ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 04:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Solve the CTAD regression & add up the Shell file for the docker management in testing (#3634)

Browse Source 
* Finished the work

* Fix the pipeline
This commit is contained in:
Thomas Ning
2026-01-26 10:29:28 -08:00
committed by GitHub
parent 834642202c
commit 3900e1e7ce
12 changed files with 1150 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck_tile/ops/flatmm/pipeline/mx_flatmm_pipeline_agmem_bgmem_creg_v1_policy.hpp
View File

@@ -227,7 +227,7 @@ struct MXFlatmmPipelineAgBgCrPolicy : UniversalFlatmmPipelineAgBgCrPolicy
sequence<1>>{});
else
return make_static_tile_distribution(
tile_distribution_encoding< //
tile_distribution_encoding<
sequence<NWarps>,
tuple<sequence<MWarps, MXdlPack, MPerXdl>,
sequence<K_Thread / AK1, K_Lane, AK1 / APackedSize>>,

4
include/ck_tile/ops/reduce/block/block_reduce.hpp
View File

@@ -392,8 +392,4 @@ struct BlockReduce2D
InDataType reduce_init;
};
// deduction guide
template <typename T>
CK_TILE_HOST_DEVICE_EXTERN BlockReduce2D(const T&, const typename T::DataType&) -> BlockReduce2D<T>;
} // namespace ck_tile

2
include/ck_tile/ops/softmax/block/block_softmax_2d.hpp
View File

@@ -40,7 +40,7 @@ struct BlockSoftmax2D
#endif
// compute row max
auto reduce_row_max = BlockReduce2D{x, -numeric<DataType>::infinity()};
auto reduce_row_max = BlockReduce2D<decltype(x)>{x, -numeric<DataType>::infinity()};
#if _BLOCK_SOFTMAX_USE_UNPACK2
auto row_max = reduce_row_max(f_max3, f_max, sequence<1, 2>{});
#else