mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2025-09-14 10:59:41 +00:00
merged but incoherent
This commit is contained in:
Concedo
commit
abf9531d08
6 changed files with 145814 additions and 145104 deletions
9
Makefile
9
Makefile
|
|
@ -505,7 +505,7 @@ 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/src/ggml-vulkan.cpp ggml/include/ggml-vulkan.h
|
||||
ggml-vulkan.o: ggml/src/ggml-vulkan.cpp ggml/include/ggml-vulkan.h ggml/src/ggml-vulkan-shaders.hpp ggml/src/ggml-vulkan-shaders.cpp
|
||||
$(CXX) $(CXXFLAGS) $(VULKAN_FLAGS) -c $< -o $@
|
||||
|
||||
# intermediate objects
|
||||
|
|
@ -554,7 +554,7 @@ gpttype_adapter_vulkan_noavx2.o: $(GPTTYPE_ADAPTER)
|
|||
$(CXX) $(CXXFLAGS) $(FAILSAFE_FLAGS) $(VULKAN_FLAGS) -c $< -o $@
|
||||
|
||||
clean:
|
||||
rm -vf *.o main sdmain whispermain quantize_gguf quantize_clip quantize_gpt2 quantize_gptj quantize_neox quantize_mpt quantize-stats perplexity embedding benchmark-matmult save-load-state gguf imatrix imatrix.exe gguf.exe main.exe quantize_clip.exe quantize_gguf.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_vulkan_noavx2.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 koboldcpp_vulkan_noavx2.so
|
||||
rm -vf *.o main sdmain whispermain quantize_gguf quantize_clip quantize_gpt2 quantize_gptj quantize_neox quantize_mpt quantize-stats perplexity embedding benchmark-matmult save-load-state gguf imatrix vulkan-shaders-gen vulkan-shaders-gen.exe imatrix.exe gguf.exe main.exe quantize_clip.exe quantize_gguf.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_vulkan_noavx2.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 koboldcpp_vulkan_noavx2.so
|
||||
rm -vrf ggml/src/ggml-cuda/*.o
|
||||
rm -vrf ggml/src/ggml-cuda/template-instances/*.o
|
||||
|
||||
|
|
@ -573,6 +573,11 @@ gguf: examples/gguf/gguf.cpp build-info.h ggml.o llama.o llavaclip_default.o lla
|
|||
gguf-split: examples/gguf-split/gguf-split.cpp ggml.o llama.o build-info.h llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
|
||||
vulkan-shaders-gen: ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
|
||||
@echo 'This command can be MANUALLY run to regenerate vulkan shaders. Normally concedo will do it, so you do not have to.'
|
||||
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
|
||||
@echo 'Now rebuilding vulkan shaders...'
|
||||
$(shell) vulkan-shaders-gen --glslc glslc --input-dir ggml/src/vulkan-shaders --target-hpp ggml/src/ggml-vulkan-shaders.hpp --target-cpp ggml/src/ggml-vulkan-shaders.cpp
|
||||
|
||||
#generated libraries
|
||||
koboldcpp_default: ggml.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter.o sdcpp_default.o whispercpp_default.o llavaclip_default.o llava.o ggml-backend_default.o $(OBJS_FULL) $(OBJS)
|
||||
|
|
|
|||
|
|
@ -1,220 +0,0 @@
|
|||
#!/usr/bin/env python
|
||||
|
||||
import logging
|
||||
import argparse
|
||||
import asyncio
|
||||
import os
|
||||
from tempfile import gettempdir
|
||||
|
||||
logger = logging.getLogger("ggml-vk-generate-shaders")
|
||||
|
||||
GLSLC = "glslc"
|
||||
|
||||
type_names = [
|
||||
"f32",
|
||||
"f16",
|
||||
"q4_0",
|
||||
"q4_1",
|
||||
"q5_0",
|
||||
"q5_1",
|
||||
"q8_0",
|
||||
"q2_k",
|
||||
"q3_k",
|
||||
"q4_k",
|
||||
"q5_k",
|
||||
"q6_k",
|
||||
]
|
||||
|
||||
ASYNCIO_CONCURRENCY = 64
|
||||
|
||||
input_dir = "vulkan-shaders"
|
||||
output_dir = gettempdir()
|
||||
|
||||
lock = asyncio.Lock()
|
||||
shader_fnames = []
|
||||
|
||||
|
||||
async def string_to_spv(name, in_fname, defines, fp16=True):
|
||||
name = f"{name}{'_fp32' if not fp16 else ''}"
|
||||
