Skip to content

For OuterProductAccumulate, matrix layout must be outerproductoptimal and matrix stride must be zero #7412

Closed
anupamachandra wants to merge 7 commits intomicrosoft:staging-sm6.9from
anupamachandra:anupamac/outer-prod-acc-matrix-layout
Closed

For OuterProductAccumulate, matrix layout must be outerproductoptimal and matrix stride must be zero #7412
anupamachandra wants to merge 7 commits intomicrosoft:staging-sm6.9from
anupamachandra:anupamac/outer-prod-acc-matrix-layout

Conversation

@anupamachandra
Copy link
Copy Markdown
Collaborator

Implements the DXIL portion of microsoft/hlsl-specs#494, The HLSL checks will be a part of the HLSL validation checks, did not add it to this PR due to shared infrastructure.

@anupamachandra anupamachandra changed the base branch from staging-sm6.9 to main May 1, 2025 16:45
@anupamachandra anupamachandra changed the base branch from main to staging-sm6.9 May 1, 2025 16:46
Comment on lines +11 to +44
enum CompType {
Invalid = 0,
I1 = 1,
I16 = 2,
U16 = 3,
I32 = 4,
U32 = 5,
I64 = 6,
U64 = 7,
F16 = 8,
F32 = 9,
F64 = 10,
SNormF16 = 11,
UNormF16 = 12,
SNormF32 = 13,
UNormF32 = 14,
SNormF64 = 15,
UNormF64 = 16,
PackedS8x32 = 17,
PackedU8x32 = 18,

// BEGIN NEW FOR SM 6.9
U8 = 19,
I8 = 20,
F8_E4M3 = 21,
F8_E5M2 = 22,
};

enum MatLayout {
RowMajor = 0,
ColumnMajor = 1,
MulOptimal = 2,
OuterProductOptimal = 3,
};
Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

I'm not super keen on having the full enums duplicated here. I may have missed a discussion about it though...

Since only one value is could we instead do something like:

  const uint matrix_interpretation = 8; // ComponentType::F16;

That way grepping through the source code will find the right things.

@anupamachandra
Copy link
Copy Markdown
Collaborator Author

Opened a new PR targetting master branch: #7417

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Archived in project

Development

Successfully merging this pull request may close these issues.

2 participants