ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-12 09:16:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

tuning on vega 20

Browse Source
This commit is contained in:
Chao Liu
2019-04-25 17:28:59 -05:00
parent a903146427
commit 2603bb0fe3
24 changed files with 659 additions and 654 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
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);
}
}
@@ -425,13 +425,13 @@ int main(int argc, char* argv[])
constexpr index_t WPad = 0;
#elif 0
// 3x3, 56x56
constexpr index_t N = 64;
constexpr index_t C = 64;
constexpr index_t N = 64;
constexpr index_t C = 64;
constexpr index_t HI = 56;
constexpr index_t WI = 56;
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;
@@ -603,9 +603,9 @@ int main(int argc, char* argv[])
device_direct_convolution_2_nchw_kcyx_nkhw
#elif 0
device_direct_convolution_2_vectorized_nchw_kcyx_nkhw
#elif 0
#elif 1
device_convolution_implicit_gemm_v1_chwn_cyxk_khwn
#elif 0
#elif 1
device_convolution_implicit_gemm_v1_nchw_cyxk_khwn
#elif 1
device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw