frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Perf] Tunings for SM100 FP8 CUTLASS kernel (#18778)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-06-04 13:46:28 -04:00
committed by GitHub
parent d459fae0a2
commit 53a5a0ce30
1 changed files with 51 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm100_fp8_dispatch.cuh
View File

@ -15,6 +15,7 @@ using c3x::cutlass_gemm_caller;
template <typename InType, typename OutType,
template <typename, typename, typename> typename Epilogue>
struct sm100_fp8_config_default {
// M in (128, inf)
static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
@ -25,6 +26,34 @@ struct sm100_fp8_config_default {
KernelSchedule, EpilogueSchedule>;
};
template <typename InType, typename OutType,
template <typename, typename, typename> typename Epilogue>
struct sm100_fp8_config_M128 {
// M in (64, 128]
static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
using TileShape = Shape<_128, _128, _64>;
using ClusterShape = Shape<_2, _2, _1>;
using Cutlass3xGemm =
cutlass_3x_gemm_sm100<InType, OutType, Epilogue, TileShape, ClusterShape,
KernelSchedule, EpilogueSchedule>;
};
template <typename InType, typename OutType,
template <typename, typename, typename> typename Epilogue>
struct sm100_fp8_config_M64 {
// M in [1, 64]
static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
using KernelSchedule = cutlass::gemm::collective::KernelScheduleAuto;
using EpilogueSchedule = cutlass::epilogue::collective::EpilogueScheduleAuto;
using TileShape = Shape<_64, _64, _256>;
using ClusterShape = Shape<_1, _8, _1>;
using Cutlass3xGemm =
cutlass_3x_gemm_sm100<InType, OutType, Epilogue, TileShape, ClusterShape,
KernelSchedule, EpilogueSchedule>;
};
template <typename InType, typename OutType,
template <typename, typename, typename> typename Epilogue,
typename... EpilogueArgs>
@ -39,8 +68,28 @@ inline void cutlass_gemm_sm100_fp8_dispatch(torch::Tensor& out,
using Cutlass3xGemmDefault =
typename sm100_fp8_config_default<InType, OutType,
Epilogue>::Cutlass3xGemm;
return cutlass_gemm_caller<Cutlass3xGemmDefault>(
out, a, b, std::forward<EpilogueArgs>(args)...);
using Cutlass3xGemmM64 =
typename sm100_fp8_config_M64<InType, OutType, Epilogue>::Cutlass3xGemm;
using Cutlass3xGemmM128 =
typename sm100_fp8_config_M128<InType, OutType, Epilogue>::Cutlass3xGemm;
uint32_t const m = a.size(0);
uint32_t const mp2 =
std::max(static_cast<uint32_t>(64), next_pow_2(m)); // next power of 2
if (mp2 <= 64) {
// m in [1, 64]
return cutlass_gemm_caller<Cutlass3xGemmM64>(
out, a, b, std::forward<EpilogueArgs>(args)...);
} else if (mp2 <= 128) {
// m in (64, 128]
return cutlass_gemm_caller<Cutlass3xGemmM128>(
out, a, b, std::forward<EpilogueArgs>(args)...);
} else {
// m in (128, inf)
return cutlass_gemm_caller<Cutlass3xGemmDefault>(
out, a, b, std::forward<EpilogueArgs>(args)...);
}
}
template <template <typename, typename, typename> typename Epilogue,