VIDEO_H264_PICTURE_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_PICTURE_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264PictureType;
typedef enum StdVideoH264NonVclNaluType {
STD_VIDEO_H264_NON_VCL_NALU_TYPE_SPS = 0,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_PPS = 1,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_AUD = 2,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_PREFIX = 3,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_END_OF_SEQUENCE = 4,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_END_OF_STREAM = 5,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_PRECODED = 6,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H264_NON_VCL_NALU_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH264NonVclNaluType;
typedef struct StdVideoH264SpsVuiFlags {
uint32_t aspect_ratio_info_present_flag : 1;
uint32_t overscan_info_present_flag : 1;
uint32_t overscan_appropriate_flag : 1;
uint32_t video_signal_type_present_flag : 1;
uint32_t video_full_range_flag : 1;
uint32_t color_description_present_flag : 1;
uint32_t chroma_loc_info_present_flag : 1;
uint32_t timing_info_present_flag : 1;
uint32_t fixed_frame_rate_flag : 1;
uint32_t bitstream_restriction_flag : 1;
uint32_t nal_hrd_parameters_present_flag : 1;
uint32_t vcl_hrd_parameters_present_flag : 1;
} StdVideoH264SpsVuiFlags;
typedef struct StdVideoH264HrdParameters {
uint8_t cpb_cnt_minus1;
uint8_t bit_rate_scale;
uint8_t cpb_size_scale;
uint8_t reserved1;
uint32_t bit_rate_value_minus1[STD_VIDEO_H264_CPB_CNT_LIST_SIZE];
uint32_t cpb_size_value_minus1[STD_VIDEO_H264_CPB_CNT_LIST_SIZE];
uint8_t cbr_flag[STD_VIDEO_H264_CPB_CNT_LIST_SIZE];
uint32_t initial_cpb_removal_delay_length_minus1;
uint32_t cpb_removal_delay_length_minus1;
uint32_t dpb_output_delay_length_minus1;
uint32_t time_offset_length;
} StdVideoH264HrdParameters;
typedef struct StdVideoH264SequenceParameterSetVui {
StdVideoH264SpsVuiFlags flags;
StdVideoH264AspectRatioIdc aspect_ratio_idc;
uint16_t sar_width;
uint16_t sar_height;
uint8_t video_format;
uint8_t colour_primaries;
uint8_t transfer_characteristics;
uint8_t matrix_coefficients;
uint32_t num_units_in_tick;
uint32_t time_scale;
uint8_t max_num_reorder_frames;
uint8_t max_dec_frame_buffering;
uint8_t chroma_sample_loc_type_top_field;
uint8_t chroma_sample_loc_type_bottom_field;
uint32_t reserved1;
const StdVideoH264HrdParameters* pHrdParameters;
} StdVideoH264SequenceParameterSetVui;
typedef struct StdVideoH264SpsFlags {
uint32_t constraint_set0_flag : 1;
uint32_t constraint_set1_flag : 1;
uint32_t constraint_set2_flag : 1;
uint32_t constraint_set3_flag : 1;
uint32_t constraint_set4_flag : 1;
uint32_t constraint_set5_flag : 1;
uint32_t direct_8x8_inference_flag : 1;
uint32_t mb_adaptive_frame_field_flag : 1;
uint32_t frame_mbs_only_flag : 1;
uint32_t delta_pic_order_always_zero_flag : 1;
uint32_t separate_colour_plane_flag : 1;
uint32_t gaps_in_frame_num_value_allowed_flag : 1;
uint32_t qpprime_y_zero_transform_bypass_flag : 1;
uint32_t frame_cropping_flag : 1;
uint32_t seq_scaling_matrix_present_flag : 1;
uint32_t vui_parameters_present_flag : 1;
} StdVideoH264SpsFlags;
typedef struct StdVideoH264ScalingLists {
uint16_t scaling_list_present_mask;
uint16_t use_default_scaling_matrix_mask;
uint8_t ScalingList4x4[STD_VIDEO_H264_SCALING_LIST_4X4_NUM_LISTS][STD_VIDEO_H264_SCALING_LIST_4X4_NUM_ELEMENTS];
uint8_t ScalingList8x8[STD_VIDEO_H264_SCALING_LIST_8X8_NUM_LISTS][STD_VIDEO_H264_SCALING_LIST_8X8_NUM_ELEMENTS];
} StdVideoH264ScalingLists;
typedef struct StdVideoH264SequenceParameterSet {
StdVideoH264SpsFlags flags;
StdVideoH264ProfileIdc profile_idc;
StdVideoH264LevelIdc level_idc;
StdVideoH264ChromaFormatIdc chroma_format_idc;
uint8_t seq_parameter_set_id;
uint8_t bit_depth_luma_minus8;
uint8_t bit_depth_chroma_minus8;
uint8_t log2_max_frame_num_minus4;
StdVideoH264PocType pic_order_cnt_type;
int32_t offset_for_non_ref_pic;
int32_t offset_for_top_to_bottom_field;
uint8_t log2_max_pic_order_cnt_lsb_minus4;
uint8_t num_ref_frames_in_pic_order_cnt_cycle;
uint8_t max_num_ref_frames;
uint8_t reserved1;
uint32_t pic_width_in_mbs_minus1;
uint32_t pic_height_in_map_units_minus1;
uint32_t frame_crop_left_offset;
uint32_t frame_crop_right_offset;
uint32_t frame_crop_top_offset;
uint32_t frame_crop_bottom_offset;
uint32_t reserved2;
const int32_t* pOffsetForRefFrame;
const StdVideoH264ScalingLists* pScalingLists;
const StdVideoH264SequenceParameterSetVui* pSequenceParameterSetVui;
} StdVideoH264SequenceParameterSet;
typedef struct StdVideoH264PpsFlags {
uint32_t transform_8x8_mode_flag : 1;
uint32_t redundant_pic_cnt_present_flag : 1;
uint32_t constrained_intra_pred_flag : 1;
uint32_t deblocking_filter_control_present_flag : 1;
uint32_t weighted_pred_flag : 1;
uint32_t bottom_field_pic_order_in_frame_present_flag : 1;
uint32_t entropy_coding_mode_flag : 1;
uint32_t pic_scaling_matrix_present_flag : 1;
} StdVideoH264PpsFlags;
typedef struct StdVideoH264PictureParameterSet {
StdVideoH264PpsFlags flags;
uint8_t seq_parameter_set_id;
uint8_t pic_parameter_set_id;
uint8_t num_ref_idx_l0_default_active_minus1;
uint8_t num_ref_idx_l1_default_active_minus1;
StdVideoH264WeightedBipredIdc weighted_bipred_idc;
int8_t pic_init_qp_minus26;
int8_t pic_init_qs_minus26;
int8_t chroma_qp_index_offset;
int8_t second_chroma_qp_index_offset;
const StdVideoH264ScalingLists* pScalingLists;
} StdVideoH264PictureParameterSet;
#ifdef __cplusplus
}
#endif
#endif

75
include/vulkan/vk_video/vulkan_video_codec_h264std_decode.h Normal file
View file

@ -0,0 +1,75 @@
#ifndef VULKAN_VIDEO_CODEC_H264STD_DECODE_H_
#define VULKAN_VIDEO_CODEC_H264STD_DECODE_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h264std_decode 1
#define VK_STD_VULKAN_VIDEO_CODEC_H264_DECODE_API_VERSION_1_0_0 VK_MAKE_VIDEO_STD_VERSION(1, 0, 0)
#define STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_LIST_SIZE 2
#define VK_STD_VULKAN_VIDEO_CODEC_H264_DECODE_SPEC_VERSION VK_STD_VULKAN_VIDEO_CODEC_H264_DECODE_API_VERSION_1_0_0
#define VK_STD_VULKAN_VIDEO_CODEC_H264_DECODE_EXTENSION_NAME "VK_STD_vulkan_video_codec_h264_decode"
typedef enum StdVideoDecodeH264FieldOrderCount {
STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_TOP = 0,
STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_BOTTOM = 1,
STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_INVALID = 0x7FFFFFFF,
STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_MAX_ENUM = 0x7FFFFFFF
} StdVideoDecodeH264FieldOrderCount;
typedef struct StdVideoDecodeH264PictureInfoFlags {
uint32_t field_pic_flag : 1;
uint32_t is_intra : 1;
uint32_t IdrPicFlag : 1;
uint32_t bottom_field_flag : 1;
uint32_t is_reference : 1;
uint32_t complementary_field_pair : 1;
} StdVideoDecodeH264PictureInfoFlags;
typedef struct StdVideoDecodeH264PictureInfo {
StdVideoDecodeH264PictureInfoFlags flags;
uint8_t seq_parameter_set_id;
uint8_t pic_parameter_set_id;
uint8_t reserved1;
uint8_t reserved2;
uint16_t frame_num;
uint16_t idr_pic_id;
int32_t PicOrderCnt[STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_LIST_SIZE];
} StdVideoDecodeH264PictureInfo;
typedef struct StdVideoDecodeH264ReferenceInfoFlags {
uint32_t top_field_flag : 1;
uint32_t bottom_field_flag : 1;
uint32_t used_for_long_term_reference : 1;
uint32_t is_non_existing : 1;
} StdVideoDecodeH264ReferenceInfoFlags;
typedef struct StdVideoDecodeH264ReferenceInfo {
StdVideoDecodeH264ReferenceInfoFlags flags;
uint16_t FrameNum;
uint16_t reserved;
int32_t PicOrderCnt[STD_VIDEO_DECODE_H264_FIELD_ORDER_COUNT_LIST_SIZE];
} StdVideoDecodeH264ReferenceInfo;
#ifdef __cplusplus
}
#endif
#endif

141
include/vulkan/vk_video/vulkan_video_codec_h264std_encode.h Normal file
View file

@ -0,0 +1,141 @@
#ifndef VULKAN_VIDEO_CODEC_H264STD_ENCODE_H_
#define VULKAN_VIDEO_CODEC_H264STD_ENCODE_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h264std_encode 1
// Vulkan 0.9 provisional Vulkan video H.264 encode std specification version number
#define VK_STD_VULKAN_VIDEO_CODEC_H264_ENCODE_API_VERSION_0_9_9 VK_MAKE_VIDEO_STD_VERSION(0, 9, 9)
#define VK_STD_VULKAN_VIDEO_CODEC_H264_ENCODE_SPEC_VERSION VK_STD_VULKAN_VIDEO_CODEC_H264_ENCODE_API_VERSION_0_9_9
#define VK_STD_VULKAN_VIDEO_CODEC_H264_ENCODE_EXTENSION_NAME "VK_STD_vulkan_video_codec_h264_encode"
typedef struct StdVideoEncodeH264WeightTableFlags {
uint32_t luma_weight_l0_flag;
uint32_t chroma_weight_l0_flag;
uint32_t luma_weight_l1_flag;
uint32_t chroma_weight_l1_flag;
} StdVideoEncodeH264WeightTableFlags;
typedef struct StdVideoEncodeH264WeightTable {
StdVideoEncodeH264WeightTableFlags flags;
uint8_t luma_log2_weight_denom;
uint8_t chroma_log2_weight_denom;
int8_t luma_weight_l0[STD_VIDEO_H264_MAX_NUM_LIST_REF];
int8_t luma_offset_l0[STD_VIDEO_H264_MAX_NUM_LIST_REF];
int8_t chroma_weight_l0[STD_VIDEO_H264_MAX_NUM_LIST_REF][STD_VIDEO_H264_MAX_CHROMA_PLANES];
int8_t chroma_offset_l0[STD_VIDEO_H264_MAX_NUM_LIST_REF][STD_VIDEO_H264_MAX_CHROMA_PLANES];
int8_t luma_weight_l1[STD_VIDEO_H264_MAX_NUM_LIST_REF];
int8_t luma_offset_l1[STD_VIDEO_H264_MAX_NUM_LIST_REF];
int8_t chroma_weight_l1[STD_VIDEO_H264_MAX_NUM_LIST_REF][STD_VIDEO_H264_MAX_CHROMA_PLANES];
int8_t chroma_offset_l1[STD_VIDEO_H264_MAX_NUM_LIST_REF][STD_VIDEO_H264_MAX_CHROMA_PLANES];
} StdVideoEncodeH264WeightTable;
typedef struct StdVideoEncodeH264SliceHeaderFlags {
uint32_t direct_spatial_mv_pred_flag : 1;
uint32_t num_ref_idx_active_override_flag : 1;
uint32_t no_output_of_prior_pics_flag : 1;
uint32_t adaptive_ref_pic_marking_mode_flag : 1;
uint32_t no_prior_references_available_flag : 1;
} StdVideoEncodeH264SliceHeaderFlags;
typedef struct StdVideoEncodeH264PictureInfoFlags {
uint32_t idr_flag : 1;
uint32_t is_reference_flag : 1;
uint32_t used_for_long_term_reference : 1;
} StdVideoEncodeH264PictureInfoFlags;
typedef struct StdVideoEncodeH264ReferenceInfoFlags {
uint32_t used_for_long_term_reference : 1;
} StdVideoEncodeH264ReferenceInfoFlags;
typedef struct StdVideoEncodeH264ReferenceListsInfoFlags {
uint32_t ref_pic_list_modification_flag_l0 : 1;
uint32_t ref_pic_list_modification_flag_l1 : 1;
} StdVideoEncodeH264ReferenceListsInfoFlags;
typedef struct StdVideoEncodeH264RefListModEntry {
StdVideoH264ModificationOfPicNumsIdc modification_of_pic_nums_idc;
uint16_t abs_diff_pic_num_minus1;
uint16_t long_term_pic_num;
} StdVideoEncodeH264RefListModEntry;
typedef struct StdVideoEncodeH264RefPicMarkingEntry {
StdVideoH264MemMgmtControlOp operation;
uint16_t difference_of_pic_nums_minus1;
uint16_t long_term_pic_num;
uint16_t long_term_frame_idx;
uint16_t max_long_term_frame_idx_plus1;
} StdVideoEncodeH264RefPicMarkingEntry;
typedef struct StdVideoEncodeH264ReferenceListsInfo {
StdVideoEncodeH264ReferenceListsInfoFlags flags;
uint8_t refPicList0EntryCount;
uint8_t refPicList1EntryCount;
uint8_t refList0ModOpCount;
uint8_t refList1ModOpCount;
uint8_t refPicMarkingOpCount;
uint8_t reserved1[7];
const uint8_t* pRefPicList0Entries;
const uint8_t* pRefPicList1Entries;
const StdVideoEncodeH264RefListModEntry* pRefList0ModOperations;
const StdVideoEncodeH264RefListModEntry* pRefList1ModOperations;
const StdVideoEncodeH264RefPicMarkingEntry* pRefPicMarkingOperations;
} StdVideoEncodeH264ReferenceListsInfo;
typedef struct StdVideoEncodeH264PictureInfo {
StdVideoEncodeH264PictureInfoFlags flags;
uint8_t seq_parameter_set_id;
uint8_t pic_parameter_set_id;
uint16_t reserved1;
StdVideoH264PictureType pictureType;
uint32_t frame_num;
int32_t PicOrderCnt;
} StdVideoEncodeH264PictureInfo;
typedef struct StdVideoEncodeH264ReferenceInfo {
StdVideoEncodeH264ReferenceInfoFlags flags;
StdVideoH264PictureType pictureType;
uint32_t FrameNum;
int32_t PicOrderCnt;
uint16_t long_term_pic_num;
uint16_t long_term_frame_idx;
} StdVideoEncodeH264ReferenceInfo;
typedef struct StdVideoEncodeH264SliceHeader {
StdVideoEncodeH264SliceHeaderFlags flags;
uint32_t first_mb_in_slice;
StdVideoH264SliceType slice_type;
uint16_t idr_pic_id;
uint8_t num_ref_idx_l0_active_minus1;
uint8_t num_ref_idx_l1_active_minus1;
StdVideoH264CabacInitIdc cabac_init_idc;
StdVideoH264DisableDeblockingFilterIdc disable_deblocking_filter_idc;
int8_t slice_alpha_c0_offset_div2;
int8_t slice_beta_offset_div2;
uint16_t reserved1;
uint32_t reserved2;
const StdVideoEncodeH264WeightTable* pWeightTable;
} StdVideoEncodeH264SliceHeader;
#ifdef __cplusplus
}
#endif
#endif

