Skip to content

GitLab

Unverified Commit 500fa995 authored by Chao Liu's avatar Chao Liu Committed by GitHub
Browse files

Clean up conv example, Instances, profiler and test (#324) 

* convnd_fwd fp16 example

* update example

* update example

* update instance

* updating refernce conv

* update reference conv

* update conv fwd profiler

* update conv 1d and 3d instance

* update include path

* clean

* update profiler for conv bwd data and weight

* update conv bwd weight

* clean

* update conv example

* update profiler for conv bwd weight

* update ckprofiler for conv bwd data

* fix reference conv bwd data bug; update conv bwd data test

* update examples

* fix initialization issue

* update test for conv fwd

* clean

* clean

* remove test case too sensitive to error threshhold

* fix test

* clean

* fix build

* adding conv multiple d

* adding conv multiple D

* add matrix padder

* add gemm padding to convnd

* adding group conv

* update gemm multi-d

* refactor

* refactor

* refactor

* clean

* clean

* refactor

* refactor

* reorg

* add ds

* add bias

* clean

* add G

* adding group

* adding group

* adding group

* update Tensor

* clean

* update example

* update DeviceGemmMultipleD_Xdl_CShuffle

* update conv bwd-data and bwd-weight

* upate contraction example

* update gemm and batch gemm with e permute

* fix example build

* instance for grouped conv1d

* update example

* adding group conv instance

* update gemm bilinear instance

* update gemm+add+add+fastgelu instance

* update profiler

* update profiler

* update test

* update test and client example

* clean

* add grouped conv into profiler

* update profiler

* clean

* add test grouped conv, update all conv test to gtest

* update test
parent 85978e02
Hide whitespace changes
Inline Side-by-side
Showing with 745 additions and 152 deletions
example/23_softmax/softmax_blockwise.cpp
View file @ 500fa995
...@@ -13,8 +13,8 @@ ...@@ -13,8 +13,8 @@
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp" #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_common_util.hpp" #include "ck/library/utility/host_common_util.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_softmax.hpp"
using namespace ck; using namespace ck;
...@@ -177,7 +177,7 @@ int main(int argc, char* argv[]) ...@@ -177,7 +177,7 @@ int main(int argc, char* argv[])
} }
if(beta != 0.0f) if(beta != 0.0f)
for(size_t i = 0; i < out_ref.mDesc.GetElementSpace(); i++) for(size_t i = 0; i < out_ref.mDesc.GetElementSpaceSize(); i++)
out.mData[i] = out_ref.mData[i]; out.mData[i] = out_ref.mData[i];
}; };
// std::cout << "beta = " << beta << std::endl; // std::cout << "beta = " << beta << std::endl;
...@@ -185,8 +185,8 @@ int main(int argc, char* argv[]) ...@@ -185,8 +185,8 @@ int main(int argc, char* argv[])
// LogRangeAsType<float>(std::cout << "tensor prior out: " , out.mData, ",") << std::endl; // LogRangeAsType<float>(std::cout << "tensor prior out: " , out.mData, ",") << std::endl;
// these buffers are usually provided by the user application // these buffers are usually provided by the user application
DeviceMem in_dev(sizeof(InDataType) * in.mDesc.GetElementSpace()); DeviceMem in_dev(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
DeviceMem out_dev(sizeof(OutDataType) * out.mDesc.GetElementSpace()); DeviceMem out_dev(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
in_dev.ToDevice(in.mData.data()); in_dev.ToDevice(in.mData.data());
......
example/24_batched_gemm_c_permute/CMakeLists.txt deleted 100644 → 0
View file @ 85978e02
add_example_executable(example_batched_gemm_c_permute_xdl_fp16 batched_gemm_c_permute_xdl_fp16.cpp)
example/24_batched_gemm_e_permute/CMakeLists.txt 0 → 100644
View file @ 500fa995
add_example_executable(example_batched_gemm_e_permute_xdl_fp16 batched_gemm_e_permute_xdl_fp16.cpp)
example/24_batched_gemm_c_permute/batched_gemm_c_permute_xdl_fp16.cpp → example/24_batched_gemm_e_permute/batched_gemm_e_permute_xdl_fp16.cpp
View file @ 500fa995
...@@ -6,13 +6,13 @@ ...@@ -6,13 +6,13 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_batched_gemm_c_permute_xdl.hpp" #include "ck/tensor_operation/gpu/device/device_batched_gemm_e_permute_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -30,7 +30,6 @@ using ADataType = F16; ...@@ -30,7 +30,6 @@ using ADataType = F16;
using BDataType = F16; using BDataType = F16;
using AccDataType = F32; using AccDataType = F32;
using CShuffleDataType = F16; using CShuffleDataType = F16;
using DsDataType = ck::Tuple<>;
using EDataType = F16; using EDataType = F16;
using ALayout = Row; using ALayout = Row;
...@@ -42,16 +41,14 @@ using BElementOp = PassThrough; ...@@ -42,16 +41,14 @@ using BElementOp = PassThrough;
using CDEElementOp = PassThrough; using CDEElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// static constexpr auto MNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
// static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmEPermuteXdl
// clang-format off
// clang-format off //######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| EData| A| B| CDE| GEMM| NumGemmK| Block| 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|
using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmCPermuteXdl //######| | | | Type| Type| Type| DataType| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector|
//######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| //######| | | | | | | | | 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|
//######| | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_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| < ALayout, BLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>;
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>;
// clang-format on // clang-format on
using ReferenceBatchedGemmInstance = ck::tensor_operation::host:: using ReferenceBatchedGemmInstance = ck::tensor_operation::host::
...@@ -99,7 +96,7 @@ int main(int argc, char* argv[]) ...@@ -99,7 +96,7 @@ int main(int argc, char* argv[])
} }
// GEMM shape // GEMM shape
ck::tensor_operation::device::BatchedGemmCPermuteDesc batched_gemm_c_permute_desc{ ck::tensor_operation::device::BatchedGemmEPermuteDesc batched_gemm_e_permute_desc{
G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N}; G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N};
