Merge commit 'de6a9590abe907283e189abba1b487f8e5562d1b' into develop

This commit is contained in:
assistant-librarian[bot]
2025-11-24 21:29:18 +00:00
parent 5297edb40c
commit 4aaa8c92bb
4 changed files with 52 additions and 17 deletions

View File

@@ -227,9 +227,15 @@ struct UniversalFlatmmPipelineAgBgCrPolicy
}
template <typename Problem>
CK_TILE_HOST_DEVICE static constexpr auto GetSmemPackA()
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemPackA()
{
return Problem::VectorLoadSize / sizeof(typename Problem::ADataType);
using A = remove_cvref_t<typename Problem::ADataType>;
using BlockGemm = remove_cvref_t<decltype(GetBlockGemm<Problem>())>;
constexpr index_t KPack = static_cast<index_t>(BlockGemm::Traits::KPack);
constexpr index_t VecElems = static_cast<index_t>(Problem::VectorLoadSize / sizeof(A));
return (KPack < VecElems) ? KPack : VecElems;
}
template <typename Problem>