Skip to content

GitLab

"git@developer.sourcefind.cn:yangql/composable_kernel.git" did not exist on "70456328851dea8e7c6d908193e2d32c2d7c509c"
Unverified Commit 98e1e2d0 authored by rocking5566's avatar rocking5566 Committed by GitHub
Browse files

Refine kernel parameter of int8 (ScalarPerVector) (#155) 

* Change int8 ScalarPerVector

* Modify vector width of C
parent 0536f2b3
Hide whitespace changes
Inline Side-by-side
Showing with 22 additions and 22 deletions
example/01_gemm/gemm_xdl_int8.cpp
View file @ 98e1e2d0
...@@ -53,9 +53,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle ...@@ -53,9 +53,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle
256, // BlockSize 256, // BlockSize
256, // MPerBlock 256, // MPerBlock
128, // NPerBlock 128, // NPerBlock
32, // KPerBlock 64, // KPerBlock
8, // AK1 16, // AK1
8, // BK1 16, // BK1
32, // MPerXDL 32, // MPerXDL
32, // NPerXDL 32, // NPerXDL
4, // MXdlPerWave 4, // MXdlPerWave
...@@ -64,15 +64,15 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle ...@@ -64,15 +64,15 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim 2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector 16, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_K1 16, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM true, // ABlockLdsAddExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1 S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // BBlockTransferSrcAccessOrder S<1, 0, 2>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim 2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector 16, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_K1 16, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle 1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle 1, // CShuffleNXdlPerWavePerShuffle
......
example/14_gemm_xdl_requant_relu_requant/gemm_xdl_requant_relu_requant_int8.cpp
View file @ 98e1e2d0
...@@ -28,11 +28,11 @@ using Col = ck::tensor_layout::gemm::ColumnMajor; ...@@ -28,11 +28,11 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
using RequantReluRequant = ck::tensor_operation::element_wise::RequantReluRequant; using RequantReluRequant = ck::tensor_operation::element_wise::RequantReluRequant;
using ADataType = int8_t; using ADataType = int8_t;
using BDataType = int8_t; using BDataType = int8_t;
using CDataType = int8_t; using CDataType = int8_t;
using AccDataType = int32_t; using AccDataType = int32_t;
using ShuffleDataType = int32_t; using CShuffleDataType = int32_t;
using ALayout = ck::tensor_layout::gemm::RowMajor; using ALayout = ck::tensor_layout::gemm::RowMajor;
using BLayout = ck::tensor_layout::gemm::ColumnMajor; using BLayout = ck::tensor_layout::gemm::ColumnMajor;
...@@ -44,7 +44,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle ...@@ -44,7 +44,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle
BDataType, // BDataType BDataType, // BDataType
CDataType, // CDataType CDataType, // CDataType
AccDataType, // AccDataType AccDataType, // AccDataType
ShuffleDataType, // ShuffleDataType CShuffleDataType, // CShuffleDataType
ALayout, // ALayout ALayout, // ALayout
BLayout, // BLayout BLayout, // BLayout
CLayout, // CLayout CLayout, // CLayout
...@@ -54,9 +54,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle ...@@ -54,9 +54,9 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle
256, // BlockSize 256, // BlockSize
256, // MPerBlock 256, // MPerBlock
128, // NPerBlock 128, // NPerBlock
32, // KPerBlock 64, // KPerBlock
8, // AK1 16, // AK1
8, // BK1 16, // BK1
32, // MPerXDL 32, // MPerXDL
32, // NPerXDL 32, // NPerXDL
4, // MXdlPerWave 4, // MXdlPerWave
...@@ -65,20 +65,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle ...@@ -65,20 +65,20 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdl_C_Shuffle
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim 2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector 16, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_K1 16, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM true, // ABlockLdsAddExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1 S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // BBlockTransferSrcAccessOrder S<1, 0, 2>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim 2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector 16, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_K1 16, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle 1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle 1, // CShuffleNXdlPerWavePerShuffle
S<1, 1, 32, 1, 1, 8>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl S<1, 1, 64, 1, 1, 4>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
8>; // CBlockTransferScalarPerVector_NWaveNPerXdl 16>; // CBlockTransferScalarPerVector_NWaveNPerXdl
// clang-format on // clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
......
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