auto f_host_tensor_descriptor = [](std::size_t batch_count_, auto f_host_tensor_descriptor = [](std::size_t batch_count_,
...@@ -125,7 +122,7 @@ int main(int argc, char* argv[]) ...@@ -125,7 +122,7 @@ int main(int argc, char* argv[])
Tensor<BDataType> b_g_k_n( Tensor<BDataType> b_g_k_n(
f_host_tensor_descriptor(batch_count, K, N, stride_B, batch_stride_B, BLayout{})); f_host_tensor_descriptor(batch_count, K, N, stride_B, batch_stride_B, BLayout{}));
auto f_host_c_tensor_descriptor = [](std::size_t G0_, auto f_host_e_tensor_descriptor = [](std::size_t G0_,
std::size_t G1_, std::size_t G1_,
std::size_t M_, std::size_t M_,
std::size_t N_, std::size_t N_,
...@@ -138,15 +135,15 @@ int main(int argc, char* argv[]) ...@@ -138,15 +135,15 @@ int main(int argc, char* argv[])
std::vector<std::size_t>({stride_G0_, stride_G1_, stride_M_, stride_N_})); std::vector<std::size_t>({stride_G0_, stride_G1_, stride_M_, stride_N_}));
}; };
Tensor<EDataType> c_g0_g1_m_n_host_result( Tensor<EDataType> e_g0_g1_m_n_host_result(
f_host_c_tensor_descriptor(G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N)); f_host_e_tensor_descriptor(G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N));
Tensor<EDataType> c_g0_g1_m_n_device_result( Tensor<EDataType> e_g0_g1_m_n_device_result(
f_host_c_tensor_descriptor(G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N)); f_host_e_tensor_descriptor(G0, G1, M, N, stride_G0, stride_G1, stride_M, stride_N));
std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl; std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl;
std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl; std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl;
std::cout << "c_g0_g1_m_n: " << c_g0_g1_m_n_host_result.mDesc << std::endl; std::cout << "e_g0_g1_m_n: " << e_g0_g1_m_n_host_result.mDesc << std::endl;
switch(init_method) switch(init_method)
{ {
...@@ -161,9 +158,10 @@ int main(int argc, char* argv[]) ...@@ -161,9 +158,10 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpaceSize());
DeviceMem c_device_buf(sizeof(EDataType) * c_g0_g1_m_n_device_result.mDesc.GetElementSpace()); DeviceMem e_device_buf(sizeof(EDataType) *
e_g0_g1_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_g_m_k.mData.data()); a_device_buf.ToDevice(a_g_m_k.mData.data());
b_device_buf.ToDevice(b_g_k_n.mData.data()); b_device_buf.ToDevice(b_g_k_n.mData.data());
...@@ -178,7 +176,7 @@ int main(int argc, char* argv[]) ...@@ -178,7 +176,7 @@ int main(int argc, char* argv[])
// do GEM // do GEM
auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()), auto argument = gemm.MakeArgument(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()), static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
static_cast<EDataType*>(c_device_buf.GetDeviceBuffer()), static_cast<EDataType*>(e_device_buf.GetDeviceBuffer()),
M, M,
N, N,
K, K,
...@@ -186,7 +184,7 @@ int main(int argc, char* argv[]) ...@@ -186,7 +184,7 @@ int main(int argc, char* argv[])
stride_B, stride_B,
batch_stride_A, batch_stride_A,
batch_stride_B, batch_stride_B,
batched_gemm_c_permute_desc, batched_gemm_e_permute_desc,
batch_count, batch_count,
a_element_op, a_element_op,
b_element_op, b_element_op,
...@@ -217,7 +215,7 @@ int main(int argc, char* argv[]) ...@@ -217,7 +215,7 @@ int main(int argc, char* argv[])
if(do_verification) if(do_verification)
{ {
c_device_buf.FromDevice(c_g0_g1_m_n_device_result.mData.data()); e_device_buf.FromDevice(e_g0_g1_m_n_device_result.mData.data());
auto ref_batched_gemm = ReferenceBatchedGemmInstance{}; auto ref_batched_gemm = ReferenceBatchedGemmInstance{};
auto ref_invoker = ref_batched_gemm.MakeInvoker(); auto ref_invoker = ref_batched_gemm.MakeInvoker();
...@@ -238,15 +236,16 @@ int main(int argc, char* argv[]) ...@@ -238,15 +236,16 @@ int main(int argc, char* argv[])
{ {
for(int n = 0; n < N; n++) for(int n = 0; n < N; n++)
{ {
int g = g0 * G1 + g1; int g = g0 * G1 + g1;
c_g0_g1_m_n_host_result(g0, g1, m, n) = c_g_m_n_host_result(g, m, n);
e_g0_g1_m_n_host_result(g0, g1, m, n) = c_g_m_n_host_result(g, m, n);
} }
} }
} }
} }
pass = ck::utils::check_err(c_g0_g1_m_n_host_result.mData, pass = ck::utils::check_err(e_g0_g1_m_n_host_result.mData,
c_g0_g1_m_n_device_result.mData, e_g0_g1_m_n_device_result.mData,
"Error: Incorrect results c"); "Error: Incorrect results c");
} }
......
example/25_gemm_bias_c_permute/CMakeLists.txt deleted 100644 → 0
View file @ 85978e02
add_example_executable(example_gemm_bias_c_permute_xdl_fp16 gemm_bias_c_permute_xdl_fp16.cpp)
example/25_gemm_bias_e_permute/CMakeLists.txt 0 → 100644
View file @ 500fa995
add_example_executable(example_gemm_bias_e_permute_xdl_fp16 gemm_bias_e_permute_xdl_fp16.cpp)
example/25_gemm_bias_c_permute/gemm_bias_c_permute_xdl_fp16.cpp → example/25_gemm_bias_e_permute/gemm_bias_e_permute_xdl_fp16.cpp
View file @ 500fa995
...@@ -9,12 +9,12 @@ ...@@ -9,12 +9,12 @@
#include "ck/ck.hpp" #include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp" #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp" #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/tensor_operation/gpu/device/device_gemm_bias_c_permute_xdl.hpp" #include "ck/tensor_operation/gpu/device/device_gemm_bias_e_permute_xdl.hpp"
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
...@@ -49,7 +49,7 @@ using CDEElementOp = Add; ...@@ -49,7 +49,7 @@ using CDEElementOp = Add;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// clang-format off // clang-format off
using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmBiasCPermute_Xdl using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmBiasEPermute_Xdl
//######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| //######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| //######| | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_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|
...@@ -186,12 +186,12 @@ int main(int argc, char* argv[]) ...@@ -186,12 +186,12 @@ int main(int argc, char* argv[])
d_m0_m1_m2_n0_n1.GenerateTensorValue(GeneratorTensor_3<DDataType>{0.0, 1.0}); d_m0_m1_m2_n0_n1.GenerateTensorValue(GeneratorTensor_3<DDataType>{0.0, 1.0});
} }
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace()); DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace()); DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem d_m0_m1_m2_n0_n1_device_buf(sizeof(DDataType) * DeviceMem d_m0_m1_m2_n0_n1_device_buf(sizeof(DDataType) *
