[ROCm/composable_kernel commit: f7be86b9e4]
This commit is contained in:
Chao Liu
2019-09-16 22:47:55 -05:00
parent ff99bc281e
commit 2b26fb76ca
6 changed files with 110 additions and 87 deletions

View File

@@ -70,7 +70,7 @@ void device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw(InDesc,
using InBlockCopySubLengths_E_N1_B_N2 = Sequence<1, 1, 1, 4>;
using InBlockCopyClusterLengths_E_N1_B_N2 = Sequence<8, 2, 16, 1>;
using InBlockCopyThreadClusterArrangeOrder = Sequence<0, 1, 3, 2>; // [E, N1, N2, B]
using InBlockCopySrcAccessOrder = Sequence<0, 1, 3, 2>; // [E, N1, N2, B]
using InBlockCopySrcAccessOrder = Sequence<0, 2, 1, 3>; // [E, B, N1, N2]
using InBlockCopyDstAccessOrder = Sequence<0, 1, 2, 3>; // [E, N1, B, N2]
constexpr index_t InBlockCopySrcDataPerRead_B = 1;

View File

@@ -74,7 +74,7 @@ void device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw_padded(InDesc,
using InBlockCopySubLengths_E_N1_B_N2 = Sequence<1, 1, 1, 4>;
using InBlockCopyClusterLengths_E_N1_B_N2 = Sequence<8, 2, 16, 1>;
using InBlockCopyThreadClusterArrangeOrder = Sequence<0, 1, 3, 2>; // [E, N1, N2, B]
using InBlockCopySrcAccessOrder = Sequence<0, 1, 3, 2>; // [E, N1, N2, B]
using InBlockCopySrcAccessOrder = Sequence<0, 2, 1, 3>; // [E, B, N1, N2]
using InBlockCopyDstAccessOrder = Sequence<0, 1, 2, 3>; // [E, N1, B, N2]
constexpr index_t InBlockCopySrcDataPerRead_B = 1;