out_fname = os.path.join(output_dir, f"{name}.spv")
|
||||
|
||||
in_path = os.path.join(input_dir, in_fname)
|
||||
|
||||
cmd = [GLSLC, "-fshader-stage=compute", "--target-env=vulkan1.2", "-O", in_path, "-o", out_fname]
|
||||
|
||||
cmd.extend([f"-D{key}={value}" for key, value in defines.items()])
|
||||
|
||||
proc = await asyncio.create_subprocess_exec(*cmd, stdout=asyncio.subprocess.PIPE, stderr=asyncio.subprocess.PIPE)
|
||||
|
||||
stdout, stderr = await proc.communicate()
|
||||
|
||||
stdout = stdout.decode()
|
||||
error = stderr.decode()
|
||||
|
||||
if proc.returncode:
|
||||
cmd = " ".join(cmd)
|
||||
logger.error(f"cannot compile {name}\n\n{cmd}\n\n{error}")
|
||||
return
|
||||
|
||||
async with lock:
|
||||
shader_fnames.append((name, out_fname))
|
||||
|
||||
|
||||
def matmul_shaders(tasks, fp16, matmul_id):
|
||||
if fp16:
|
||||
load_vec = "8"
|
||||
aligned_b_type_f32 = "mat2x4"
|
||||
aligned_b_type_f16 = "f16mat2x4"
|
||||
else:
|
||||
load_vec = "4"
|
||||
aligned_b_type_f32 = "vec4"
|
||||
aligned_b_type_f16 = "f16vec4"
|
||||
|
||||
base_dict = {"FLOAT_TYPE": "float" if not fp16 else "float16_t"}
|
||||
shader_name = "matmul"
|
||||
|
||||
if matmul_id:
|
||||
base_dict["MUL_MAT_ID"] = "1"
|
||||
shader_name = "matmul_id"
|
||||
|
||||
if fp16:
|
||||
base_dict["FLOAT16"] = "1"
|
||||
|
||||
# Shaders with f16 B_TYPE
|
||||
tasks.append(string_to_spv(f"{shader_name}_f32_f16", "mul_mm.comp", base_dict | {"DATA_A_F32": "1", "B_TYPE": "float16_t", "D_TYPE": "float"}, fp16))
|
||||
tasks.append(string_to_spv(f"{shader_name}_f32_f16_aligned", "mul_mm.comp", base_dict | {"DATA_A_F32": "1", "LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "B_TYPE": aligned_b_type_f16, "D_TYPE": "float"}, fp16))
|
||||
|
||||
tasks.append(string_to_spv(f"{shader_name}_f16", "mul_mm.comp", base_dict | {"DATA_A_F16": "1", "B_TYPE": "float16_t", "D_TYPE": "float"}, fp16))
|
||||
tasks.append(string_to_spv(f"{shader_name}_f16_aligned", "mul_mm.comp", base_dict | {"DATA_A_F16": "1", "LOAD_VEC_A": load_vec, "LOAD_VEC_B": load_vec, "B_TYPE": aligned_b_type_f16, "D_TYPE": "float"}, fp16))
|
||||
|
||||
for tname in type_names:
|
||||
data_a_key = f"DATA_A_{tname.upper()}"
|
||||
load_vec_a = load_vec if tname in ("f32", "f16") else "2"
|
||||
tasks.append(string_to_spv(f"{shader_name}_{tname}_f32", "mul_mm.comp", base_dict | {data_a_key: "1", "B_TYPE": "float", "D_TYPE": "float"}, fp16))
|
||||
tasks.append(string_to_spv(f"{shader_name}_{tname}_f32_aligned", "mul_mm.comp", base_dict | {data_a_key: "2", "LOAD_VEC_A": load_vec_a, "LOAD_VEC_B": load_vec, "B_TYPE": aligned_b_type_f32, "D_TYPE": "float"}, fp16))
|
||||
|
||||
|
||||
async def main():
|
||||
logger.info("ggml_vulkan: Generating and compiling shaders to SPIR-V")
|
||||
|
||||
tasks = []
|
||||
|
||||
base_dict = {"FLOAT_TYPE": "float"}
|
||||
|
||||
for fp16 in (False, True):
|
||||
# MUL_MAT
|
||||
matmul_shaders(tasks, fp16, False)
|
||||
# MUL_MAT_ID
|
||||
matmul_shaders(tasks, fp16, True)
|
||||
|
||||
for tname in type_names:
|
||||
# mul mat vec
|
||||
data_a_key = f"DATA_A_{tname.upper()}"
|
||||
shader = f"mul_mat_vec_{tname}.comp" if tname.endswith("_k") else "mul_mat_vec.comp"
|
||||
|
||||
tasks.append(string_to_spv(f"mul_mat_vec_{tname}_f32_f32", shader, base_dict | {data_a_key: "1", "B_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv(f"mul_mat_vec_{tname}_f16_f32", shader, base_dict | {data_a_key: "1", "B_TYPE": "float16_t", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv(f"mul_mat_vec_id_{tname}_f32", shader, base_dict | {"MUL_MAT_ID": "1", data_a_key: "1", "B_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