d_m0_m1_m2_n0_n1.mDesc.GetElementSpace()); d_m0_m1_m2_n0_n1.mDesc.GetElementSpaceSize());
DeviceMem e_m0_m1_m2_n0_n1_device_buf(sizeof(EDataType) * DeviceMem e_m0_m1_m2_n0_n1_device_buf(
e_m0_m1_m2_n0_n1_device_result.mDesc.GetElementSpace()); sizeof(EDataType) * e_m0_m1_m2_n0_n1_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data()); a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data()); b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/26_contraction/contraction_bilinear_xdl_fp32.cpp
View file @ 500fa995
...@@ -12,9 +12,9 @@ ...@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
...@@ -324,10 +324,10 @@ int main(int argc, char* argv[]) ...@@ -324,10 +324,10 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpace()); DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpace()); DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_ms_ks.mData.data()); a_device_buf.ToDevice(a_ms_ks.mData.data());
b_device_buf.ToDevice(b_ns_ks.mData.data()); b_device_buf.ToDevice(b_ns_ks.mData.data());
......
example/26_contraction/contraction_scale_xdl_fp32.cpp
View file @ 500fa995
...@@ -12,9 +12,9 @@ ...@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
using S = ck::Sequence<Is...>; using S = ck::Sequence<Is...>;
...@@ -260,16 +260,16 @@ int main(int argc, char* argv[]) ...@@ -260,16 +260,16 @@ int main(int argc, char* argv[])
e_ms_ns_lengths = {M0, M1, N0, N1}; e_ms_ns_lengths = {M0, M1, N0, N1};
e_ms_ns_strides = { e_ms_ns_strides = {
std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])}; std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
scale = std::stof(argv[26]); scale = std::stof(argv[22]);
} }
else else
{ {
printf("arg1: verification (0=no, 1=yes)\n"); printf("arg1: verification (0=no, 1=yes)\n");
printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"); printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
printf("arg3: time kernel (0=no, 1=yes)\n"); printf("arg3: time kernel (0=no, 1=yes)\n");
printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n"); printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n"); printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n"); printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n"); printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
...@@ -307,9 +307,9 @@ int main(int argc, char* argv[]) ...@@ -307,9 +307,9 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpace()); DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_ms_ks.mData.data()); a_device_buf.ToDevice(a_ms_ks.mData.data());
b_device_buf.ToDevice(b_ns_ks.mData.data()); b_device_buf.ToDevice(b_ns_ks.mData.data());
......
example/27_layernorm/layernorm_blockwise.cpp
View file @ 500fa995
...@@ -13,10 +13,10 @@ ...@@ -13,10 +13,10 @@
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp" #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_common_util.hpp" #include "ck/library/utility/host_common_util.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
using XDataType = ck::half_t; using XDataType = ck::half_t;
...@@ -75,10 +75,10 @@ int main() ...@@ -75,10 +75,10 @@ int main()
gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0}); gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0});
beta.GenerateTensorValue(GeneratorTensor_3<BetaDataType>{0.0, 1.0}); beta.GenerateTensorValue(GeneratorTensor_3<BetaDataType>{0.0, 1.0});
DeviceMem x_dev(sizeof(XDataType) * x.mDesc.GetElementSpace()); DeviceMem x_dev(sizeof(XDataType) * x.mDesc.GetElementSpaceSize());
DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpace()); DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpace()); DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpace()); DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
x_dev.ToDevice(x.mData.data()); x_dev.ToDevice(x.mData.data());
gamma_dev.ToDevice(gamma.mData.data()); gamma_dev.ToDevice(gamma.mData.data());
......
example/28_grouped_gemm_bias/grouped_gemm_bias_xdl_fp16.cpp
View file @ 500fa995
...@@ -13,9 +13,9 @@ ...@@ -13,9 +13,9 @@
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -34,13 +34,15 @@ using ADataType = F16; ...@@ -34,13 +34,15 @@ using ADataType = F16;
using BDataType = F16; using BDataType = F16;
using AccDataType = F32; using AccDataType = F32;
using CShuffleDataType = F16; using CShuffleDataType = F16;
using D0DataType = F16; using DDataType = F16;
using DsDataType = ck::Tuple<D0DataType>; using DsDataType = ck::Tuple<DDataType>;
using EDataType = F16; using EDataType = F16;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
using ELayout = Row; using DLayout = Row;
using DsLayout = ck::Tuple<DLayout>;
using ELayout = Row;
using AElementOp = PassThrough; using AElementOp = PassThrough;
using BElementOp = PassThrough; using BElementOp = PassThrough;
...@@ -48,13 +50,13 @@ using CDEElementOp = Add; ...@@ -48,13 +50,13 @@ using CDEElementOp = Add;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdl using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl
// clang-format off // clang-format off
//######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| //######| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_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|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>;
// clang-format on // clang-format on
int main(int argc, char* argv[]) int main(int argc, char* argv[])
...@@ -118,24 +120,24 @@ int main(int argc, char* argv[]) ...@@ -118,24 +120,24 @@ int main(int argc, char* argv[])
std::vector<Tensor<ADataType>> a_tensors; std::vector<Tensor<ADataType>> a_tensors;
std::vector<Tensor<BDataType>> b_tensors; std::vector<Tensor<BDataType>> b_tensors;
std::vector<Tensor<D0DataType>> d0_tensors; std::vector<Tensor<DDataType>> d_tensors;
std::vector<Tensor<EDataType>> e_host_tensors; std::vector<Tensor<EDataType>> e_host_tensors;
std::vector<Tensor<EDataType>> e_device_tensors; std::vector<Tensor<EDataType>> e_device_tensors;
a_tensors.reserve(group_count); a_tensors.reserve(group_count);
b_tensors.reserve(group_count); b_tensors.reserve(group_count);
d0_tensors.reserve(group_count); d_tensors.reserve(group_count);
e_host_tensors.reserve(group_count); e_host_tensors.reserve(group_count);
e_device_tensors.reserve(group_count); e_device_tensors.reserve(group_count);