443
include/vulkan/vk_video/vulkan_video_codec_h265std.h Normal file
View file

@ -0,0 +1,443 @@
#ifndef VULKAN_VIDEO_CODEC_H265STD_H_
#define VULKAN_VIDEO_CODEC_H265STD_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h265std 1
#define STD_VIDEO_H265_SUBLAYERS_LIST_SIZE 7
#define STD_VIDEO_H265_CPB_CNT_LIST_SIZE 32
#define STD_VIDEO_H265_SCALING_LIST_4X4_NUM_LISTS 6
#define STD_VIDEO_H265_SCALING_LIST_4X4_NUM_ELEMENTS 16
#define STD_VIDEO_H265_SCALING_LIST_8X8_NUM_LISTS 6
#define STD_VIDEO_H265_SCALING_LIST_8X8_NUM_ELEMENTS 64
#define STD_VIDEO_H265_SCALING_LIST_16X16_NUM_LISTS 6
#define STD_VIDEO_H265_SCALING_LIST_16X16_NUM_ELEMENTS 64
#define STD_VIDEO_H265_SCALING_LIST_32X32_NUM_LISTS 2
#define STD_VIDEO_H265_SCALING_LIST_32X32_NUM_ELEMENTS 64
#define STD_VIDEO_H265_PREDICTOR_PALETTE_COMPONENTS_LIST_SIZE 3
#define STD_VIDEO_H265_PREDICTOR_PALETTE_COMP_ENTRIES_LIST_SIZE 128
#define STD_VIDEO_H265_MAX_DPB_SIZE 16
#define STD_VIDEO_H265_MAX_LONG_TERM_REF_PICS_SPS 32
#define STD_VIDEO_H265_CHROMA_QP_OFFSET_LIST_SIZE 6
#define STD_VIDEO_H265_CHROMA_QP_OFFSET_TILE_COLS_LIST_SIZE 19
#define STD_VIDEO_H265_CHROMA_QP_OFFSET_TILE_ROWS_LIST_SIZE 21
#define STD_VIDEO_H265_MAX_NUM_LIST_REF 15
#define STD_VIDEO_H265_MAX_CHROMA_PLANES 2
#define STD_VIDEO_H265_MAX_SHORT_TERM_REF_PIC_SETS 64
#define STD_VIDEO_H265_MAX_LONG_TERM_PICS 16
#define STD_VIDEO_H265_MAX_DELTA_POC 48
typedef enum StdVideoH265ChromaFormatIdc {
STD_VIDEO_H265_CHROMA_FORMAT_IDC_MONOCHROME = 0,
STD_VIDEO_H265_CHROMA_FORMAT_IDC_420 = 1,
STD_VIDEO_H265_CHROMA_FORMAT_IDC_422 = 2,
STD_VIDEO_H265_CHROMA_FORMAT_IDC_444 = 3,
STD_VIDEO_H265_CHROMA_FORMAT_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_CHROMA_FORMAT_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265ChromaFormatIdc;
typedef enum StdVideoH265ProfileIdc {
STD_VIDEO_H265_PROFILE_IDC_MAIN = 1,
STD_VIDEO_H265_PROFILE_IDC_MAIN_10 = 2,
STD_VIDEO_H265_PROFILE_IDC_MAIN_STILL_PICTURE = 3,
STD_VIDEO_H265_PROFILE_IDC_FORMAT_RANGE_EXTENSIONS = 4,
STD_VIDEO_H265_PROFILE_IDC_SCC_EXTENSIONS = 9,
STD_VIDEO_H265_PROFILE_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_PROFILE_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265ProfileIdc;
typedef enum StdVideoH265LevelIdc {
STD_VIDEO_H265_LEVEL_IDC_1_0 = 0,
STD_VIDEO_H265_LEVEL_IDC_2_0 = 1,
STD_VIDEO_H265_LEVEL_IDC_2_1 = 2,
STD_VIDEO_H265_LEVEL_IDC_3_0 = 3,
STD_VIDEO_H265_LEVEL_IDC_3_1 = 4,
STD_VIDEO_H265_LEVEL_IDC_4_0 = 5,
STD_VIDEO_H265_LEVEL_IDC_4_1 = 6,
STD_VIDEO_H265_LEVEL_IDC_5_0 = 7,
STD_VIDEO_H265_LEVEL_IDC_5_1 = 8,
STD_VIDEO_H265_LEVEL_IDC_5_2 = 9,
STD_VIDEO_H265_LEVEL_IDC_6_0 = 10,
STD_VIDEO_H265_LEVEL_IDC_6_1 = 11,
STD_VIDEO_H265_LEVEL_IDC_6_2 = 12,
STD_VIDEO_H265_LEVEL_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_LEVEL_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265LevelIdc;
typedef enum StdVideoH265SliceType {
STD_VIDEO_H265_SLICE_TYPE_B = 0,
STD_VIDEO_H265_SLICE_TYPE_P = 1,
STD_VIDEO_H265_SLICE_TYPE_I = 2,
STD_VIDEO_H265_SLICE_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_SLICE_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265SliceType;
typedef enum StdVideoH265PictureType {
STD_VIDEO_H265_PICTURE_TYPE_P = 0,
STD_VIDEO_H265_PICTURE_TYPE_B = 1,
STD_VIDEO_H265_PICTURE_TYPE_I = 2,
STD_VIDEO_H265_PICTURE_TYPE_IDR = 3,
STD_VIDEO_H265_PICTURE_TYPE_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_PICTURE_TYPE_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265PictureType;
typedef enum StdVideoH265AspectRatioIdc {
STD_VIDEO_H265_ASPECT_RATIO_IDC_UNSPECIFIED = 0,
STD_VIDEO_H265_ASPECT_RATIO_IDC_SQUARE = 1,
STD_VIDEO_H265_ASPECT_RATIO_IDC_12_11 = 2,
STD_VIDEO_H265_ASPECT_RATIO_IDC_10_11 = 3,
STD_VIDEO_H265_ASPECT_RATIO_IDC_16_11 = 4,
STD_VIDEO_H265_ASPECT_RATIO_IDC_40_33 = 5,
STD_VIDEO_H265_ASPECT_RATIO_IDC_24_11 = 6,
STD_VIDEO_H265_ASPECT_RATIO_IDC_20_11 = 7,
STD_VIDEO_H265_ASPECT_RATIO_IDC_32_11 = 8,
STD_VIDEO_H265_ASPECT_RATIO_IDC_80_33 = 9,
STD_VIDEO_H265_ASPECT_RATIO_IDC_18_11 = 10,
STD_VIDEO_H265_ASPECT_RATIO_IDC_15_11 = 11,
STD_VIDEO_H265_ASPECT_RATIO_IDC_64_33 = 12,
STD_VIDEO_H265_ASPECT_RATIO_IDC_160_99 = 13,
STD_VIDEO_H265_ASPECT_RATIO_IDC_4_3 = 14,
STD_VIDEO_H265_ASPECT_RATIO_IDC_3_2 = 15,
STD_VIDEO_H265_ASPECT_RATIO_IDC_2_1 = 16,
STD_VIDEO_H265_ASPECT_RATIO_IDC_EXTENDED_SAR = 255,
STD_VIDEO_H265_ASPECT_RATIO_IDC_INVALID = 0x7FFFFFFF,
STD_VIDEO_H265_ASPECT_RATIO_IDC_MAX_ENUM = 0x7FFFFFFF
} StdVideoH265AspectRatioIdc;
typedef struct StdVideoH265DecPicBufMgr {
uint32_t max_latency_increase_plus1[STD_VIDEO_H265_SUBLAYERS_LIST_SIZE];
uint8_t max_dec_pic_buffering_minus1[STD_VIDEO_H265_SUBLAYERS_LIST_SIZE];
uint8_t max_num_reorder_pics[STD_VIDEO_H265_SUBLAYERS_LIST_SIZE];
} StdVideoH265DecPicBufMgr;
typedef struct StdVideoH265SubLayerHrdParameters {
uint32_t bit_rate_value_minus1[STD_VIDEO_H265_CPB_CNT_LIST_SIZE];
uint32_t cpb_size_value_minus1[STD_VIDEO_H265_CPB_CNT_LIST_SIZE];
uint32_t cpb_size_du_value_minus1[STD_VIDEO_H265_CPB_CNT_LIST_SIZE];
uint32_t bit_rate_du_value_minus1[STD_VIDEO_H265_CPB_CNT_LIST_SIZE];
uint32_t cbr_flag;
} StdVideoH265SubLayerHrdParameters;
typedef struct StdVideoH265HrdFlags {
uint32_t nal_hrd_parameters_present_flag : 1;
uint32_t vcl_hrd_parameters_present_flag : 1;
uint32_t sub_pic_hrd_params_present_flag : 1;
uint32_t sub_pic_cpb_params_in_pic_timing_sei_flag : 1;
uint32_t fixed_pic_rate_general_flag : 8;
uint32_t fixed_pic_rate_within_cvs_flag : 8;
uint32_t low_delay_hrd_flag : 8;
} StdVideoH265HrdFlags;
typedef struct StdVideoH265HrdParameters {
StdVideoH265HrdFlags flags;
uint8_t tick_divisor_minus2;
uint8_t du_cpb_removal_delay_increment_length_minus1;
uint8_t dpb_output_delay_du_length_minus1;
uint8_t bit_rate_scale;
uint8_t cpb_size_scale;
uint8_t cpb_size_du_scale;
uint8_t initial_cpb_removal_delay_length_minus1;
uint8_t au_cpb_removal_delay_length_minus1;
uint8_t dpb_output_delay_length_minus1;
uint8_t cpb_cnt_minus1[STD_VIDEO_H265_SUBLAYERS_LIST_SIZE];
uint16_t elemental_duration_in_tc_minus1[STD_VIDEO_H265_SUBLAYERS_LIST_SIZE];
uint16_t reserved[3];
const StdVideoH265SubLayerHrdParameters* pSubLayerHrdParametersNal;
const StdVideoH265SubLayerHrdParameters* pSubLayerHrdParametersVcl;
} StdVideoH265HrdParameters;
typedef struct StdVideoH265VpsFlags {
uint32_t vps_temporal_id_nesting_flag : 1;
uint32_t vps_sub_layer_ordering_info_present_flag : 1;
uint32_t vps_timing_info_present_flag : 1;
uint32_t vps_poc_proportional_to_timing_flag : 1;
} StdVideoH265VpsFlags;
typedef struct StdVideoH265ProfileTierLevelFlags {
uint32_t general_tier_flag : 1;
uint32_t general_progressive_source_flag : 1;
uint32_t general_interlaced_source_flag : 1;
uint32_t general_non_packed_constraint_flag : 1;
uint32_t general_frame_only_constraint_flag : 1;
} StdVideoH265ProfileTierLevelFlags;
typedef struct StdVideoH265ProfileTierLevel {
StdVideoH265ProfileTierLevelFlags flags;
StdVideoH265ProfileIdc general_profile_idc;
StdVideoH265LevelIdc general_level_idc;
} StdVideoH265ProfileTierLevel;
typedef struct StdVideoH265VideoParameterSet {
StdVideoH265VpsFlags flags;
uint8_t vps_video_parameter_set_id;
uint8_t vps_max_sub_layers_minus1;
uint8_t reserved1;
uint8_t reserved2;
uint32_t vps_num_units_in_tick;
uint32_t vps_time_scale;
uint32_t vps_num_ticks_poc_diff_one_minus1;
uint32_t reserved3;
const StdVideoH265DecPicBufMgr* pDecPicBufMgr;
const StdVideoH265HrdParameters* pHrdParameters;
const StdVideoH265ProfileTierLevel* pProfileTierLevel;
} StdVideoH265VideoParameterSet;
typedef struct StdVideoH265ScalingLists {
uint8_t ScalingList4x4[STD_VIDEO_H265_SCALING_LIST_4X4_NUM_LISTS][STD_VIDEO_H265_SCALING_LIST_4X4_NUM_ELEMENTS];
uint8_t ScalingList8x8[STD_VIDEO_H265_SCALING_LIST_8X8_NUM_LISTS][STD_VIDEO_H265_SCALING_LIST_8X8_NUM_ELEMENTS];
uint8_t ScalingList16x16[STD_VIDEO_H265_SCALING_LIST_16X16_NUM_LISTS][STD_VIDEO_H265_SCALING_LIST_16X16_NUM_ELEMENTS];
uint8_t ScalingList32x32[STD_VIDEO_H265_SCALING_LIST_32X32_NUM_LISTS][STD_VIDEO_H265_SCALING_LIST_32X32_NUM_ELEMENTS];
uint8_t ScalingListDCCoef16x16[STD_VIDEO_H265_SCALING_LIST_16X16_NUM_LISTS];
uint8_t ScalingListDCCoef32x32[STD_VIDEO_H265_SCALING_LIST_32X32_NUM_LISTS];
} StdVideoH265ScalingLists;
typedef struct StdVideoH265SpsVuiFlags {
uint32_t aspect_ratio_info_present_flag : 1;
uint32_t overscan_info_present_flag : 1;
uint32_t overscan_appropriate_flag : 1;
uint32_t video_signal_type_present_flag : 1;
uint32_t video_full_range_flag : 1;
uint32_t colour_description_present_flag : 1;
uint32_t chroma_loc_info_present_flag : 1;
uint32_t neutral_chroma_indication_flag : 1;
uint32_t field_seq_flag : 1;
uint32_t frame_field_info_present_flag : 1;
uint32_t default_display_window_flag : 1;
uint32_t vui_timing_info_present_flag : 1;
uint32_t vui_poc_proportional_to_timing_flag : 1;
uint32_t vui_hrd_parameters_present_flag : 1;
uint32_t bitstream_restriction_flag : 1;
uint32_t tiles_fixed_structure_flag : 1;
uint32_t motion_vectors_over_pic_boundaries_flag : 1;
uint32_t restricted_ref_pic_lists_flag : 1;
} StdVideoH265SpsVuiFlags;
typedef struct StdVideoH265SequenceParameterSetVui {
StdVideoH265SpsVuiFlags flags;
StdVideoH265AspectRatioIdc aspect_ratio_idc;
uint16_t sar_width;
uint16_t sar_height;
uint8_t video_format;
uint8_t colour_primaries;
uint8_t transfer_characteristics;
uint8_t matrix_coeffs;
uint8_t chroma_sample_loc_type_top_field;
uint8_t chroma_sample_loc_type_bottom_field;
uint8_t reserved1;
uint8_t reserved2;
uint16_t def_disp_win_left_offset;
uint16_t def_disp_win_right_offset;
uint16_t def_disp_win_top_offset;
uint16_t def_disp_win_bottom_offset;
uint32_t vui_num_units_in_tick;
uint32_t vui_time_scale;
uint32_t vui_num_ticks_poc_diff_one_minus1;
uint16_t min_spatial_segmentation_idc;
uint16_t reserved3;
uint8_t max_bytes_per_pic_denom;
uint8_t max_bits_per_min_cu_denom;
uint8_t log2_max_mv_length_horizontal;
uint8_t log2_max_mv_length_vertical;
const StdVideoH265HrdParameters* pHrdParameters;
} StdVideoH265SequenceParameterSetVui;
typedef struct StdVideoH265PredictorPaletteEntries {
uint16_t PredictorPaletteEntries[STD_VIDEO_H265_PREDICTOR_PALETTE_COMPONENTS_LIST_SIZE][STD_VIDEO_H265_PREDICTOR_PALETTE_COMP_ENTRIES_LIST_SIZE];
} StdVideoH265PredictorPaletteEntries;
typedef struct StdVideoH265SpsFlags {
uint32_t sps_temporal_id_nesting_flag : 1;
uint32_t separate_colour_plane_flag : 1;
uint32_t conformance_window_flag : 1;
uint32_t sps_sub_layer_ordering_info_present_flag : 1;
uint32_t scaling_list_enabled_flag : 1;
uint32_t sps_scaling_list_data_present_flag : 1;
uint32_t amp_enabled_flag : 1;
uint32_t sample_adaptive_offset_enabled_flag : 1;
uint32_t pcm_enabled_flag : 1;
uint32_t pcm_loop_filter_disabled_flag : 1;
uint32_t long_term_ref_pics_present_flag : 1;
uint32_t sps_temporal_mvp_enabled_flag : 1;
uint32_t strong_intra_smoothing_enabled_flag : 1;
uint32_t vui_parameters_present_flag : 1;
uint32_t sps_extension_present_flag : 1;
uint32_t sps_range_extension_flag : 1;
uint32_t transform_skip_rotation_enabled_flag : 1;
uint32_t transform_skip_context_enabled_flag : 1;
uint32_t implicit_rdpcm_enabled_flag : 1;
uint32_t explicit_rdpcm_enabled_flag : 1;
uint32_t extended_precision_processing_flag : 1;
uint32_t intra_smoothing_disabled_flag : 1;
uint32_t high_precision_offsets_enabled_flag : 1;
uint32_t persistent_rice_adaptation_enabled_flag : 1;
uint32_t cabac_bypass_alignment_enabled_flag : 1;
uint32_t sps_scc_extension_flag : 1;
uint32_t sps_curr_pic_ref_enabled_flag : 1;
uint32_t palette_mode_enabled_flag : 1;
uint32_t sps_palette_predictor_initializers_present_flag : 1;
uint32_t intra_boundary_filtering_disabled_flag : 1;
} StdVideoH265SpsFlags;
typedef struct StdVideoH265ShortTermRefPicSetFlags {
uint32_t inter_ref_pic_set_prediction_flag : 1;
uint32_t delta_rps_sign : 1;
} StdVideoH265ShortTermRefPicSetFlags;
typedef struct StdVideoH265ShortTermRefPicSet {
StdVideoH265ShortTermRefPicSetFlags flags;
uint32_t delta_idx_minus1;
uint16_t use_delta_flag;
uint16_t abs_delta_rps_minus1;
uint16_t used_by_curr_pic_flag;
uint16_t used_by_curr_pic_s0_flag;
uint16_t used_by_curr_pic_s1_flag;
uint16_t reserved1;
uint8_t reserved2;
uint8_t reserved3;
uint8_t num_negative_pics;
uint8_t num_positive_pics;
uint16_t delta_poc_s0_minus1[STD_VIDEO_H265_MAX_DPB_SIZE];
uint16_t delta_poc_s1_minus1[STD_VIDEO_H265_MAX_DPB_SIZE];
} StdVideoH265ShortTermRefPicSet;
typedef struct StdVideoH265LongTermRefPicsSps {
uint32_t used_by_curr_pic_lt_sps_flag;
uint32_t lt_ref_pic_poc_lsb_sps[STD_VIDEO_H265_MAX_LONG_TERM_REF_PICS_SPS];
} StdVideoH265LongTermRefPicsSps;
typedef struct StdVideoH265SequenceParameterSet {
StdVideoH265SpsFlags flags;
StdVideoH265ChromaFormatIdc chroma_format_idc;
uint32_t pic_width_in_luma_samples;
uint32_t pic_height_in_luma_samples;
uint8_t sps_video_parameter_set_id;
uint8_t sps_max_sub_layers_minus1;
uint8_t sps_seq_parameter_set_id;
uint8_t bit_depth_luma_minus8;
uint8_t bit_depth_chroma_minus8;
uint8_t log2_max_pic_order_cnt_lsb_minus4;
uint8_t log2_min_luma_coding_block_size_minus3;
uint8_t log2_diff_max_min_luma_coding_block_size;
uint8_t log2_min_luma_transform_block_size_minus2;
uint8_t log2_diff_max_min_luma_transform_block_size;
uint8_t max_transform_hierarchy_depth_inter;
uint8_t max_transform_hierarchy_depth_intra;
uint8_t num_short_term_ref_pic_sets;
uint8_t num_long_term_ref_pics_sps;
uint8_t pcm_sample_bit_depth_luma_minus1;
uint8_t pcm_sample_bit_depth_chroma_minus1;
uint8_t log2_min_pcm_luma_coding_block_size_minus3;
uint8_t log2_diff_max_min_pcm_luma_coding_block_size;
uint8_t reserved1;
uint8_t reserved2;
uint8_t palette_max_size;
uint8_t delta_palette_max_predictor_size;
uint8_t motion_vector_resolution_control_idc;
uint8_t sps_num_palette_predictor_initializers_minus1;
uint32_t conf_win_left_offset;
uint32_t conf_win_right_offset;
uint32_t conf_win_top_offset;
uint32_t conf_win_bottom_offset;
const StdVideoH265ProfileTierLevel* pProfileTierLevel;
const StdVideoH265DecPicBufMgr* pDecPicBufMgr;
const StdVideoH265ScalingLists* pScalingLists;
const StdVideoH265ShortTermRefPicSet* pShortTermRefPicSet;
const StdVideoH265LongTermRefPicsSps* pLongTermRefPicsSps;
const StdVideoH265SequenceParameterSetVui* pSequenceParameterSetVui;
const StdVideoH265PredictorPaletteEntries* pPredictorPaletteEntries;
} StdVideoH265SequenceParameterSet;
typedef struct StdVideoH265PpsFlags {
uint32_t dependent_slice_segments_enabled_flag : 1;
uint32_t output_flag_present_flag : 1;
uint32_t sign_data_hiding_enabled_flag : 1;
uint32_t cabac_init_present_flag : 1;
uint32_t constrained_intra_pred_flag : 1;
uint32_t transform_skip_enabled_flag : 1;
uint32_t cu_qp_delta_enabled_flag : 1;
uint32_t pps_slice_chroma_qp_offsets_present_flag : 1;
uint32_t weighted_pred_flag : 1;
uint32_t weighted_bipred_flag : 1;
uint32_t transquant_bypass_enabled_flag : 1;
uint32_t tiles_enabled_flag : 1;
uint32_t entropy_coding_sync_enabled_flag : 1;
uint32_t uniform_spacing_flag : 1;
uint32_t loop_filter_across_tiles_enabled_flag : 1;
uint32_t pps_loop_filter_across_slices_enabled_flag : 1;
uint32_t deblocking_filter_control_present_flag : 1;
uint32_t deblocking_filter_override_enabled_flag : 1;
uint32_t pps_deblocking_filter_disabled_flag : 1;
uint32_t pps_scaling_list_data_present_flag : 1;
uint32_t lists_modification_present_flag : 1;
uint32_t slice_segment_header_extension_present_flag : 1;
uint32_t pps_extension_present_flag : 1;
uint32_t cross_component_prediction_enabled_flag : 1;
uint32_t chroma_qp_offset_list_enabled_flag : 1;
uint32_t pps_curr_pic_ref_enabled_flag : 1;
uint32_t residual_adaptive_colour_transform_enabled_flag : 1;
uint32_t pps_slice_act_qp_offsets_present_flag : 1;
uint32_t pps_palette_predictor_initializers_present_flag : 1;
uint32_t monochrome_palette_flag : 1;
uint32_t pps_range_extension_flag : 1;
} StdVideoH265PpsFlags;
typedef struct StdVideoH265PictureParameterSet {
StdVideoH265PpsFlags flags;
uint8_t pps_pic_parameter_set_id;
uint8_t pps_seq_parameter_set_id;
uint8_t sps_video_parameter_set_id;
uint8_t num_extra_slice_header_bits;
uint8_t num_ref_idx_l0_default_active_minus1;
uint8_t num_ref_idx_l1_default_active_minus1;
int8_t init_qp_minus26;
uint8_t diff_cu_qp_delta_depth;
int8_t pps_cb_qp_offset;
int8_t pps_cr_qp_offset;
int8_t pps_beta_offset_div2;
int8_t pps_tc_offset_div2;
uint8_t log2_parallel_merge_level_minus2;
uint8_t log2_max_transform_skip_block_size_minus2;
uint8_t diff_cu_chroma_qp_offset_depth;
uint8_t chroma_qp_offset_list_len_minus1;
int8_t cb_qp_offset_list[STD_VIDEO_H265_CHROMA_QP_OFFSET_LIST_SIZE];
int8_t cr_qp_offset_list[STD_VIDEO_H265_CHROMA_QP_OFFSET_LIST_SIZE];
uint8_t log2_sao_offset_scale_luma;
uint8_t log2_sao_offset_scale_chroma;
int8_t pps_act_y_qp_offset_plus5;
int8_t pps_act_cb_qp_offset_plus5;
int8_t pps_act_cr_qp_offset_plus3;
uint8_t pps_num_palette_predictor_initializers;
uint8_t luma_bit_depth_entry_minus8;
uint8_t chroma_bit_depth_entry_minus8;
uint8_t num_tile_columns_minus1;
uint8_t num_tile_rows_minus1;
uint8_t reserved1;
uint8_t reserved2;
uint16_t column_width_minus1[STD_VIDEO_H265_CHROMA_QP_OFFSET_TILE_COLS_LIST_SIZE];
uint16_t row_height_minus1[STD_VIDEO_H265_CHROMA_QP_OFFSET_TILE_ROWS_LIST_SIZE];
uint32_t reserved3;
const StdVideoH265ScalingLists* pScalingLists;
const StdVideoH265PredictorPaletteEntries* pPredictorPaletteEntries;
} StdVideoH265PictureParameterSet;
#ifdef __cplusplus
}
#endif
#endif

