Skip to content

[ObjectFifo][NFC] Refactor DmaUtils + SplitLogicalObjectFifoForReuse#759

Merged
Abhishek-Varma merged 3 commits intomainfrom
avarma_refactor_for_combining_objFifos
Sep 10, 2024
Merged

[ObjectFifo][NFC] Refactor DmaUtils + SplitLogicalObjectFifoForReuse#759
Abhishek-Varma merged 3 commits intomainfrom
avarma_refactor_for_combining_objFifos

Conversation

@Abhishek-Varma
Copy link
Contributor

-- This commit adds refactoring of few utilities in DmaUtils as well as a few involved in SplitLogicalObjectFifosForReuse pass.
-- This is required for the follow-up PR that adds a new pass --iree-amdaie-combine-logical-objectfifos-for-connection-reuse.

Signed-off-by: Abhishek Varma abhvarma@amd.com

-- This commit adds refactoring of few utilities in DmaUtils as well
   as a few involved in SplitLogicalObjectFifosForReuse pass.
-- This is required for the follow-up PR that adds a new pass
   `--iree-amdaie-combine-logical-objectfifos-for-connection-reuse`.

Signed-off-by: Abhishek Varma <abhvarma@amd.com>
Comment on lines 301 to 362
SmallVector<OpFoldResult, 4> staticL3AsSourceSizes =
SmallVector<OpFoldResult> staticL3AsSourceSizes =
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Relaying here the review comment : #755 (comment) (@jtuyls )

Two reasons :-

  1. The second argument to SmallVector<arg1, arg2> used to be mandatory but stopped being so - I came across this relaxed constraint in one of the upstream contribution's review comments.
  2. In case of "splitting" we start creation of a new L2 logicalobjectFifo via L2->L1's route. In case of "combing" we start it from L3->L2. The function createNewLogicalObjectFifo declaration complains because :-
    a. SmallVector<arg1> != SmallVector<arg1, 4>
    b. SmallVector<arg1, 4> != SmallVector<arg1, 6>

Therefore the only way to circumvent point 2z, that I knew of, was to remove this which aligned well with point 1.

Copy link
Collaborator

@jtuyls jtuyls Sep 10, 2024

Choose a reason for hiding this comment

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

I think it's still useful to specify the number of inlined elements in the vector as we know it will usually be less than 4 and not much more: SmallVector<OpFoldResult, 4>. createNewLogicalObjectFifo then has to accept a SmallVectorImpl instead of a SmallVector to avoid the issue in 2).

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Sure, I've addressed it in the latest push. Please take a look.

@Abhishek-Varma
Copy link
Contributor Author

Hi @jtuyls , based on #755 (review) I'm segregating #755 into two PRs.

This concerns NFC changes required to have the actual pass (#760) implemented.

@Abhishek-Varma Abhishek-Varma enabled auto-merge (squash) September 10, 2024 07:58
Copy link
Collaborator

@jtuyls jtuyls left a comment

Choose a reason for hiding this comment

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

LGTM, just one nit below

@Abhishek-Varma Abhishek-Varma enabled auto-merge (squash) September 10, 2024 09:05
@Abhishek-Varma Abhishek-Varma merged commit c041358 into main Sep 10, 2024
@Abhishek-Varma Abhishek-Varma deleted the avarma_refactor_for_combining_objFifos branch September 10, 2024 09:18
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