using DeviceMemPtr = std::unique_ptr<DeviceMem>; using DeviceMemPtr = std::unique_ptr<DeviceMem>;
std::vector<DeviceMemPtr> a_tensors_device, b_tensors_device, d0_tensors_device, std::vector<DeviceMemPtr> a_tensors_device, b_tensors_device, d_tensors_device,
e_tensors_device; e_tensors_device;
a_tensors_device.reserve(group_count); a_tensors_device.reserve(group_count);
b_tensors_device.reserve(group_count); b_tensors_device.reserve(group_count);
d0_tensors_device.reserve(group_count); d_tensors_device.reserve(group_count);
e_tensors_device.reserve(group_count); e_tensors_device.reserve(group_count);
std::size_t flop = 0, num_btype = 0; std::size_t flop = 0, num_btype = 0;
...@@ -146,7 +148,7 @@ int main(int argc, char* argv[]) ...@@ -146,7 +148,7 @@ int main(int argc, char* argv[])
gemm_descs[i].M_, gemm_descs[i].K_, gemm_descs[i].stride_A_, ALayout{}))); gemm_descs[i].M_, gemm_descs[i].K_, gemm_descs[i].stride_A_, ALayout{})));
b_tensors.push_back(Tensor<BDataType>(f_host_tensor_descriptor( b_tensors.push_back(Tensor<BDataType>(f_host_tensor_descriptor(
gemm_descs[i].K_, gemm_descs[i].N_, gemm_descs[i].stride_B_, BLayout{}))); gemm_descs[i].K_, gemm_descs[i].N_, gemm_descs[i].stride_B_, BLayout{})));
d0_tensors.push_back(Tensor<D0DataType>(f_host_tensor_descriptor( d_tensors.push_back(Tensor<DDataType>(f_host_tensor_descriptor(
gemm_descs[i].M_, gemm_descs[i].N_, gemm_descs[i].stride_Ds_[0], ELayout{}))); gemm_descs[i].M_, gemm_descs[i].N_, gemm_descs[i].stride_Ds_[0], ELayout{})));
e_host_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor( e_host_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
gemm_descs[i].M_, gemm_descs[i].N_, gemm_descs[i].stride_C_, ELayout{}))); gemm_descs[i].M_, gemm_descs[i].N_, gemm_descs[i].stride_C_, ELayout{})));
...@@ -168,38 +170,38 @@ int main(int argc, char* argv[]) ...@@ -168,38 +170,38 @@ int main(int argc, char* argv[])
case 1: case 1:
a_tensors[i].GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5}); a_tensors[i].GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5}); b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
d0_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5}); d_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
break; break;
case 2: case 2:
a_tensors[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0}); a_tensors[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
b_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}); b_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
d0_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}); d_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
break; break;
default: default:
a_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{}); a_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{}); b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
d0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{}); d_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
} }
} }
for(std::size_t i = 0; i < gemm_descs.size(); i++) for(std::size_t i = 0; i < gemm_descs.size(); i++)
{ {
a_tensors_device.emplace_back( a_tensors_device.emplace_back(std::make_unique<DeviceMem>(
std::make_unique<DeviceMem>(sizeof(ADataType) * a_tensors[i].mDesc.GetElementSpace())); sizeof(ADataType) * a_tensors[i].mDesc.GetElementSpaceSize()));
b_tensors_device.emplace_back( b_tensors_device.emplace_back(std::make_unique<DeviceMem>(
std::make_unique<DeviceMem>(sizeof(BDataType) * b_tensors[i].mDesc.GetElementSpace())); sizeof(BDataType) * b_tensors[i].mDesc.GetElementSpaceSize()));
d0_tensors_device.emplace_back(std::make_unique<DeviceMem>( d_tensors_device.emplace_back(std::make_unique<DeviceMem>(
sizeof(D0DataType) * d0_tensors[i].mDesc.GetElementSpace())); sizeof(DDataType) * d_tensors[i].mDesc.GetElementSpaceSize()));
e_tensors_device.emplace_back(std::make_unique<DeviceMem>( e_tensors_device.emplace_back(std::make_unique<DeviceMem>(
sizeof(EDataType) * e_device_tensors[i].mDesc.GetElementSpace())); sizeof(EDataType) * e_device_tensors[i].mDesc.GetElementSpaceSize()));
a_tensors_device[i]->ToDevice(a_tensors[i].mData.data()); a_tensors_device[i]->ToDevice(a_tensors[i].mData.data());
b_tensors_device[i]->ToDevice(b_tensors[i].mData.data()); b_tensors_device[i]->ToDevice(b_tensors[i].mData.data());
d0_tensors_device[i]->ToDevice(d0_tensors[i].mData.data()); d_tensors_device[i]->ToDevice(d_tensors[i].mData.data());
p_a.push_back(a_tensors_device[i]->GetDeviceBuffer()); p_a.push_back(a_tensors_device[i]->GetDeviceBuffer());
p_b.push_back(b_tensors_device[i]->GetDeviceBuffer()); p_b.push_back(b_tensors_device[i]->GetDeviceBuffer());
p_ds.push_back({d0_tensors_device[i]->GetDeviceBuffer()}); p_ds.push_back({d_tensors_device[i]->GetDeviceBuffer()});
p_c.push_back(e_tensors_device[i]->GetDeviceBuffer()); p_c.push_back(e_tensors_device[i]->GetDeviceBuffer());
} }
...@@ -266,7 +268,7 @@ int main(int argc, char* argv[]) ...@@ -266,7 +268,7 @@ int main(int argc, char* argv[])
for(int n = 0; n < gemm_descs[i].N_; ++n) for(int n = 0; n < gemm_descs[i].N_; ++n)
{ {
cde_element_op( cde_element_op(
e_host_tensors[i](m, n), e_host_tensors[i](m, n), d0_tensors[i](m, n)); e_host_tensors[i](m, n), e_host_tensors[i](m, n), d_tensors[i](m, n));
} }
} }
......
example/29_batched_gemm_multi_d/batched_gemm_bias_xdl_fp16.cpp
View file @ 500fa995
...@@ -10,9 +10,9 @@ ...@@ -10,9 +10,9 @@
#include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -37,7 +37,9 @@ using EDataType = F16; ...@@ -37,7 +37,9 @@ using EDataType = F16;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
using DELayout = Row; using DLayout = Row;
using DsLayout = ck::Tuple<DLayout>;
using ELayout = Row;
using AElementOp = PassThrough; using AElementOp = PassThrough;
using BElementOp = PassThrough; using BElementOp = PassThrough;
...@@ -48,12 +50,12 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa ...@@ -48,12 +50,12 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; // static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// clang-format off // clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmMultiDXdl using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmMultiD_Xdl
//######| ALayout| BLayout| DELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| //######| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_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|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, DELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>;