65
include/vulkan/vk_video/vulkan_video_codec_h265std_decode.h Normal file
View file

@ -0,0 +1,65 @@
#ifndef VULKAN_VIDEO_CODEC_H265STD_DECODE_H_
#define VULKAN_VIDEO_CODEC_H265STD_DECODE_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h265std_decode 1
#define VK_STD_VULKAN_VIDEO_CODEC_H265_DECODE_API_VERSION_1_0_0 VK_MAKE_VIDEO_STD_VERSION(1, 0, 0)
#define STD_VIDEO_DECODE_H265_REF_PIC_SET_LIST_SIZE 8
#define VK_STD_VULKAN_VIDEO_CODEC_H265_DECODE_SPEC_VERSION VK_STD_VULKAN_VIDEO_CODEC_H265_DECODE_API_VERSION_1_0_0
#define VK_STD_VULKAN_VIDEO_CODEC_H265_DECODE_EXTENSION_NAME "VK_STD_vulkan_video_codec_h265_decode"
typedef struct StdVideoDecodeH265PictureInfoFlags {
uint32_t IrapPicFlag : 1;
uint32_t IdrPicFlag : 1;
uint32_t IsReference : 1;
uint32_t short_term_ref_pic_set_sps_flag : 1;
} StdVideoDecodeH265PictureInfoFlags;
typedef struct StdVideoDecodeH265PictureInfo {
StdVideoDecodeH265PictureInfoFlags flags;
uint8_t sps_video_parameter_set_id;
uint8_t pps_seq_parameter_set_id;
uint8_t pps_pic_parameter_set_id;
uint8_t NumDeltaPocsOfRefRpsIdx;
int32_t PicOrderCntVal;
uint16_t NumBitsForSTRefPicSetInSlice;
uint16_t reserved;
uint8_t RefPicSetStCurrBefore[STD_VIDEO_DECODE_H265_REF_PIC_SET_LIST_SIZE];
uint8_t RefPicSetStCurrAfter[STD_VIDEO_DECODE_H265_REF_PIC_SET_LIST_SIZE];
uint8_t RefPicSetLtCurr[STD_VIDEO_DECODE_H265_REF_PIC_SET_LIST_SIZE];
} StdVideoDecodeH265PictureInfo;
typedef struct StdVideoDecodeH265ReferenceInfoFlags {
uint32_t used_for_long_term_reference : 1;
uint32_t unused_for_reference : 1;
} StdVideoDecodeH265ReferenceInfoFlags;
typedef struct StdVideoDecodeH265ReferenceInfo {
StdVideoDecodeH265ReferenceInfoFlags flags;
int32_t PicOrderCntVal;
} StdVideoDecodeH265ReferenceInfo;
#ifdef __cplusplus
}
#endif
#endif

150
include/vulkan/vk_video/vulkan_video_codec_h265std_encode.h Normal file
View file

@ -0,0 +1,150 @@
#ifndef VULKAN_VIDEO_CODEC_H265STD_ENCODE_H_
#define VULKAN_VIDEO_CODEC_H265STD_ENCODE_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codec_h265std_encode 1
// Vulkan 0.9 provisional Vulkan video H.265 encode std specification version number
#define VK_STD_VULKAN_VIDEO_CODEC_H265_ENCODE_API_VERSION_0_9_10 VK_MAKE_VIDEO_STD_VERSION(0, 9, 10)
#define VK_STD_VULKAN_VIDEO_CODEC_H265_ENCODE_SPEC_VERSION VK_STD_VULKAN_VIDEO_CODEC_H265_ENCODE_API_VERSION_0_9_10
#define VK_STD_VULKAN_VIDEO_CODEC_H265_ENCODE_EXTENSION_NAME "VK_STD_vulkan_video_codec_h265_encode"
typedef struct StdVideoEncodeH265WeightTableFlags {
uint16_t luma_weight_l0_flag;
uint16_t chroma_weight_l0_flag;
uint16_t luma_weight_l1_flag;
uint16_t chroma_weight_l1_flag;
} StdVideoEncodeH265WeightTableFlags;
typedef struct StdVideoEncodeH265WeightTable {
StdVideoEncodeH265WeightTableFlags flags;
uint8_t luma_log2_weight_denom;
int8_t delta_chroma_log2_weight_denom;
int8_t delta_luma_weight_l0[STD_VIDEO_H265_MAX_NUM_LIST_REF];
int8_t luma_offset_l0[STD_VIDEO_H265_MAX_NUM_LIST_REF];
int8_t delta_chroma_weight_l0[STD_VIDEO_H265_MAX_NUM_LIST_REF][STD_VIDEO_H265_MAX_CHROMA_PLANES];
int8_t delta_chroma_offset_l0[STD_VIDEO_H265_MAX_NUM_LIST_REF][STD_VIDEO_H265_MAX_CHROMA_PLANES];
int8_t delta_luma_weight_l1[STD_VIDEO_H265_MAX_NUM_LIST_REF];
int8_t luma_offset_l1[STD_VIDEO_H265_MAX_NUM_LIST_REF];
int8_t delta_chroma_weight_l1[STD_VIDEO_H265_MAX_NUM_LIST_REF][STD_VIDEO_H265_MAX_CHROMA_PLANES];
int8_t delta_chroma_offset_l1[STD_VIDEO_H265_MAX_NUM_LIST_REF][STD_VIDEO_H265_MAX_CHROMA_PLANES];
} StdVideoEncodeH265WeightTable;
typedef struct StdVideoEncodeH265SliceSegmentHeaderFlags {
uint32_t first_slice_segment_in_pic_flag : 1;
uint32_t no_output_of_prior_pics_flag : 1;
uint32_t dependent_slice_segment_flag : 1;
uint32_t pic_output_flag : 1;
uint32_t short_term_ref_pic_set_sps_flag : 1;
uint32_t slice_temporal_mvp_enable_flag : 1;
uint32_t slice_sao_luma_flag : 1;
uint32_t slice_sao_chroma_flag : 1;
uint32_t num_ref_idx_active_override_flag : 1;
uint32_t mvd_l1_zero_flag : 1;
uint32_t cabac_init_flag : 1;
uint32_t cu_chroma_qp_offset_enabled_flag : 1;
uint32_t deblocking_filter_override_flag : 1;
uint32_t slice_deblocking_filter_disabled_flag : 1;
uint32_t collocated_from_l0_flag : 1;
uint32_t slice_loop_filter_across_slices_enabled_flag : 1;
} StdVideoEncodeH265SliceSegmentHeaderFlags;
typedef struct StdVideoEncodeH265SliceSegmentLongTermRefPics {
uint8_t num_long_term_sps;
uint8_t num_long_term_pics;
uint8_t lt_idx_sps[STD_VIDEO_H265_MAX_LONG_TERM_REF_PICS_SPS];
uint8_t poc_lsb_lt[STD_VIDEO_H265_MAX_LONG_TERM_PICS];
uint16_t used_by_curr_pic_lt_flag;
uint8_t delta_poc_msb_present_flag[STD_VIDEO_H265_MAX_DELTA_POC];
uint8_t delta_poc_msb_cycle_lt[STD_VIDEO_H265_MAX_DELTA_POC];
} StdVideoEncodeH265SliceSegmentLongTermRefPics;
typedef struct StdVideoEncodeH265SliceSegmentHeader {
StdVideoEncodeH265SliceSegmentHeaderFlags flags;
StdVideoH265SliceType slice_type;
uint32_t slice_segment_address;
uint8_t short_term_ref_pic_set_idx;
uint8_t collocated_ref_idx;
uint8_t num_ref_idx_l0_active_minus1;
uint8_t num_ref_idx_l1_active_minus1;
uint8_t MaxNumMergeCand;
int8_t slice_cb_qp_offset;
int8_t slice_cr_qp_offset;
int8_t slice_beta_offset_div2;
int8_t slice_tc_offset_div2;
int8_t slice_act_y_qp_offset;
int8_t slice_act_cb_qp_offset;
int8_t slice_act_cr_qp_offset;
const StdVideoH265ShortTermRefPicSet* pShortTermRefPicSet;
const StdVideoEncodeH265SliceSegmentLongTermRefPics* pLongTermRefPics;
const StdVideoEncodeH265WeightTable* pWeightTable;
} StdVideoEncodeH265SliceSegmentHeader;
typedef struct StdVideoEncodeH265ReferenceListsInfoFlags {
uint32_t ref_pic_list_modification_flag_l0 : 1;
uint32_t ref_pic_list_modification_flag_l1 : 1;
} StdVideoEncodeH265ReferenceListsInfoFlags;
typedef struct StdVideoEncodeH265ReferenceListsInfo {
StdVideoEncodeH265ReferenceListsInfoFlags flags;
uint8_t num_ref_idx_l0_active_minus1;
uint8_t num_ref_idx_l1_active_minus1;
uint16_t reserved1;
const uint8_t* pRefPicList0Entries;
const uint8_t* pRefPicList1Entries;
const uint8_t* pRefList0Modifications;
const uint8_t* pRefList1Modifications;
} StdVideoEncodeH265ReferenceListsInfo;
typedef struct StdVideoEncodeH265PictureInfoFlags {
uint32_t is_reference_flag : 1;
uint32_t IrapPicFlag : 1;
uint32_t long_term_flag : 1;
uint32_t discardable_flag : 1;
uint32_t cross_layer_bla_flag : 1;
} StdVideoEncodeH265PictureInfoFlags;
typedef struct StdVideoEncodeH265PictureInfo {
StdVideoEncodeH265PictureInfoFlags flags;
StdVideoH265PictureType PictureType;
uint8_t sps_video_parameter_set_id;
uint8_t pps_seq_parameter_set_id;
uint8_t pps_pic_parameter_set_id;
uint8_t TemporalId;
int32_t PicOrderCntVal;
} StdVideoEncodeH265PictureInfo;
typedef struct StdVideoEncodeH265ReferenceInfoFlags {
uint32_t used_for_long_term_reference : 1;
uint32_t unused_for_reference : 1;
} StdVideoEncodeH265ReferenceInfoFlags;
typedef struct StdVideoEncodeH265ReferenceInfo {
StdVideoEncodeH265ReferenceInfoFlags flags;
StdVideoH265PictureType PictureType;
int32_t PicOrderCntVal;
uint8_t TemporalId;
} StdVideoEncodeH265ReferenceInfo;
#ifdef __cplusplus
}
#endif
#endif

31
include/vulkan/vk_video/vulkan_video_codecs_common.h Normal file
View file

@ -0,0 +1,31 @@
#ifndef VULKAN_VIDEO_CODECS_COMMON_H_
#define VULKAN_VIDEO_CODECS_COMMON_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define vulkan_video_codecs_common 1
#define VK_MAKE_VIDEO_STD_VERSION(major, minor, patch) \
((((uint32_t)(major)) << 22) | (((uint32_t)(minor)) << 12) | ((uint32_t)(patch)))
#ifdef __cplusplus
}
#endif
#endif

