Skip to content

[0035] Add missing SumAccumulate OpCode #690

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[0035] Add missing SumAccumulate OpCode #690

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions proposals/0035-linalg-matrix.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1149,9 +1149,10 @@ declare void @dx.op.matrixOp(
)
```

Two opcodes are available for this operation class, one for multiplying matrices
and storing the result as `C = A * B`. The second for multiply accumulation `C
+= A * B`.
Three opcodes are available for this operation class:
* Matrix Matrix Multiply: `C = A * B`
* Matrix Matrix Multiply with Accumulation: `C += A * B`
* Matrix Matrix Addition with Accumulation: `C += A + B`
Copy link
Collaborator

Choose a reason for hiding this comment

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

When I look up for the operation class, I see it would be OpCodeClass::MatrixOp.

Isn't "MatrixOp" a little too generic? The operation is limited to accepting three matrix operands, where it overwrites or accumulates into the first one. On that note, I think the operation class should be all you need to know to reason about the type of memory accesses for each matrix, but they are different for multiply vs. multiply or add with accumulate.

I think I'd split this into MatrixBinaryOp and MatrixBinaryAccumulateOp.

What do you think?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

I don't have any issues with that, but @llvm-beanz should confirm. Chris do you have any thoughts here?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Since you mentioned that you aren't blocking this PR I'll go ahead and merge as is but would love to keep the conversation going about the details here!

Further thoughts after chatting a bit with Chris:

Regardless of if we split these up or not they'll actually both end up with the same memory access annotations. MatrixBinaryOp would be readwrite* (read from A,B, write to C) and MatrixBinaryAccumulateOp would be readwrite* (read from A,B,C, write to C).

Also if we do split them, MatrixBinaryAccumulateOp should be called something like MatrixTernaryOp because its actually more of shape C' = A * B + C

*: I use readwrite but iirc the actual annotation is none/the empty string which implies readwrite


Validation rules will enforce that:
* argument A is an `A` matrix
Expand Down