ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-01 04:07:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix clang format

Browse Source
This commit is contained in:
Clement Lin
2025-05-18 18:34:17 +08:00
parent d18cda141b
commit ebda5006f8
5 changed files with 13 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
example/ck_tile/tutorial/03_flash_attention_fwd/block_gemm_areg_bsmem_creg_v1_iteratek_policy.hpp
View File

@@ -26,7 +26,8 @@ struct BlockGemmARegBSmemCRegV1K8Policy
#if !defined(TOY_FA_FWD_QK_SWIZZLE)
return make_tuple(WarpGemmMfmaF16F16F32M32N32K16TransposedCDistribution{}, 4, 1);
#else
return make_tuple(WarpGemmMfmaF16F16F32M32N32K16SwizzleBTransposedCDistribution{}, 4, 1);
return make_tuple(
WarpGemmMfmaF16F16F32M32N32K16SwizzleBTransposedCDistribution{}, 4, 1);
#endif
}
else

6
example/ck_tile/tutorial/03_flash_attention_fwd/flash_attention_fwd_impl.hpp
View File

@@ -64,9 +64,9 @@ struct FlashAttentionFwdImpl
constexpr index_t kNPerBlock = kN1PerBlock;
constexpr index_t kKPerBlock = kK1PerBlock;
#if !defined(TOY_FA_FWD_QK_SWIZZLE)
constexpr index_t kKPack = 4;
constexpr index_t kKPack = 4;
#else
constexpr index_t kKPack = 8;
constexpr index_t kKPack = 8;
#endif
constexpr auto dataTypeSize = sizeof(VDataType);
@@ -210,7 +210,7 @@ struct FlashAttentionFwdImpl
{0, 0},
make_static_tile_distribution(gemm1.MakeBBlockDistributionEncode()));
#else
auto v_lds_window = make_tile_window(
auto v_lds_window = make_tile_window(
v_lds, make_tuple(number<kN1PerBlock>{}, number<kK1PerBlock>{}), {0, 0});
#endif

3
example/ck_tile/tutorial/04_codegen_flash_attention_fwd/block_gemm_areg_bsmem_creg_v1_iteratek_policy.hpp
View File

@@ -26,7 +26,8 @@ struct BlockGemmARegBSmemCRegV1K8Policy
#if !defined(TOY_FA_FWD_QK_SWIZZLE)
return make_tuple(WarpGemmMfmaF16F16F32M32N32K16TransposedCDistribution{}, 4, 1);
#else
return make_tuple(WarpGemmMfmaF16F16F32M32N32K16SwizzleBTransposedCDistribution{}, 4, 1);
return make_tuple(
WarpGemmMfmaF16F16F32M32N32K16SwizzleBTransposedCDistribution{}, 4, 1);
#endif
}
else

6
example/ck_tile/tutorial/04_codegen_flash_attention_fwd/flash_attention_fwd.hpp
View File

@@ -90,13 +90,13 @@ struct FlashAttentionFwd
const auto f = [](index_t dividend, index_t divisor) {
index_t quotient = dividend / divisor;
index_t modulus = dividend - quotient * divisor;
return make_tuple(quotient, modulus);
};
const auto [itmp, id_tile_n] = f(id_block, num_tile_n1);
const auto [id_tile_batch, id_tile_m] = f(itmp, num_tile_m0);
const index_t iBatch = __builtin_amdgcn_readfirstlane(id_tile_batch);
const index_t iM0 = __builtin_amdgcn_readfirstlane(id_tile_m * kM0PerBlock);
const index_t iN1 = __builtin_amdgcn_readfirstlane(id_tile_n * kN1PerBlock);

6
example/ck_tile/tutorial/04_codegen_flash_attention_fwd/flash_attention_fwd_impl.hpp
View File

@@ -64,9 +64,9 @@ struct FlashAttentionFwdImpl
constexpr index_t kNPerBlock = kN1PerBlock;
constexpr index_t kKPerBlock = kK1PerBlock;
#if !defined(TOY_FA_FWD_QK_SWIZZLE)
constexpr index_t kKPack = 4;
constexpr index_t kKPack = 4;
#else
constexpr index_t kKPack = 8;
constexpr index_t kKPack = 8;
#endif
constexpr auto dataTypeSize = sizeof(VDataType);
@@ -210,7 +210,7 @@ struct FlashAttentionFwdImpl
{0, 0},
make_static_tile_distribution(gemm1.MakeBBlockDistributionEncode()));
#else
auto v_lds_window = make_tile_window(
auto v_lds_window = make_tile_window(
v_lds, make_tuple(number<kN1PerBlock>{}, number<kK1PerBlock>{}), {0, 0});
#endif