Skip to content

Configure Optimal SplitK Kernel (Block Size 256) #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
gryph6 wants to merge 2 commits into
base: develop
Choose a base branch
from
Open

Configure Optimal SplitK Kernel (Block Size 256) #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fd58e5d
modify block size to 64
May 28, 2024
4634207
optimal block 256 build
May 28, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions example/35_splitK_gemm/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
add_compile_options(--save-temps)

add_custom_target(example_splitK_gemm_xdl)
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
Expand Down
6 changes: 4 additions & 2 deletions example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,15 +42,17 @@ using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;

static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::KPadding;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto Pipeline = ck::PipelineVersion::v2;
static constexpr auto NumGemmKPrefetchStage = 1;

using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
// clang-format off
//######| AData| BData| CData| AccData| ALayout| BLayout| CLayout| A| B| C| GEMM| Block| MPer| NPer| KPer| K1| 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| | | | Elementwise| Elementwise| Elementwise| Spacialization| Size| Block| Block| Block| | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
//######| | | | | | | | Operation| Operation| Operation| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl| _NWaveNPerXdl|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 8, 8, true, 1, 1, S<1, 32, 1, 8>, 8>;
< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementOp, BElementOp, CElementOp, GemmDefault, 256, 256, 128, 4, 8, 32, 32, 4, 2, S<1, 4, 64, 1>, S<0, 2, 1, 3>, S<0, 2, 1, 3>, 3, 8, 8, true, S<1, 4, 64, 1>, S<0, 1, 3, 2>, S<0, 1, 3, 2>, 3, 2, 8, true, 1, 1, S<1, 32, 1, 8>, 8, CDataType, Pipeline, NumGemmKPrefetchStage>;
// clang-format on

#include "run_splitK_gemm_example.inc"
Expand Down
14 changes: 6 additions & 8 deletions include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,11 +58,12 @@ template <typename ADataType,
index_t CShuffleNRepeatPerShuffle,
typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
typename ComputeType = CDataType,
PipelineVersion PipelineVer = PipelineVersion::v1,
LoopScheduler LoopSched = make_default_loop_scheduler(),
typename LDSTypeA = ComputeType,
typename LDSTypeB = ComputeType>
typename ComputeType = CDataType,
PipelineVersion PipelineVer = PipelineVersion::v1,
index_t NumGemmKPrefetchStage = 1,
LoopScheduler LoopSched = make_default_loop_scheduler(),
typename LDSTypeA = ComputeType,
typename LDSTypeB = ComputeType>

struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
BLayout,
Expand All @@ -80,9 +81,6 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
static constexpr auto I2 = Number<2>{};
static constexpr auto I3 = Number<3>{};

// TODO: should be exposed as Tparams.
static constexpr index_t NumGemmKPrefetchStage = 1;

using ComputeTypeA = ComputeType;
using ComputeTypeB = ComputeType;

Expand Down
4 changes: 2 additions & 2 deletions script/cmake-ck-dev.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,12 +8,12 @@ MY_PROJECT_SOURCE=$1
if [ $# -ge 2 ] ; then
GPU_TARGETS=$2
else
GPU_TARGETS="gfx908;gfx90a;gfx940"
GPU_TARGETS="gfx908"
fi

cmake \
-D CMAKE_PREFIX_PATH=/opt/rocm \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_CXX_COMPILER=/opt/rocm-5.7.1/bin/hipcc \
-D CMAKE_CXX_FLAGS="-Xclang -mllvm -Xclang -enable-post-misched=0 -std=c++17 -O3 -ftemplate-backtrace-limit=0 -fPIE -Wno-gnu-line-marker" \
-D CMAKE_BUILD_TYPE=Release \
-D BUILD_DEV=ON \
Expand Down