Skip to content

GitLab

Commit 9ae3308a authored by Adam Osewski's avatar Adam Osewski
Browse files

Fix BlockwiseGemm. 

Use default A/B MmaTileKStride equal to KPerThread.
parent 0a65fc55
Show whitespace changes
Inline Side-by-side
Showing with 42 additions and 38 deletions
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_direct_c_write_out.hpp
View file @ 9ae3308a
......@@ -266,9 +266,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_v1
MXdlPerWave,
NXdlPerWave,
KPack,
true, // TransposeC
KPack, // A MMaTileKStride
KPack>; // B MMaTileKStride
true>; // TransposeC
// KPack // A MMaTileKStride
// KPack>; // B MMaTileKStride
return BlockwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_N2_N3_N4(c_grid_desc_m_n);
}
......@@ -425,9 +425,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_v1
MXdlPerWave,
NXdlPerWave,
KPack,
true, // TransposeC
KPack, // A MMaTileKStride
KPack>{}; // B MMaTileKStride
true>{}; // TransposeC
// KPack, // A MMaTileKStride
// KPack>{}; // B MMaTileKStride
auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
......@@ -517,8 +517,12 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_v1
n_thread_data_on_grid_to_n0_n1_n2_n3_n4_adaptor.CalculateBottomIndex(
make_multi_index(n_thread_data_on_grid));
if(threadIdx.x == 0 || threadIdx.x == 15 || threadIdx.x == 33 || threadIdx.x == 60)
{
// if(threadIdx.x == 0
// // || threadIdx.x == 15
// // || threadIdx.x == 33
// // || threadIdx.x == 60
// )
// {
auto c_thread_copy = ThreadwiseTensorSliceTransfer_v1r3<
FloatGemmAcc,
FloatC,
......@@ -547,7 +551,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_v1
c_thread_buf,
c_grid_desc_m0_n0_m1_n1_m2_n2_n3_n4,
c_grid_buf);
}
// }
// TODO: how SpaceFillingCurve works ?
// space filling curve for threadwise C in VGPR
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment