Skip to content

GitLab

Commit aea62819 authored by Chaitanya Inumella's avatar Chaitanya Inumella
Browse files

Rebase branch 'develop' of... 

Rebase branch 'develop' of https://github.com/ROCmSoftwarePlatform/composable_kernel into contraction_hipTENSOR
parents 75af5450 75ab874e
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 351 additions and 582 deletions
Dockerfile
View file @ aea62819
......@@ -2,6 +2,7 @@ FROM ubuntu:18.04
ARG ROCMVERSION=5.1
ARG OSDB_BKC_VERSION
ARG compiler_version
RUN set -xe
......@@ -23,8 +24,6 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
cmake-data=3.15.1-0kitware1 \
cmake=3.15.1-0kitware1 \
curl \
g++ \
gdb \
git \
hip-rocclr \
jq \
......@@ -61,17 +60,7 @@ ENV UBSAN_OPTIONS=print_stacktrace=1
RUN wget https://github.com/Yelp/dumb-init/releases/download/v1.2.0/dumb-init_1.2.0_amd64.deb
RUN dpkg -i dumb-init_*.deb && rm dumb-init_*.deb
# Install cget
RUN pip install cget
# Install rclone
RUN pip install https://github.com/pfultz2/rclone/archive/master.tar.gz
ARG PREFIX=/opt/rocm
# Install dependencies
RUN cget install pfultz2/rocm-recipes
# Install rbuild
RUN pip3 install https://github.com/RadeonOpenCompute/rbuild/archive/6d78a0553babdaea8d2da5de15cbda7e869594b8.tar.gz
# Install packages for processing the performance results
RUN pip3 install --upgrade pip
RUN pip3 install sqlalchemy
......@@ -84,12 +73,26 @@ ENV UBSAN_OPTIONS=print_stacktrace=1
ENV LC_ALL=C.UTF-8
ENV LANG=C.UTF-8
ADD rbuild.ini /rbuild.ini
ADD dev-requirements.txt dev-requirements.txt
RUN rbuild prepare -s develop -d $PREFIX
RUN groupadd -f render
# Install the new rocm-cmake version
RUN git clone -b master https://github.com/RadeonOpenCompute/rocm-cmake.git && \
cd rocm-cmake && mkdir build && cd build && \
cmake .. && cmake --build . && cmake --build . --target install
WORKDIR /
ENV compiler_version=$compiler_version
RUN sh -c "echo compiler version = '$compiler_version'"
RUN if [ "$compiler_version" = "9110" ]; then \
git clone -b ck-9110 https://github.com/RadeonOpenCompute/llvm-project.git && \
cd llvm-project && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
make -j 8 ; \
else echo "using the release compiler"; \
fi
#ENV HIP_CLANG_PATH='/llvm-project/build/bin'
#RUN sh -c "echo HIP_CLANG_PATH = '$HIP_CLANG_PATH'"
Jenkinsfile
View file @ aea62819
This diff is collapsed.
README.md
View file @ aea62819
......@@ -10,7 +10,7 @@ rocm/tensorflow:rocm5.1-tf2.6-dev \
/bin/bash
```
# Install the new rocm-cmake version
# Install newer version of rocm-cmake
https://github.com/RadeonOpenCompute/rocm-cmake
## Build
......@@ -54,6 +54,7 @@ make install
```
## Using CK as pre-built kernel library
Instructions for using CK as a pre-built kernel library are under ```client_example/```
## Caveat
### Kernel Timing and Verification
......
client_example/01_gemm/gemm.cpp
View file @ aea62819
......@@ -63,7 +63,7 @@ int main(int argc, char* argv[])
{
// use default case
}
else if(argc == 5)
else if(argc == 7)
{
M = std::stoi(argv[1]);
N = std::stoi(argv[2]);
......
client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
View file @ aea62819
......@@ -31,11 +31,11 @@ using D0DataType = F16;
using D1DataType = F16;
using EDataType = F16;
using ALayout = Row;
using BLayout = Col;
using DDELayout = Row;
using DDELayout = Row;
using DELayout = Row;
using ALayout = Row;
using BLayout = Col;
using D0Layout = Row;
using D1Layout = Row;
using ELayout = Row;
struct SimpleDeviceMem
{
......@@ -105,16 +105,16 @@ int main(int argc, char* argv[])
SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
f_matrix_space_size(M, N, StrideD0, DDELayout{}));
f_matrix_space_size(M, N, StrideD0, D0Layout{}));
SimpleDeviceMem d1_m_n_device_buf(sizeof(D1DataType) *
f_matrix_space_size(M, N, StrideD1, DDELayout{}));
SimpleDeviceMem e_device_buf(sizeof(EDataType) *
f_matrix_space_size(M, N, StrideE, DELayout{}));
f_matrix_space_size(M, N, StrideD1, D1Layout{}));
SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
ALayout,
BLayout,
DDELayout,
ck::Tuple<D0Layout, D1Layout>,
ELayout,
ADataType,
BDataType,
ck::Tuple<D0DataType, D1DataType>,
......
example/01_gemm/gemm_dl_fp16.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is>
......@@ -142,9 +142,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_dl_fp32.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is>
......@@ -141,9 +141,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_dl_int8.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is>
......@@ -139,9 +139,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_xdl_bf16.cpp
View file @ aea62819
......@@ -11,9 +11,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp"
......@@ -170,9 +170,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_xdl_fp16.cpp
View file @ aea62819
......@@ -13,9 +13,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is>
......@@ -155,9 +155,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_Sequential<1>{});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_xdl_fp64.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp"
......@@ -165,9 +165,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_1<BDataType>{1});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/01_gemm/gemm_xdl_int8.cpp
View file @ aea62819
......@@ -13,9 +13,9 @@
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
template <ck::index_t... Is>
......@@ -167,9 +167,9 @@ int main(int argc, char* argv[])
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
}
DeviceMem a_m_k_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_k_n_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.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.GetElementSpaceSize());
DeviceMem c_m_n_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
a_m_k_device_buf.ToDevice(a_m_k.mData.data());
b_k_n_device_buf.ToDevice(b_k_n.mData.data());
......
example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
View file @ aea62819
......@@ -11,9 +11,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp"
......@@ -51,33 +51,34 @@ using BDataType = F16;
using AccDataType = F32;
using CShuffleDataType = F32;
using DDataType = F16;
using DsDataType = ck::Tuple<DDataType>;
using EDataType = F16;
using ALayout = Row;
using BLayout = Col;
using DELayout = Row;
using ALayout = Row;
using BLayout = Col;
using DLayout = Row;
using ELayout = Row;
using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CDEElementOp = AlphaBetaAdd;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using DeviceOpInstance =
ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_CShuffle<ALayout,
BLayout,
DELayout,
ck::Tuple<DLayout>,
ELayout,
ADataType,
BDataType,
AccDataType,
CShuffleDataType,
DsDataType,
ck::Tuple<DDataType>,
EDataType,
AElementOp,
BElementOp,
CDEElementOp,
GemmDefault,
GemmSpec,
1,
256,
256,
......@@ -190,9 +191,9 @@ int main(int argc, char* argv[])
Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DELayout{}));
Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, DELayout{}));
Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, DELayout{}));
Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
......@@ -213,10 +214,10 @@ int main(int argc, char* argv[])
d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
}
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpace());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpace());
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data());
......
example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp"
......@@ -47,33 +47,34 @@ using BDataType = F16;
using AccDataType = F32;
using CShuffleDataType = F16;
using DDataType = F16;
using DsDataType = ck::Tuple<DDataType>;
using EDataType = F16;
using ALayout = Row;
using BLayout = Col;
using DLayout = Row;
using ELayout = Row;
using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CDEElementOp = AddRelu;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using DeviceOpInstance =
ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_CShuffle<ALayout,
BLayout,
ck::Tuple<DLayout>,
ELayout,
ADataType,
BDataType,
AccDataType,
CShuffleDataType,
DsDataType,
ck::Tuple<DDataType>,
EDataType,
AElementOp,
BElementOp,
CDEElementOp,
GemmDefault,
GemmSpec,
1,
256,
256,
......@@ -191,10 +192,10 @@ int main(int argc, char* argv[])
d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{0.0, 1.0});
}
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpace());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpace());
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data());
......
