ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-17 09:08:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix in comments

Browse Source
This commit is contained in:
Qianfeng Zhang
2026-02-24 07:53:50 +00:00
parent a5c29b39b2
commit d20ed807e4
3 changed files with 8 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck_tile/ops/fmha/kernel/fmha_fwd_kernel.hpp
View File

@@ -87,7 +87,7 @@ struct has_naive_hdim_load_flag<
template <typename T>
static inline constexpr bool is_naive_hdim_load_v = has_naive_hdim_load_flag<T>::value;
// A helper struct for detechting kUseTrLoad
// A helper struct for detecting kUseTrLoad
template <typename T, typename = void>
struct has_use_trload_flag : std::false_type
{

2
include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs_whole_k_prefetch.hpp
View File

@@ -318,7 +318,7 @@ struct BlockFmhaPipelineQRKSVSWholeKPrefetch
Policy::template MakeBiasDramTileDistribution<Problem>());
// assuming no random values need be saved, this is true when the pipeline is called from
// xformers, since we have a separate kernel to generated randomm values
// xformers, since we have a separate kernel to generatd random values
auto null_randval_window = [&]() {
if constexpr(kHasDropout)
{

11
include/ck_tile/ops/fmha/pipeline/tile_fmha_shape.hpp
View File

@@ -52,9 +52,10 @@ struct TileFmhaShape
static constexpr index_t kM0 = BlockTile::at(number<0>{}); // tile size along q seqlen
static constexpr index_t kN0 = BlockTile::at(number<1>{}); // tile size along k seqlen
static constexpr index_t kK0 = BlockTile::at(number<2>{}); // tile size along qk gemm unroll
static constexpr index_t kN0Sub = BlockTile::at(number<2>{}); // tile size for dividing kN0
static constexpr index_t kN1 = BlockTile::at(number<3>{}); // tile size along v head_dim
static constexpr index_t kK1 = BlockTile::at(number<4>{}); // tile size along kv gemm unroll
static constexpr index_t kN0Sub = BlockTile::at(
number<2>{}); // same index as kK0; used as subdivision factor when dividing kN0
static constexpr index_t kN1 = BlockTile::at(number<3>{}); // tile size along v head_dim
static constexpr index_t kK1 = BlockTile::at(number<4>{}); // tile size along kv gemm unroll
static constexpr index_t kQKHeaddim =
BlockTile::at(number<5>{}); // total length of K0, used for pipeline that need load Q at
// once (or repeately load Q as a whole tile)
@@ -65,8 +66,8 @@ struct TileFmhaShape
// v, rowmajor : seqlen*hdim, colmajor : hdim*seqlen
static constexpr bool IsVLayoutRowMajor = IsVLayoutRowMajor_;
using VLayout = std::conditional_t<IsVLayoutRowMajor,
ck_tile::tensor_layout::gemm::RowMajor,
ck_tile::tensor_layout::gemm::ColumnMajor>;
ck_tile::tensor_layout::gemm::RowMajor,
ck_tile::tensor_layout::gemm::ColumnMajor>;
};
template <typename BlockTile_, // sequence<...