10862
include/vulkan/vulkan/vk_enum_string_helper.h Normal file
View file

File diff suppressed because it is too large Load diff

258
include/vulkan/vulkan/vk_icd.h Normal file
View file

@ -0,0 +1,258 @@
//
// File: vk_icd.h
//
/*
* Copyright (c) 2015-2023 LunarG, Inc.
* Copyright (c) 2015-2023 The Khronos Group Inc.
* Copyright (c) 2015-2023 Valve Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
#pragma once
#include "vulkan.h"
#include <stdbool.h>
// Loader-ICD version negotiation API. Versions add the following features:
// Version 0 - Initial. Doesn't support vk_icdGetInstanceProcAddr
// or vk_icdNegotiateLoaderICDInterfaceVersion.
// Version 1 - Add support for vk_icdGetInstanceProcAddr.
// Version 2 - Add Loader/ICD Interface version negotiation
// via vk_icdNegotiateLoaderICDInterfaceVersion.
// Version 3 - Add ICD creation/destruction of KHR_surface objects.
// Version 4 - Add unknown physical device extension querying via
// vk_icdGetPhysicalDeviceProcAddr.
// Version 5 - Tells ICDs that the loader is now paying attention to the
// application version of Vulkan passed into the ApplicationInfo
// structure during vkCreateInstance. This will tell the ICD
// that if the loader is older, it should automatically fail a
// call for any API version > 1.0. Otherwise, the loader will
// manually determine if it can support the expected version.
// Version 6 - Add support for vk_icdEnumerateAdapterPhysicalDevices.
// Version 7 - If an ICD supports any of the following functions, they must be
// queryable with vk_icdGetInstanceProcAddr:
// vk_icdNegotiateLoaderICDInterfaceVersion
// vk_icdGetPhysicalDeviceProcAddr
// vk_icdEnumerateAdapterPhysicalDevices (Windows only)
// In addition, these functions no longer need to be exported directly.
// This version allows drivers provided through the extension
// VK_LUNARG_direct_driver_loading be able to support the entire
// Driver-Loader interface.
#define CURRENT_LOADER_ICD_INTERFACE_VERSION 7
#define MIN_SUPPORTED_LOADER_ICD_INTERFACE_VERSION 0
#define MIN_PHYS_DEV_EXTENSION_ICD_INTERFACE_VERSION 4
// Old typedefs that don't follow a proper naming convention but are preserved for compatibility
typedef VkResult(VKAPI_PTR *PFN_vkNegotiateLoaderICDInterfaceVersion)(uint32_t *pVersion);
// This is defined in vk_layer.h which will be found by the loader, but if an ICD is building against this
// file directly, it won't be found.
#ifndef PFN_GetPhysicalDeviceProcAddr
typedef PFN_vkVoidFunction(VKAPI_PTR *PFN_GetPhysicalDeviceProcAddr)(VkInstance instance, const char *pName);
#endif
// Typedefs for loader/ICD interface
typedef VkResult (VKAPI_PTR *PFN_vk_icdNegotiateLoaderICDInterfaceVersion)(uint32_t* pVersion);
typedef PFN_vkVoidFunction (VKAPI_PTR *PFN_vk_icdGetInstanceProcAddr)(VkInstance instance, const char* pName);
typedef PFN_vkVoidFunction (VKAPI_PTR *PFN_vk_icdGetPhysicalDeviceProcAddr)(VkInstance instance, const char* pName);
#if defined(VK_USE_PLATFORM_WIN32_KHR)
typedef VkResult (VKAPI_PTR *PFN_vk_icdEnumerateAdapterPhysicalDevices)(VkInstance instance, LUID adapterLUID,
uint32_t* pPhysicalDeviceCount, VkPhysicalDevice* pPhysicalDevices);
#endif
// Prototypes for loader/ICD interface
#if !defined(VK_NO_PROTOTYPES)
#ifdef __cplusplus
extern "C" {
#endif
VKAPI_ATTR VkResult VKAPI_CALL vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pVersion);
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(VkInstance instance, const char* pName);
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetPhysicalDeviceProcAddr(VkInstance instance, const char* pName);
#if defined(VK_USE_PLATFORM_WIN32_KHR)
VKAPI_ATTR VkResult VKAPI_CALL vk_icdEnumerateAdapterPhysicalDevices(VkInstance instance, LUID adapterLUID,
uint32_t* pPhysicalDeviceCount, VkPhysicalDevice* pPhysicalDevices);
#endif
#ifdef __cplusplus
}
#endif
#endif
/*
* The ICD must reserve space for a pointer for the loader's dispatch
* table, at the start of <each object>.
* The ICD must initialize this variable using the SET_LOADER_MAGIC_VALUE macro.
*/
#define ICD_LOADER_MAGIC 0x01CDC0DE
typedef union {
uintptr_t loaderMagic;
void *loaderData;
} VK_LOADER_DATA;
static inline void set_loader_magic_value(void *pNewObject) {
VK_LOADER_DATA *loader_info = (VK_LOADER_DATA *)pNewObject;
loader_info->loaderMagic = ICD_LOADER_MAGIC;
}
static inline bool valid_loader_magic_value(void *pNewObject) {
const VK_LOADER_DATA *loader_info = (VK_LOADER_DATA *)pNewObject;
return (loader_info->loaderMagic & 0xffffffff) == ICD_LOADER_MAGIC;
}
/*
* Windows and Linux ICDs will treat VkSurfaceKHR as a pointer to a struct that
* contains the platform-specific connection and surface information.
*/
typedef enum {
VK_ICD_WSI_PLATFORM_MIR,
VK_ICD_WSI_PLATFORM_WAYLAND,
VK_ICD_WSI_PLATFORM_WIN32,
VK_ICD_WSI_PLATFORM_XCB,
VK_ICD_WSI_PLATFORM_XLIB,
VK_ICD_WSI_PLATFORM_ANDROID,
VK_ICD_WSI_PLATFORM_MACOS,
VK_ICD_WSI_PLATFORM_IOS,
VK_ICD_WSI_PLATFORM_DISPLAY,
VK_ICD_WSI_PLATFORM_HEADLESS,
VK_ICD_WSI_PLATFORM_METAL,
VK_ICD_WSI_PLATFORM_DIRECTFB,
VK_ICD_WSI_PLATFORM_VI,
VK_ICD_WSI_PLATFORM_GGP,
VK_ICD_WSI_PLATFORM_SCREEN,
VK_ICD_WSI_PLATFORM_FUCHSIA,
} VkIcdWsiPlatform;
typedef struct {
VkIcdWsiPlatform platform;
} VkIcdSurfaceBase;
#ifdef VK_USE_PLATFORM_MIR_KHR
typedef struct {
VkIcdSurfaceBase base;
MirConnection *connection;
MirSurface *mirSurface;
} VkIcdSurfaceMir;
#endif // VK_USE_PLATFORM_MIR_KHR
#ifdef VK_USE_PLATFORM_WAYLAND_KHR
typedef struct {
VkIcdSurfaceBase base;
struct wl_display *display;
struct wl_surface *surface;
} VkIcdSurfaceWayland;
#endif // VK_USE_PLATFORM_WAYLAND_KHR
#ifdef VK_USE_PLATFORM_WIN32_KHR
typedef struct {
VkIcdSurfaceBase base;
HINSTANCE hinstance;
HWND hwnd;
} VkIcdSurfaceWin32;
#endif // VK_USE_PLATFORM_WIN32_KHR
#ifdef VK_USE_PLATFORM_XCB_KHR
typedef struct {
VkIcdSurfaceBase base;
xcb_connection_t *connection;
xcb_window_t window;
} VkIcdSurfaceXcb;
#endif // VK_USE_PLATFORM_XCB_KHR
#ifdef VK_USE_PLATFORM_XLIB_KHR
typedef struct {
VkIcdSurfaceBase base;
Display *dpy;
Window window;
} VkIcdSurfaceXlib;
#endif // VK_USE_PLATFORM_XLIB_KHR
#ifdef VK_USE_PLATFORM_DIRECTFB_EXT
typedef struct {
VkIcdSurfaceBase base;
IDirectFB *dfb;
IDirectFBSurface *surface;
} VkIcdSurfaceDirectFB;
#endif // VK_USE_PLATFORM_DIRECTFB_EXT
#ifdef VK_USE_PLATFORM_ANDROID_KHR
typedef struct {
VkIcdSurfaceBase base;
struct ANativeWindow *window;
} VkIcdSurfaceAndroid;
#endif // VK_USE_PLATFORM_ANDROID_KHR
#ifdef VK_USE_PLATFORM_MACOS_MVK
typedef struct {
VkIcdSurfaceBase base;
const void *pView;
} VkIcdSurfaceMacOS;
#endif // VK_USE_PLATFORM_MACOS_MVK
#ifdef VK_USE_PLATFORM_IOS_MVK
typedef struct {
VkIcdSurfaceBase base;
const void *pView;
} VkIcdSurfaceIOS;
#endif // VK_USE_PLATFORM_IOS_MVK
#ifdef VK_USE_PLATFORM_GGP
typedef struct {
VkIcdSurfaceBase base;
GgpStreamDescriptor streamDescriptor;
} VkIcdSurfaceGgp;
#endif // VK_USE_PLATFORM_GGP
typedef struct {
VkIcdSurfaceBase base;
VkDisplayModeKHR displayMode;
uint32_t planeIndex;
uint32_t planeStackIndex;
VkSurfaceTransformFlagBitsKHR transform;
float globalAlpha;
VkDisplayPlaneAlphaFlagBitsKHR alphaMode;
VkExtent2D imageExtent;
} VkIcdSurfaceDisplay;
typedef struct {
VkIcdSurfaceBase base;
} VkIcdSurfaceHeadless;
#ifdef VK_USE_PLATFORM_METAL_EXT
typedef struct {
VkIcdSurfaceBase base;
const CAMetalLayer *pLayer;
} VkIcdSurfaceMetal;
#endif // VK_USE_PLATFORM_METAL_EXT
#ifdef VK_USE_PLATFORM_VI_NN
typedef struct {
VkIcdSurfaceBase base;
void *window;
} VkIcdSurfaceVi;
#endif // VK_USE_PLATFORM_VI_NN
#ifdef VK_USE_PLATFORM_SCREEN_QNX
typedef struct {
VkIcdSurfaceBase base;
struct _screen_context *context;
struct _screen_window *window;
} VkIcdSurfaceScreen;
#endif // VK_USE_PLATFORM_SCREEN_QNX
#ifdef VK_USE_PLATFORM_FUCHSIA
typedef struct {
VkIcdSurfaceBase base;
} VkIcdSurfaceImagePipe;
#endif // VK_USE_PLATFORM_FUCHSIA

203
include/vulkan/vulkan/vk_layer.h Normal file
View file

@ -0,0 +1,203 @@
//
// File: vk_layer.h
//
/*
* Copyright (c) 2015-2023 LunarG, Inc.
* Copyright (c) 2015-2023 The Khronos Group Inc.
* Copyright (c) 2015-2023 Valve Corporation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*/
#pragma once
/* Need to define dispatch table
* Core struct can then have ptr to dispatch table at the top
* Along with object ptrs for current and next OBJ
*/
#include "vulkan_core.h"
#define MAX_NUM_UNKNOWN_EXTS 250
// Loader-Layer version negotiation API. Versions add the following features:
// Versions 0/1 - Initial. Doesn't support vk_layerGetPhysicalDeviceProcAddr
// or vk_icdNegotiateLoaderLayerInterfaceVersion.
// Version 2 - Add support for vk_layerGetPhysicalDeviceProcAddr and
// vk_icdNegotiateLoaderLayerInterfaceVersion.
#define CURRENT_LOADER_LAYER_INTERFACE_VERSION 2
#define MIN_SUPPORTED_LOADER_LAYER_INTERFACE_VERSION 1
#define VK_CURRENT_CHAIN_VERSION 1
// Typedef for use in the interfaces below
typedef PFN_vkVoidFunction (VKAPI_PTR *PFN_GetPhysicalDeviceProcAddr)(VkInstance instance, const char* pName);
// Version negotiation values
typedef enum VkNegotiateLayerStructType {
LAYER_NEGOTIATE_UNINTIALIZED = 0,
LAYER_NEGOTIATE_INTERFACE_STRUCT = 1,
} VkNegotiateLayerStructType;
// Version negotiation structures
typedef struct VkNegotiateLayerInterface {
VkNegotiateLayerStructType sType;
void *pNext;
uint32_t loaderLayerInterfaceVersion;
PFN_vkGetInstanceProcAddr pfnGetInstanceProcAddr;
PFN_vkGetDeviceProcAddr pfnGetDeviceProcAddr;
PFN_GetPhysicalDeviceProcAddr pfnGetPhysicalDeviceProcAddr;
} VkNegotiateLayerInterface;
// Version negotiation functions
typedef VkResult (VKAPI_PTR *PFN_vkNegotiateLoaderLayerInterfaceVersion)(VkNegotiateLayerInterface *pVersionStruct);
// Function prototype for unknown physical device extension command
typedef VkResult(VKAPI_PTR *PFN_PhysDevExt)(VkPhysicalDevice phys_device);
// ------------------------------------------------------------------------------------------------
// CreateInstance and CreateDevice support structures
/* Sub type of structure for instance and device loader ext of CreateInfo.
* When sType == VK_STRUCTURE_TYPE_LOADER_INSTANCE_CREATE_INFO
* or sType == VK_STRUCTURE_TYPE_LOADER_DEVICE_CREATE_INFO
* then VkLayerFunction indicates struct type pointed to by pNext
*/
typedef enum VkLayerFunction_ {
VK_LAYER_LINK_INFO = 0,
VK_LOADER_DATA_CALLBACK = 1,
VK_LOADER_LAYER_CREATE_DEVICE_CALLBACK = 2,
VK_LOADER_FEATURES = 3,
} VkLayerFunction;
typedef struct VkLayerInstanceLink_ {
struct VkLayerInstanceLink_ *pNext;
PFN_vkGetInstanceProcAddr pfnNextGetInstanceProcAddr;
PFN_GetPhysicalDeviceProcAddr pfnNextGetPhysicalDeviceProcAddr;
} VkLayerInstanceLink;
/*
* When creating the device chain the loader needs to pass
* down information about it's device structure needed at
* the end of the chain. Passing the data via the
* VkLayerDeviceInfo avoids issues with finding the
* exact instance being used.
*/
typedef struct VkLayerDeviceInfo_ {
void *device_info;
PFN_vkGetInstanceProcAddr pfnNextGetInstanceProcAddr;
} VkLayerDeviceInfo;
typedef VkResult (VKAPI_PTR *PFN_vkSetInstanceLoaderData)(VkInstance instance,
void *object);
typedef VkResult (VKAPI_PTR *PFN_vkSetDeviceLoaderData)(VkDevice device,
void *object);
typedef VkResult (VKAPI_PTR *PFN_vkLayerCreateDevice)(VkInstance instance, VkPhysicalDevice physicalDevice, const VkDeviceCreateInfo *pCreateInfo,
const VkAllocationCallbacks *pAllocator, VkDevice *pDevice, PFN_vkGetInstanceProcAddr layerGIPA, PFN_vkGetDeviceProcAddr *nextGDPA);
typedef void (VKAPI_PTR *PFN_vkLayerDestroyDevice)(VkDevice physicalDevice, const VkAllocationCallbacks *pAllocator, PFN_vkDestroyDevice destroyFunction);
typedef enum VkLoaderFeastureFlagBits {
VK_LOADER_FEATURE_PHYSICAL_DEVICE_SORTING = 0x00000001,
} VkLoaderFlagBits;
typedef VkFlags VkLoaderFeatureFlags;
typedef struct {
VkStructureType sType; // VK_STRUCTURE_TYPE_LOADER_INSTANCE_CREATE_INFO
const void *pNext;
VkLayerFunction function;
union {
VkLayerInstanceLink *pLayerInfo;
PFN_vkSetInstanceLoaderData pfnSetInstanceLoaderData;
struct {
PFN_vkLayerCreateDevice pfnLayerCreateDevice;
PFN_vkLayerDestroyDevice pfnLayerDestroyDevice;
} layerDevice;
VkLoaderFeatureFlags loaderFeatures;
} u;
} VkLayerInstanceCreateInfo;
typedef struct VkLayerDeviceLink_ {
struct VkLayerDeviceLink_ *pNext;
PFN_vkGetInstanceProcAddr pfnNextGetInstanceProcAddr;
PFN_vkGetDeviceProcAddr pfnNextGetDeviceProcAddr;
} VkLayerDeviceLink;
typedef struct {
VkStructureType sType; // VK_STRUCTURE_TYPE_LOADER_DEVICE_CREATE_INFO
const void *pNext;
VkLayerFunction function;
union {
VkLayerDeviceLink *pLayerInfo;
PFN_vkSetDeviceLoaderData pfnSetDeviceLoaderData;
} u;
} VkLayerDeviceCreateInfo;
#ifdef __cplusplus
extern "C" {
#endif
VKAPI_ATTR VkResult VKAPI_CALL vkNegotiateLoaderLayerInterfaceVersion(VkNegotiateLayerInterface *pVersionStruct);
typedef enum VkChainType {
VK_CHAIN_TYPE_UNKNOWN = 0,
VK_CHAIN_TYPE_ENUMERATE_INSTANCE_EXTENSION_PROPERTIES = 1,
VK_CHAIN_TYPE_ENUMERATE_INSTANCE_LAYER_PROPERTIES = 2,
VK_CHAIN_TYPE_ENUMERATE_INSTANCE_VERSION = 3,
} VkChainType;
typedef struct VkChainHeader {
VkChainType type;
uint32_t version;
uint32_t size;
} VkChainHeader;
typedef struct VkEnumerateInstanceExtensionPropertiesChain {
VkChainHeader header;
VkResult(VKAPI_PTR *pfnNextLayer)(const struct VkEnumerateInstanceExtensionPropertiesChain *, const char *, uint32_t *,
VkExtensionProperties *);
const struct VkEnumerateInstanceExtensionPropertiesChain *pNextLink;
#if defined(__cplusplus)
inline VkResult CallDown(const char *pLayerName, uint32_t *pPropertyCount, VkExtensionProperties *pProperties) const {
return pfnNextLayer(pNextLink, pLayerName, pPropertyCount, pProperties);
}
#endif
} VkEnumerateInstanceExtensionPropertiesChain;
typedef struct VkEnumerateInstanceLayerPropertiesChain {
VkChainHeader header;
VkResult(VKAPI_PTR *pfnNextLayer)(const struct VkEnumerateInstanceLayerPropertiesChain *, uint32_t *, VkLayerProperties *);
const struct VkEnumerateInstanceLayerPropertiesChain *pNextLink;
#if defined(__cplusplus)
inline VkResult CallDown(uint32_t *pPropertyCount, VkLayerProperties *pProperties) const {
return pfnNextLayer(pNextLink, pPropertyCount, pProperties);
}
#endif
} VkEnumerateInstanceLayerPropertiesChain;
typedef struct VkEnumerateInstanceVersionChain {
VkChainHeader header;
VkResult(VKAPI_PTR *pfnNextLayer)(const struct VkEnumerateInstanceVersionChain *, uint32_t *);
const struct VkEnumerateInstanceVersionChain *pNextLink;
#if defined(__cplusplus)
inline VkResult CallDown(uint32_t *pApiVersion) const {
return pfnNextLayer(pNextLink, pApiVersion);
}
#endif
} VkEnumerateInstanceVersionChain;
#ifdef __cplusplus
}
#endif