example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
View file @ aea62819
......@@ -12,9 +12,9 @@
#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.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/reference_tensor_operation/cpu/reference_gemm.hpp"
#include "ck/library/utility/check_err.hpp"
......@@ -43,6 +43,7 @@ using ALayout = Row;
using BLayout = Col;
using D0Layout = Row;
using D1Layout = Row;
using DsLayout = ck::Tuple<D0Layout, D1Layout>;
using ELayout = Row;
using AElementOp = PassThrough;
......@@ -53,11 +54,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
// clang-format off
using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleD_Xdl_CShuffle
//######| 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|
//######| | | | | | | | | | 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| 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|
//######| | | | | | | | | | | 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, 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
int main(int argc, char* argv[])
......@@ -156,11 +157,11 @@ int main(int argc, char* argv[])
d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
}
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpace());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpace());
DeviceMem d0_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpace());
DeviceMem d1_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpace());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpace());
DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
DeviceMem d0_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpaceSize());
DeviceMem d1_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data());
......
example/06_conv2d_fwd_bias_relu/CMakeLists.txt deleted 100644 → 0
View file @ 75af5450
add_example_executable(example_conv2d_fwd_xdl_bias_relu conv2d_fwd_xdl_bias_relu.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu PRIVATE conv_util)
example/06_conv2d_fwd_bias_relu/README.md deleted 100644 → 0
View file @ 75af5450
# Instructions for ```example_conv_xdl_bias_relu```
## Run ```example_conv_xdl_bias_relu```
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./bin/example_conv_xdl_bias_relu 0 1 5
```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
```
in_n_c_hi_wi: dim 4, lengths {128, 192, 71, 71}, strides {967872, 1, 13632, 192}
wei_k_c_y_x: dim 4, lengths {256, 192, 3, 3}, strides {1728, 1, 576, 192}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
bias_k: dim 1, lengths {256}, strides {1}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 5 times...
Perf: 1.39009 ms, 105.581 TFlops, 239.981 GB/s
```
example/06_conv2d_fwd_bias_relu/conv2d_fwd_xdl_bias_relu.cpp deleted 100644 → 0
View file @ 75af5450
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/conv_util.hpp"
#include "ck/library/host_tensor/device_memory.hpp"
#include "ck/library/host_tensor/host_tensor.hpp"
#include "ck/library/host_tensor/host_tensor_generator.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp"
namespace {
using InDataType = ck::half_t;
using WeiDataType = ck::half_t;
using OutDataType = ck::half_t;
using AccDataType = float;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
using InLayout = ck::tensor_layout::convolution::NHWC;
using WeiLayout = ck::tensor_layout::convolution::KYXC;
using OutLayout = ck::tensor_layout::convolution::NHWK;
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 MemorySet = ck::InMemoryDataOperationEnum::Set;
static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
// clang-format off
using DeviceConvFwdInstance = ck::tensor_operation::device::
DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
InDataType, // InDataType
WeiDataType, // WeiDataType
OutDataType, // OutDataType
AccDataType, // AccDataType
InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation
MemorySet, // OutGlobalMemoryDataOperation
ConvFwdDefault, // ConvForwardSpecialization
256, // BlockSize
128, // MPerBlock
256, // NPerBlock
4, // K0PerBlock
8, // K1
32, // MPerXdl
32, // NPerXdl
2, // MXdlPerWave
4, // NXdlPerWave
