Skip to content

GitLab

Commit a422df82 authored by Jing Zhang's avatar Jing Zhang
Browse files

clean tuning paramters

parent f39f7d79
Hide whitespace changes
Inline Side-by-side
Showing with 18 additions and 33 deletions
example/01_gemm/gemm_dl_fp16.cpp
View file @ a422df82
...@@ -23,15 +23,12 @@ static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpeciali ...@@ -23,15 +23,12 @@ static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpeciali
// clang-format off // clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDl
// ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer| // ######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| K0Per| K1| M1Per| N1Per| KPer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| BBlockTransfer| CThreadTransfer| BThreadTransfer| CThreadTransfer| CThreadTransfer| CThreadTransfer|
// ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector| // ######| Type| Type| Type| Type| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | ThreadM111| ThreadN111| Thread| ThreadSliceLengths| ThreadClusterLengths| ThreadCluster| SrcAccess| SrcVectorTensor| SrcVectorTensor| DstVectorTensor| SrcAccess| SrcVectorDim| SrcScalarPerVector| SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
// ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| K0_N0_N1_K1| K0_N0_N1_K1| ArrangeOrder| Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1| Order| | | // ######| | | | | | | | Operation| Operation| Operation| | | | | | | | | | K0_M0_M1_K1| K0_M0_M1_K1| ArrangeOrder| Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1| Order| | | Order| | |
// ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | // ######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
//< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 64, 64, 16, 2, 8, 8, 1, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>; < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 4, 64, 16, 2, 1, 4, 1, S<1, 1, 1, 2>, S<16, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, 3, 2, S<0, 1, 2, 3, 4, 5>, 5, 4>;
//< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 64, 64, 16, 2, 4, 4, 1, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>; //< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 1, 64, 32, 2, 1, 1, 1, S<1, 1, 1, 2>, S<32, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, 3, 2, S<0, 1, 2, 3, 4, 5>, 5, 1>;
//< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 16, 64, 16, 2, 2, 8, 1, S<4, 1, 1, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>;
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 4, 64, 16, 2, 1, 4, 1, S<1, 1, 1, 2>, S<16, 1, 4, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 4, 2>, S<4, 1, 16, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 4>;
//< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmMNPadding, 64, 1, 64, 32, 2, 1, 1, 1, S<1, 1, 1, 2>, S<32, 1, 1, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<4, 1, 8, 2>, S<8, 1, 8, 1>, S<1, 2, 0, 3>, S<1, 2, 0, 3>, S<4, 1, 1, 2>, S<1, 2, 0, 3>, S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>, 5, 1>;
// clang-format on // clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
......
include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
View file @ a422df82
...@@ -47,13 +47,9 @@ template < ...@@ -47,13 +47,9 @@ template <
typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1, typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
typename ABlockTransferSrcVectorTensorContiguousDimOrder, typename ABlockTransferSrcVectorTensorContiguousDimOrder,
typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1, typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1, typename BThreadTransferSrcDstAccessOrder,
typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1, index_t BThreadTransferSrcVectorDim,
typename BBlockTransferThreadClusterArrangeOrder, index_t BThreadTransferSrcScalarPerVector,
typename BBlockTransferSrcAccessOrder,
typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
typename BBlockTransferSrcVectorTensorContiguousDimOrder,
typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
typename CThreadTransferSrcDstAccessOrder, typename CThreadTransferSrcDstAccessOrder,
index_t CThreadTransferSrcDstVectorDim, index_t CThreadTransferSrcDstVectorDim,
index_t CThreadTransferDstScalarPerVector, index_t CThreadTransferDstScalarPerVector,
...@@ -105,13 +101,9 @@ struct DeviceGemmDl : public DeviceGemm<ALayout, ...@@ -105,13 +101,9 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1, ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
ABlockTransferSrcVectorTensorContiguousDimOrder, ABlockTransferSrcVectorTensorContiguousDimOrder,
ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1, ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
BBlockTransferThreadSliceLengths_K0_N0_N1_K1, BThreadTransferSrcDstAccessOrder,
BBlockTransferThreadClusterLengths_K0_N0_N1_K1, BThreadTransferSrcVectorDim,
BBlockTransferThreadClusterArrangeOrder, BThreadTransferSrcScalarPerVector,
BBlockTransferSrcAccessOrder,
BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
BBlockTransferSrcVectorTensorContiguousDimOrder,
BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
CThreadTransferSrcDstAccessOrder, CThreadTransferSrcDstAccessOrder,
CThreadTransferSrcDstVectorDim, CThreadTransferSrcDstVectorDim,
CThreadTransferDstScalarPerVector>; CThreadTransferDstScalarPerVector>;
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
View file @ a422df82
...@@ -90,13 +90,9 @@ template <index_t BlockSize, ...@@ -90,13 +90,9 @@ template <index_t BlockSize,
typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1, typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
typename ABlockTransferSrcVectorTensorContiguousDimOrder, typename ABlockTransferSrcVectorTensorContiguousDimOrder,
typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1, typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1, typename BThreadTransferSrcDstAccessOrder,
typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1, index_t BThreadTransferSrcVectorDim,
typename BBlockTransferThreadClusterArrangeOrder, index_t BThreadTransferSrcScalarPerVector,
typename BBlockTransferSrcAccessOrder,
typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
typename BBlockTransferSrcVectorTensorContiguousDimOrder,
typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
typename CThreadTransferSrcDstAccessOrder, typename CThreadTransferSrcDstAccessOrder,
index_t CThreadTransferSrcDstVectorDim, index_t CThreadTransferSrcDstVectorDim,
index_t CThreadTransferDstScalarPerVector> index_t CThreadTransferDstScalarPerVector>
...@@ -460,9 +456,9 @@ struct GridwiseGemmDl_km_kn_mn_v1r3 ...@@ -460,9 +456,9 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
remove_reference_t<decltype(b_grid_desc_k0_n0_n1_k1)>, remove_reference_t<decltype(b_grid_desc_k0_n0_n1_k1)>,
decltype(b_thread_desc_k0_n0_n1_k1), decltype(b_thread_desc_k0_n0_n1_k1),
Sequence<K0PerBlock, 1, NPerThread, K1.value>, Sequence<K0PerBlock, 1, NPerThread, K1.value>,
Sequence<0, 1, 2, 3>, // BBlockTransferSrcAccessOrder, BThreadTransferSrcDstAccessOrder,
3, BThreadTransferSrcVectorDim,
K1, BThreadTransferSrcScalarPerVector,
1, 1,
false, false,
true>( true>(
......
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