# Dequant shaders
|
||||
if tname != "f16":
|
||||
tasks.append(string_to_spv(f"dequant_{tname}", f"dequant_{tname}.comp", base_dict | {data_a_key: "1", "D_TYPE": "float16_t"}))
|
||||
|
||||
# get_rows
|
||||
if not tname.endswith("_k"):
|
||||
shader = "get_rows.comp" if tname in ("f32", "f16") else "get_rows_quant.comp"
|
||||
|
||||
if tname == "f16":
|
||||
tasks.append(string_to_spv(f"get_rows_{tname}", shader, {data_a_key: "1", "B_TYPE": "int", "D_TYPE": "float16_t", "OPTIMIZATION_ERROR_WORKAROUND": "1"}))
|
||||
else:
|
||||
tasks.append(string_to_spv(f"get_rows_{tname}", shader, {data_a_key: "1", "B_TYPE": "int", "D_TYPE": "float16_t"}))
|
||||
tasks.append(string_to_spv(f"get_rows_{tname}_f32", shader, {data_a_key: "1", "B_TYPE": "int", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("mul_mat_vec_p021_f16_f32", "mul_mat_vec_p021.comp", {"A_TYPE": "float16_t", "B_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("mul_mat_vec_nc_f16_f32", "mul_mat_vec_nc.comp", {"A_TYPE": "float16_t", "B_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
# Norms
|
||||
tasks.append(string_to_spv("norm_f32", "norm.comp", base_dict | {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("rms_norm_f32", "rms_norm.comp", base_dict | {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("cpy_f32_f32", "copy.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("cpy_f32_f16", "copy.comp", {"A_TYPE": "float", "D_TYPE": "float16_t"}))
|
||||
tasks.append(string_to_spv("cpy_f16_f16", "copy.comp", {"A_TYPE": "float16_t", "D_TYPE": "float16_t", "OPTIMIZATION_ERROR_WORKAROUND": "1"}))
|
||||
|
||||
tasks.append(string_to_spv("add_f32", "add.comp", {"A_TYPE": "float", "B_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {}))
|
||||
|
||||
tasks.append(string_to_spv("mul_f32", "mul.comp", {"A_TYPE": "float", "B_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("div_f32", "div.comp", {"A_TYPE": "float", "B_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("scale_f32", "scale.comp", {"A_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("sqr_f32", "square.comp", {"A_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("clamp_f32", "clamp.comp", {"A_TYPE": "float", "D_TYPE": "float", "FLOAT_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("gelu_f32", "gelu.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("silu_f32", "silu.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("relu_f32", "relu.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("diag_mask_inf_f32", "diag_mask_inf.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("soft_max_f32", "soft_max.comp", base_dict | {"A_TYPE": "float", "B_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("soft_max_f32_f16", "soft_max.comp", base_dict | {"A_TYPE": "float", "B_TYPE": "float16_t", "D_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("rope_norm_f32", "rope_norm.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("rope_norm_f16", "rope_norm.comp", {"A_TYPE": "float16_t", "D_TYPE": "float16_t"}))
|
||||
|
||||
tasks.append(string_to_spv("rope_neox_f32", "rope_neox.comp", {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
tasks.append(string_to_spv("rope_neox_f16", "rope_neox.comp", {"A_TYPE": "float16_t", "D_TYPE": "float16_t"}))
|
||||
|
||||
tasks.append(string_to_spv("argsort_f32", "argsort.comp", {"A_TYPE": "float"}))
|
||||
|
||||
tasks.append(string_to_spv("sum_rows_f32", "sum_rows.comp", base_dict | {"A_TYPE": "float", "D_TYPE": "float"}))