S<4, 64, 1>, // ABlockTransferThreadClusterLengths_K0_M_K1
S<1, 0, 2>, // ABlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // ABlockTransferSrcAccessOrder
2, // ABlockTransferSrcVectorDim
8, // ABlockTransferSrcScalarPerVector
8, // ABlockTransferDstScalarPerVector_K1
true, // ABlockLdsAddExtraM
S<4, 64, 1>, // BBlockTransferThreadClusterLengths_K0_N_K1
S<1, 0, 2>, // BBlockTransferThreadClusterArrangeOrder
S<1, 0, 2>, // BBlockTransferSrcAccessOrder
2, // BBlockTransferSrcVectorDim
8, // BBlockTransferSrcScalarPerVector
8, // BBlockTransferDstScalarPerVector_K1
true, // BBlockLdsAddExtraN
1, // CShuffleMXdlPerWavePerShuffle
1, // CShuffleNXdlPerWavePerShuffle
S<1, 1, 32, 1, 1, 8>, // CBlockTransferClusterLengths_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl
8>; // CBlockTransferScalarPerVector_NWaveNPerXdl
// clang-format on
using ReferenceConvFwdInstance =
ck::tensor_operation::host::ReferenceConvFwd_Bias_Activation<InDataType,
WeiDataType,
OutDataType,
InElementOp,
WeiElementOp,
OutElementOp>;
void PrintUseMsg()
{
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=n0, 1=yes)\n"
<< "Following arguments:\n"
<< " N, K, C, \n"
<< " <filter spatial dimensions>, (ie Y, X for 2D)\n"
<< " <input image spatial dimensions>, (ie Hi, Wi for 2D)\n"
<< " <strides>, (ie Sy, Sx for 2D)\n"
<< " <dilations>, (ie Dy, Dx for 2D)\n"
<< " <left padding>, (ie LeftPy, LeftPx for 2D)\n"
<< " <right padding>, (ie RightPy, RightPx for 2D)\n"
<< std::endl;
}
ck::utils::conv::ConvParams ParseConvParams(int argc, char* argv[])
{
// (N, K, C) + num_dim_spatial * 6 (filter, input, strides, dilations, pad left, pad right)
int num_dim_spatial = 2;
int conv_args = 3 + num_dim_spatial * 6;
int cmdline_nargs = conv_args + 4;
if(cmdline_nargs != argc)
{
PrintUseMsg();
exit(0);
}
ck::utils::conv::ConvParams params;
int arg_idx = 4;
params.num_dim_spatial_ = num_dim_spatial;
params.N_ = std::stoi(argv[arg_idx++]);
params.K_ = std::stoi(argv[arg_idx++]);
params.C_ = std::stoi(argv[arg_idx++]);
params.filter_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.filter_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.input_spatial_lengths_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_spatial_lengths_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_strides_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_strides_[i] = std::stoi(argv[arg_idx++]);
}
params.conv_filter_dilations_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.conv_filter_dilations_[i] = std::stoi(argv[arg_idx++]);
}
params.input_left_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_left_pads_[i] = std::stoi(argv[arg_idx++]);
}
params.input_right_pads_.resize(num_dim_spatial);
for(int i = 0; i < num_dim_spatial; ++i)
{
params.input_right_pads_[i] = std::stoi(argv[arg_idx++]);
}
return params;
}
} // anonymous namespace
int main(int argc, char* argv[])
{
using namespace ck::utils::conv;
bool do_verification = true;
int init_method = 1;
bool time_kernel = false;
const int num_dim_spatial = 2;
ck::utils::conv::ConvParams params;
if(argc >= 4)
{
do_verification = std::stoi(argv[1]);
init_method = std::stoi(argv[2]);
time_kernel = std::stoi(argv[3]);
}
if(argc >= 5)
{
params = ParseConvParams(argc, argv);
}
std::vector<std::size_t> input_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.C_)};
input_dims.insert(std::end(input_dims),
std::begin(params.input_spatial_lengths_),
std::end(params.input_spatial_lengths_));
std::vector<std::size_t> filter_dims{static_cast<std::size_t>(params.K_),
static_cast<std::size_t>(params.C_)};
filter_dims.insert(std::end(filter_dims),
std::begin(params.filter_spatial_lengths_),
std::end(params.filter_spatial_lengths_));
const std::vector<ck::index_t>& output_spatial_lengths = params.GetOutputSpatialLengths();
std::vector<std::size_t> output_dims{static_cast<std::size_t>(params.N_),
static_cast<std::size_t>(params.K_)};
output_dims.insert(std::end(output_dims),
std::begin(output_spatial_lengths),
std::end(output_spatial_lengths));
Tensor<InDataType> input(get_input_host_tensor_descriptor(input_dims, num_dim_spatial));
