ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-20 04:49:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Grouped Gemm with looping over the tiles. (#788)" (#982)

Browse Source 
This reverts commit 43fe5037d4ff9d07365e5d3b8f5b31676a8ff9da.

[ROCm/composable_kernel commit: c99323be6e]
This commit is contained in:
zjing14
2023-10-11 14:27:29 -05:00
committed by GitHub
parent 34b77070f3
commit 91e1cf6750
21 changed files with 517 additions and 854 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
View File

@@ -14,27 +14,27 @@ namespace ck {
namespace tensor_operation {
namespace host {
///
/// @brief Reference implementation for forward convolution.
///
/// @paragraph
/// Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
/// Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
/// as long as dimensions in tensor descriptor is in GNCHW order
///
/// @tparam InDataType Input tensor data type.
/// @tparam WeiDataType Weights tensor data type.
/// @tparam OutDataType Output tensor data type.
/// @tparam InElementwiseOperation Functor for input tensor elementwise
/// operation.
/// @tparam WeiElementwiseOperation Functor for weights tensor elementwise
/// operation.
/// @tparam NDimSpatial Number of spatial dimensions.
///
/// input descriptor in [G, N, C, Do, Ho, Wo] order
/// weight descriptor in [G, K, C, Z, Y, X] order
/// output descriptor in [G, N, K, Di, Hi, Wi] order
/// phyiscal layout is irrelavent
//
// @brief Reference implementation for forward convolution.
//
// @paragraph
// Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
// Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
// as long as dimensions in tensor descriptor is in GNCHW order
//
// @tparam InDataType Input tensor data type.
// @tparam WeiDataType Weights tensor data type.
// @tparam OutDataType Output tensor data type.
// @tparam InElementwiseOperation Functor for input tensor elementwise
// operation.
// @tparam WeiElementwiseOperation Functor for weights tensor elementwise
// operation.
// @tparam NDimSpatial Number of spatial dimensions.
//
// input descriptor in [G, N, C, Do, Ho, Wo] order
// weight descriptor in [G, K, C, Z, Y, X] order
// output descriptor in [G, N, K, Di, Hi, Wi] order
// phyiscal layout is irrelavent
template <ck::index_t NDimSpatial,
typename InDataType,
typename WeiDataType,