Skip to content

GitLab

Commit cf226747 authored by Chao Liu's avatar Chao Liu
Browse files

tweak

parent 20718690
Hide whitespace changes
Inline Side-by-side
Showing with 17 additions and 1 deletion
example/01_gemm/gemm_xdl_fp16.cpp
View file @ cf226747
...@@ -64,6 +64,16 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle ...@@ -64,6 +64,16 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
//######| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //######| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //######| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// all thread
< Row, Col, Row, F16, F16, F16, F32, F32, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 0, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>;
// < Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 0, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8>;
#elif 1
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle_v2
//######| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle| A| B| C| GEMM| NumGemmK| ABBlockTransfer| BlockGemm| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer|
//######| | | | Type| Type| Type| Type| DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| ThreadGroupSize| ThreadGroupSize| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
// producer & consumer
< Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 8>, 8>; < Row, Col, Row, F16, F16, F16, F32, F16, PassThrough, PassThrough, PassThrough, GemmDefault, 1, 256, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 8>, 8>;
// < Row, Col, Row, F16, F16, F16, F32, F16, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 8>, 8>; // < Row, Col, Row, F16, F16, F16, F32, F16, AElementOp, BElementOp, CElementOp, GemmDefault, 1, 256, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 64, 1, 8>, 8>;
#elif 1 #elif 1
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v2.hpp
View file @ cf226747
...@@ -117,6 +117,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2 ...@@ -117,6 +117,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2
using ThisThreadBlock = using ThisThreadBlock =
AnyThreadBlock<ABBlockTransferThreadGroupSize + BlockGemmThreadGroupSize>; AnyThreadBlock<ABBlockTransferThreadGroupSize + BlockGemmThreadGroupSize>;
#if 0
struct ABBlockTransferThreadGroup struct ABBlockTransferThreadGroup
{ {
__device__ static constexpr index_t GetNumOfThread() __device__ static constexpr index_t GetNumOfThread()
...@@ -151,6 +152,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2 ...@@ -151,6 +152,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2
}; };
using CShuffleBlockTransferThreadGroup = ThisThreadBlock; using CShuffleBlockTransferThreadGroup = ThisThreadBlock;
#else
using ABBlockTransferThreadGroup = ThisThreadBlock;
using BlockGemmThreadGroup = ThisThreadBlock;
using CShuffleBlockTransferThreadGroup = ThisThreadBlock;
#endif
__host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1() __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
{ {
...@@ -487,7 +493,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2 ...@@ -487,7 +493,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v2
(a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) / (a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2)) /
KPerBlock); KPerBlock);
#if 0 #if 1
// gridwise GEMM pipeline // gridwise GEMM pipeline
const auto gridwise_gemm_pipeline = const auto gridwise_gemm_pipeline =
GridwiseGemmPipeline_v1<remove_cvref_t<decltype(a_grid_desc_ak0_m_ak1)>, GridwiseGemmPipeline_v1<remove_cvref_t<decltype(a_grid_desc_ak0_m_ak1)>,
......
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