// clang-format on // clang-format on
int main(int argc, char* argv[]) int main(int argc, char* argv[])
...@@ -117,10 +119,10 @@ int main(int argc, char* argv[]) ...@@ -117,10 +119,10 @@ int main(int argc, char* argv[])
f_host_tensor_descriptor(batch_count, K, N, stride_B, batch_stride_B, BLayout{})); f_host_tensor_descriptor(batch_count, K, N, stride_B, batch_stride_B, BLayout{}));
Tensor<DDataType> d_g_m_n( Tensor<DDataType> d_g_m_n(
f_host_tensor_descriptor(batch_count, M, N, stride_D, batch_stride_D, DELayout{})); f_host_tensor_descriptor(batch_count, M, N, stride_D, batch_stride_D, DLayout{}));
Tensor<EDataType> e_g_m_n_device_result( Tensor<EDataType> e_g_m_n_device_result(
f_host_tensor_descriptor(batch_count, M, N, stride_E, batch_stride_E, DELayout{})); f_host_tensor_descriptor(batch_count, M, N, stride_E, batch_stride_E, ELayout{}));
std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl; std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl;
std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl; std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl;
...@@ -142,10 +144,10 @@ int main(int argc, char* argv[]) ...@@ -142,10 +144,10 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf(sizeof(DDataType) * d_g_m_n.mDesc.GetElementSpace()); DeviceMem d_device_buf(sizeof(DDataType) * d_g_m_n.mDesc.GetElementSpaceSize());
DeviceMem c_device_buf(sizeof(EDataType) * e_g_m_n_device_result.mDesc.GetElementSpace()); DeviceMem c_device_buf(sizeof(EDataType) * e_g_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_g_m_k.mData.data()); a_device_buf.ToDevice(a_g_m_k.mData.data());
b_device_buf.ToDevice(b_g_k_n.mData.data()); b_device_buf.ToDevice(b_g_k_n.mData.data());
...@@ -166,6 +168,7 @@ int main(int argc, char* argv[]) ...@@ -166,6 +168,7 @@ int main(int argc, char* argv[])
M, M,
N, N,
K, K,
batch_count,
stride_A, stride_A,
stride_B, stride_B,
{stride_D}, {stride_D},
...@@ -174,7 +177,6 @@ int main(int argc, char* argv[]) ...@@ -174,7 +177,6 @@ int main(int argc, char* argv[])
batch_stride_B, batch_stride_B,
{batch_stride_D}, {batch_stride_D},
batch_stride_E, batch_stride_E,
batch_count,
a_element_op, a_element_op,
b_element_op, b_element_op,
cde_element_op); cde_element_op);
...@@ -218,7 +220,7 @@ int main(int argc, char* argv[]) ...@@ -218,7 +220,7 @@ int main(int argc, char* argv[])
auto ref_invoker = ref_batched_gemm.MakeInvoker(); auto ref_invoker = ref_batched_gemm.MakeInvoker();
Tensor<EDataType> e_g_m_n_host_result( Tensor<EDataType> e_g_m_n_host_result(
f_host_tensor_descriptor(batch_count, M, N, stride_E, batch_stride_E, DELayout{})); f_host_tensor_descriptor(batch_count, M, N, stride_E, batch_stride_E, ELayout{}));
auto ref_argument = ref_batched_gemm.MakeArgument( auto ref_argument = ref_batched_gemm.MakeArgument(
a_g_m_k, b_g_k_n, e_g_m_n_host_result, a_element_op, b_element_op, PassThrough{}); a_g_m_k, b_g_k_n, e_g_m_n_host_result, a_element_op, b_element_op, PassThrough{});
......
example/29_batched_gemm_multi_d/batched_gemm_xdl_fp16.cpp
View file @ 500fa995
...@@ -10,9 +10,9 @@ ...@@ -10,9 +10,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp" #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp" #include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp" #include "ck/library/utility/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp" #include "ck/library/utility/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp" #include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp" #include "ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp"
template <ck::index_t... Is> template <ck::index_t... Is>
...@@ -33,9 +33,10 @@ using CShuffleDataType = F16; ...@@ -33,9 +33,10 @@ using CShuffleDataType = F16;
using DsDataType = ck::Tuple<>; using DsDataType = ck::Tuple<>;
using EDataType = F16; using EDataType = F16;
using ALayout = Row; using ALayout = Row;
using BLayout = Col; using BLayout = Col;
using ELayout = Row; using DsLayout = ck::Tuple<>;
using ELayout = Row;
using AElementOp = PassThrough; using AElementOp = PassThrough;
using BElementOp = PassThrough; using BElementOp = PassThrough;
...@@ -46,12 +47,12 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa ...@@ -46,12 +47,12 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding; // static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// clang-format off // clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmMultiDXdl using DeviceGemmInstance = ck::tensor_operation::device::DeviceBatchedGemmMultiD_Xdl
//######| ALayout| BLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| //######| ALayout| BLayout| DsLayout| ELayout| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| 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| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_MWaveMPerXdl| ScalarPerVector| //######| | | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| 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_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|
//######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //######| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
< ALayout, BLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>; < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, AElementOp, BElementOp, CDEElementOp, GemmDefault, 1, 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, 32, 1, 8>, 8>;
// clang-format on // clang-format on
using ReferenceBatchedGemmInstance = ck::tensor_operation::host:: using ReferenceBatchedGemmInstance = ck::tensor_operation::host::
...@@ -135,9 +136,9 @@ int main(int argc, char* argv[]) ...@@ -135,9 +136,9 @@ int main(int argc, char* argv[])
break; break;
} }
DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpace()); DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpace()); DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpaceSize());
DeviceMem c_device_buf(sizeof(EDataType) * e_g_m_n_device_result.mDesc.GetElementSpace()); DeviceMem c_device_buf(sizeof(EDataType) * e_g_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_g_m_k.mData.data()); a_device_buf.ToDevice(a_g_m_k.mData.data());
b_device_buf.ToDevice(b_g_k_n.mData.data()); b_device_buf.ToDevice(b_g_k_n.mData.data());