84
include/vulkan/vulkan/vk_platform.h Normal file
View file

@ -0,0 +1,84 @@
//
// File: vk_platform.h
//
/*
** Copyright 2014-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
#ifndef VK_PLATFORM_H_
#define VK_PLATFORM_H_
#ifdef __cplusplus
extern "C"
{
#endif // __cplusplus
/*
***************************************************************************************************
* Platform-specific directives and type declarations
***************************************************************************************************
*/
/* Platform-specific calling convention macros.
*
* Platforms should define these so that Vulkan clients call Vulkan commands
* with the same calling conventions that the Vulkan implementation expects.
*
* VKAPI_ATTR - Placed before the return type in function declarations.
* Useful for C++11 and GCC/Clang-style function attribute syntax.
* VKAPI_CALL - Placed after the return type in function declarations.
* Useful for MSVC-style calling convention syntax.
* VKAPI_PTR - Placed between the '(' and '*' in function pointer types.
*
* Function declaration: VKAPI_ATTR void VKAPI_CALL vkCommand(void);
* Function pointer type: typedef void (VKAPI_PTR *PFN_vkCommand)(void);
*/
#if defined(_WIN32)
// On Windows, Vulkan commands use the stdcall convention
#define VKAPI_ATTR
#define VKAPI_CALL __stdcall
#define VKAPI_PTR VKAPI_CALL
#elif defined(__ANDROID__) && defined(__ARM_ARCH) && __ARM_ARCH < 7
#error "Vulkan is not supported for the 'armeabi' NDK ABI"
#elif defined(__ANDROID__) && defined(__ARM_ARCH) && __ARM_ARCH >= 7 && defined(__ARM_32BIT_STATE)
// On Android 32-bit ARM targets, Vulkan functions use the "hardfloat"
// calling convention, i.e. float parameters are passed in registers. This
// is true even if the rest of the application passes floats on the stack,
// as it does by default when compiling for the armeabi-v7a NDK ABI.
#define VKAPI_ATTR __attribute__((pcs("aapcs-vfp")))
#define VKAPI_CALL
#define VKAPI_PTR VKAPI_ATTR
#else
// On other platforms, use the default calling convention
#define VKAPI_ATTR
#define VKAPI_CALL
#define VKAPI_PTR
#endif
#if !defined(VK_NO_STDDEF_H)
#include <stddef.h>
#endif // !defined(VK_NO_STDDEF_H)
#if !defined(VK_NO_STDINT_H)
#if defined(_MSC_VER) && (_MSC_VER < 1600)
typedef signed __int8 int8_t;
typedef unsigned __int8 uint8_t;
typedef signed __int16 int16_t;
typedef unsigned __int16 uint16_t;
typedef signed __int32 int32_t;
typedef unsigned __int32 uint32_t;
typedef signed __int64 int64_t;
typedef unsigned __int64 uint64_t;
#else
#include <stdint.h>
#endif
#endif // !defined(VK_NO_STDINT_H)
#ifdef __cplusplus
} // extern "C"
#endif // __cplusplus
#endif

99
include/vulkan/vulkan/vulkan.h Normal file
View file

@ -0,0 +1,99 @@
#ifndef VULKAN_H_
#define VULKAN_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
#include "vk_platform.h"
#include "vulkan_core.h"
#ifdef VK_USE_PLATFORM_ANDROID_KHR
#include "vulkan_android.h"
#endif
#ifdef VK_USE_PLATFORM_FUCHSIA
#include <zircon/types.h>
#include "vulkan_fuchsia.h"
#endif
#ifdef VK_USE_PLATFORM_IOS_MVK
#include "vulkan_ios.h"
#endif
#ifdef VK_USE_PLATFORM_MACOS_MVK
#include "vulkan_macos.h"
#endif
#ifdef VK_USE_PLATFORM_METAL_EXT
#include "vulkan_metal.h"
#endif
#ifdef VK_USE_PLATFORM_VI_NN
#include "vulkan_vi.h"
#endif
#ifdef VK_USE_PLATFORM_WAYLAND_KHR
#include "vulkan_wayland.h"
#endif
#ifdef VK_USE_PLATFORM_WIN32_KHR
#include <windows.h>
#include "vulkan_win32.h"
#endif
#ifdef VK_USE_PLATFORM_XCB_KHR
#include <xcb/xcb.h>
#include "vulkan_xcb.h"
#endif
#ifdef VK_USE_PLATFORM_XLIB_KHR
#include <X11/Xlib.h>
#include "vulkan_xlib.h"
#endif
#ifdef VK_USE_PLATFORM_DIRECTFB_EXT
#include <directfb.h>
#include "vulkan_directfb.h"
#endif
#ifdef VK_USE_PLATFORM_XLIB_XRANDR_EXT
#include <X11/Xlib.h>
#include <X11/extensions/Xrandr.h>
#include "vulkan_xlib_xrandr.h"
#endif
#ifdef VK_USE_PLATFORM_GGP
#include <ggp_c/vulkan_types.h>
#include "vulkan_ggp.h"
#endif
#ifdef VK_USE_PLATFORM_SCREEN_QNX
#include <screen/screen.h>
#include "vulkan_screen.h"
#endif
#ifdef VK_USE_PLATFORM_SCI
#include <nvscisync.h>
#include <nvscibuf.h>
#include "vulkan_sci.h"
#endif
#ifdef VK_ENABLE_BETA_EXTENSIONS
#include "vulkan_beta.h"
#endif
#endif // VULKAN_H_

16171
include/vulkan/vulkan/vulkan.hpp Normal file
View file

File diff suppressed because it is too large Load diff

125
include/vulkan/vulkan/vulkan_android.h Normal file
View file

@ -0,0 +1,125 @@
#ifndef VULKAN_ANDROID_H_
#define VULKAN_ANDROID_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_KHR_android_surface 1
struct ANativeWindow;
#define VK_KHR_ANDROID_SURFACE_SPEC_VERSION 6
#define VK_KHR_ANDROID_SURFACE_EXTENSION_NAME "VK_KHR_android_surface"
typedef VkFlags VkAndroidSurfaceCreateFlagsKHR;
typedef struct VkAndroidSurfaceCreateInfoKHR {
VkStructureType sType;
const void* pNext;
VkAndroidSurfaceCreateFlagsKHR flags;
struct ANativeWindow* window;
} VkAndroidSurfaceCreateInfoKHR;
typedef VkResult (VKAPI_PTR *PFN_vkCreateAndroidSurfaceKHR)(VkInstance instance, const VkAndroidSurfaceCreateInfoKHR* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateAndroidSurfaceKHR(
VkInstance instance,
const VkAndroidSurfaceCreateInfoKHR* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#define VK_ANDROID_external_memory_android_hardware_buffer 1
struct AHardwareBuffer;
#define VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_SPEC_VERSION 5
#define VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_EXTENSION_NAME "VK_ANDROID_external_memory_android_hardware_buffer"
typedef struct VkAndroidHardwareBufferUsageANDROID {
VkStructureType sType;
void* pNext;
uint64_t androidHardwareBufferUsage;
} VkAndroidHardwareBufferUsageANDROID;
typedef struct VkAndroidHardwareBufferPropertiesANDROID {
VkStructureType sType;
void* pNext;
VkDeviceSize allocationSize;
uint32_t memoryTypeBits;
} VkAndroidHardwareBufferPropertiesANDROID;
typedef struct VkAndroidHardwareBufferFormatPropertiesANDROID {
VkStructureType sType;
void* pNext;
VkFormat format;
uint64_t externalFormat;
VkFormatFeatureFlags formatFeatures;
VkComponentMapping samplerYcbcrConversionComponents;
VkSamplerYcbcrModelConversion suggestedYcbcrModel;
VkSamplerYcbcrRange suggestedYcbcrRange;
VkChromaLocation suggestedXChromaOffset;
VkChromaLocation suggestedYChromaOffset;
} VkAndroidHardwareBufferFormatPropertiesANDROID;
typedef struct VkImportAndroidHardwareBufferInfoANDROID {
VkStructureType sType;
const void* pNext;
struct AHardwareBuffer* buffer;
} VkImportAndroidHardwareBufferInfoANDROID;
typedef struct VkMemoryGetAndroidHardwareBufferInfoANDROID {
VkStructureType sType;
const void* pNext;
VkDeviceMemory memory;
} VkMemoryGetAndroidHardwareBufferInfoANDROID;
typedef struct VkExternalFormatANDROID {
VkStructureType sType;
void* pNext;
uint64_t externalFormat;
} VkExternalFormatANDROID;
typedef struct VkAndroidHardwareBufferFormatProperties2ANDROID {
VkStructureType sType;
void* pNext;
VkFormat format;
uint64_t externalFormat;
VkFormatFeatureFlags2 formatFeatures;
VkComponentMapping samplerYcbcrConversionComponents;
VkSamplerYcbcrModelConversion suggestedYcbcrModel;
VkSamplerYcbcrRange suggestedYcbcrRange;
VkChromaLocation suggestedXChromaOffset;
VkChromaLocation suggestedYChromaOffset;
} VkAndroidHardwareBufferFormatProperties2ANDROID;
typedef VkResult (VKAPI_PTR *PFN_vkGetAndroidHardwareBufferPropertiesANDROID)(VkDevice device, const struct AHardwareBuffer* buffer, VkAndroidHardwareBufferPropertiesANDROID* pProperties);
typedef VkResult (VKAPI_PTR *PFN_vkGetMemoryAndroidHardwareBufferANDROID)(VkDevice device, const VkMemoryGetAndroidHardwareBufferInfoANDROID* pInfo, struct AHardwareBuffer** pBuffer);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkGetAndroidHardwareBufferPropertiesANDROID(
VkDevice device,
const struct AHardwareBuffer* buffer,
VkAndroidHardwareBufferPropertiesANDROID* pProperties);
VKAPI_ATTR VkResult VKAPI_CALL vkGetMemoryAndroidHardwareBufferANDROID(
VkDevice device,
const VkMemoryGetAndroidHardwareBufferInfoANDROID* pInfo,
struct AHardwareBuffer** pBuffer);
#endif
#ifdef __cplusplus
}
#endif
#endif

