ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix bugs

Browse Source
This commit is contained in:
aska-0096
2025-08-08 15:42:15 +00:00
parent 250dc13c75
commit 729e8785fb
2 changed files with 10 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/ck_tile/ops/elementwise/unary_element_wise_operation.hpp
View File

@@ -330,13 +330,6 @@ struct PassThrough
y = type_convert<float>(x);
}
template <>
CK_TILE_HOST_DEVICE void
operator()<ck_tile::bf16_t, ck_tile::fp16_t>(ck_tile::bf16_t& y, const ck_tile::fp16_t& x) const
{
y = type_convert<ck_tile::bf16_t>(x);
}
template <>
CK_TILE_HOST_DEVICE void operator()<float, ck_tile::fp16_t>(float& y,
const ck_tile::fp16_t& x) const

20
include/ck_tile/ops/fmha/pipeline/block_fmha_bwd_pipeline_default_policy.hpp
View File

@@ -1028,7 +1028,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<NIterPerWarp, NWarp>, sequence<KIterPerWarp>>,
tuple<sequence<0, 1>>,
tuple<sequence<0, 1>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto k_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1086,7 +1086,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<NIterPerWarp, NWarp>, sequence<KIterPerWarp>>,
tuple<sequence<0, 1>>,
tuple<sequence<0, 1>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto v_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1180,7 +1180,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<NIterPerWarp, NWarp>, sequence<KIterPerWarp>>,
tuple<sequence<0, 1>>,
tuple<sequence<0, 1>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto kt_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1239,7 +1239,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<MIterPerWarp, MWarp>, sequence<KIterPerWarp>>,
tuple<sequence<1, 0>>,
tuple<sequence<1, 0>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto q_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1334,7 +1334,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<NIterPerWarp, NWarp>, sequence<KIterPerWarp>>,
tuple<sequence<0, 1>>,
tuple<sequence<0, 1>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto qt_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1369,7 +1369,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<MIterPerWarp, MWarp>, sequence<KIterPerWarp>>,
tuple<sequence<1, 0>>,
tuple<sequence<1, 0>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto dst_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1480,7 +1480,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<MIterPerWarp, MWarp>, sequence<KIterPerWarp>>,
tuple<sequence<1, 0>>,
tuple<sequence<1, 0>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto do_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1579,7 +1579,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<NIterPerWarp, NWarp>, sequence<KIterPerWarp>>,
tuple<sequence<0, 1>>,
tuple<sequence<0, 1>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto dot_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1613,7 +1613,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<MIterPerWarp, MWarp>, sequence<KIterPerWarp>>,
tuple<sequence<1, 0>>,
tuple<sequence<1, 0>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto pt_block_dstr_encode = detail::make_embed_tile_distribution_encoding(
@@ -1657,7 +1657,7 @@ struct BlockFmhaBwdPipelineDefaultPolicy
tuple<sequence<MIterPerWarp, MWarp>, sequence<KIterPerWarp>>,
tuple<sequence<1, 0>>,
tuple<sequence<1, 0>>,
sequence<1, 2>,
sequence<2, 1>,
sequence<0, 0>>{};
constexpr auto ds_block_dstr_encode = detail::make_embed_tile_distribution_encoding(