|
||||
|
||||
# Helper to decorate tasks with semaphore acquisition.
|
||||
async def withSemaphore(sem, task):
|
||||
async with sem:
|
||||
return await task
|
||||
|
||||
# Run tasks concurrently guarded by a concurrency limit.
|
||||
sem = asyncio.Semaphore(ASYNCIO_CONCURRENCY)
|
||||
await asyncio.gather(*(withSemaphore(sem, task) for task in tasks))
|
||||
|
||||
with open("ggml-vulkan-shaders.hpp", "w") as f:
|
||||
f.write("#include <cstdint>\n\n")
|
||||
for name, path in sorted(shader_fnames):
|
||||
|
||||
with open(path, "rb") as spv:
|
||||
counter = 0
|
||||
newline_counter = 0
|
||||
f.write(f"unsigned char {name}_data[] = {{\n")
|
||||
for val in spv.read():
|
||||
f.write(f"0x{val:02x},")
|
||||
newline_counter += 1
|
||||
counter += 1
|
||||
if newline_counter >= 12:
|
||||
newline_counter = 0
|
||||
f.write("\n")
|
||||
f.write("\n};\n")
|
||||
f.write(f"const uint64_t {name}_len = {counter};\n\n")
|
||||
os.remove(path)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
parser = argparse.ArgumentParser(description="GGML Vulkan Shader Generator")
|
||||
|
||||
parser.add_argument("--glslc", help="Path to glslc")
|
||||
parser.add_argument("--verbose", action="store_true", help="increase output verbosity")
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
logging.basicConfig(level=logging.DEBUG if args.verbose else logging.INFO)
|
||||
|
||||
if args.glslc:
|
||||
GLSLC = args.glslc
|
||||
|
||||
asyncio.run(main())
|
||||
144758
ggml/src/ggml-vulkan-shaders.cpp
Normal file
144758
ggml/src/ggml-vulkan-shaders.cpp
Normal file
File diff suppressed because it is too large
Load diff
File diff suppressed because it is too large
Load diff
|
|
@ -23,7 +23,7 @@
|
|||
#include "ggml.h"
|
||||
#include "ggml-backend-impl.h"
|
||||
|
||||
#include "ggml-vulkan-shaders.hpp"
|
||||
#include "ggml-vulkan-shaders.cpp"
|
||||
|
||||
#define VK_API_VERSION VK_API_VERSION_1_2
|
||||
|
||||
|
|
|
|||
524
ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
Normal file
524
ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
Normal file
|
|
@ -0,0 +1,524 @@
|
|||
|
||||
|
||||
#include <iostream>
|
||||
#include <fstream>
|
||||
#include <sstream>
|
||||
#include <string>
|
||||
#include <stdexcept>
|
||||
#include <array>
|
||||
#include <vector>
|
||||
#include <map>
|
||||
#include <thread>
|
||||
#include <mutex>
|
||||
#include <future>
|
||||
#include <queue>
|
||||
#include <condition_variable>
|
||||
#include <cstdio>
|
||||
#include <cstring>
|
||||
#include <cstdlib>
|
||||
#include <sys/stat.h>
|
||||
#include <sys/types.h>
|
||||
|
||||
#ifdef _WIN32
|
||||
#include <windows.h>
|
||||
#include <direct.h> // For _mkdir on Windows
|
||||
#else
|
||||
#include <unistd.h>
|
||||
#include <sys/wait.h>
|
||||
#include <fcntl.h>
|
||||
#endif
|
||||
|
||||
#define ASYNCIO_CONCURRENCY 64
|
||||
|
||||
std::mutex lock;
|
||||
std::vector<std::pair<std::string, std::string>> shader_fnames;
|
||||
|
||||
std::string GLSLC = "glslc";
|
||||
std::string input_dir = "vulkan-shaders";
|
||||
std::string output_dir = "/tmp";
|
||||
std::string target_hpp = "ggml-vulkan-shaders.hpp";
|
||||
std::string target_cpp = "ggml-vulkan-shaders.cpp";
|
||||
bool no_clean = false;
|
||||
|
||||
const std::vector<std::string> type_names = {
|
||||
"f32",
|
||||
"f16",
|
||||
"q4_0",
|
||||
"q4_1",
|
||||
"q5_0",
|
||||
"q5_1",
|
||||
"q8_0",
|
||||
"q2_k",
|
||||
"q3_k",
|
||||
"q4_k",
|
||||
"q5_k",
|
||||
"q6_k"
|
||||
};
|
||||
|
||||
void execute_command(const std::string& command, std::string& stdout_str, std::string& stderr_str) {
|
||||
#ifdef _WIN32
|
||||
HANDLE stdout_read, stdout_write;
|
||||
HANDLE stderr_read, stderr_write;
|
||||
SECURITY_ATTRIBUTES sa = { sizeof(SECURITY_ATTRIBUTES), NULL, TRUE };
|
||||
|
||||
