ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 13:11:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add batched attention special kernel instances (#424)

Browse Source 
* sanity check

* add attribution

* add irrgular k tile size for batched attention

* format
This commit is contained in:
Anthony Chang
2022-09-20 08:20:54 +08:00
committed by GitHub
parent c6b8b472a7
commit 7c788e10ce
5 changed files with 51 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
View File

@@ -649,6 +649,9 @@ struct BlockwiseGemmXdlops_v2
static constexpr index_t MWaves = MPerBlock / (MRepeat * MPerXDL);
static constexpr index_t NWaves = NPerBlock / (NRepeat * NPerXDL);
static_assert(KPerThread % KPack == 0,
"Wrong KPack setting; try increasing KPerThread or decreasing KPack");
StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr,
FloatAcc,
MRepeat * NRepeat,