-
Notifications
You must be signed in to change notification settings - Fork 13.4k
Swedev 414443 #65947
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.
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
Swedev 414443 #65947
Conversation
SWEDEV-414443. Fixes performance regression introduced by D155343. Allows sinking in a simple case when there are no loops.
@@ -171,6 +171,39 @@ bool SIInstrInfo::isIgnorableUse(const MachineOperand &MO) const { | |||
isVALU(*MO.getParent()) && !resultDependsOnExec(*MO.getParent()); | |||
} | |||
|
|||
bool SIInstrInfo::modifiesRegisterImplicitly( |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We have an ever growing set of convoluted callbacks to deal with this problem. I think we need to look into moving some of the modeling of this into the CFG itself. I think we need to mark block edges as convergent/divergent/something
// SI_END_CF restores exec mask as it was before SI_IF (unchanged) | ||
Register EndCF = ModifierInstr->getOperand(0).getReg(); | ||
MachineInstr *SIIF = MRI.getVRegDef(EndCF); | ||
if (SIIF->getOpcode() == AMDGPU::SI_IF && MRI.hasOneUse(EndCF) && |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This doesn't even work after lowering
Description needs to be what this is doing, the swdev thing should be a comment footer |
Is this still relevant? I thought a newer patch fixed the same issue? |
Implemented in #67456 |
No description provided.