...@@ -157,6 +158,7 @@ int main(int argc, char* argv[]) ...@@ -157,6 +158,7 @@ int main(int argc, char* argv[])
M, M,
N, N,
K, K,
batch_count,
stride_A, stride_A,
stride_B, stride_B,
{}, {},
...@@ -165,7 +167,6 @@ int main(int argc, char* argv[]) ...@@ -165,7 +167,6 @@ int main(int argc, char* argv[])
batch_stride_B, batch_stride_B,
{}, {},
batch_stride_C, batch_stride_C,
batch_count,
a_element_op, a_element_op,
b_element_op, b_element_op,
cde_element_op); cde_element_op);
......
example/30_grouped_convnd_fwd_bias_relu/CMakeLists.txt 0 → 100644
View file @ 500fa995
add_example_executable(example_grouped_convnd_fwd_bias_relu_xdl_fp16 grouped_convnd_fwd_bias_relu_xdl_fp16.cpp)
target_link_libraries(example_grouped_convnd_fwd_bias_relu_xdl_fp16 PRIVATE utility)
example/30_grouped_convnd_fwd_bias_relu/README.md 0 → 100644
View file @ 500fa995
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: time kernel (0=no, 1=yes)
#Following arguments (depending on number of spatial dims):
# N spatial dimensions
# G, N, K, C,
# <filter spatial dimensions>, (ie Y, X for 2D)
# <input image spatial dimensions>, (ie Hi, Wi for 2D)
# <strides>, (ie Sy, Sx for 2D)
# <dilations>, (ie Dy, Dx for 2D)
# <left padding>, (ie LeftPy, LeftPx for 2D)
# <right padding>, (ie RightPy, RightPx for 2D)
bin/example_grouped_convnd_fwd_bias_relu_xdl_fp16 1 1 1
```
Result (MI100)
```
in: dim 5, lengths {1, 128, 192, 71, 71}, strides {6912, 967872, 1, 13632, 192}
wei: dim 5, lengths {1, 256, 192, 3, 3}, strides {192, 1728, 1, 576, 192}
bias: dim 5, lengths {1, 128, 256, 36, 36}, strides {256, 0, 1, 0, 0}
out: dim 5, lengths {1, 128, 256, 36, 36}, strides {256, 331776, 1, 9216, 256}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up 1 time
Start running 10 times...
Perf: 1.19215 ms, 123.112 TFlops, 279.827 GB/s, DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<256, 128, 256, 32, Default>
```
example/30_grouped_convnd_fwd_bias_relu/grouped_convnd_fwd_bias_common.hpp 0 → 100644
View file @ 500fa995
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <numeric>
#include <type_traits>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/convolution_parameter.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
void print_helper_msg()
{
std::cout << "arg1: verification (0=no, 1=yes)\n"
<< "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
<< "arg3: time kernel (0=no, 1=yes)\n"
<< ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
}
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,
typename OutDataType,
typename InElementOp,
typename WeiElementOp,
typename OutElementOp,
typename DeviceConvNDFwdInstance>
int run_grouped_conv_fwd_bias(bool do_verification,
int init_method,
bool time_kernel,
const ck::utils::conv::ConvParam& conv_param,
const HostTensorDescriptor& in_g_n_c_wis_desc,
const HostTensorDescriptor& wei_g_k_c_xs_desc,
const HostTensorDescriptor& bias_g_n_k_wos_desc,
const HostTensorDescriptor& out_g_n_k_wos_desc,
const InElementOp& in_element_op,
const WeiElementOp& wei_element_op,
const OutElementOp& out_element_op)
{
Tensor<InDataType> in(in_g_n_c_wis_desc);
Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
Tensor<OutDataType> bias(bias_g_n_k_wos_desc);
Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
std::cout << "in: " << in.mDesc << std::endl;
std::cout << "wei: " << wei.mDesc << std::endl;
std::cout << "bias: " << bias.mDesc << std::endl;
std::cout << "out: " << out_host.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
bias.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
break;
default:
in.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
bias.GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.5, 0.5});
}
DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
DeviceMem bias_device_buf(sizeof(OutDataType) * bias.mDesc.GetElementSpaceSize());
DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
in_device_buf.ToDevice(in.mData.data());
wei_device_buf.ToDevice(wei.mData.data());
bias_device_buf.ToDevice(bias.mData.data());
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
std::array<ck::index_t, NDimSpatial + 3> d_g_n_k_wos_lengths{};
std::array<ck::index_t, NDimSpatial + 3> d_g_n_k_wos_strides{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{};
std::array<ck::index_t, NDimSpatial> input_right_pads{};
auto copy = [](auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
copy(bias_g_n_k_wos_desc.GetLengths(), d_g_n_k_wos_lengths);
copy(bias_g_n_k_wos_desc.GetStrides(), d_g_n_k_wos_strides);
copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
copy(conv_param.conv_filter_strides_, conv_filter_strides);
copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
copy(conv_param.input_left_pads_, input_left_pads);
copy(conv_param.input_right_pads_, input_right_pads);
// do Conv
auto conv = DeviceConvNDFwdInstance{};
auto invoker = conv.MakeInvoker();
auto argument = conv.MakeArgument(
in_device_buf.GetDeviceBuffer(),
wei_device_buf.GetDeviceBuffer(),
std::array<const void*, 1>{bias_device_buf.GetDeviceBuffer()},
out_device_buf.GetDeviceBuffer(),
a_g_n_c_wis_lengths,
a_g_n_c_wis_strides,
b_g_k_c_xs_lengths,
b_g_k_c_xs_strides,
std::array<std::array<ck::index_t, NDimSpatial + 3>, 1>{{d_g_n_k_wos_lengths}},
std::array<std::array<ck::index_t, NDimSpatial + 3>, 1>{{d_g_n_k_wos_strides}},
e_g_n_k_wos_lengths,
e_g_n_k_wos_strides,
conv_filter_strides,
conv_filter_dilations,
input_left_pads,
input_right_pads,
in_element_op,
wei_element_op,
out_element_op);
if(!conv.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device_conv with the specified compilation parameters does "
"not support this Conv problem");
}
float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = conv_param.GetFlops();
std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
float gb_per_sec = num_btype / 1.E6 / avg_time;
std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
<< conv.GetTypeString() << std::endl;
if(do_verification)
{
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
Tensor<OutDataType> c_host(out_g_n_k_wos_desc);
auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
PassThrough>();
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(in,
wei,
c_host,
conv_param.conv_filter_strides_,
conv_param.conv_filter_dilations_,
conv_param.input_left_pads_,
conv_param.input_right_pads_,
in_element_op,
wei_element_op,
PassThrough{});
ref_invoker.Run(ref_argument);
