ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-24 06:44:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '4a49dac7c6fff9ffe4d275bed761a79e51188f3c' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-08-28 13:19:19 +00:00
parent b36f983558
commit c8234bd387
6 changed files with 30 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
example/ck_tile/39_copy/copy_basic.hpp
View File

@@ -27,8 +27,9 @@ struct TileCopyShape
static constexpr index_t ThreadTile_N = ThreadTile::at(number<1>{});
// Wave tile dimensions
static constexpr index_t Wave_Tile_M = WaveTile::at(number<0>{});
static constexpr index_t WaveSize = get_warp_size();
static constexpr index_t Wave_Tile_N = WaveTile::at(number<1>{});
static constexpr index_t Wave_Tile_M = ThreadTile_M * ThreadTile_N * WaveSize / Wave_Tile_N;
// Block tile dimensions
static constexpr index_t Block_Tile_M = BlockTile::at(number<0>{});
@@ -45,7 +46,6 @@ struct TileCopyShape
Block_Tile_N / (Waves_Per_Block_N * Wave_Tile_N);
// Hardware configuration
static constexpr index_t WaveSize = get_warp_size();
static constexpr index_t BlockSize = Waves_Per_Block_M * Waves_Per_Block_N * WaveSize;
// Configuration validation
@@ -60,8 +60,10 @@ struct TileCopyShape
"Invalid wave configuration for N dimension");
// Ensure wave tile dimensions align with wave size
#if defined(__HIP_DEVICE_COMPILE__)
static_assert(Wave_Tile_M / ThreadTile_M * Wave_Tile_N / ThreadTile_N == WaveSize,
"(Wave_Tile_M/ThreadTile_M) * (Wave_Tile_N/ThreadTile_N) != WaveSize");
#endif
};
/**
@@ -200,6 +202,19 @@ struct ElementWiseTileCopyKernel
using XDataType = typename Problem::XDataType;
using Policy = ck_tile::remove_cvref_t<Policy_>;
static constexpr index_t kBlockSize = Problem::BlockShape::BlockSize;
CK_TILE_HOST static auto BlockSize()
{
if(ck_tile::is_wave32())
{
return kBlockSize / 2;
}
else
{
return kBlockSize;
}
}
CK_TILE_DEVICE void operator()(const XDataType* p_x, XDataType* p_y, index_t M, index_t N) const
{
using S = typename Problem::BlockShape;