ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 21:51:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

NHWC Conv2d Bwd weight fp16 ckprofiler and test (#166)

Browse Source 
* change backward weight name

* start add bwd weight lib and profiler

* change tuning paramter

* change output info

* add bwd weight test

* change test info

* using conv_util

* change wgt to weight

* add }

* add fp32
This commit is contained in:
ltqin
2022-04-05 09:32:00 +08:00
committed by GitHub
parent 82c8b9f8ee
commit 781cacd2e6
19 changed files with 814 additions and 42 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
example/11_conv2d_bwd_weight/README.md Normal file
View File

@@ -0,0 +1,25 @@
# Instructions for ```example_conv2d_bwd_weight_xdl``` Example
## Run ```example_conv2d_bwd_weight_xdl```
```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: is show log (0=no, 1=yes)
#arg5 to 19: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, RightPx, split-k
./bin/example_conv2d_bwd_weight_xdl 0 1 5 0 4
```
Result
```
in_n_c_hi_wi: dim 4, lengths {128, 1024, 14, 14}, strides {200704, 1, 14336, 1024}
wei_k_c_y_x: dim 4, lengths {256, 1024, 3, 3}, strides {9216, 1, 3072, 1024}
out_n_k_ho_wo: dim 4, lengths {128, 256, 6, 6}, strides {9216, 1, 1536, 256}
arg.a_grid_desc_kbatch_k0_m_k1_{4, 144, 256, 8}
arg.b_grid_desc_kbatch_k0_n_k1_{4, 144, 9216, 8}
arg.c_grid_desc_m_n_{ 256, 9216}
launch_and_time_kernel: grid_dim {576, 1, 1}, block_dim {256, 1, 1}
Warm up
Start running 5 times...
Perf: 0.401084 ms, 54.2112 TFlops, 145.75 GB/s
```