ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-14 02:57:45 +00:00
Code Issues Packages Projects Releases Wiki Activity

refine

Browse Source
This commit is contained in:
Jin Zhou
2025-09-11 07:42:54 +00:00
parent 0754ddc285
commit e52448b80e
1 changed files with 7 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v3_multi_d_b_preshuffle.hpp
View File

@@ -569,6 +569,11 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3_b_preshuffle
using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N(0, 0, 0, 0, {}))>;
// This struct is used to encapsulate functions related to preload Ds:
// 1. set SrcSliceOrigin,
// 2. copy data to vgpr,
// 3. return origin when doing cshuffle,
// 4. generate vgpr buffer description.
template <index_t idx, typename DGridDesc, typename SfcCdeBlock>
struct DsBlockTransfer
{
@@ -597,8 +602,6 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3_b_preshuffle
}
}();
__device__ static constexpr auto GetBuffSize() { return d_buff_desc.GetElementSpaceSize(); }
__device__ constexpr DsBlockTransfer(const DGridDesc& d_grid_desc, const Index& src_origin)
: d_buffer_transfer_(d_grid_desc, src_origin)
{
@@ -1874,13 +1877,13 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3_b_preshuffle
make_multi_index(block_m_id, 0, block_n_id, 0));
auto d0_thread_buf =
make_static_buffer<AddressSpaceEnum::Vgpr, typename D0BlockTransfer::DDataType>(
d0_block_copy_to_vgpr.GetBuffSize());
d0_block_copy_to_vgpr.d_buff_desc.GetElementSpaceSize());
D1BlockTransfer d1_block_copy_to_vgpr(ds_grid_desc_mblock_mperblock_nblock_nperblock[I1],
make_multi_index(block_m_id, 0, block_n_id, 0));
auto d1_thread_buf =
make_static_buffer<AddressSpaceEnum::Vgpr, typename D1BlockTransfer::DDataType>(
d1_block_copy_to_vgpr.GetBuffSize());
d1_block_copy_to_vgpr.d_buff_desc.GetElementSpaceSize());
const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
(a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) /
@@ -2079,7 +2082,6 @@ struct GridwiseGemmMultiD_xdl_cshuffle_v3_b_preshuffle
static_for<0, num_access, 1>{}([&](auto access_id) {
// make sure it's safe to write to LDS
block_sync_lds();
// each thread write its data from VGPR to LDS
c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
sfc_c_vgpr.GetIndexTupleOfNumber(access_id),