Merge commit '7cb1f30cfb6045bccbbd484c5e8e4715e2ebc2f3' into develop

This commit is contained in:
assistant-librarian[bot]
2025-10-01 14:12:07 +00:00
parent f10bae28c9
commit 58353e999c

View File

@@ -1586,7 +1586,7 @@ struct ConvBwdDataImplicitGemmOutTransform
Tuple<index_t, index_t, index_t, index_t>
low_lengths_magic_divisor_shift_; // XDotSlice_K_, K_, TildeSlice_, WTildeSlice_
__host__ __device__ constexpr ConvBwdDataImplicitGemmOutTransform() = default;
__host__ __device__ ConvBwdDataImplicitGemmOutTransform() = default;
__host__ __device__ constexpr ConvBwdDataImplicitGemmOutTransform(index_t N,
index_t Ho,
@@ -1645,7 +1645,7 @@ struct ConvBwdDataImplicitGemmOutTransform
template <typename UpIdx>
__host__ __device__ constexpr auto CalculateLowerIndexN(const UpIdx& idx_up) const
{
index_t NStep, HStep, WStep;
index_t NStep{0}, HStep{0}, WStep{0};
// Merge
// NStep = M_id / TildeSlice_
NStep = MagicDivision::DoMagicDivision(idx_up[I1],