vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 17:44:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Vulkan: Add DP4A MMQ and Q8_1 quantization shader (#12135)

Browse source 
* Vulkan: Add DP4A MMQ and Q8_1 quantization shader

* Add q4_0 x q8_1 matrix matrix multiplication support

* Vulkan: Add int8 coopmat MMQ support

* Vulkan: Add q4_1, q5_0 and q5_1 quants, improve integer dot code

* Add GL_EXT_integer_dot_product check

* Remove ggml changes, fix mmq pipeline picker

* Remove ggml changes, restore Intel coopmat behaviour

* Fix glsl compile attempt when integer vec dot is not supported

* Remove redundant code, use non-saturating integer dot, enable all matmul sizes for mmq

* Remove redundant comment

* Fix integer dot check

* Fix compile issue with unsupported int dot glslc

* Update Windows build Vulkan SDK version
This commit is contained in:
0cc4m 2025-03-31 14:37:01 +02:00 • committed by GitHub
parent 1790e73157
commit a8a1f33567
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 1146 additions and 95 deletions
Download patch file Download diff file Expand all files Collapse all files

24
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View file

@ -295,7 +295,10 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
std::string aligned_b_type_f32 = coopmat2 ? "float" : fp16 ? "mat2x4" : "vec4";
std::string aligned_b_type_f16 = coopmat2 ? "float16_t" : fp16 ? "f16mat2x4" : "f16vec4";
std::map<std::string, std::string> base_dict = {{"FLOAT_TYPE", (coopmat2 || fp16) ? "float16_t" : "float"}};
std::map<std::string, std::string> base_dict = {
{"FLOAT_TYPE", (coopmat2 || fp16) ? "float16_t" : "float"},
{"FLOAT_TYPE_VEC2", (coopmat2 || fp16) ? "f16vec2" : "vec2"},
};
std::string shader_name = "matmul";
if (matmul_id) {
@ -313,9 +316,7 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
base_dict["COOPMAT"] = "1";
}
base_dict["ACC_TYPE"] = f16acc ? "float16_t" : "float";
std::string source_name = coopmat2 ? "mul_mm_cm2.comp" : "mul_mm.comp";
const std::string source_name = coopmat2 ? "mul_mm_cm2.comp" : "mul_mm.comp";
// Shaders with f16 B_TYPE
string_to_spv(shader_name + "_f32_f16", source_name, merge_maps(base_dict, {{"DATA_A_F32", "1"}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}, }), fp16, coopmat, coopmat2, f16acc);
@ -339,14 +340,20 @@ void matmul_shaders(bool fp16, bool matmul_id, bool coopmat, bool coopmat2, bool
// don't generate f32 variants for coopmat2
if (!coopmat2) {
string_to_spv(shader_name + "_" + tname + "_f32", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f32", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f32_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
}
if (tname != "f16" && tname != "f32") {
string_to_spv(shader_name + "_" + tname + "_f16", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f16_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"B_IS_FLOAT", "1"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f16", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a_unaligned}, {"B_TYPE", "float16_t"}, {"D_TYPE", "float"}}), fp16, coopmat, coopmat2, f16acc);
string_to_spv(shader_name + "_" + tname + "_f16_aligned", source_name, merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f16}, {"D_TYPE", "float"}, {"ALIGNED", "1"}}), fp16, coopmat, coopmat2, f16acc);
}
#if defined(GGML_VULKAN_INTEGER_DOT_GLSLC_SUPPORT)
if (!coopmat && !coopmat2 && !matmul_id && (tname == "q4_0" || tname == "q4_1" || tname == "q5_0" || tname == "q5_1" || tname == "q8_0")) {
string_to_spv(shader_name + "_" + tname + "_q8_1", "mul_mmq.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"D_TYPE", "float"},}), fp16, coopmat, coopmat2, f16acc);
}
#endif
}
}
@ -458,6 +465,7 @@ void process_shaders() {
string_to_spv("acc_f32", "acc.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});
string_to_spv("split_k_reduce", "mul_mat_split_k_reduce.comp", {});
string_to_spv("quantize_q8_1", "quantize_q8_1.comp", {});
string_to_spv("mul_f32", "mul.comp", {{"A_TYPE", "float"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}, {"FLOAT_TYPE", "float"}});