537
include/vulkan/vulkan/vulkan_beta.h Normal file
View file

@ -0,0 +1,537 @@
#ifndef VULKAN_BETA_H_
#define VULKAN_BETA_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_KHR_portability_subset 1
#define VK_KHR_PORTABILITY_SUBSET_SPEC_VERSION 1
#define VK_KHR_PORTABILITY_SUBSET_EXTENSION_NAME "VK_KHR_portability_subset"
typedef struct VkPhysicalDevicePortabilitySubsetFeaturesKHR {
VkStructureType sType;
void* pNext;
VkBool32 constantAlphaColorBlendFactors;
VkBool32 events;
VkBool32 imageViewFormatReinterpretation;
VkBool32 imageViewFormatSwizzle;
VkBool32 imageView2DOn3DImage;
VkBool32 multisampleArrayImage;
VkBool32 mutableComparisonSamplers;
VkBool32 pointPolygons;
VkBool32 samplerMipLodBias;
VkBool32 separateStencilMaskRef;
VkBool32 shaderSampleRateInterpolationFunctions;
VkBool32 tessellationIsolines;
VkBool32 tessellationPointMode;
VkBool32 triangleFans;
VkBool32 vertexAttributeAccessBeyondStride;
} VkPhysicalDevicePortabilitySubsetFeaturesKHR;
typedef struct VkPhysicalDevicePortabilitySubsetPropertiesKHR {
VkStructureType sType;
void* pNext;
uint32_t minVertexInputBindingStrideAlignment;
} VkPhysicalDevicePortabilitySubsetPropertiesKHR;
#define VK_KHR_video_encode_queue 1
#define VK_KHR_VIDEO_ENCODE_QUEUE_SPEC_VERSION 8
#define VK_KHR_VIDEO_ENCODE_QUEUE_EXTENSION_NAME "VK_KHR_video_encode_queue"
typedef enum VkVideoEncodeTuningModeKHR {
VK_VIDEO_ENCODE_TUNING_MODE_DEFAULT_KHR = 0,
VK_VIDEO_ENCODE_TUNING_MODE_HIGH_QUALITY_KHR = 1,
VK_VIDEO_ENCODE_TUNING_MODE_LOW_LATENCY_KHR = 2,
VK_VIDEO_ENCODE_TUNING_MODE_ULTRA_LOW_LATENCY_KHR = 3,
VK_VIDEO_ENCODE_TUNING_MODE_LOSSLESS_KHR = 4,
VK_VIDEO_ENCODE_TUNING_MODE_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeTuningModeKHR;
typedef VkFlags VkVideoEncodeFlagsKHR;
typedef enum VkVideoEncodeCapabilityFlagBitsKHR {
VK_VIDEO_ENCODE_CAPABILITY_PRECEDING_EXTERNALLY_ENCODED_BYTES_BIT_KHR = 0x00000001,
VK_VIDEO_ENCODE_CAPABILITY_FLAG_BITS_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeCapabilityFlagBitsKHR;
typedef VkFlags VkVideoEncodeCapabilityFlagsKHR;
typedef enum VkVideoEncodeRateControlModeFlagBitsKHR {
VK_VIDEO_ENCODE_RATE_CONTROL_MODE_DEFAULT_KHR = 0,
VK_VIDEO_ENCODE_RATE_CONTROL_MODE_DISABLED_BIT_KHR = 0x00000001,
VK_VIDEO_ENCODE_RATE_CONTROL_MODE_CBR_BIT_KHR = 0x00000002,
VK_VIDEO_ENCODE_RATE_CONTROL_MODE_VBR_BIT_KHR = 0x00000004,
VK_VIDEO_ENCODE_RATE_CONTROL_MODE_FLAG_BITS_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeRateControlModeFlagBitsKHR;
typedef VkFlags VkVideoEncodeRateControlModeFlagsKHR;
typedef enum VkVideoEncodeFeedbackFlagBitsKHR {
VK_VIDEO_ENCODE_FEEDBACK_BITSTREAM_BUFFER_OFFSET_BIT_KHR = 0x00000001,
VK_VIDEO_ENCODE_FEEDBACK_BITSTREAM_BYTES_WRITTEN_BIT_KHR = 0x00000002,
VK_VIDEO_ENCODE_FEEDBACK_FLAG_BITS_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeFeedbackFlagBitsKHR;
typedef VkFlags VkVideoEncodeFeedbackFlagsKHR;
typedef enum VkVideoEncodeUsageFlagBitsKHR {
VK_VIDEO_ENCODE_USAGE_DEFAULT_KHR = 0,
VK_VIDEO_ENCODE_USAGE_TRANSCODING_BIT_KHR = 0x00000001,
VK_VIDEO_ENCODE_USAGE_STREAMING_BIT_KHR = 0x00000002,
VK_VIDEO_ENCODE_USAGE_RECORDING_BIT_KHR = 0x00000004,
VK_VIDEO_ENCODE_USAGE_CONFERENCING_BIT_KHR = 0x00000008,
VK_VIDEO_ENCODE_USAGE_FLAG_BITS_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeUsageFlagBitsKHR;
typedef VkFlags VkVideoEncodeUsageFlagsKHR;
typedef enum VkVideoEncodeContentFlagBitsKHR {
VK_VIDEO_ENCODE_CONTENT_DEFAULT_KHR = 0,
VK_VIDEO_ENCODE_CONTENT_CAMERA_BIT_KHR = 0x00000001,
VK_VIDEO_ENCODE_CONTENT_DESKTOP_BIT_KHR = 0x00000002,
VK_VIDEO_ENCODE_CONTENT_RENDERED_BIT_KHR = 0x00000004,
VK_VIDEO_ENCODE_CONTENT_FLAG_BITS_MAX_ENUM_KHR = 0x7FFFFFFF
} VkVideoEncodeContentFlagBitsKHR;
typedef VkFlags VkVideoEncodeContentFlagsKHR;
typedef VkFlags VkVideoEncodeRateControlFlagsKHR;
typedef struct VkVideoEncodeInfoKHR {
VkStructureType sType;
const void* pNext;
VkVideoEncodeFlagsKHR flags;
uint32_t qualityLevel;
VkBuffer dstBuffer;
VkDeviceSize dstBufferOffset;
VkDeviceSize dstBufferRange;
VkVideoPictureResourceInfoKHR srcPictureResource;
const VkVideoReferenceSlotInfoKHR* pSetupReferenceSlot;
uint32_t referenceSlotCount;
const VkVideoReferenceSlotInfoKHR* pReferenceSlots;
uint32_t precedingExternallyEncodedBytes;
} VkVideoEncodeInfoKHR;
typedef struct VkVideoEncodeCapabilitiesKHR {
VkStructureType sType;
void* pNext;
VkVideoEncodeCapabilityFlagsKHR flags;
VkVideoEncodeRateControlModeFlagsKHR rateControlModes;
uint32_t maxRateControlLayers;
uint32_t maxQualityLevels;
VkExtent2D inputImageDataFillAlignment;
VkVideoEncodeFeedbackFlagsKHR supportedEncodeFeedbackFlags;
} VkVideoEncodeCapabilitiesKHR;
typedef struct VkQueryPoolVideoEncodeFeedbackCreateInfoKHR {
VkStructureType sType;
const void* pNext;
VkVideoEncodeFeedbackFlagsKHR encodeFeedbackFlags;
} VkQueryPoolVideoEncodeFeedbackCreateInfoKHR;
typedef struct VkVideoEncodeUsageInfoKHR {
VkStructureType sType;
const void* pNext;
VkVideoEncodeUsageFlagsKHR videoUsageHints;
VkVideoEncodeContentFlagsKHR videoContentHints;
VkVideoEncodeTuningModeKHR tuningMode;
} VkVideoEncodeUsageInfoKHR;
typedef struct VkVideoEncodeRateControlLayerInfoKHR {
VkStructureType sType;
const void* pNext;
uint64_t averageBitrate;
uint64_t maxBitrate;
uint32_t frameRateNumerator;
uint32_t frameRateDenominator;
uint32_t virtualBufferSizeInMs;
uint32_t initialVirtualBufferSizeInMs;
} VkVideoEncodeRateControlLayerInfoKHR;
typedef struct VkVideoEncodeRateControlInfoKHR {
VkStructureType sType;
const void* pNext;
VkVideoEncodeRateControlFlagsKHR flags;
VkVideoEncodeRateControlModeFlagBitsKHR rateControlMode;
uint32_t layerCount;
const VkVideoEncodeRateControlLayerInfoKHR* pLayers;
} VkVideoEncodeRateControlInfoKHR;
typedef void (VKAPI_PTR *PFN_vkCmdEncodeVideoKHR)(VkCommandBuffer commandBuffer, const VkVideoEncodeInfoKHR* pEncodeInfo);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR void VKAPI_CALL vkCmdEncodeVideoKHR(
VkCommandBuffer commandBuffer,
const VkVideoEncodeInfoKHR* pEncodeInfo);
#endif
#define VK_EXT_video_encode_h264 1
#include "vk_video/vulkan_video_codec_h264std.h"
#include "vk_video/vulkan_video_codec_h264std_encode.h"
#define VK_EXT_VIDEO_ENCODE_H264_SPEC_VERSION 10
#define VK_EXT_VIDEO_ENCODE_H264_EXTENSION_NAME "VK_EXT_video_encode_h264"
typedef enum VkVideoEncodeH264RateControlStructureEXT {
VK_VIDEO_ENCODE_H264_RATE_CONTROL_STRUCTURE_UNKNOWN_EXT = 0,
VK_VIDEO_ENCODE_H264_RATE_CONTROL_STRUCTURE_FLAT_EXT = 1,
VK_VIDEO_ENCODE_H264_RATE_CONTROL_STRUCTURE_DYADIC_EXT = 2,
VK_VIDEO_ENCODE_H264_RATE_CONTROL_STRUCTURE_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH264RateControlStructureEXT;
typedef enum VkVideoEncodeH264CapabilityFlagBitsEXT {
VK_VIDEO_ENCODE_H264_CAPABILITY_DIRECT_8X8_INFERENCE_ENABLED_BIT_EXT = 0x00000001,
VK_VIDEO_ENCODE_H264_CAPABILITY_DIRECT_8X8_INFERENCE_DISABLED_BIT_EXT = 0x00000002,
VK_VIDEO_ENCODE_H264_CAPABILITY_SEPARATE_COLOUR_PLANE_BIT_EXT = 0x00000004,
VK_VIDEO_ENCODE_H264_CAPABILITY_QPPRIME_Y_ZERO_TRANSFORM_BYPASS_BIT_EXT = 0x00000008,
VK_VIDEO_ENCODE_H264_CAPABILITY_SCALING_LISTS_BIT_EXT = 0x00000010,
VK_VIDEO_ENCODE_H264_CAPABILITY_HRD_COMPLIANCE_BIT_EXT = 0x00000020,
VK_VIDEO_ENCODE_H264_CAPABILITY_CHROMA_QP_OFFSET_BIT_EXT = 0x00000040,
VK_VIDEO_ENCODE_H264_CAPABILITY_SECOND_CHROMA_QP_OFFSET_BIT_EXT = 0x00000080,
VK_VIDEO_ENCODE_H264_CAPABILITY_PIC_INIT_QP_MINUS26_BIT_EXT = 0x00000100,
VK_VIDEO_ENCODE_H264_CAPABILITY_WEIGHTED_PRED_BIT_EXT = 0x00000200,
VK_VIDEO_ENCODE_H264_CAPABILITY_WEIGHTED_BIPRED_EXPLICIT_BIT_EXT = 0x00000400,
VK_VIDEO_ENCODE_H264_CAPABILITY_WEIGHTED_BIPRED_IMPLICIT_BIT_EXT = 0x00000800,
VK_VIDEO_ENCODE_H264_CAPABILITY_WEIGHTED_PRED_NO_TABLE_BIT_EXT = 0x00001000,
VK_VIDEO_ENCODE_H264_CAPABILITY_TRANSFORM_8X8_BIT_EXT = 0x00002000,
VK_VIDEO_ENCODE_H264_CAPABILITY_CABAC_BIT_EXT = 0x00004000,
VK_VIDEO_ENCODE_H264_CAPABILITY_CAVLC_BIT_EXT = 0x00008000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DEBLOCKING_FILTER_DISABLED_BIT_EXT = 0x00010000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DEBLOCKING_FILTER_ENABLED_BIT_EXT = 0x00020000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DEBLOCKING_FILTER_PARTIAL_BIT_EXT = 0x00040000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DISABLE_DIRECT_SPATIAL_MV_PRED_BIT_EXT = 0x00080000,
VK_VIDEO_ENCODE_H264_CAPABILITY_MULTIPLE_SLICE_PER_FRAME_BIT_EXT = 0x00100000,
VK_VIDEO_ENCODE_H264_CAPABILITY_SLICE_MB_COUNT_BIT_EXT = 0x00200000,
VK_VIDEO_ENCODE_H264_CAPABILITY_ROW_UNALIGNED_SLICE_BIT_EXT = 0x00400000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DIFFERENT_SLICE_TYPE_BIT_EXT = 0x00800000,
VK_VIDEO_ENCODE_H264_CAPABILITY_B_FRAME_IN_L1_LIST_BIT_EXT = 0x01000000,
VK_VIDEO_ENCODE_H264_CAPABILITY_DIFFERENT_REFERENCE_FINAL_LISTS_BIT_EXT = 0x02000000,
VK_VIDEO_ENCODE_H264_CAPABILITY_FLAG_BITS_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH264CapabilityFlagBitsEXT;
typedef VkFlags VkVideoEncodeH264CapabilityFlagsEXT;
typedef struct VkVideoEncodeH264CapabilitiesEXT {
VkStructureType sType;
void* pNext;
VkVideoEncodeH264CapabilityFlagsEXT flags;
uint32_t maxPPictureL0ReferenceCount;
uint32_t maxBPictureL0ReferenceCount;
uint32_t maxL1ReferenceCount;
VkBool32 motionVectorsOverPicBoundariesFlag;
uint32_t maxBytesPerPicDenom;
uint32_t maxBitsPerMbDenom;
uint32_t log2MaxMvLengthHorizontal;
uint32_t log2MaxMvLengthVertical;
} VkVideoEncodeH264CapabilitiesEXT;
typedef struct VkVideoEncodeH264SessionParametersAddInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t stdSPSCount;
const StdVideoH264SequenceParameterSet* pStdSPSs;
uint32_t stdPPSCount;
const StdVideoH264PictureParameterSet* pStdPPSs;
} VkVideoEncodeH264SessionParametersAddInfoEXT;
typedef struct VkVideoEncodeH264SessionParametersCreateInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t maxStdSPSCount;
uint32_t maxStdPPSCount;
const VkVideoEncodeH264SessionParametersAddInfoEXT* pParametersAddInfo;
} VkVideoEncodeH264SessionParametersCreateInfoEXT;
typedef struct VkVideoEncodeH264NaluSliceInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t mbCount;
const StdVideoEncodeH264ReferenceListsInfo* pStdReferenceFinalLists;
const StdVideoEncodeH264SliceHeader* pStdSliceHeader;
} VkVideoEncodeH264NaluSliceInfoEXT;
typedef struct VkVideoEncodeH264VclFrameInfoEXT {
VkStructureType sType;
const void* pNext;
const StdVideoEncodeH264ReferenceListsInfo* pStdReferenceFinalLists;
uint32_t naluSliceEntryCount;
const VkVideoEncodeH264NaluSliceInfoEXT* pNaluSliceEntries;
const StdVideoEncodeH264PictureInfo* pStdPictureInfo;
} VkVideoEncodeH264VclFrameInfoEXT;
typedef struct VkVideoEncodeH264DpbSlotInfoEXT {
VkStructureType sType;
const void* pNext;
const StdVideoEncodeH264ReferenceInfo* pStdReferenceInfo;
} VkVideoEncodeH264DpbSlotInfoEXT;
typedef struct VkVideoEncodeH264ProfileInfoEXT {
VkStructureType sType;
const void* pNext;
StdVideoH264ProfileIdc stdProfileIdc;
} VkVideoEncodeH264ProfileInfoEXT;
typedef struct VkVideoEncodeH264RateControlInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t gopFrameCount;
uint32_t idrPeriod;
uint32_t consecutiveBFrameCount;
VkVideoEncodeH264RateControlStructureEXT rateControlStructure;
uint32_t temporalLayerCount;
} VkVideoEncodeH264RateControlInfoEXT;
typedef struct VkVideoEncodeH264QpEXT {
int32_t qpI;
int32_t qpP;
int32_t qpB;
} VkVideoEncodeH264QpEXT;
typedef struct VkVideoEncodeH264FrameSizeEXT {
uint32_t frameISize;
uint32_t framePSize;
uint32_t frameBSize;
} VkVideoEncodeH264FrameSizeEXT;
typedef struct VkVideoEncodeH264RateControlLayerInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t temporalLayerId;
VkBool32 useInitialRcQp;
VkVideoEncodeH264QpEXT initialRcQp;
VkBool32 useMinQp;
VkVideoEncodeH264QpEXT minQp;
VkBool32 useMaxQp;
VkVideoEncodeH264QpEXT maxQp;
VkBool32 useMaxFrameSize;
VkVideoEncodeH264FrameSizeEXT maxFrameSize;
} VkVideoEncodeH264RateControlLayerInfoEXT;
#define VK_EXT_video_encode_h265 1
#include "vk_video/vulkan_video_codec_h265std.h"
#include "vk_video/vulkan_video_codec_h265std_encode.h"
#define VK_EXT_VIDEO_ENCODE_H265_SPEC_VERSION 10
#define VK_EXT_VIDEO_ENCODE_H265_EXTENSION_NAME "VK_EXT_video_encode_h265"
typedef enum VkVideoEncodeH265RateControlStructureEXT {
VK_VIDEO_ENCODE_H265_RATE_CONTROL_STRUCTURE_UNKNOWN_EXT = 0,
VK_VIDEO_ENCODE_H265_RATE_CONTROL_STRUCTURE_FLAT_EXT = 1,
VK_VIDEO_ENCODE_H265_RATE_CONTROL_STRUCTURE_DYADIC_EXT = 2,
VK_VIDEO_ENCODE_H265_RATE_CONTROL_STRUCTURE_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH265RateControlStructureEXT;
typedef enum VkVideoEncodeH265CapabilityFlagBitsEXT {
VK_VIDEO_ENCODE_H265_CAPABILITY_SEPARATE_COLOUR_PLANE_BIT_EXT = 0x00000001,
VK_VIDEO_ENCODE_H265_CAPABILITY_SCALING_LISTS_BIT_EXT = 0x00000002,
VK_VIDEO_ENCODE_H265_CAPABILITY_SAMPLE_ADAPTIVE_OFFSET_ENABLED_BIT_EXT = 0x00000004,
VK_VIDEO_ENCODE_H265_CAPABILITY_PCM_ENABLE_BIT_EXT = 0x00000008,
VK_VIDEO_ENCODE_H265_CAPABILITY_SPS_TEMPORAL_MVP_ENABLED_BIT_EXT = 0x00000010,
VK_VIDEO_ENCODE_H265_CAPABILITY_HRD_COMPLIANCE_BIT_EXT = 0x00000020,
VK_VIDEO_ENCODE_H265_CAPABILITY_INIT_QP_MINUS26_BIT_EXT = 0x00000040,
VK_VIDEO_ENCODE_H265_CAPABILITY_LOG2_PARALLEL_MERGE_LEVEL_MINUS2_BIT_EXT = 0x00000080,
VK_VIDEO_ENCODE_H265_CAPABILITY_SIGN_DATA_HIDING_ENABLED_BIT_EXT = 0x00000100,
VK_VIDEO_ENCODE_H265_CAPABILITY_TRANSFORM_SKIP_ENABLED_BIT_EXT = 0x00000200,
VK_VIDEO_ENCODE_H265_CAPABILITY_TRANSFORM_SKIP_DISABLED_BIT_EXT = 0x00000400,
VK_VIDEO_ENCODE_H265_CAPABILITY_PPS_SLICE_CHROMA_QP_OFFSETS_PRESENT_BIT_EXT = 0x00000800,
VK_VIDEO_ENCODE_H265_CAPABILITY_WEIGHTED_PRED_BIT_EXT = 0x00001000,
VK_VIDEO_ENCODE_H265_CAPABILITY_WEIGHTED_BIPRED_BIT_EXT = 0x00002000,
VK_VIDEO_ENCODE_H265_CAPABILITY_WEIGHTED_PRED_NO_TABLE_BIT_EXT = 0x00004000,
VK_VIDEO_ENCODE_H265_CAPABILITY_TRANSQUANT_BYPASS_ENABLED_BIT_EXT = 0x00008000,
VK_VIDEO_ENCODE_H265_CAPABILITY_ENTROPY_CODING_SYNC_ENABLED_BIT_EXT = 0x00010000,
VK_VIDEO_ENCODE_H265_CAPABILITY_DEBLOCKING_FILTER_OVERRIDE_ENABLED_BIT_EXT = 0x00020000,
VK_VIDEO_ENCODE_H265_CAPABILITY_MULTIPLE_TILE_PER_FRAME_BIT_EXT = 0x00040000,
VK_VIDEO_ENCODE_H265_CAPABILITY_MULTIPLE_SLICE_PER_TILE_BIT_EXT = 0x00080000,
VK_VIDEO_ENCODE_H265_CAPABILITY_MULTIPLE_TILE_PER_SLICE_BIT_EXT = 0x00100000,
VK_VIDEO_ENCODE_H265_CAPABILITY_SLICE_SEGMENT_CTB_COUNT_BIT_EXT = 0x00200000,
VK_VIDEO_ENCODE_H265_CAPABILITY_ROW_UNALIGNED_SLICE_SEGMENT_BIT_EXT = 0x00400000,
VK_VIDEO_ENCODE_H265_CAPABILITY_DEPENDENT_SLICE_SEGMENT_BIT_EXT = 0x00800000,
VK_VIDEO_ENCODE_H265_CAPABILITY_DIFFERENT_SLICE_TYPE_BIT_EXT = 0x01000000,
VK_VIDEO_ENCODE_H265_CAPABILITY_B_FRAME_IN_L1_LIST_BIT_EXT = 0x02000000,
VK_VIDEO_ENCODE_H265_CAPABILITY_DIFFERENT_REFERENCE_FINAL_LISTS_BIT_EXT = 0x04000000,
VK_VIDEO_ENCODE_H265_CAPABILITY_FLAG_BITS_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH265CapabilityFlagBitsEXT;
typedef VkFlags VkVideoEncodeH265CapabilityFlagsEXT;
typedef enum VkVideoEncodeH265CtbSizeFlagBitsEXT {
VK_VIDEO_ENCODE_H265_CTB_SIZE_16_BIT_EXT = 0x00000001,
VK_VIDEO_ENCODE_H265_CTB_SIZE_32_BIT_EXT = 0x00000002,
VK_VIDEO_ENCODE_H265_CTB_SIZE_64_BIT_EXT = 0x00000004,
VK_VIDEO_ENCODE_H265_CTB_SIZE_FLAG_BITS_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH265CtbSizeFlagBitsEXT;
typedef VkFlags VkVideoEncodeH265CtbSizeFlagsEXT;
typedef enum VkVideoEncodeH265TransformBlockSizeFlagBitsEXT {
VK_VIDEO_ENCODE_H265_TRANSFORM_BLOCK_SIZE_4_BIT_EXT = 0x00000001,
VK_VIDEO_ENCODE_H265_TRANSFORM_BLOCK_SIZE_8_BIT_EXT = 0x00000002,
VK_VIDEO_ENCODE_H265_TRANSFORM_BLOCK_SIZE_16_BIT_EXT = 0x00000004,
VK_VIDEO_ENCODE_H265_TRANSFORM_BLOCK_SIZE_32_BIT_EXT = 0x00000008,
VK_VIDEO_ENCODE_H265_TRANSFORM_BLOCK_SIZE_FLAG_BITS_MAX_ENUM_EXT = 0x7FFFFFFF
} VkVideoEncodeH265TransformBlockSizeFlagBitsEXT;
typedef VkFlags VkVideoEncodeH265TransformBlockSizeFlagsEXT;
typedef struct VkVideoEncodeH265CapabilitiesEXT {
VkStructureType sType;
void* pNext;
VkVideoEncodeH265CapabilityFlagsEXT flags;
VkVideoEncodeH265CtbSizeFlagsEXT ctbSizes;
VkVideoEncodeH265TransformBlockSizeFlagsEXT transformBlockSizes;
uint32_t maxPPictureL0ReferenceCount;
uint32_t maxBPictureL0ReferenceCount;
uint32_t maxL1ReferenceCount;
uint32_t maxSubLayersCount;
uint32_t minLog2MinLumaCodingBlockSizeMinus3;
uint32_t maxLog2MinLumaCodingBlockSizeMinus3;
uint32_t minLog2MinLumaTransformBlockSizeMinus2;
uint32_t maxLog2MinLumaTransformBlockSizeMinus2;
uint32_t minMaxTransformHierarchyDepthInter;
uint32_t maxMaxTransformHierarchyDepthInter;
uint32_t minMaxTransformHierarchyDepthIntra;
uint32_t maxMaxTransformHierarchyDepthIntra;
uint32_t maxDiffCuQpDeltaDepth;
uint32_t minMaxNumMergeCand;
uint32_t maxMaxNumMergeCand;
} VkVideoEncodeH265CapabilitiesEXT;
typedef struct VkVideoEncodeH265SessionParametersAddInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t stdVPSCount;
const StdVideoH265VideoParameterSet* pStdVPSs;
uint32_t stdSPSCount;
const StdVideoH265SequenceParameterSet* pStdSPSs;
uint32_t stdPPSCount;
const StdVideoH265PictureParameterSet* pStdPPSs;
} VkVideoEncodeH265SessionParametersAddInfoEXT;
typedef struct VkVideoEncodeH265SessionParametersCreateInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t maxStdVPSCount;
uint32_t maxStdSPSCount;
uint32_t maxStdPPSCount;
const VkVideoEncodeH265SessionParametersAddInfoEXT* pParametersAddInfo;
} VkVideoEncodeH265SessionParametersCreateInfoEXT;
typedef struct VkVideoEncodeH265NaluSliceSegmentInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t ctbCount;
const StdVideoEncodeH265ReferenceListsInfo* pStdReferenceFinalLists;
const StdVideoEncodeH265SliceSegmentHeader* pStdSliceSegmentHeader;
} VkVideoEncodeH265NaluSliceSegmentInfoEXT;
typedef struct VkVideoEncodeH265VclFrameInfoEXT {
VkStructureType sType;
const void* pNext;
const StdVideoEncodeH265ReferenceListsInfo* pStdReferenceFinalLists;
uint32_t naluSliceSegmentEntryCount;
const VkVideoEncodeH265NaluSliceSegmentInfoEXT* pNaluSliceSegmentEntries;
const StdVideoEncodeH265PictureInfo* pStdPictureInfo;
} VkVideoEncodeH265VclFrameInfoEXT;
typedef struct VkVideoEncodeH265DpbSlotInfoEXT {
VkStructureType sType;
const void* pNext;
const StdVideoEncodeH265ReferenceInfo* pStdReferenceInfo;
} VkVideoEncodeH265DpbSlotInfoEXT;
typedef struct VkVideoEncodeH265ProfileInfoEXT {
VkStructureType sType;
const void* pNext;
StdVideoH265ProfileIdc stdProfileIdc;
} VkVideoEncodeH265ProfileInfoEXT;
typedef struct VkVideoEncodeH265RateControlInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t gopFrameCount;
uint32_t idrPeriod;
uint32_t consecutiveBFrameCount;
VkVideoEncodeH265RateControlStructureEXT rateControlStructure;
uint32_t subLayerCount;
} VkVideoEncodeH265RateControlInfoEXT;
typedef struct VkVideoEncodeH265QpEXT {
int32_t qpI;
int32_t qpP;
int32_t qpB;
} VkVideoEncodeH265QpEXT;
typedef struct VkVideoEncodeH265FrameSizeEXT {
uint32_t frameISize;
uint32_t framePSize;
uint32_t frameBSize;
} VkVideoEncodeH265FrameSizeEXT;
typedef struct VkVideoEncodeH265RateControlLayerInfoEXT {
VkStructureType sType;
const void* pNext;
uint32_t temporalId;
VkBool32 useInitialRcQp;
VkVideoEncodeH265QpEXT initialRcQp;
VkBool32 useMinQp;
VkVideoEncodeH265QpEXT minQp;
VkBool32 useMaxQp;
VkVideoEncodeH265QpEXT maxQp;
VkBool32 useMaxFrameSize;
VkVideoEncodeH265FrameSizeEXT maxFrameSize;
} VkVideoEncodeH265RateControlLayerInfoEXT;
#define VK_NV_displacement_micromap 1
#define VK_NV_DISPLACEMENT_MICROMAP_SPEC_VERSION 1
#define VK_NV_DISPLACEMENT_MICROMAP_EXTENSION_NAME "VK_NV_displacement_micromap"
typedef enum VkDisplacementMicromapFormatNV {
VK_DISPLACEMENT_MICROMAP_FORMAT_64_TRIANGLES_64_BYTES_NV = 1,
VK_DISPLACEMENT_MICROMAP_FORMAT_256_TRIANGLES_128_BYTES_NV = 2,
VK_DISPLACEMENT_MICROMAP_FORMAT_1024_TRIANGLES_128_BYTES_NV = 3,
VK_DISPLACEMENT_MICROMAP_FORMAT_MAX_ENUM_NV = 0x7FFFFFFF
} VkDisplacementMicromapFormatNV;
typedef struct VkPhysicalDeviceDisplacementMicromapFeaturesNV {
VkStructureType sType;
void* pNext;
VkBool32 displacementMicromap;
} VkPhysicalDeviceDisplacementMicromapFeaturesNV;
typedef struct VkPhysicalDeviceDisplacementMicromapPropertiesNV {
VkStructureType sType;
void* pNext;
uint32_t maxDisplacementMicromapSubdivisionLevel;
} VkPhysicalDeviceDisplacementMicromapPropertiesNV;
typedef struct VkAccelerationStructureTrianglesDisplacementMicromapNV {
VkStructureType sType;
void* pNext;
VkFormat displacementBiasAndScaleFormat;
VkFormat displacementVectorFormat;
VkDeviceOrHostAddressConstKHR displacementBiasAndScaleBuffer;
VkDeviceSize displacementBiasAndScaleStride;
VkDeviceOrHostAddressConstKHR displacementVectorBuffer;
VkDeviceSize displacementVectorStride;
VkDeviceOrHostAddressConstKHR displacedMicromapPrimitiveFlags;
VkDeviceSize displacedMicromapPrimitiveFlagsStride;
VkIndexType indexType;
VkDeviceOrHostAddressConstKHR indexBuffer;
VkDeviceSize indexStride;
uint32_t baseTriangle;
uint32_t usageCountsCount;
const VkMicromapUsageEXT* pUsageCounts;
const VkMicromapUsageEXT* const* ppUsageCounts;
VkMicromapEXT micromap;
} VkAccelerationStructureTrianglesDisplacementMicromapNV;
#ifdef __cplusplus
}
#endif
#endif