// TODO: implement elementwise operation for host
out_host.ForEach(
[&](auto&, auto idx) { out_element_op(out_host(idx), c_host(idx), bias(idx)); });
out_device_buf.FromDevice(out_device.mData.data());
return ck::utils::check_err(
out_device.mData, out_host.mData, "Error: incorrect results!", 1e-5f, 1e-4f)
? 0
: 1;
}
return 0;
}
example/30_grouped_convnd_fwd_bias_relu/grouped_convnd_fwd_bias_relu_xdl_fp16.cpp 0 → 100644
View file @ 500fa995
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "grouped_convnd_fwd_bias_common.hpp"
#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using AccDataType = float;
using CShuffleDataType = ck::half_t;
using BiasDataType = ck::half_t;
using OutDataType = ck::half_t;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
static constexpr auto ConvSpec =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
template <ck::index_t NDimSpatial,
typename InLayout,
typename WeiLayout,
typename BiasLayout,
typename OutLayout>
using DeviceGroupedConvNDFwdInstance =
ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
NDimSpatial,
InLayout,
WeiLayout,
ck::Tuple<BiasLayout>,
OutLayout,
InDataType,
WeiDataType,
AccDataType,
CShuffleDataType,
ck::Tuple<BiasDataType>,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
ConvSpec, // ConvForwardSpecialization
GemmSpec, // GemmSpecialization
1, //
256, // BlockSize
128, // MPerBlock
256, // NPerBlock
32, // KPerBlock
8, // AK1
8, // BK1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
4, // NXdlPerWave
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_AK1
1, // ABlockLdsExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_BK1
1, // BBlockLdsExtraN
1,
1,
S<1, 32, 1, 8>,
8>;
int main(int argc, char* argv[])
{
namespace ctc = ck::tensor_layout::convolution;
print_helper_msg();
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
ck::utils::conv::ConvParam conv_param{
2, 2, 128, 256, 192, {3, 3}, {71, 71}, {2, 2}, {1, 1}, {1, 1}, {1, 1}};
if(argc == 1)
{
// use default
}
else if(argc == 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
}
else
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
const ck::index_t num_dim_spatial = std::stoi(argv[4]);
conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
}
const auto in_element_op = InElementOp{};
const auto wei_element_op = WeiElementOp{};
const auto out_element_op = OutElementOp{};
if(conv_param.num_dim_spatial_ == 1)
{
using InLayout = ctc::G_NW_C;
using WeiLayout = ctc::G_K_X_C;
using BiasLayout = ctc::G_NW_K;
using OutLayout = ctc::G_NW_K;
const auto in_g_n_c_wis_desc = HostTensorDescriptor(
{conv_param.G_, conv_param.N_, conv_param.C_, conv_param.input_spatial_lengths_[0]},
{
conv_param.C_, // g
conv_param.input_spatial_lengths_[0] * conv_param.G_ * conv_param.C_, // n
1, // c
conv_param.G_ * conv_param.C_ // wi
});
const auto wei_g_k_c_xs_desc = HostTensorDescriptor(
{conv_param.G_, conv_param.K_, conv_param.C_, conv_param.filter_spatial_lengths_[0]},
{
conv_param.C_, // g
conv_param.filter_spatial_lengths_[0] * conv_param.G_ * conv_param.C_, // k
1, // c
conv_param.G_ * conv_param.C_ // x
});
const auto bias_g_n_k_wos_desc = HostTensorDescriptor(
{conv_param.G_, conv_param.N_, conv_param.K_, conv_param.output_spatial_lengths_[0]},
{
conv_param.K_, // g
0, // k
1, // c
0 // x
});
const auto out_g_n_k_wos_desc = HostTensorDescriptor(
{conv_param.G_, conv_param.N_, conv_param.K_, conv_param.output_spatial_lengths_[0]},
{
conv_param.K_, // g
conv_param.output_spatial_lengths_[0] * conv_param.G_ * conv_param.K_, // n
1, // k
conv_param.G_ * conv_param.K_ // wo
});
return run_grouped_conv_fwd_bias<
1,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<1, InLayout, WeiLayout, BiasLayout, OutLayout>>(
do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
bias_g_n_k_wos_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
else if(conv_param.num_dim_spatial_ == 2)
{
using InLayout = ctc::G_NHW_C;
using WeiLayout = ctc::G_K_YX_C;
using BiasLayout = ctc::G_NHW_K;
using OutLayout = ctc::G_NHW_K;
const auto in_g_n_c_wis_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.N_,
conv_param.C_,
conv_param.input_spatial_lengths_[0],
conv_param.input_spatial_lengths_[1]},
{
conv_param.output_spatial_lengths_[0] * conv_param.C_, // g
conv_param.input_spatial_lengths_[0] * conv_param.input_spatial_lengths_[1] *
conv_param.G_ * conv_param.C_, // n
1, // c
conv_param.input_spatial_lengths_[1] * conv_param.G_ * conv_param.C_, // hi
conv_param.G_ * conv_param.C_ // wi
});
const auto wei_g_k_c_xs_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.K_,
conv_param.C_,
conv_param.filter_spatial_lengths_[0],
conv_param.filter_spatial_lengths_[1]},
{
conv_param.C_, // g
conv_param.filter_spatial_lengths_[0] * conv_param.filter_spatial_lengths_[1] *
conv_param.G_ * conv_param.C_, // k
1, // c
conv_param.filter_spatial_lengths_[1] * conv_param.G_ * conv_param.C_, // y
conv_param.G_ * conv_param.C_ // x
});
const auto bias_g_n_k_wos_desc =
HostTensorDescriptor({conv_param.G_,
conv_param.N_,
conv_param.K_,
conv_param.output_spatial_lengths_[0],
conv_param.output_spatial_lengths_[1]},
{
conv_param.K_, // g
0, // n
1, // k
0, // ho
0 // wo
});
const auto out_g_n_k_wos_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.N_,
conv_param.K_,
conv_param.output_spatial_lengths_[0],
conv_param.output_spatial_lengths_[1]},
{
conv_param.K_, // g
conv_param.output_spatial_lengths_[0] * conv_param.output_spatial_lengths_[1] *
conv_param.G_ * conv_param.K_, // n
1, // k
conv_param.output_spatial_lengths_[1] * conv_param.G_ * conv_param.K_, // ho
conv_param.G_ * conv_param.K_ // wo
});
return run_grouped_conv_fwd_bias<
2,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<2, InLayout, WeiLayout, BiasLayout, OutLayout>>(
do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
bias_g_n_k_wos_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
else if(conv_param.num_dim_spatial_ == 3)
{
using InLayout = ctc::G_NDHW_C;
using WeiLayout = ctc::G_K_ZYX_C;
using BiasLayout = ctc::G_NDHW_K;
using OutLayout = ctc::G_NDHW_K;
const auto in_g_n_c_wis_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.N_,
conv_param.C_,
conv_param.input_spatial_lengths_[0],
conv_param.input_spatial_lengths_[1],
conv_param.input_spatial_lengths_[2]},
{
conv_param.output_spatial_lengths_[0] * conv_param.C_, // g
