Skip to content

GitLab

"...python/kernel/test_tilelang_kernel_dequantize_gemm.py" did not exist on "34e0883db09c4a66827126aeac5cf0dd66b5f1ef"
Unverified Commit cd167e49 authored by Chao Liu's avatar Chao Liu Committed by GitHub
Browse files

Compile for gfx908 and gfx90a (#130) 

* adding compilation for multiple targets

* fix build

* clean

* update Jekinsfile

* update readme

* update Jenkins

* use ck::half_t instead of ushort for bf16

* rename enum classes

* clean

* rename

* clean
parent ecf337ba
Hide whitespace changes
Inline Side-by-side
Showing with 95 additions and 390 deletions
Jenkinsfile
View file @ cd167e49
...@@ -182,7 +182,7 @@ pipeline { ...@@ -182,7 +182,7 @@ pipeline {
{ {
agent { label rocmnode("nogpu")} agent { label rocmnode("nogpu")}
environment{ environment{
setup_args = """ -D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " -DBUILD_DEV=On """ setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
} }
steps{ steps{
buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release') buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release')
...@@ -192,7 +192,7 @@ pipeline { ...@@ -192,7 +192,7 @@ pipeline {
{ {
agent { label rocmnode("nogpu")} agent { label rocmnode("nogpu")}
environment{ environment{
setup_args = """ -D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " -DBUILD_DEV=On """ setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
} }
steps{ steps{
// until we stabilize debug build due to compiler crashes // until we stabilize debug build due to compiler crashes
...@@ -228,7 +228,7 @@ pipeline { ...@@ -228,7 +228,7 @@ pipeline {
{ {
agent{ label rocmnode("gfx908")} agent{ label rocmnode("gfx908")}
environment{ environment{
setup_args = """ -D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " -DBUILD_DEV=On """ setup_args = """ -D CMAKE_CXX_FLAGS="--offload-arch=gfx908 -O3 " -DBUILD_DEV=On """
} }
steps{ steps{
buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "check", no_reboot:true, build_type: 'Release') buildHipClangJobAndReboot(setup_args:setup_args, config_targets: "check", no_reboot:true, build_type: 'Release')
......
README.md
View file @ cd167e49
## Docker script
```bash
docker run \
-it \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908 and gfx90a
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS=" --offload-arch=gfx908 --offload-arch=gfx90a -O3 \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
### Build and Run Examples
```bash
make -j examples
```
Instructions for running each individual examples are under ```example/```
## Tests
```bash
make -j tests
make test
```
## Build ckProfiler
```bash
make -j ckProfiler
```
Instructions for running ckProfiler are under ```profiler/```
example/01_gemm/README.md
View file @ cd167e49
# Instructions for ```gemm_xdl``` Example # Instructions for ```example_gemm_xdl```
## Docker script ## Run ```example_gemm_xdl```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```gemm_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j gemm_xdl
```
## Run ```gemm_xdl```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
./example/gemm_xdl 0 1 5 ./bin/example_gemm_xdl 0 1 5
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
......
example/01_gemm/gemm_xdl_fp16.cpp
View file @ cd167e49
...@@ -40,7 +40,7 @@ using AElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -40,7 +40,7 @@ using AElementOp = ck::tensor_operation::element_wise::PassThrough;
using BElementOp = ck::tensor_operation::element_wise::PassThrough; using BElementOp = ck::tensor_operation::element_wise::PassThrough;
using CElementOp = ck::tensor_operation::element_wise::PassThrough; using CElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization_t::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
// clang-format off // clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
......
example/02_gemm_alpha_beta/README.md
View file @ cd167e49
# Instructions for ```gemm_xdl_alpha_beta``` Example # Instructions for ```example_gemm_xdl_alpha_beta```
## Docker script ## Run ```example_gemm_xdl_alpha_beta```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```gemm_xdl_alpha_beta```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j gemm_xdl_alpha_beta
```
## Run ```gemm_xdl_alpha_beta```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
./example/gemm_xdl_alpha_beta 1 1 1 0.5 0.5 ./bin/example_gemm_xdl_alpha_beta 1 1 1 0.5 0.5
``` ```
Result (MI100 @ 1502Mhz, 184.6TFlops peak FP16) Result (MI100 @ 1502Mhz, 184.6TFlops peak FP16)
``` ```
......
example/03_gemm_bias_relu/README.md
View file @ cd167e49
