Skip to content

Conversation

F-Stuckmann
Copy link
Collaborator

Put the zero-offset part last, since it doesn't need the pointer to survive.

Note: This is a preparation PR for GEMM.

Core_Compute_Insn_Count Core_PMSize_absolute Core_StackSize_absolute

case AIE2P::G_AIE_POSTINC_3D_STORE: {
for (unsigned SubRegIdx = 0; SubRegIdx < SplitFactor; ++SubRegIdx) {
for (unsigned Part = 0; Part < SplitFactor; ++Part) {
const unsigned SubRegIdx = AMI.MemI.getOpcode() == AIE2P::G_STORE
Copy link
Collaborator

Choose a reason for hiding this comment

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

Why do we need to handle differently S_STORE? They will expand to offset stores anyway... If we have a strong objection we could also have a comment.

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

good catch, we can simplify it

case AIE2P::G_AIE_POSTINC_2D_LOAD:
case AIE2P::G_AIE_POSTINC_3D_LOAD: {
for (unsigned SubRegIdx = 0; SubRegIdx < SplitFactor; ++SubRegIdx) {
for (unsigned Part = 0; Part < SplitFactor; ++Part) {
Copy link
Collaborator

Choose a reason for hiding this comment

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

Same comment as the G_STORE one.

@F-Stuckmann F-Stuckmann force-pushed the stuckmann.split.large.ls branch from 3d93257 to d2025b0 Compare October 7, 2025 15:01
@andcarminati
Copy link
Collaborator

Do you think we could have another QoR run because of the G_LOAD/G_STORE change?

@F-Stuckmann
Copy link
Collaborator Author

Before force pushing the change, I ran QoR to verify nothing degraded.

Do you think we could have another QoR run because of the G_LOAD/G_STORE change?

@andcarminati
Copy link
Collaborator

This looks good to me, I only recommend to fix the commit message before approval.

Put the zero-offset part last, since it doesn't need the pointer to survive.
@F-Stuckmann F-Stuckmann force-pushed the stuckmann.split.large.ls branch from d2025b0 to 1817126 Compare October 9, 2025 08:21
@F-Stuckmann
Copy link
Collaborator Author

I updated the commit message

@F-Stuckmann F-Stuckmann changed the title [AIE2P] Better postincrment splitting [AIE2P] split offsets before postincrment in large Load/Stores Oct 9, 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