if (!CreatePipe(&stdout_read, &stdout_write, &sa, 0) ||
|
||||
!SetHandleInformation(stdout_read, HANDLE_FLAG_INHERIT, 0)) {
|
||||
throw std::runtime_error("Failed to create stdout pipe");
|
||||
}
|
||||
|
||||
if (!CreatePipe(&stderr_read, &stderr_write, &sa, 0) ||
|
||||
!SetHandleInformation(stderr_read, HANDLE_FLAG_INHERIT, 0)) {
|
||||
throw std::runtime_error("Failed to create stderr pipe");
|
||||
}
|
||||
|
||||
PROCESS_INFORMATION pi;
|
||||
STARTUPINFOA si = { sizeof(STARTUPINFOA) };
|
||||
si.dwFlags = STARTF_USESTDHANDLES;
|
||||
si.hStdOutput = stdout_write;
|
||||
si.hStdError = stderr_write;
|
||||
|
||||
std::vector<char> cmd(command.begin(), command.end());
|
||||
cmd.push_back('\0');
|
||||
|
||||
if (!CreateProcessA(NULL, cmd.data(), NULL, NULL, TRUE, 0, NULL, NULL, &si, &pi)) {
|
||||
throw std::runtime_error("Failed to create process");
|
||||
}
|
||||
|
||||
CloseHandle(stdout_write);
|
||||
CloseHandle(stderr_write);
|
||||
|
||||
std::array<char, 128> buffer;
|
||||
DWORD bytes_read;
|
||||
|
||||
while (ReadFile(stdout_read, buffer.data(), buffer.size(), &bytes_read, NULL) && bytes_read > 0) {
|
||||
stdout_str.append(buffer.data(), bytes_read);
|
||||
}
|
||||
|
||||
while (ReadFile(stderr_read, buffer.data(), buffer.size(), &bytes_read, NULL) && bytes_read > 0) {
|
||||
stderr_str.append(buffer.data(), bytes_read);
|
||||
}
|
||||
|
||||
CloseHandle(stdout_read);
|
||||
CloseHandle(stderr_read);
|
||||
WaitForSingleObject(pi.hProcess, INFINITE);
|
||||
CloseHandle(pi.hProcess);
|
||||
CloseHandle(pi.hThread);
|
||||
#else
|
||||
int stdout_pipe[2];
|
||||
int stderr_pipe[2];
|
||||
|
||||
if (pipe(stdout_pipe) != 0 || pipe(stderr_pipe) != 0) {
|
||||
throw std::runtime_error("Failed to create pipes");
|
||||
}
|
||||
|
||||
pid_t pid = fork();
|
||||
if (pid < 0) {
|
||||
throw std::runtime_error("Failed to fork process");
|
||||
}
|
||||
|
||||
if (pid == 0) {
|
||||
close(stdout_pipe[0]);
|
||||
close(stderr_pipe[0]);
|
||||
dup2(stdout_pipe[1], STDOUT_FILENO);
|
||||
dup2(stderr_pipe[1], STDERR_FILENO);
|
||||
close(stdout_pipe[1]);
|
||||
close(stderr_pipe[1]);
|
||||
execl("/bin/sh", "sh", "-c", command.c_str(), (char*) nullptr);
|
||||
_exit(EXIT_FAILURE);
|
||||
} else {
|
||||
close(stdout_pipe[1]);
|
||||
close(stderr_pipe[1]);
|
||||
|
||||
std::array<char, 128> buffer;
|
||||
ssize_t bytes_read;
|
||||
|
||||
while ((bytes_read = read(stdout_pipe[0], buffer.data(), buffer.size())) > 0) {
|
||||
stdout_str.append(buffer.data(), bytes_read);
|
||||
}
|
||||
|
||||
while ((bytes_read = read(stderr_pipe[0], buffer.data(), buffer.size())) > 0) {
|
||||
stderr_str.append(buffer.data(), bytes_read);
|
||||
}
|
||||
|
||||
close(stdout_pipe[0]);
|
||||
close(stderr_pipe[0]);
|
||||
waitpid(pid, nullptr, 0);
|
||||
}
|
||||
#endif
|
||||
}
|
||||
|
||||
bool directory_exists(const std::string& path) {
|
||||
struct stat info;
|
||||
if (stat(path.c_str(), &info) != 0) {
|
||||
return false; // Path doesn't exist or can't be accessed
|
||||
}
|
||||
return (info.st_mode & S_IFDIR) != 0; // Check if it is a directory
|
||||
}
|
||||
|
||||
bool create_directory(const std::string& path) {
|
||||
#ifdef _WIN32
|
||||
return _mkdir(path.c_str()) == 0 || errno == EEXIST; // EEXIST means the directory already exists
|
||||
#else
|
||||
return mkdir(path.c_str(), 0755) == 0 || errno == EEXIST; // 0755 is the directory permissions
|
||||
#endif
|
||||
}
|
||||
|
||||
std::string to_uppercase(const std::string& input) {
|
||||
std::string result = input;
|
||||
for (char& c : result) {
|
||||
c = std::toupper(c);
|
||||
}
|
||||
return result;
|
||||
}
|
||||
|
||||
bool string_ends_with(const std::string& str, const std::string& suffix) {
|
||||
if (suffix.size() > str.size()) {
|
||||
return false;
|
||||
}
|
||||