# Instructions for ```gemm_xdl_bias_relu_add``` Example # Instructions for ```example_gemm_xdl_bias_relu_add```
## Docker script ## Run ```example_gemm_xdl_bias_relu_add```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```gemm_xdl_bias_relu_add```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j gemm_xdl_bias_relu_add
```
## Run ```gemm_xdl_bias_relu_add```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC #arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC
./example/gemm_xdl_bias_relu_add 0 1 5 3840 4096 4096 4096 4096 4096 ./bin/example_gemm_xdl_bias_relu_add 0 1 5 3840 4096 4096 4096 4096 4096
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
......
example/04_gemm_bias_relu_add/README.md
View file @ cd167e49
# Instructions for ```gemm_xdl_bias_relu_add``` Example # Instructions for ```example_gemm_xdl_bias_relu_add```
## Docker script ## Run ```example_gemm_xdl_bias_relu_add```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```gemm_xdl_bias_relu_add```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j gemm_xdl_bias_relu_add
```
## Run ```gemm_xdl_bias_relu_add```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC #arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC
./example/gemm_xdl_bias_relu_add 0 1 5 3840 4096 4096 4096 4096 4096 ./bin/example_gemm_xdl_bias_relu_add 0 1 5 3840 4096 4096 4096 4096 4096
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
......
example/05_conv2d_fwd/README.md
View file @ cd167e49
# Instructions for ```conv2d_fwd_xdl``` Example # Instructions for ```example_conv2d_fwd_xdl```
## Docker script ## Run ```example_conv2d_fwd_xdl```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv2d_fwd_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv2d_fwd_xdl
```
## Run ```conv2d_fwd_xdl```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx #arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./example/conv2d_fwd_xdl 0 1 5 ./bin/example_conv2d_fwd_xdl 0 1 5
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
......
example/05_conv2d_fwd/conv2d_fwd_xdl_fp16.cpp
View file @ cd167e49
...@@ -34,7 +34,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -34,7 +34,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough; using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
using DeviceConvFwdInstance = ck::tensor_operation::device:: using DeviceConvFwdInstance = ck::tensor_operation::device::
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
......
example/05_conv2d_fwd/conv2d_fwd_xdl_int8.cpp
View file @ cd167e49
...@@ -35,7 +35,7 @@ using OutElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -35,7 +35,7 @@ using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
using PassThrough = ck::tensor_operation::element_wise::PassThrough; using PassThrough = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
using DeviceConvFwdInstance = ck::tensor_operation::device:: using DeviceConvFwdInstance = ck::tensor_operation::device::
DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
......
example/06_conv2d_fwd_bias_relu/README.md
View file @ cd167e49
# Instructions for ```conv_xdl_bias_relu_add``` Example # Instructions for ```example_conv_xdl_bias_relu```
## Docker script ## Run ```example_conv_xdl_bias_relu```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv_xdl_bias_relu_add```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv_xdl_bias_relu_add
```
## Run ```conv_xdl_bias_relu_add```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx #arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./example/conv_xdl_bias_relu_add 0 1 5 ./bin/example_conv_xdl_bias_relu 0 1 5
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
...@@ -48,14 +15,8 @@ in_n_c_hi_wi: dim 4, lengths {128, 192, 71, 71}, strides {967872, 1, 13632, 192} ...@@ -48,14 +15,8 @@ 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} 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} 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} 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}
arg.a_grid_desc_k0_m_k1_{216, 165888, 8}
arg.b_grid_desc_k0_n_k1_{216, 256, 8}
arg.c_grid_desc_m_n_{ 165888, 256}
arg.c0_grid_desc_m_n_{ 165888, 256}
arg.c1_grid_desc_m_n_{ 165888, 256}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1} launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up Warm up
Start running 5 times... Start running 5 times...
Perf: 1.71779 ms, 85.4396 TFlops, 194.2 GB/s Perf: 1.39009 ms, 105.581 TFlops, 239.981 GB/s
``` ```
example/06_conv2d_fwd_bias_relu/conv2d_fwd_xdl_bias_relu.cpp
View file @ cd167e49
