ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 20:51:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

aggregate device macros in ck_tile config header (#1297)

Browse Source
This commit is contained in:
Illia Silin
2024-05-20 08:34:45 -07:00
committed by GitHub
parent 1274861a9d
commit 06b891c5c2
10 changed files with 56 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -36,8 +36,7 @@ struct WarpGemmAttributeMfmaImplF16F16F32M32N32K8
CK_TILE_DEVICE void
operator()(CVecType& c_vec, const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
defined(__gfx942__)
#if defined(__gfx9__)
c_vec = __builtin_amdgcn_mfma_f32_32x32x8f16(a_vec, b_vec, c_vec, 0, 0, 0);
#else
ck_tile::ignore = c_vec;
@@ -49,8 +48,7 @@ struct WarpGemmAttributeMfmaImplF16F16F32M32N32K8
// c_vec = a_vec * b_vec
CK_TILE_DEVICE CVecType operator()(const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
defined(__gfx942__)
#if defined(__gfx9__)
return bit_cast<CVecType>(
__builtin_amdgcn_mfma_f32_32x32x8f16(a_vec, b_vec, fp32x16_t{0.f}, 0, 0, 0));
#else
@@ -89,8 +87,7 @@ struct WarpGemmAttributeMfmaImplF16F16F32M16N16K16
CK_TILE_DEVICE void
operator()(CVecType& c_vec, const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
defined(__gfx942__)
#if defined(__gfx9__)
c_vec = __builtin_amdgcn_mfma_f32_16x16x16f16(a_vec, b_vec, c_vec, 0, 0, 0);
#else
ck_tile::ignore = c_vec;
@@ -102,8 +99,7 @@ struct WarpGemmAttributeMfmaImplF16F16F32M16N16K16
// c_vec = a_vec * b_vec
CK_TILE_DEVICE CVecType operator()(const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
defined(__gfx942__)
#if defined(__gfx9__)
return bit_cast<CVecType>(
__builtin_amdgcn_mfma_f32_16x16x16f16(a_vec, b_vec, fp32x4_t{0.f}, 0, 0, 0));
#else
@@ -143,7 +139,7 @@ struct WarpGemmAttributeMfmaImplBf16Bf16F32M32N32K8
CK_TILE_DEVICE void
operator()(CVecType& c_vec, const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx90a__) || defined(__gfx94__)
c_vec = __builtin_amdgcn_mfma_f32_32x32x8bf16_1k(a_vec, b_vec, c_vec, 0, 0, 0);
#elif defined(__gfx908__)
static_for<0, 2, 1>{}([&](auto k) {
@@ -167,7 +163,7 @@ struct WarpGemmAttributeMfmaImplBf16Bf16F32M32N32K8
// c_vec = a_vec * b_vec
CK_TILE_DEVICE CVecType operator()(const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx90a__) || defined(__gfx94__)
return bit_cast<CVecType>(
__builtin_amdgcn_mfma_f32_32x32x8bf16_1k(a_vec, b_vec, fp32x16_t{0.f}, 0, 0, 0));
#elif defined(__gfx908__)
@@ -220,7 +216,7 @@ struct WarpGemmAttributeMfmaImplBf16Bf16F32M16N16K16
CK_TILE_DEVICE void
operator()(CVecType& c_vec, const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx90a__) || defined(__gfx94__)
c_vec = __builtin_amdgcn_mfma_f32_16x16x16bf16_1k(a_vec, b_vec, c_vec, 0, 0, 0);
#elif defined(__gfx908__)
static_for<0, 2, 1>{}([&](auto k) {
@@ -244,7 +240,7 @@ struct WarpGemmAttributeMfmaImplBf16Bf16F32M16N16K16
// c_vec = a_vec * b_vec
CK_TILE_DEVICE CVecType operator()(const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx90a__) || defined(__gfx94__)
return bit_cast<CVecType>(
__builtin_amdgcn_mfma_f32_16x16x16bf16_1k(a_vec, b_vec, fp32x4_t{0.f}, 0, 0, 0));
#elif defined(__gfx908__)
@@ -299,7 +295,7 @@ struct WarpGemmAttributeMfmaImpl_f32_32x32x16_f8_base
CK_TILE_DEVICE void
operator()(CVecType& c_vec, const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx94__)
if constexpr(std::is_same_v<ADataType, fp8_t> && std::is_same_v<BDataType, fp8_t>)
c_vec = __builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8(
bit_cast<long>(a_vec), bit_cast<long>(b_vec), c_vec, 0, 0, 0);
@@ -333,7 +329,7 @@ struct WarpGemmAttributeMfmaImpl_f32_32x32x16_f8_base
// c_vec = a_vec * b_vec
CK_TILE_DEVICE CVecType operator()(const AVecType& a_vec, const BVecType& b_vec) const
{
#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
#if defined(__gfx94__)
if constexpr(std::is_same_v<ADataType, fp8_t> && std::is_same_v<BDataType, fp8_t>)
return bit_cast<CVecType>(__builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8(
bit_cast<long>(a_vec), bit_cast<long>(b_vec), CVecType{0.f}, 0, 0, 0));