mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-03 05:01:25 +00:00
0536f2b312
* start convnd bwd data * add 3d laoyout name * add conv1d reference * add con3d reference * finished example client code * conv1d kernel finished * fix input error * add conv3d * add 3d layout in conv_utils.hpp * fix sepecial check * addconvnd lib * add test for bwd data * finished test * add check slice length * convnd bwd data start * profiler can be compiled * fix some bug * set input to zero * modify readme for example * fix test_convnd_bwd_data bug * test_convnd_bwd_data parameter desc * workaround for 1d * workaroud for 2d * change init value * workaround for 3d int8 * fix init value bug * remove workaround * fix acc data type * add int32 * change select function to template * tilda to tilde * remove int32 instance * fix commit for device hpp * fix comments for profiler * using profile imp to test * add pass verification * fix conv2d reference * fix conflict * remove double batched_gemm * fix exampel conv2d data and test convnd * format * change conv2d_bwd_data return value * remove repeat = 1 * remove conv bwd data Co-authored-by: ltqin <letaoqin@amd.com> Co-authored-by: Chao Liu <chao.liu2@amd.com>
Instructions for convnd_bwd_data_xdl Example
Docker script
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_bwd_data_xdl
mkdir build && cd build
# 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 \
..
make -j convnd_bwd_data_xdl
Run example_convnd_bwd_data_xdl
#arg1: verification (0=no, 1=yes)
#arg2: initialization (0=no init, 1=integer value, 2=decimal value)
#arg3: run kernel # of times (>1)
#arg4: num_dim_spatial(1|2|3)
#arg5 to ...: N, K, C, [Z,] [Y,] X, [Di,] [Hi,] Wi, S[z,] [Sy,] Sx, [Dz,] [Dy,] Dx, [LeftPz,] [LeftPy,] LeftPx, [RightPy,] [RightPy,] RightPx
./bin/convnd_bwd_data_xdl 0 1 5
Result
in_n_c_hi_wi: dim 4, lengths {128, 128, 71, 71}, strides {645248, 1, 9088, 128}
wei_k_c_y_x: dim 4, lengths {256, 128, 3, 3}, strides {1152, 1, 384, 128}
out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
arg.a_grid_desc_k0_m_k1_container_{128, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{128, 128, 8}
arg.c_grid_desc_m_n_container_{ 175232, 128}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 2, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {1369, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 128, 8}
arg.c_grid_desc_m_n_container_{ 175232, 128}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 2, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {1369, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{64, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{64, 128, 8}
arg.c_grid_desc_m_n_container_{ 175232, 128}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 2, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {1369, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
arg.a_grid_desc_k0_m_k1_container_{32, 175232, 8}
arg.b_grid_desc_k0_n_k1_container_{32, 128, 8}
arg.c_grid_desc_m_n_container_{ 175232, 128}
arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( 2738, 2, 2, 2, 4, 2 )
launch_and_time_kernel: grid_dim {1369, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 1 times...
Perf: 1.40031 ms, 69.8734 TFlops, 179.037 GB/s