return std::equal(suffix.rbegin(), suffix.rend(), str.rbegin());
|
||||
}
|
||||
|
||||
#ifdef _WIN32
|
||||
static const char path_separator = '\\';
|
||||
#else
|
||||
static const char path_separator = '/';
|
||||
#endif
|
||||
|
||||
std::string join_paths(const std::string& path1, const std::string& path2) {
|
||||
return path1 + path_separator + path2;
|
||||
}
|
||||
|
||||
std::string basename(const std::string &path) {
|
||||
return path.substr(path.find_last_of("/\\") + 1);
|
||||
}
|
||||
|
||||
void string_to_spv(const std::string& _name, const std::string& in_fname, const std::map<std::string, std::string>& defines, bool fp16 = true) {
|
||||
std::string name = _name + (fp16 ? "" : "_fp32");
|
||||
std::string out_fname = join_paths(output_dir, name + ".spv");
|
||||
std::string in_path = join_paths(input_dir, in_fname);
|
||||
|
||||
std::vector<std::string> cmd = {GLSLC, "-fshader-stage=compute", "--target-env=vulkan1.2", "-O", in_path, "-o", out_fname};
|
||||
for (const auto& define : defines) {
|
||||
cmd.push_back("-D" + define.first + "=" + define.second);
|
||||
}
|
||||
|
||||
std::string command;
|
||||
for (const auto& part : cmd) {
|
||||
command += part + " ";
|
||||
}
|
||||
|
||||
std::string stdout_str, stderr_str;
|
||||
try {
|
||||
// std::cout << "Executing command: ";
|
||||
// for (const auto& part : cmd) {
|
||||
// std::cout << part << " ";
|
||||
// }
|
||||
// std::cout << std::endl;
|
||||
|
||||
execute_command(command, stdout_str, stderr_str);
|
||||
if (!stderr_str.empty()) {
|
||||
std::cerr << "cannot compile " << name << "\n\n" << command << "\n\n" << stderr_str << std::endl;
|
||||
return;
|
||||
}
|
||||
|
||||
std::lock_guard<std::mutex> guard(lock);
|
||||
shader_fnames.push_back(std::make_pair(name, out_fname));
|
||||
} catch (const std::exception& e) {
|
||||
std::cerr << "Error executing command for " << name << ": " << e.what() << std::endl;
|
||||
}
|
||||
}
|
||||
|
||||
std::map<std::string, std::string> merge_maps(const std::map<std::string, std::string>& a, const std::map<std::string, std::string>& b) {
|
||||
std::map<std::string, std::string> result = a;
|
||||
result.insert(b.begin(), b.end());
|
||||
return result;
|
||||
}
|
||||
|
||||
void matmul_shaders(std::vector<std::future<void>>& tasks, bool fp16, bool matmul_id) {
|
||||
std::string load_vec = fp16 ? "8" : "4";
|
||||
std::string aligned_b_type_f32 = fp16 ? "mat2x4" : "vec4";
|
||||
std::string aligned_b_type_f16 = fp16 ? "f16mat2x4" : "f16vec4";
|
||||
|
||||
std::map<std::string, std::string> base_dict = {{"FLOAT_TYPE", fp16 ? "float16_t" : "float"}};
|
||||
std::string shader_name = "matmul";
|
||||
|
||||
if (matmul_id) {
|
||||
base_dict["MUL_MAT_ID"] = "1";
|
||||
shader_name = "matmul_id";
|
||||
}
|
||||
|
||||
if (fp16) {
|
||||
base_dict["FLOAT16"] = "1";
|
||||
}
|
||||
|
||||
// Shaders with f16 B_TYPE
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_f32_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_f32_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_f16", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_f16_aligned", "mul_mm.comp", merge_maps(base_dict, {{"DATA_A_F16", "1"}, {"LOAD_VEC_A", load_vec}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
|
||||
for (const auto& tname : type_names) {
|
||||
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
|
||||
std::string load_vec_a = (tname == "f32" || tname == "f16") ? load_vec : "2";
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_" + tname + "_f32", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv(shader_name + "_" + tname + "_f32_aligned", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "2"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}}), fp16);
|
||||
}));
|
||||
}
|
||||
}
|
||||
|
||||
