ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 01:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

v1r3 nchw*cyxk=nkhw lds double buffer

Browse Source
This commit is contained in:
Chao Liu
2019-04-27 15:06:15 -05:00
parent 63cdc6d2a4
commit d2436a58c4
7 changed files with 729 additions and 158 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
driver/driver.hip.cpp
View File

@@ -371,7 +371,7 @@ void host_winograd_3x3_convolution(const Tensor<TIn>& in_nchw,
std::size_t ho = HoPerTile * htile + j;
for(int i = 0; i < WoPerTile; ++i)
{
std::size_t wo = WoPerTile * wtile + i;
std::size_t wo = WoPerTile * wtile + i;
out_nkhw(n, k, ho, wo) = out_hold(n, k, htile, wtile, j, i);
}
}
@@ -413,13 +413,13 @@ int main(int argc, char* argv[])
{
#if 1
// 3x3, 34x34
constexpr index_t N = 64;
constexpr index_t C = 256;
constexpr index_t N = 64;
constexpr index_t C = 256;
constexpr index_t HI = 34;
constexpr index_t WI = 34;
constexpr index_t K = 128;
constexpr index_t Y = 3;
constexpr index_t X = 3;
constexpr index_t K = 128;
constexpr index_t Y = 3;
constexpr index_t X = 3;
constexpr index_t HPad = 0;
constexpr index_t WPad = 0;