...@@ -32,10 +32,10 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -32,10 +32,10 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::AddRelu; using OutElementOp = ck::tensor_operation::element_wise::AddRelu;
static constexpr auto MemorySet = ck::InMemoryDataOperationEnum_t::Set; static constexpr auto MemorySet = ck::InMemoryDataOperationEnum::Set;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
// clang-format off // clang-format off
using DeviceConvFwdInstance = ck::tensor_operation::device:: using DeviceConvFwdInstance = ck::tensor_operation::device::
......
example/07_conv2d_fwd_bias_relu_add/README.md
View file @ cd167e49
# Instructions for ```conv_xdl_bias_relu_add``` Example # Instructions for ```example_conv_xdl_bias_relu_add```
## Docker script
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv_xdl_bias_relu_add```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv_xdl_bias_relu_add
```
## Run ```conv_xdl_bias_relu_add``` ## Run ```example_conv_xdl_bias_relu_add```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx #arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./example/conv_xdl_bias_relu_add 0 1 5 ./bin/example_conv_xdl_bias_relu_add 0 1 5
``` ```
Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
...@@ -49,13 +17,8 @@ wei_k_c_y_x: dim 4, lengths {256, 192, 3, 3}, strides {1728, 1, 576, 192} ...@@ -49,13 +17,8 @@ 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} 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} 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} resi_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
arg.a_grid_desc_k0_m_k1_{216, 165888, 8}
arg.b_grid_desc_k0_n_k1_{216, 256, 8}
arg.c_grid_desc_m_n_{ 165888, 256}
arg.c0_grid_desc_m_n_{ 165888, 256}
arg.c1_grid_desc_m_n_{ 165888, 256}
launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1} launch_and_time_kernel: grid_dim {1296, 1, 1}, block_dim {256, 1, 1}
Warm up Warm up
Start running 5 times... Start running 5 times...
Perf: 1.71779 ms, 85.4396 TFlops, 194.2 GB/s Perf: 1.44711 ms, 101.421 TFlops, 289.218 GB/s
``` ```
example/07_conv2d_fwd_bias_relu_add/conv2d_fwd_xdl_bias_relu_add.cpp
View file @ cd167e49
...@@ -33,7 +33,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -33,7 +33,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd; using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
// clang-format off // clang-format off
using DeviceConvFwdInstance = ck::tensor_operation::device:: using DeviceConvFwdInstance = ck::tensor_operation::device::
......
example/08_conv3d_fwd/README.md
View file @ cd167e49
# Instructions for ```conv3d_fwd_xdl``` Example # Instructions for ```example_conv3d_fwd_xdl```
## Docker script ## Run ```example_conv3d_fwd_xdl```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv3d_fwd_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv3d_fwd_xdl
```
## Run ```conv3d_fwd_xdl```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 24: N, K, C, Z, Y, X, Di, Hi, Wi, Sz, Sy, Sx, Dz, Dy, Dx, leftPz, LeftPy, LeftPx, RightPz, RightPy, RightPx #arg4 to 24: N, K, C, Z, Y, X, Di, Hi, Wi, Sz, Sy, Sx, Dz, Dy, Dx, leftPz, LeftPy, LeftPx, RightPz, RightPy, RightPx
./example/conv3d_fwd_xdl 0 1 5 ./bin/example_conv3d_fwd_xdl 0 1 5
``` ```
Result (MI100 dynamic frequency) Result (MI100 @ 1087Mhz, 133.5TFlops peak FP16)
``` ```
in: dim 5, lengths {4, 71, 71, 71, 192}, strides {68718912, 967872, 13632, 192, 1}
wei: dim 5, lengths {256, 3, 3, 3, 192}, strides {5184, 1728, 576, 192, 1} wei: dim 5, lengths {256, 3, 3, 3, 192}, strides {5184, 1728, 576, 192, 1}
out: dim 5, lengths {4, 36, 36, 36, 256}, strides {11943936, 331776, 9216, 256, 1} out: dim 5, lengths {4, 36, 36, 36, 256}, strides {11943936, 331776, 9216, 256, 1}
a_grid_desc_b_k0_m_k1{1, 648, 186624, 8} num_batches_of_GEMM = 1
b_grid_desc_b_k0_n_k1{1, 648, 256, 8} a_grid_desc_k0_m_k1{648, 186624, 8}
b_grid_desc_k0_n_k1{648, 256, 8}
c_grid_desc_m_n{ 186624, 256}
launch_and_time_kernel: grid_dim {1458, 1, 1}, block_dim {256, 1, 1} launch_and_time_kernel: grid_dim {1458, 1, 1}, block_dim {256, 1, 1}
Warm up Warm up
Start running 5 times... Start running 5 times...
Perf: 4.49466 ms, 110.206 TFlops, 144.161 GB/s Perf: 4.58795 ms, 107.965 TFlops, 141.23 GB/s
``` ```
example/08_conv3d_fwd/conv3d_fwd_xdl.cpp
View file @ cd167e49
...@@ -37,7 +37,7 @@ using WeiLayout = ck::tensor_layout::convolution::KZYXC; ...@@ -37,7 +37,7 @@ using WeiLayout = ck::tensor_layout::convolution::KZYXC;
using OutLayout = ck::tensor_layout::convolution::NDHWK; using OutLayout = ck::tensor_layout::convolution::NDHWK;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
using DeviceConv3dFwdInstance = ck::tensor_operation::device:: using DeviceConv3dFwdInstance = ck::tensor_operation::device::
DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K< DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_K<
......
example/09_convnd_fwd/README.md
View file @ cd167e49
# Instructions for ```convnd_fwd_xdl``` Example # Instructions for ```example_convnd_fwd_xdl```
## Docker script ## Run ```example_convnd_fwd_xdl```
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```convnd_fwd_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j convnd_fwd_xdl
```
## Run ```convnd_fwd_xdl```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
...@@ -47,7 +14,7 @@ cmake \ ...@@ -47,7 +14,7 @@ cmake \
# <dilations>, (ie Dy, Dx for 2D) # <dilations>, (ie Dy, Dx for 2D)
# <left padding>, (ie LeftPy, LeftPx for 2D) # <left padding>, (ie LeftPy, LeftPx for 2D)
# <right padding>, (ie RightPy, RightPx for 2D) # <right padding>, (ie RightPy, RightPx for 2D)
./example/convnd_fwd_xdl 0 1 100 ./bin/example_convnd_fwd_xdl 0 1 100
``` ```
Result (MI100 @ 1087Mhz, 33.4TFlops peak FP32) Result (MI100 @ 1087Mhz, 33.4TFlops peak FP32)
......
example/09_convnd_fwd/convnd_fwd_xdl.cpp
View file @ cd167e49
...@@ -26,7 +26,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -26,7 +26,7 @@ using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough; using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvFwdDefault = static constexpr auto ConvFwdDefault =
ck::tensor_operation::device::ConvolutionForwardSpecialization_t::Default; ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
using DeviceConvFwdBasePtr = using DeviceConvFwdBasePtr =
ck::tensor_operation::device::DeviceConvFwdPtr<InElementOp, WeiElementOp, OutElementOp>; ck::tensor_operation::device::DeviceConvFwdPtr<InElementOp, WeiElementOp, OutElementOp>;
......
example/10_conv2d_bwd_data/README.md
View file @ cd167e49
# Instructions for ```conv2d_bwd_data_xdl``` Example # Instructions for ```example_conv2d_bwd_data_xdl``` Example
## Docker script
```bash
docker run \
-it \
--rm \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
rocm/tensorflow:rocm4.3.1-tf2.6-dev \
/bin/bash
```
## Build ```conv2d_bwd_data_xdl```
```bash
mkdir build && cd build
```
```bash
# Need to specify target ID, example below is gfx908
cmake \
-D BUILD_DEV=OFF \
-D CMAKE_BUILD_TYPE=Release \
-D CMAKE_CXX_FLAGS="-DCK_AMD_GPU_GFX908 --amdgpu-target=gfx908 -O3 " \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_PREFIX_PATH=/opt/rocm \
..
```
```bash
make -j conv2d_bwd_data_xdl
```
## Run ```conv2d_bwd_data_xdl``` ## Run ```example_conv2d_bwd_data_xdl```
```bash ```bash
#arg1: verification (0=no, 1=yes) #arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value) #arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1) #arg3: run kernel # of times (>1)
#arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx #arg4 to 18: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx
./bin/conv2d_bwd_data_xdl 0 1 5 ./bin/example_conv2d_bwd_data_xdl 0 1 5
``` ```
Result Result
......
example/10_conv2d_bwd_data/conv2d_bwd_data_xdl.cpp
View file @ cd167e49
...@@ -27,7 +27,7 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough; ...@@ -27,7 +27,7 @@ using InElementOp = ck::tensor_operation::element_wise::PassThrough;
using WeiElementOp = ck::tensor_operation::element_wise::PassThrough; using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
using OutElementOp = ck::tensor_operation::element_wise::PassThrough; using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
static constexpr auto ConvBwdDefault = static constexpr auto ConvBwdDefault =
ck::tensor_operation::device::ConvolutionBackwardDataSpecialization_t::Default; ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
using DeviceConvBwdDataInstance = ck::tensor_operation::device:: using DeviceConvBwdDataInstance = ck::tensor_operation::device::
DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K< DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K<
...@@ -38,7 +38,7 @@ using DeviceConvBwdDataInstance = ck::tensor_operation::device:: ...@@ -38,7 +38,7 @@ using DeviceConvBwdDataInstance = ck::tensor_operation::device::
InElementOp, // InElementwiseOperation InElementOp, // InElementwiseOperation
WeiElementOp, // WeiElementwiseOperation WeiElementOp, // WeiElementwiseOperation
OutElementOp, // OutElementwiseOperation OutElementOp, // OutElementwiseOperation
ConvBwdDefault, // ConvolutionBackwardDataSpecialization_t ConvBwdDefault, // ConvolutionBackwardDataSpecialization
256, // BlockSize 256, // BlockSize
128, // MPerBlock 128, // MPerBlock
128, // NPerBlock 128, // NPerBlock
......
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