void process_shaders(std::vector<std::future<void>>& tasks) {
|
||||
std::cout << "ggml_vulkan: Generating and compiling shaders to SPIR-V" << std::endl;
|
||||
std::map<std::string, std::string> base_dict = {{"FLOAT_TYPE", "float"}};
|
||||
|
||||
for (const auto& fp16 : {false, true}) {
|
||||
matmul_shaders(tasks, fp16, false);
|
||||
matmul_shaders(tasks, fp16, true);
|
||||
}
|
||||
|
||||
for (const auto& tname : type_names) {
|
||||
// mul mat vec
|
||||
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
|
||||
std::string shader = (string_ends_with(tname, "_k")) ? "mul_mat_vec_" + tname + ".comp" : "mul_mat_vec.comp";
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("mul_mat_vec_" + tname + "_f32_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("mul_mat_vec_" + tname + "_f16_f32", shader, merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("mul_mat_vec_id_" + tname + "_f32", shader, merge_maps(base_dict, {{"MUL_MAT_ID", "1"}, {data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
|
||||
// Dequant shaders
|
||||
if (tname != "f16") {
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("dequant_" + tname, "dequant_" + tname + ".comp", merge_maps(base_dict, {{data_a_key, "1"}, {"D_TYPE", "float16_t"}}));
|
||||
}));
|
||||
}
|
||||
|
||||
if (!string_ends_with(tname, "_k")) {
|
||||
shader = (tname == "f32" || tname == "f16") ? "get_rows.comp" : "get_rows_quant.comp";
|
||||
|
||||
if (tname == "f16") {
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
|
||||
}));
|
||||
} else {
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("get_rows_" + tname, shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float16_t"}});
|
||||
}));
|
||||
}
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("get_rows_" + tname + "_f32", shader, {{data_a_key, "1"}, {"B_TYPE", "int"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
}
|
||||
}
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("mul_mat_vec_p021_f16_f32", "mul_mat_vec_p021.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("mul_mat_vec_nc_f16_f32", "mul_mat_vec_nc.comp", {{"A_TYPE", "float16_t"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
// Norms
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("norm_f32", "norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("rms_norm_f32", "rms_norm.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("cpy_f32_f32", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("cpy_f32_f16", "copy.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float16_t"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("cpy_f16_f16", "copy.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}, {"OPTIMIZATION_ERROR_WORKAROUND", "1"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("add_f32", "add.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("mul_f32", "mul.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("div_f32", "div.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("scale_f32", "scale.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("sqr_f32", "square.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("clamp_f32", "clamp.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("gelu_f32", "gelu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("silu_f32", "silu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("relu_f32", "relu.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("diag_mask_inf_f32", "diag_mask_inf.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("soft_max_f32", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("soft_max_f32_f16", "soft_max.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("rope_norm_f32", "rope_norm.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("rope_norm_f16", "rope_norm.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("rope_neox_f32", "rope_neox.comp", {{"A_TYPE", "float"}, {"D_TYPE", "float"}});
