ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 09:45:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

tweaked params for direct conv; added a dummy winograd

Browse Source
This commit is contained in:
Chao Liu
2018-11-24 03:55:36 -06:00
parent dbffe05a98
commit 8732ea04fb
8 changed files with 529 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/include/threadwise_direct_convolution.cuh
View File

@@ -44,17 +44,11 @@ __device__ void threadwise_direct_convolution(InDesc,
const unsigned hi = ho + s;
const unsigned wi = wo + r;
const unsigned in_index =
in_desc.GetStride(I0) * n + in_desc.GetStride(I1) * c +
in_desc.GetStride(I2) * hi + in_desc.GetStride(I3) * wi;
const unsigned in_index = in_desc.Get1dIndex(n, c, hi, wi);
const unsigned wei_index =
wei_desc.GetStride(I0) * k + wei_desc.GetStride(I1) * c +
wei_desc.GetStride(I2) * s + in_desc.GetStride(I3) * r;
const unsigned wei_index = wei_desc.Get1dIndex(k, c, s, r);
const unsigned out_index =
out_desc.GetStride(I0) * n + out_desc.GetStride(I1) * k +
out_desc.GetStride(I2) * ho + out_desc.GetStride(I3) * wo;
const unsigned out_index = out_desc.Get1dIndex(n, k, ho, wo);
p_out[out_index] += p_wei[wei_index] * p_in[in_index];