17211
include/vulkan/vulkan/vulkan_core.h Normal file
View file

File diff suppressed because it is too large Load diff

54
include/vulkan/vulkan/vulkan_directfb.h Normal file
View file

@ -0,0 +1,54 @@
#ifndef VULKAN_DIRECTFB_H_
#define VULKAN_DIRECTFB_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_EXT_directfb_surface 1
#define VK_EXT_DIRECTFB_SURFACE_SPEC_VERSION 1
#define VK_EXT_DIRECTFB_SURFACE_EXTENSION_NAME "VK_EXT_directfb_surface"
typedef VkFlags VkDirectFBSurfaceCreateFlagsEXT;
typedef struct VkDirectFBSurfaceCreateInfoEXT {
VkStructureType sType;
const void* pNext;
VkDirectFBSurfaceCreateFlagsEXT flags;
IDirectFB* dfb;
IDirectFBSurface* surface;
} VkDirectFBSurfaceCreateInfoEXT;
typedef VkResult (VKAPI_PTR *PFN_vkCreateDirectFBSurfaceEXT)(VkInstance instance, const VkDirectFBSurfaceCreateInfoEXT* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
typedef VkBool32 (VKAPI_PTR *PFN_vkGetPhysicalDeviceDirectFBPresentationSupportEXT)(VkPhysicalDevice physicalDevice, uint32_t queueFamilyIndex, IDirectFB* dfb);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateDirectFBSurfaceEXT(
VkInstance instance,
const VkDirectFBSurfaceCreateInfoEXT* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
VKAPI_ATTR VkBool32 VKAPI_CALL vkGetPhysicalDeviceDirectFBPresentationSupportEXT(
VkPhysicalDevice physicalDevice,
uint32_t queueFamilyIndex,
IDirectFB* dfb);
#endif
#ifdef __cplusplus
}
#endif
#endif

