ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-16 08:44:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix group 128 for both decode and prefill shapes

Browse Source
This commit is contained in:
khuagarw
2025-12-19 01:05:01 +00:00
parent 39afffac85
commit 5b4a67ec6d
10 changed files with 148 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/38_block_scale_gemm/CMakeLists.txt
View File

@@ -18,7 +18,7 @@ if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx12")
# gemm_bquant_quantgrouped_fp8i4.cpp
# gemm_bquant_quantgrouped_bf16mxfp4.cpp
# gemm_bquant_quantgrouped_bf8.cpp
# gemm_bquant_quantgrouped_fp8.cpp
gemm_bquant_quantgrouped_fp8.cpp
# gemm_bquant_quantgrouped_preshuffleb.cpp
gemm_bquant_quantgrouped_preshufflequant.cpp
# gemm_bquant_quantgrouped_preshuffleb_preshufflequant.cpp

8
example/ck_tile/38_block_scale_gemm/gemm_bquant_quantgrouped_fp8.cpp
View File

@@ -4,7 +4,7 @@
#include "run_gemm_quant_example.inc"
template <typename T>
using GemmConfig = GemmConfigQuantPrefill<T>;
using GemmConfig = GemmConfigQuantDecode<T>;
#define RUN_GEMM_EXAMPLE_PREC_TYPE \
run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>, \
@@ -49,4 +49,10 @@ void bquant_quantgrouped_fp8_instance_factory(
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 64, 128>>;
return RUN_GEMM_EXAMPLE_PREC_TYPE;
};
lut[hash_multiple_strings(
{"fp8", "bquant", "non-preshuffleb", "non-preshufflequant", "1x128x128"})] =
[](const ck_tile::ArgParser& arg_parser) {
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 128, 128>>;
return RUN_GEMM_EXAMPLE_PREC_TYPE;
};
}

83
example/ck_tile/38_block_scale_gemm/gemm_bquant_quantgrouped_preshufflequant.cpp
View File

@@ -4,8 +4,8 @@
#include "run_gemm_quant_example.inc"
template <typename T>
using GemmConfig = GemmConfigPreshuffleBQuantPrefill<T>; // GemmConfigPreshuffleQuantDecode<T>;
// //GemmConfigPreshuffleBQuantPrefill<T>;
using GemmConfig = GemmConfigPreshuffleQuantDecode<T>;
// //GemmConfigPreshuffleBQuantPrefill<T>;
void bquant_quantgrouped_preshufflequant_instance_factory(
std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut)
@@ -24,31 +24,33 @@ void bquant_quantgrouped_preshufflequant_instance_factory(
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
lut[hash_multiple_strings({"fp8", "bquant", "non-preshuffleb", "preshufflequant", "1x8x128"})] =
[](const ck_tile::ArgParser& arg_parser) {
using TypeConfig = decltype(GemmQuantTypeConfig<ck_tile::fp8_t,
ck_tile::fp8_t,
ck_tile::half_t,
float>{});
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 8, 128>>;
return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
TypeConfig,
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
lut[hash_multiple_strings({"fp8",
"bquant",
"non-preshuffleb",
"preshufflequant",
"1x16x128"})] = [](const ck_tile::ArgParser& arg_parser) {
using TypeConfig =
decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t, float>{});
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 16, 128>>;
return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
TypeConfig,
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
// lut[hash_multiple_strings({"fp8", "bquant", "non-preshuffleb", "preshufflequant",
// "1x8x128"})] =
// [](const ck_tile::ArgParser& arg_parser) {
// using TypeConfig = decltype(GemmQuantTypeConfig<ck_tile::fp8_t,
// ck_tile::fp8_t,
// ck_tile::half_t,
// float>{});
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 8, 128>>;
// return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
// TypeConfig,
// QuantGroupSize,
// ck_tile::QuantType::BQuantGrouped>(arg_parser);
// };
// lut[hash_multiple_strings({"fp8",
// "bquant",
// "non-preshuffleb",
// "preshufflequant",
// "1x16x128"})] = [](const ck_tile::ArgParser& arg_parser) {
// using TypeConfig =
// decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t,
// float>{});
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 16, 128>>;
// return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
// TypeConfig,
// QuantGroupSize,
// ck_tile::QuantType::BQuantGrouped>(arg_parser);
// };
lut[hash_multiple_strings({"fp8",
"bquant",
"non-preshuffleb",
@@ -75,20 +77,19 @@ void bquant_quantgrouped_preshufflequant_instance_factory(
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
// lut[hash_multiple_strings({"fp8",
// "bquant",
// "non-preshuffleb",
// "preshufflequant",
// "1x128x128"})] = [](const ck_tile::ArgParser& arg_parser) {
// using TypeConfig =
// decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t,
// float>{});
// using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 128, 128>>;
// return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
// TypeConfig,
// QuantGroupSize,
// ck_tile::QuantType::BQuantGrouped>(arg_parser);
//};
lut[hash_multiple_strings({"fp8",
"bquant",
"non-preshuffleb",
"preshufflequant",
"1x128x128"})] = [](const ck_tile::ArgParser& arg_parser) {
using TypeConfig =
decltype(GemmQuantTypeConfig<ck_tile::fp8_t, ck_tile::fp8_t, ck_tile::half_t, float>{});
using QuantGroupSize = ck_tile::QuantGroupShape<ck_tile::sequence<1, 128, 128>>;
return run_gemm_example_prec_type<GemmConfig<ck_tile::fp8_t>,
TypeConfig,
QuantGroupSize,
ck_tile::QuantType::BQuantGrouped>(arg_parser);
};
// lut[hash_multiple_strings({"bf8", "bquant", "non-preshuffleb", "preshufflequant",
// "1x1x128"})] =
// [](const ck_tile::ArgParser& arg_parser) {

6
example/ck_tile/38_block_scale_gemm/gemm_quant.cpp
View File

@@ -89,8 +89,8 @@ auto gen_lut_key(const ck_tile::ArgParser& arg_parser)
// std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut);
// void aquant_quantgrouped_preshufflequant_instance_factory(
// std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut);
// void bquant_quantgrouped_fp8_instance_factory(
// std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut);
void bquant_quantgrouped_fp8_instance_factory(
std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut);
// void bquant_quantgrouped_bf8_instance_factory(
// std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>>& lut);
// void bquant_quantgrouped_fp8i4_instance_factory(
@@ -126,7 +126,7 @@ int main(int argc, char* argv[])
std::unordered_map<size_t, std::function<int(const ck_tile::ArgParser&)>> lut;
// aquant_quantgrouped_instance_factory(lut);
// aquant_quantgrouped_preshufflequant_instance_factory(lut);
// bquant_quantgrouped_fp8_instance_factory(lut);
bquant_quantgrouped_fp8_instance_factory(lut);
// bquant_quantgrouped_bf8_instance_factory(lut);
// bquant_quantgrouped_fp8i4_instance_factory(lut);
// bquant_quantgrouped_bf8i4_instance_factory(lut);