ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-18 06:00:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

terminology clean-up (#1792)

Browse Source
This commit is contained in:
Illia Silin
2025-01-03 16:38:22 -08:00
committed by GitHub
parent 6df5fe2ad8
commit 8ea375bb58
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
View File

@@ -1544,7 +1544,7 @@ struct ThreadwiseTensorSliceTransfer_StaticToStatic
ElementwiseOperation element_op_;
};
// Specilized for WMMA-Navi3
// Specialized for gfx11
// A single Wave32 is composed by double row
// Data exchange allowed between these two rows
// This RowLane Dst buf will be filled from two Src buf
@@ -1679,7 +1679,7 @@ struct ThreadwiseTensorSliceTransfer_StaticToStatic_InterRow
ElementwiseOperation element_op_{};
};
// Specilized for WMMA-Navi4
// Specialized for gfx12
template <typename SrcData,
typename DstData,
typename SrcDesc,

2
include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
View File

@@ -307,7 +307,7 @@ struct wmma_type<WmmaInstr::wmma_f32_16x16x16_f16_gfx12,
// Wave mode dependent propety
static constexpr index_t wave_size = Number<WaveSize>{};
// * Fixed in Navi3x, Will be wave mode dependent on Navi4x
// * Fixed for gfx11, Will be wave mode dependent on gfx12
// static constexpr index_t num_src_a_vgprs_per_wave = k_per_wmma / 2 * src_a_data_size / 4;
// static constexpr index_t num_src_b_vgprs_per_wave = k_per_wmma / 2 * src_b_data_size / 4;
// * num_acc_vgprs_per_wave alone M direction