6571
include/vulkan/vulkan/vulkan_enums.hpp Normal file
View file

File diff suppressed because it is too large Load diff

1586
include/vulkan/vulkan/vulkan_extension_inspection.hpp Normal file
View file

File diff suppressed because it is too large Load diff

7614
include/vulkan/vulkan/vulkan_format_traits.hpp Normal file
View file

File diff suppressed because it is too large Load diff

258
include/vulkan/vulkan/vulkan_fuchsia.h Normal file
View file

@ -0,0 +1,258 @@
#ifndef VULKAN_FUCHSIA_H_
#define VULKAN_FUCHSIA_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_FUCHSIA_imagepipe_surface 1
#define VK_FUCHSIA_IMAGEPIPE_SURFACE_SPEC_VERSION 1
#define VK_FUCHSIA_IMAGEPIPE_SURFACE_EXTENSION_NAME "VK_FUCHSIA_imagepipe_surface"
typedef VkFlags VkImagePipeSurfaceCreateFlagsFUCHSIA;
typedef struct VkImagePipeSurfaceCreateInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkImagePipeSurfaceCreateFlagsFUCHSIA flags;
zx_handle_t imagePipeHandle;
} VkImagePipeSurfaceCreateInfoFUCHSIA;
typedef VkResult (VKAPI_PTR *PFN_vkCreateImagePipeSurfaceFUCHSIA)(VkInstance instance, const VkImagePipeSurfaceCreateInfoFUCHSIA* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateImagePipeSurfaceFUCHSIA(
VkInstance instance,
const VkImagePipeSurfaceCreateInfoFUCHSIA* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#define VK_FUCHSIA_external_memory 1
#define VK_FUCHSIA_EXTERNAL_MEMORY_SPEC_VERSION 1
#define VK_FUCHSIA_EXTERNAL_MEMORY_EXTENSION_NAME "VK_FUCHSIA_external_memory"
typedef struct VkImportMemoryZirconHandleInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkExternalMemoryHandleTypeFlagBits handleType;
zx_handle_t handle;
} VkImportMemoryZirconHandleInfoFUCHSIA;
typedef struct VkMemoryZirconHandlePropertiesFUCHSIA {
VkStructureType sType;
void* pNext;
uint32_t memoryTypeBits;
} VkMemoryZirconHandlePropertiesFUCHSIA;
typedef struct VkMemoryGetZirconHandleInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkDeviceMemory memory;
VkExternalMemoryHandleTypeFlagBits handleType;
} VkMemoryGetZirconHandleInfoFUCHSIA;
typedef VkResult (VKAPI_PTR *PFN_vkGetMemoryZirconHandleFUCHSIA)(VkDevice device, const VkMemoryGetZirconHandleInfoFUCHSIA* pGetZirconHandleInfo, zx_handle_t* pZirconHandle);
typedef VkResult (VKAPI_PTR *PFN_vkGetMemoryZirconHandlePropertiesFUCHSIA)(VkDevice device, VkExternalMemoryHandleTypeFlagBits handleType, zx_handle_t zirconHandle, VkMemoryZirconHandlePropertiesFUCHSIA* pMemoryZirconHandleProperties);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkGetMemoryZirconHandleFUCHSIA(
VkDevice device,
const VkMemoryGetZirconHandleInfoFUCHSIA* pGetZirconHandleInfo,
zx_handle_t* pZirconHandle);
VKAPI_ATTR VkResult VKAPI_CALL vkGetMemoryZirconHandlePropertiesFUCHSIA(
VkDevice device,
VkExternalMemoryHandleTypeFlagBits handleType,
zx_handle_t zirconHandle,
VkMemoryZirconHandlePropertiesFUCHSIA* pMemoryZirconHandleProperties);
#endif
#define VK_FUCHSIA_external_semaphore 1
#define VK_FUCHSIA_EXTERNAL_SEMAPHORE_SPEC_VERSION 1
#define VK_FUCHSIA_EXTERNAL_SEMAPHORE_EXTENSION_NAME "VK_FUCHSIA_external_semaphore"
typedef struct VkImportSemaphoreZirconHandleInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkSemaphore semaphore;
VkSemaphoreImportFlags flags;
VkExternalSemaphoreHandleTypeFlagBits handleType;
zx_handle_t zirconHandle;
} VkImportSemaphoreZirconHandleInfoFUCHSIA;
typedef struct VkSemaphoreGetZirconHandleInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkSemaphore semaphore;
VkExternalSemaphoreHandleTypeFlagBits handleType;
} VkSemaphoreGetZirconHandleInfoFUCHSIA;
typedef VkResult (VKAPI_PTR *PFN_vkImportSemaphoreZirconHandleFUCHSIA)(VkDevice device, const VkImportSemaphoreZirconHandleInfoFUCHSIA* pImportSemaphoreZirconHandleInfo);
typedef VkResult (VKAPI_PTR *PFN_vkGetSemaphoreZirconHandleFUCHSIA)(VkDevice device, const VkSemaphoreGetZirconHandleInfoFUCHSIA* pGetZirconHandleInfo, zx_handle_t* pZirconHandle);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkImportSemaphoreZirconHandleFUCHSIA(
VkDevice device,
const VkImportSemaphoreZirconHandleInfoFUCHSIA* pImportSemaphoreZirconHandleInfo);
VKAPI_ATTR VkResult VKAPI_CALL vkGetSemaphoreZirconHandleFUCHSIA(
VkDevice device,
const VkSemaphoreGetZirconHandleInfoFUCHSIA* pGetZirconHandleInfo,
zx_handle_t* pZirconHandle);
#endif
#define VK_FUCHSIA_buffer_collection 1
VK_DEFINE_NON_DISPATCHABLE_HANDLE(VkBufferCollectionFUCHSIA)
#define VK_FUCHSIA_BUFFER_COLLECTION_SPEC_VERSION 2
#define VK_FUCHSIA_BUFFER_COLLECTION_EXTENSION_NAME "VK_FUCHSIA_buffer_collection"
typedef VkFlags VkImageFormatConstraintsFlagsFUCHSIA;
typedef enum VkImageConstraintsInfoFlagBitsFUCHSIA {
VK_IMAGE_CONSTRAINTS_INFO_CPU_READ_RARELY_FUCHSIA = 0x00000001,
VK_IMAGE_CONSTRAINTS_INFO_CPU_READ_OFTEN_FUCHSIA = 0x00000002,
VK_IMAGE_CONSTRAINTS_INFO_CPU_WRITE_RARELY_FUCHSIA = 0x00000004,
VK_IMAGE_CONSTRAINTS_INFO_CPU_WRITE_OFTEN_FUCHSIA = 0x00000008,
VK_IMAGE_CONSTRAINTS_INFO_PROTECTED_OPTIONAL_FUCHSIA = 0x00000010,
VK_IMAGE_CONSTRAINTS_INFO_FLAG_BITS_MAX_ENUM_FUCHSIA = 0x7FFFFFFF
} VkImageConstraintsInfoFlagBitsFUCHSIA;
typedef VkFlags VkImageConstraintsInfoFlagsFUCHSIA;
typedef struct VkBufferCollectionCreateInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
zx_handle_t collectionToken;
} VkBufferCollectionCreateInfoFUCHSIA;
typedef struct VkImportMemoryBufferCollectionFUCHSIA {
VkStructureType sType;
const void* pNext;
VkBufferCollectionFUCHSIA collection;
uint32_t index;
} VkImportMemoryBufferCollectionFUCHSIA;
typedef struct VkBufferCollectionImageCreateInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkBufferCollectionFUCHSIA collection;
uint32_t index;
} VkBufferCollectionImageCreateInfoFUCHSIA;
typedef struct VkBufferCollectionConstraintsInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
uint32_t minBufferCount;
uint32_t maxBufferCount;
uint32_t minBufferCountForCamping;
uint32_t minBufferCountForDedicatedSlack;
uint32_t minBufferCountForSharedSlack;
} VkBufferCollectionConstraintsInfoFUCHSIA;
typedef struct VkBufferConstraintsInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkBufferCreateInfo createInfo;
VkFormatFeatureFlags requiredFormatFeatures;
VkBufferCollectionConstraintsInfoFUCHSIA bufferCollectionConstraints;
} VkBufferConstraintsInfoFUCHSIA;
typedef struct VkBufferCollectionBufferCreateInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkBufferCollectionFUCHSIA collection;
uint32_t index;
} VkBufferCollectionBufferCreateInfoFUCHSIA;
typedef struct VkSysmemColorSpaceFUCHSIA {
VkStructureType sType;
const void* pNext;
uint32_t colorSpace;
} VkSysmemColorSpaceFUCHSIA;
typedef struct VkBufferCollectionPropertiesFUCHSIA {
VkStructureType sType;
void* pNext;
uint32_t memoryTypeBits;
uint32_t bufferCount;
uint32_t createInfoIndex;
uint64_t sysmemPixelFormat;
VkFormatFeatureFlags formatFeatures;
VkSysmemColorSpaceFUCHSIA sysmemColorSpaceIndex;
VkComponentMapping samplerYcbcrConversionComponents;
VkSamplerYcbcrModelConversion suggestedYcbcrModel;
VkSamplerYcbcrRange suggestedYcbcrRange;
VkChromaLocation suggestedXChromaOffset;
VkChromaLocation suggestedYChromaOffset;
} VkBufferCollectionPropertiesFUCHSIA;
typedef struct VkImageFormatConstraintsInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
VkImageCreateInfo imageCreateInfo;
VkFormatFeatureFlags requiredFormatFeatures;
VkImageFormatConstraintsFlagsFUCHSIA flags;
uint64_t sysmemPixelFormat;
uint32_t colorSpaceCount;
const VkSysmemColorSpaceFUCHSIA* pColorSpaces;
} VkImageFormatConstraintsInfoFUCHSIA;
typedef struct VkImageConstraintsInfoFUCHSIA {
VkStructureType sType;
const void* pNext;
uint32_t formatConstraintsCount;
const VkImageFormatConstraintsInfoFUCHSIA* pFormatConstraints;
VkBufferCollectionConstraintsInfoFUCHSIA bufferCollectionConstraints;
VkImageConstraintsInfoFlagsFUCHSIA flags;
} VkImageConstraintsInfoFUCHSIA;
typedef VkResult (VKAPI_PTR *PFN_vkCreateBufferCollectionFUCHSIA)(VkDevice device, const VkBufferCollectionCreateInfoFUCHSIA* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkBufferCollectionFUCHSIA* pCollection);
typedef VkResult (VKAPI_PTR *PFN_vkSetBufferCollectionImageConstraintsFUCHSIA)(VkDevice device, VkBufferCollectionFUCHSIA collection, const VkImageConstraintsInfoFUCHSIA* pImageConstraintsInfo);
typedef VkResult (VKAPI_PTR *PFN_vkSetBufferCollectionBufferConstraintsFUCHSIA)(VkDevice device, VkBufferCollectionFUCHSIA collection, const VkBufferConstraintsInfoFUCHSIA* pBufferConstraintsInfo);
typedef void (VKAPI_PTR *PFN_vkDestroyBufferCollectionFUCHSIA)(VkDevice device, VkBufferCollectionFUCHSIA collection, const VkAllocationCallbacks* pAllocator);
typedef VkResult (VKAPI_PTR *PFN_vkGetBufferCollectionPropertiesFUCHSIA)(VkDevice device, VkBufferCollectionFUCHSIA collection, VkBufferCollectionPropertiesFUCHSIA* pProperties);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateBufferCollectionFUCHSIA(
VkDevice device,
const VkBufferCollectionCreateInfoFUCHSIA* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkBufferCollectionFUCHSIA* pCollection);
VKAPI_ATTR VkResult VKAPI_CALL vkSetBufferCollectionImageConstraintsFUCHSIA(
VkDevice device,
VkBufferCollectionFUCHSIA collection,
const VkImageConstraintsInfoFUCHSIA* pImageConstraintsInfo);
VKAPI_ATTR VkResult VKAPI_CALL vkSetBufferCollectionBufferConstraintsFUCHSIA(
VkDevice device,
VkBufferCollectionFUCHSIA collection,
const VkBufferConstraintsInfoFUCHSIA* pBufferConstraintsInfo);
VKAPI_ATTR void VKAPI_CALL vkDestroyBufferCollectionFUCHSIA(
VkDevice device,
VkBufferCollectionFUCHSIA collection,
const VkAllocationCallbacks* pAllocator);
VKAPI_ATTR VkResult VKAPI_CALL vkGetBufferCollectionPropertiesFUCHSIA(
VkDevice device,
VkBufferCollectionFUCHSIA collection,
VkBufferCollectionPropertiesFUCHSIA* pProperties);
#endif
#ifdef __cplusplus
}
#endif
#endif

21949
include/vulkan/vulkan/vulkan_funcs.hpp Normal file
View file

File diff suppressed because it is too large Load diff

58
include/vulkan/vulkan/vulkan_ggp.h Normal file
View file

@ -0,0 +1,58 @@
#ifndef VULKAN_GGP_H_
#define VULKAN_GGP_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_GGP_stream_descriptor_surface 1
#define VK_GGP_STREAM_DESCRIPTOR_SURFACE_SPEC_VERSION 1
#define VK_GGP_STREAM_DESCRIPTOR_SURFACE_EXTENSION_NAME "VK_GGP_stream_descriptor_surface"
typedef VkFlags VkStreamDescriptorSurfaceCreateFlagsGGP;
typedef struct VkStreamDescriptorSurfaceCreateInfoGGP {
VkStructureType sType;
const void* pNext;
VkStreamDescriptorSurfaceCreateFlagsGGP flags;
GgpStreamDescriptor streamDescriptor;
} VkStreamDescriptorSurfaceCreateInfoGGP;
typedef VkResult (VKAPI_PTR *PFN_vkCreateStreamDescriptorSurfaceGGP)(VkInstance instance, const VkStreamDescriptorSurfaceCreateInfoGGP* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateStreamDescriptorSurfaceGGP(
VkInstance instance,
const VkStreamDescriptorSurfaceCreateInfoGGP* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#define VK_GGP_frame_token 1
#define VK_GGP_FRAME_TOKEN_SPEC_VERSION 1
#define VK_GGP_FRAME_TOKEN_EXTENSION_NAME "VK_GGP_frame_token"
typedef struct VkPresentFrameTokenGGP {
VkStructureType sType;
const void* pNext;
GgpFrameToken frameToken;
} VkPresentFrameTokenGGP;
#ifdef __cplusplus
}
#endif
#endif

14943
include/vulkan/vulkan/vulkan_handles.hpp Normal file
View file

File diff suppressed because it is too large Load diff

14739
include/vulkan/vulkan/vulkan_hash.hpp Normal file
View file

File diff suppressed because it is too large Load diff

47
include/vulkan/vulkan/vulkan_ios.h Normal file
View file

@ -0,0 +1,47 @@
#ifndef VULKAN_IOS_H_
#define VULKAN_IOS_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_MVK_ios_surface 1
#define VK_MVK_IOS_SURFACE_SPEC_VERSION 3
#define VK_MVK_IOS_SURFACE_EXTENSION_NAME "VK_MVK_ios_surface"
typedef VkFlags VkIOSSurfaceCreateFlagsMVK;
typedef struct VkIOSSurfaceCreateInfoMVK {
VkStructureType sType;
const void* pNext;
VkIOSSurfaceCreateFlagsMVK flags;
const void* pView;
} VkIOSSurfaceCreateInfoMVK;
typedef VkResult (VKAPI_PTR *PFN_vkCreateIOSSurfaceMVK)(VkInstance instance, const VkIOSSurfaceCreateInfoMVK* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateIOSSurfaceMVK(
VkInstance instance,
const VkIOSSurfaceCreateInfoMVK* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#ifdef __cplusplus
}
#endif
#endif

47
include/vulkan/vulkan/vulkan_macos.h Normal file
View file

@ -0,0 +1,47 @@
#ifndef VULKAN_MACOS_H_
#define VULKAN_MACOS_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_MVK_macos_surface 1
#define VK_MVK_MACOS_SURFACE_SPEC_VERSION 3
#define VK_MVK_MACOS_SURFACE_EXTENSION_NAME "VK_MVK_macos_surface"
typedef VkFlags VkMacOSSurfaceCreateFlagsMVK;
typedef struct VkMacOSSurfaceCreateInfoMVK {
VkStructureType sType;
const void* pNext;
VkMacOSSurfaceCreateFlagsMVK flags;
const void* pView;
} VkMacOSSurfaceCreateInfoMVK;
typedef VkResult (VKAPI_PTR *PFN_vkCreateMacOSSurfaceMVK)(VkInstance instance, const VkMacOSSurfaceCreateInfoMVK* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateMacOSSurfaceMVK(
VkInstance instance,
const VkMacOSSurfaceCreateInfoMVK* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#ifdef __cplusplus
}
#endif
#endif

193
include/vulkan/vulkan/vulkan_metal.h Normal file
View file

@ -0,0 +1,193 @@
#ifndef VULKAN_METAL_H_
#define VULKAN_METAL_H_ 1
/*
** Copyright 2015-2023 The Khronos Group Inc.
**
** SPDX-License-Identifier: Apache-2.0
*/
/*
** This header is generated from the Khronos Vulkan XML API Registry.
**
*/
#ifdef __cplusplus
extern "C" {
#endif
#define VK_EXT_metal_surface 1
#ifdef __OBJC__
@class CAMetalLayer;
#else
typedef void CAMetalLayer;
#endif
#define VK_EXT_METAL_SURFACE_SPEC_VERSION 1
#define VK_EXT_METAL_SURFACE_EXTENSION_NAME "VK_EXT_metal_surface"
typedef VkFlags VkMetalSurfaceCreateFlagsEXT;
typedef struct VkMetalSurfaceCreateInfoEXT {
VkStructureType sType;
const void* pNext;
VkMetalSurfaceCreateFlagsEXT flags;
const CAMetalLayer* pLayer;
} VkMetalSurfaceCreateInfoEXT;
typedef VkResult (VKAPI_PTR *PFN_vkCreateMetalSurfaceEXT)(VkInstance instance, const VkMetalSurfaceCreateInfoEXT* pCreateInfo, const VkAllocationCallbacks* pAllocator, VkSurfaceKHR* pSurface);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR VkResult VKAPI_CALL vkCreateMetalSurfaceEXT(
VkInstance instance,
const VkMetalSurfaceCreateInfoEXT* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkSurfaceKHR* pSurface);
#endif
#define VK_EXT_metal_objects 1
#ifdef __OBJC__
@protocol MTLDevice;
typedef id<MTLDevice> MTLDevice_id;
#else
typedef void* MTLDevice_id;
#endif
#ifdef __OBJC__
@protocol MTLCommandQueue;
typedef id<MTLCommandQueue> MTLCommandQueue_id;
#else
typedef void* MTLCommandQueue_id;
#endif
#ifdef __OBJC__
@protocol MTLBuffer;
typedef id<MTLBuffer> MTLBuffer_id;
#else
typedef void* MTLBuffer_id;
#endif
#ifdef __OBJC__
@protocol MTLTexture;
typedef id<MTLTexture> MTLTexture_id;
#else
typedef void* MTLTexture_id;
#endif
typedef struct __IOSurface* IOSurfaceRef;
#ifdef __OBJC__
@protocol MTLSharedEvent;
typedef id<MTLSharedEvent> MTLSharedEvent_id;
#else
typedef void* MTLSharedEvent_id;
#endif
#define VK_EXT_METAL_OBJECTS_SPEC_VERSION 1
#define VK_EXT_METAL_OBJECTS_EXTENSION_NAME "VK_EXT_metal_objects"
typedef enum VkExportMetalObjectTypeFlagBitsEXT {
VK_EXPORT_METAL_OBJECT_TYPE_METAL_DEVICE_BIT_EXT = 0x00000001,
VK_EXPORT_METAL_OBJECT_TYPE_METAL_COMMAND_QUEUE_BIT_EXT = 0x00000002,
VK_EXPORT_METAL_OBJECT_TYPE_METAL_BUFFER_BIT_EXT = 0x00000004,
VK_EXPORT_METAL_OBJECT_TYPE_METAL_TEXTURE_BIT_EXT = 0x00000008,
VK_EXPORT_METAL_OBJECT_TYPE_METAL_IOSURFACE_BIT_EXT = 0x00000010,
VK_EXPORT_METAL_OBJECT_TYPE_METAL_SHARED_EVENT_BIT_EXT = 0x00000020,
VK_EXPORT_METAL_OBJECT_TYPE_FLAG_BITS_MAX_ENUM_EXT = 0x7FFFFFFF
} VkExportMetalObjectTypeFlagBitsEXT;
typedef VkFlags VkExportMetalObjectTypeFlagsEXT;
typedef struct VkExportMetalObjectCreateInfoEXT {
VkStructureType sType;
const void* pNext;
VkExportMetalObjectTypeFlagBitsEXT exportObjectType;
} VkExportMetalObjectCreateInfoEXT;
typedef struct VkExportMetalObjectsInfoEXT {
VkStructureType sType;
const void* pNext;
} VkExportMetalObjectsInfoEXT;
typedef struct VkExportMetalDeviceInfoEXT {
VkStructureType sType;
const void* pNext;
MTLDevice_id mtlDevice;
} VkExportMetalDeviceInfoEXT;
typedef struct VkExportMetalCommandQueueInfoEXT {
VkStructureType sType;
const void* pNext;
VkQueue queue;
MTLCommandQueue_id mtlCommandQueue;
} VkExportMetalCommandQueueInfoEXT;
typedef struct VkExportMetalBufferInfoEXT {
VkStructureType sType;
const void* pNext;
VkDeviceMemory memory;
MTLBuffer_id mtlBuffer;
} VkExportMetalBufferInfoEXT;
typedef struct VkImportMetalBufferInfoEXT {
VkStructureType sType;
const void* pNext;
MTLBuffer_id mtlBuffer;
} VkImportMetalBufferInfoEXT;
typedef struct VkExportMetalTextureInfoEXT {
VkStructureType sType;
const void* pNext;
VkImage image;
VkImageView imageView;
VkBufferView bufferView;
VkImageAspectFlagBits plane;
MTLTexture_id mtlTexture;
} VkExportMetalTextureInfoEXT;
typedef struct VkImportMetalTextureInfoEXT {
VkStructureType sType;
const void* pNext;
VkImageAspectFlagBits plane;
MTLTexture_id mtlTexture;
} VkImportMetalTextureInfoEXT;
typedef struct VkExportMetalIOSurfaceInfoEXT {
VkStructureType sType;
const void* pNext;
VkImage image;
IOSurfaceRef ioSurface;
} VkExportMetalIOSurfaceInfoEXT;
typedef struct VkImportMetalIOSurfaceInfoEXT {
VkStructureType sType;
const void* pNext;
IOSurfaceRef ioSurface;
} VkImportMetalIOSurfaceInfoEXT;
typedef struct VkExportMetalSharedEventInfoEXT {
VkStructureType sType;
const void* pNext;
VkSemaphore semaphore;
VkEvent event;
MTLSharedEvent_id mtlSharedEvent;
} VkExportMetalSharedEventInfoEXT;
typedef struct VkImportMetalSharedEventInfoEXT {
VkStructureType sType;
const void* pNext;
MTLSharedEvent_id mtlSharedEvent;
} VkImportMetalSharedEventInfoEXT;
typedef void (VKAPI_PTR *PFN_vkExportMetalObjectsEXT)(VkDevice device, VkExportMetalObjectsInfoEXT* pMetalObjectsInfo);
#ifndef VK_NO_PROTOTYPES
VKAPI_ATTR void VKAPI_CALL vkExportMetalObjectsEXT(
VkDevice device,
VkExportMetalObjectsInfoEXT* pMetalObjectsInfo);
#endif
#ifdef __cplusplus
}
#endif
#endif

Some files were not shown because too many files have changed in this diff Show more