Tensor<WeiDataType> weights(get_filters_host_tensor_descriptor(filter_dims, num_dim_spatial));
Tensor<OutDataType> host_output(
get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
Tensor<OutDataType> device_output(
get_output_host_tensor_descriptor(output_dims, num_dim_spatial));
// bias: assume contiguous 1d vector
Tensor<OutDataType> bias(
HostTensorDescriptor(std::vector<std::size_t>({static_cast<std::size_t>(params.K_)})));
std::cout << "input: " << input.mDesc << std::endl;
std::cout << "weights: " << weights.mDesc << std::endl;
std::cout << "output: " << host_output.mDesc << std::endl;
std::cout << "bias: " << bias.mDesc << std::endl;
switch(init_method)
{
case 0: break;
case 1:
input.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
weights.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
bias.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
break;
default:
input.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
weights.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
bias.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
}
DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
DeviceMem out_device_buf(sizeof(OutDataType) * device_output.mDesc.GetElementSpace());
DeviceMem bias_device_buf(sizeof(OutDataType) * bias.mDesc.GetElementSpace());
in_device_buf.ToDevice(input.mData.data());
wei_device_buf.ToDevice(weights.mData.data());
bias_device_buf.ToDevice(bias.mData.data());
auto conv = DeviceConvFwdInstance{};
auto invoker = conv.MakeInvoker();
auto argument =
conv.MakeArgument(static_cast<const InDataType*>(in_device_buf.GetDeviceBuffer()),
static_cast<const WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
static_cast<const OutDataType*>(bias_device_buf.GetDeviceBuffer()),
params.N_,
params.K_,
params.C_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
if(!conv.IsSupportedArgument(argument))
{
throw std::runtime_error(
"wrong! device operator with the specified compilation parameters does "
"not support this problem");
}
float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
std::size_t flop = get_flops(
params.N_, params.C_, params.K_, params.filter_spatial_lengths_, output_spatial_lengths);
std::size_t num_btype =
get_btype<InDataType, WeiDataType, OutDataType>(params.N_,
params.C_,
params.K_,
params.input_spatial_lengths_,
params.filter_spatial_lengths_,
output_spatial_lengths) +
sizeof(OutDataType) * (params.K_);
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
<< std::endl;
if(do_verification)
{
auto ref_conv = ReferenceConvFwdInstance{};
auto ref_invoker = ref_conv.MakeInvoker();
auto ref_argument = ref_conv.MakeArgument(input,
weights,
host_output,
bias,
params.conv_filter_strides_,
params.conv_filter_dilations_,
params.input_left_pads_,
params.input_right_pads_,
InElementOp{},
WeiElementOp{},
OutElementOp{});
ref_invoker.Run(ref_argument);
out_device_buf.FromDevice(device_output.mData.data());
return ck::utils::check_err(device_output.mData, host_output.mData) ? 0 : 1;
}
return 0;
}
example/07_conv2d_fwd_bias_relu_add/CMakeLists.txt deleted 100644 → 0
View file @ 75af5450
# FIXME: should fix validation failure
add_example_executable_no_testing(example_conv2d_fwd_xdl_bias_relu_add conv2d_fwd_xdl_bias_relu_add.cpp)
target_link_libraries(example_conv2d_fwd_xdl_bias_relu_add PRIVATE conv_util)
example/07_conv2d_fwd_bias_relu_add/README.md deleted 100644 → 0
View file @ 75af5450
# Instructions for ```example_conv_xdl_bias_relu_add```
## Run ```example_conv_xdl_bias_relu_add```
```bash
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./bin/example_conv_xdl_bias_relu_add 0 1 5
```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
```
in_n_c_hi_wi: dim 4, lengths {128, 192, 71, 71}, strides {967872, 1, 13632, 192}
wei_k_c_y_x: dim 4, lengths {256, 192, 3, 3}, strides {1728, 1, 576, 192}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
bias_k: dim 1, lengths {256}, strides {1}
resi_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 5 times...
Perf: 1.44711 ms, 101.421 TFlops, 289.218 GB/s
```
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