Skip to content

Conversation

@anamikac-intel
Copy link

No description provided.

// 0t_mma_atom.md#TiledMMAs for more info. Sub-groups are arranged row-major (stride 4,1,0) for
// performance reasons.
using TiledMma = typename TiledMMAHelper<MMA_Atom<XE_DPAS_TT<8, float, cute::bfloat16_t>>, Layout<TileShape>, Layout<Shape<_8, _4, _1>, Stride<_4, _1, _0>>>::TiledMMA;
using TiledMma = typename TiledMMAHelper<MMA_Atom<XE_DPAS_NN<8, float, cute::bfloat16_t>>, Layout<TileShape>, Layout<Shape<_8, _4, _1>, Stride<_1, _4, _0>>>::TiledMMA;

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't stride be 8 instead of 4 ?

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes that's what I thought col major should be for subgroup layout... but Peter suggested this stride

@anamikac-intel anamikac-intel marked this pull request as ready for review December 4, 2025 09:10
@anamikac-intel anamikac-intel changed the title [WIP] Support column major MMA Atoms for Xe2. Support column major MMA Atoms for Xe2. Dec 4, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants