ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-14 02:02:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK TILE GEMM] Fix building issues (#2772)

Browse Source 
- Add `WarpGemmMfma_f32_16x16x128_[fp8|bf8]_[fp8|bf8]_CTransposed`
- Replace `__gfx950__` with `CK_GFX950_SUPPORT`

[ROCm/composable_kernel commit: e1ab460d2d]
This commit is contained in:
Cong Ma
2025-09-02 23:40:18 -06:00
committed by GitHub
parent 84d188b137
commit 1ed315638e
4 changed files with 31 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/ck_tile/38_block_scale_gemm/gemm_utils.hpp
View File

@@ -18,7 +18,7 @@
template <typename PrecType, ck_tile::index_t M_Warp_Tile>
constexpr ck_tile::index_t get_k_warp_tile()
{
#if defined(__gfx950__)
#if defined(CK_GFX950_SUPPORT)
constexpr bool is_8bit_float =
std::is_same_v<PrecType, ck_tile::fp8_t> || std::is_same_v<PrecType, ck_tile::bf8_t>;
if constexpr(M_Warp_Tile == 32)
@@ -35,7 +35,7 @@ constexpr ck_tile::index_t get_k_warp_tile()
template <typename PrecType, ck_tile::index_t M_Warp_Tile>
constexpr ck_tile::index_t get_k_from_preshuffled_warp_tile()
{
#if defined(__gfx950__)
#if defined(CK_GFX950_SUPPORT)
if constexpr(M_Warp_Tile == 32)
return sizeof(PrecType) == 2 ? 16 : 64;
else

24
include/ck_tile/ops/gemm/warp/warp_gemm.hpp
View File

@@ -301,6 +301,30 @@ using WarpGemmMfma_f32_16x16x128_bf8_bf8 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_bf8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_fp8_CTransposed =
WarpGemmImpl<WarpGemmAttributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_fp8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_fp8_bf8_CTransposed =
WarpGemmImpl<WarpGemmAttributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImpl_f32_16x16x128_fp8_bf8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_bf8_fp8_CTransposed =
WarpGemmImpl<WarpGemmAttributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_fp8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_16x16x128_bf8_bf8_CTransposed =
WarpGemmImpl<WarpGemmAttributeMfmaTransposedCDistribution<
WarpGemmAttributeMfmaImpl_f32_16x16x128_bf8_bf8<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfma_f32_32x32x64_fp8_fp8 = WarpGemmImpl<
WarpGemmAttributeMfma<WarpGemmAttributeMfmaImpl_f32_32x32x64_fp8_fp8<WGAttrCtlEnum::Default_>,

4
include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp
View File

@@ -112,6 +112,10 @@ template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16,
template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_fp8_bf8<>; };
template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8<>; };
template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 128, false> { using Type = WarpGemmMfma_f32_16x16x128_bf8_bf8<>; };
template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 16, 16, 128, true> { using Type = WarpGemmMfma_f32_16x16x128_fp8_fp8_CTransposed<>; };
template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 16, 16, 128, true> { using Type = WarpGemmMfma_f32_16x16x128_fp8_bf8_CTransposed<>; };
template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::fp8_t, float, 16, 16, 128, true> { using Type = WarpGemmMfma_f32_16x16x128_bf8_fp8_CTransposed<>; };
template<> struct WarpGemmDispatcher<ck_tile::bf8_t, ck_tile::bf8_t, float, 16, 16, 128, true> { using Type = WarpGemmMfma_f32_16x16x128_bf8_bf8_CTransposed<>; };
template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::fp8_t, float, 32, 32, 64, false> { using Type = WarpGemmMfma_f32_32x32x64_fp8_fp8<>; };
template<> struct WarpGemmDispatcher<ck_tile::fp8_t, ck_tile::bf8_t, float, 32, 32, 64, false> { using Type = WarpGemmMfma_f32_32x32x64_fp8_bf8<>; };

2
test/ck_tile/gemm_block_scale/test_gemm_aquant_utils.hpp
View File

@@ -18,7 +18,7 @@
template <typename PrecType, ck_tile::index_t M_Warp_Tile>
constexpr ck_tile::index_t get_k_warp_tile()
{
#if defined(__gfx950__)
#if defined(CK_GFX950_SUPPORT)
constexpr bool is_8bit_float =
std::is_same_v<PrecType, ck_tile::fp8_t> || std::is_same_v<PrecType, ck_tile::bf8_t>;
if constexpr(M_Warp_Tile == 32)