conv_param.input_spatial_lengths_[0] * conv_param.input_spatial_lengths_[1] *
conv_param.input_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // n
1, // c
conv_param.input_spatial_lengths_[1] * conv_param.input_spatial_lengths_[2] *
conv_param.G_ * conv_param.C_, // di
conv_param.input_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // hi
conv_param.G_ * conv_param.C_ // wi
});
const auto wei_g_k_c_xs_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.K_,
conv_param.C_,
conv_param.filter_spatial_lengths_[0],
conv_param.filter_spatial_lengths_[1],
conv_param.filter_spatial_lengths_[2]},
{
conv_param.C_, // g
conv_param.filter_spatial_lengths_[0] * conv_param.filter_spatial_lengths_[1] *
conv_param.filter_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // k
1, // c
conv_param.filter_spatial_lengths_[1] * conv_param.filter_spatial_lengths_[2] *
conv_param.G_ * conv_param.C_, // z
conv_param.filter_spatial_lengths_[2] * conv_param.G_ * conv_param.C_, // y
conv_param.G_ * conv_param.C_ // x
});
const auto bias_g_n_k_wos_desc =
HostTensorDescriptor({conv_param.G_,
conv_param.N_,
conv_param.K_,
conv_param.output_spatial_lengths_[0],
conv_param.output_spatial_lengths_[1],
conv_param.output_spatial_lengths_[2]},
{
conv_param.K_, // g
0, // n
1, // k
0, // z
0, // y
0 // x
});
const auto out_g_n_k_wos_desc = HostTensorDescriptor(
{conv_param.G_,
conv_param.N_,
conv_param.K_,
conv_param.output_spatial_lengths_[0],
conv_param.output_spatial_lengths_[1],
conv_param.output_spatial_lengths_[2]},
{
conv_param.K_, // g
conv_param.output_spatial_lengths_[0] * conv_param.output_spatial_lengths_[1] *
conv_param.output_spatial_lengths_[2] * conv_param.G_ * conv_param.K_, // n
1, // k
conv_param.output_spatial_lengths_[1] * conv_param.output_spatial_lengths_[2] *
conv_param.G_ * conv_param.K_, // do
conv_param.output_spatial_lengths_[2] * conv_param.G_ * conv_param.K_, // ho
conv_param.G_ * conv_param.K_ // wo
});
return run_grouped_conv_fwd_bias<
3,
InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp,
DeviceGroupedConvNDFwdInstance<3, InLayout, WeiLayout, BiasLayout, OutLayout>>(
do_verification,
init_method,
time_kernel,
conv_param,
in_g_n_c_wis_desc,
wei_g_k_c_xs_desc,
bias_g_n_k_wos_desc,
out_g_n_k_wos_desc,
in_element_op,
wei_element_op,
out_element_op);
}
return 0;
}
example/CMakeLists.txt
View file @ 500fa995
...@@ -8,7 +8,7 @@ add_custom_target(examples) ...@@ -8,7 +8,7 @@ add_custom_target(examples)
function(add_example_executable EXAMPLE_NAME FILE_NAME) function(add_example_executable EXAMPLE_NAME FILE_NAME)
message("adding example ${EXAMPLE_NAME}") message("adding example ${EXAMPLE_NAME}")
add_executable(${EXAMPLE_NAME} ${FILE_NAME}) add_executable(${EXAMPLE_NAME} ${FILE_NAME})
target_link_libraries(${EXAMPLE_NAME} PRIVATE host_tensor) target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
add_test(NAME ${EXAMPLE_NAME} COMMAND $<TARGET_FILE:${EXAMPLE_NAME}> ${ARGN}) add_test(NAME ${EXAMPLE_NAME} COMMAND $<TARGET_FILE:${EXAMPLE_NAME}> ${ARGN})
add_dependencies(examples ${EXAMPLE_NAME}) add_dependencies(examples ${EXAMPLE_NAME})
add_dependencies(check ${EXAMPLE_NAME}) add_dependencies(check ${EXAMPLE_NAME})
...@@ -17,7 +17,7 @@ endfunction(add_example_executable EXAMPLE_NAME) ...@@ -17,7 +17,7 @@ endfunction(add_example_executable EXAMPLE_NAME)
function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME) function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
message("adding example ${EXAMPLE_NAME}") message("adding example ${EXAMPLE_NAME}")
add_executable(${EXAMPLE_NAME} ${FILE_NAME}) add_executable(${EXAMPLE_NAME} ${FILE_NAME})
target_link_libraries(${EXAMPLE_NAME} PRIVATE host_tensor) target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
add_dependencies(examples ${EXAMPLE_NAME}) add_dependencies(examples ${EXAMPLE_NAME})
endfunction(add_example_executable_no_testing EXAMPLE_NAME) endfunction(add_example_executable_no_testing EXAMPLE_NAME)
...@@ -25,26 +25,23 @@ add_subdirectory(01_gemm) ...@@ -25,26 +25,23 @@ add_subdirectory(01_gemm)
add_subdirectory(02_gemm_bilinear) add_subdirectory(02_gemm_bilinear)
add_subdirectory(03_gemm_bias_relu) add_subdirectory(03_gemm_bias_relu)
add_subdirectory(04_gemm_add_add_fastgelu) add_subdirectory(04_gemm_add_add_fastgelu)
add_subdirectory(06_conv2d_fwd_bias_relu)
add_subdirectory(07_conv2d_fwd_bias_relu_add)
add_subdirectory(09_convnd_fwd) add_subdirectory(09_convnd_fwd)
add_subdirectory(10_conv2d_bwd_data)
add_subdirectory(11_conv2d_bwd_weight)
add_subdirectory(12_reduce) add_subdirectory(12_reduce)
add_subdirectory(13_pool2d_fwd) add_subdirectory(13_pool2d_fwd)
add_subdirectory(14_gemm_xdl_requant_relu_requant) add_subdirectory(14_gemm_xdl_requant_relu_requant)
add_subdirectory(15_grouped_gemm) add_subdirectory(15_grouped_gemm)
add_subdirectory(16_gemm_reduce) add_subdirectory(16_gemm_reduce)
add_subdirectory(17_convnd_bwd_data_xdl) add_subdirectory(17_convnd_bwd_data)
add_subdirectory(18_batched_gemm_reduce) add_subdirectory(18_batched_gemm_reduce)
add_subdirectory(19_binary_elementwise) add_subdirectory(19_binary_elementwise)
add_subdirectory(20_convnd_bwd_weight_xdl) add_subdirectory(20_convnd_bwd_weight)
add_subdirectory(21_gemm_layernorm) add_subdirectory(21_gemm_layernorm)
add_subdirectory(22_cgemm) add_subdirectory(22_cgemm)
add_subdirectory(23_softmax) add_subdirectory(23_softmax)
add_subdirectory(24_batched_gemm_c_permute) add_subdirectory(24_batched_gemm_e_permute)
add_subdirectory(25_gemm_bias_c_permute) add_subdirectory(25_gemm_bias_e_permute)
add_subdirectory(26_contraction) add_subdirectory(26_contraction)
add_subdirectory(27_layernorm) add_subdirectory(27_layernorm)
add_subdirectory(28_grouped_gemm_bias) add_subdirectory(28_grouped_gemm_bias)
add_subdirectory(29_batched_gemm_multi_d) add_subdirectory(29_batched_gemm_multi_d)
\ No newline at end of file add_subdirectory(30_grouped_convnd_fwd_bias_relu)
include/ck/ck.hpp
View file @ 500fa995
...@@ -146,7 +146,7 @@ ...@@ -146,7 +146,7 @@
// workaround: verifaction failure, due to compiler regression, for conv bwd-data fp16 using some // workaround: verifaction failure, due to compiler regression, for conv bwd-data fp16 using some
// tuning parameter // tuning parameter
#define CK_WORKAROUND_SWDEV_325164 1 #define CK_WORKAROUND_SWDEV_325164 0
namespace ck { namespace ck {
......
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