|
||||
}));
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("rope_neox_f16", "rope_neox.comp", {{"A_TYPE", "float16_t"}, {"D_TYPE", "float16_t"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [] {
|
||||
string_to_spv("argsort_f32", "argsort.comp", {{"A_TYPE", "float"}});
|
||||
}));
|
||||
|
||||
tasks.push_back(std::async(std::launch::async, [=] {
|
||||
string_to_spv("sum_rows_f32", "sum_rows.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
|
||||
}));
|
||||
}
|
||||
|
||||
void write_output_files() {
|
||||
FILE* hdr = fopen(target_hpp.c_str(), "w");
|
||||
FILE* src = fopen(target_cpp.c_str(), "w");
|
||||
|
||||
fprintf(hdr, "#include <cstdint>\n\n");
|
||||
fprintf(src, "#include \"%s\"\n\n", basename(target_hpp).c_str());
|
||||
|
||||
for (const auto& pair : shader_fnames) {
|
||||
const std::string& name = pair.first;
|
||||
const std::string& path = pair.second;
|
||||
FILE* spv = fopen(path.c_str(), "rb");
|
||||
if (!spv) {
|
||||
std::cerr << "Error opening SPIR-V file: " << path << "\n";
|
||||
continue;
|
||||
}
|
||||
|
||||
fseek(spv, 0, SEEK_END);
|
||||
size_t size = ftell(spv);
|
||||
fseek(spv, 0, SEEK_SET);
|
||||
|
||||
std::vector<unsigned char> data(size);
|
||||
size_t read_size = fread(data.data(), 1, size, spv);
|
||||
fclose(spv);
|
||||
if (read_size != size) {
|
||||
std::cerr << "Error reading SPIR-V file: " << path << "\n";
|
||||
continue;
|
||||
}
|
||||
|
||||
fprintf(hdr, "extern unsigned char %s_data[%zu];\n", name.c_str(), size);
|
||||
fprintf(hdr, "const uint64_t %s_len = %zu;\n\n", name.c_str(), size);
|
||||
|
||||
fprintf(src, "unsigned char %s_data[%zu] = {\n", name.c_str(), size);
|
||||
for (size_t i = 0; i < size; ++i) {
|
||||
fprintf(src, "0x%02x,", data[i]);
|
||||
if ((i + 1) % 12 == 0) fprintf(src, "\n");
|
||||
}
|
||||
fprintf(src, "\n};\n\n");
|
||||
|
||||
if (!no_clean) {
|
||||
std::remove(path.c_str());
|
||||
}
|
||||
}
|
||||
|
||||
fclose(hdr);
|
||||
fclose(src);
|
||||
}
|
||||
|
||||
int main(int argc, char** argv) {
|
||||
std::map<std::string, std::string> args;
|
||||
for (int i = 1; i < argc; i += 2) {
|
||||
if (i + 1 < argc) {
|
||||
args[argv[i]] = argv[i + 1];
|
||||
}
|
||||
}
|
||||
|
||||
if (args.find("--glslc") != args.end()) {
|
||||
GLSLC = args["--glslc"]; // Path to glslc
|
||||
}
|
||||
if (args.find("--input-dir") != args.end()) {
|
||||
input_dir = args["--input-dir"]; // Directory containing shader sources
|
||||
}
|
||||
if (args.find("--output-dir") != args.end()) {
|
||||
output_dir = args["--output-dir"]; // Directory for containing SPIR-V output
|
||||
}
|
||||
if (args.find("--target-hpp") != args.end()) {
|
||||
target_hpp = args["--target-hpp"]; // Path to generated header file
|
||||
}
|
||||
if (args.find("--target-cpp") != args.end()) {
|
||||
target_cpp = args["--target-cpp"]; // Path to generated cpp file
|
||||
}
|
||||
if (args.find("--no-clean") != args.end()) {
|
||||
no_clean = true; // Keep temporary SPIR-V files in output-dir after build
|
||||
}
|
||||
|
||||
if (!directory_exists(input_dir)) {
|
||||
std::cerr << "\"" << input_dir << "\" must be a valid directory containing shader sources" << std::endl;
|
||||
return EXIT_FAILURE;
|
||||
}
|
||||
|
||||
if (!directory_exists(output_dir)) {
|
||||
if (!create_directory(output_dir)) {
|
||||
std::cerr << "Error creating output directory: " << output_dir << "\n";
|
||||
return EXIT_FAILURE;
|
||||
}
|
||||
}
|
||||
|
||||
std::vector<std::future<void>> tasks;
|
||||
process_shaders(tasks);
|
||||
|
||||
for (auto& task : tasks) {
|
||||
task.get();
|
||||
}
|
||||
|
||||
write_output_files();
|
||||
|
||||
return EXIT_SUCCESS;
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue