ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-15 03:30:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

simplify code

Browse Source
This commit is contained in:
Lin, Qun
2025-06-11 00:39:42 -05:00
parent 88992a7856
commit 005d1c0b8c
1 changed files with 8 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
example/30_grouped_conv_fwd_multiple_d/device_grouped_conv_fwd_dl_v4.hpp
View File

@@ -324,17 +324,18 @@ struct GridwiseGroupedConv2DFwdDlV4
{
static_for<0, SubTileW, 2>{}([&](auto wo) {
InData2 indata;
//InData2 indata;
static_for<0, Filter_Y, 1>{}([&](auto y) {
#if 1
static_for<0, Filter_X_Pack, 1>{}([&](auto x_pack) {
constexpr auto in_idx0 = (wo + x_pack * 2)/ InScalarPerVector_Internal;
constexpr auto in_idx1 = (wo + x_pack * 2)% InScalarPerVector_Internal;
indata[0] = tmp_in[(ho + y) % Filter_Y][in_idx0][in_idx1];
indata[1] = tmp_in[(ho + y) % Filter_Y][in_idx0][in_idx1 + 1];
//constexpr auto in_idx0 = (wo + x_pack * 2)/ InScalarPerVector_Internal;
//constexpr auto in_idx1 = (wo + x_pack * 2)% InScalarPerVector_Internal;
//indata[0] = tmp_in[(ho + y) % Filter_Y][in_idx0][in_idx1];
//indata[1] = tmp_in[(ho + y) % Filter_Y][in_idx0][in_idx1 + 1];
//ignore = p_in;
inner_product(indata, p_wei_even[y * Filter_X_Pack + x_pack], tmp_out[wo.value]);
inner_product(indata, p_wei_odd[y * Filter_X_Pack + x_pack], tmp_out[wo.value + 1]);
const InData2* p_in = reinterpret_cast<InData2*>(tmp_in[(ho + y) % Filter_Y]) + wo * Stride_W / 2+ x_pack;
inner_product(*p_in, p_wei_even[y * Filter_X_Pack + x_pack], tmp_out[wo.value]);
inner_product(*p_in, p_wei_odd[y * Filter_X_Pack + x_pack], tmp_out[wo.value + 1]);
//tmp_out[wo.value] += type_convert<float>( indata[0]) * type_convert<float>( p_wei_even[y * Filter_X_Pack + x_pack][0]) +
// type_convert<float>( indata[1]) * type_convert<float>( p_wei_even[y * Filter_X_Pack + x_pack][1]);
//tmp_out[wo.value + 1] += type_convert<float>( indata[0]) * type_convert<float>( p_wei_odd[y * Filter_X_Pack + x_pack][0]) +