Skip to content

[SDAG] Fix CSE for ADDRSPACECAST nodes #122912

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

Merged
merged 2 commits into from
Jan 20, 2025
Merged

[SDAG] Fix CSE for ADDRSPACECAST nodes #122912

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7176ce0
[SDAG] Fix CSE for ADDRSPACECAST nodes
AlexMaclean Jan 14, 2025
9a367ce
address comments -- fixup test somewhat
AlexMaclean Jan 20, 2025
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
6 changes: 6 additions & 0 deletions llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -954,6 +954,12 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) {
ID.AddInteger(M);
break;
}
case ISD::ADDRSPACECAST: {
const AddrSpaceCastSDNode *ASC = cast<AddrSpaceCastSDNode>(N);
ID.AddInteger(ASC->getSrcAddressSpace());
ID.AddInteger(ASC->getDestAddressSpace());
break;
}
case ISD::TargetBlockAddress:
case ISD::BlockAddress: {
const BlockAddressSDNode *BA = cast<BlockAddressSDNode>(N);
Expand Down
23 changes: 23 additions & 0 deletions llvm/test/CodeGen/NVPTX/addrspacecast-cse.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
; RUN: llc < %s -O0 -debug-only=isel -o /dev/null 2>&1 | FileCheck %s

; REQUIRES: asserts

target triple = "nvptx64-nvidia-cuda"

;; Selection DAG CSE is hard to test since we run CSE/GVN on the IR before and
;; after selection DAG ISel so most cases will be handled by one of these.
Comment on lines +7 to +8
Copy link
Contributor

Choose a reason for hiding this comment

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

You could also hack around this with -O0, the CSE always happens at any opt level (a dirtier hack would be use -start-before/after). I'm not sure I understand how stackrestore helps here.

Copy link
Member Author

Choose a reason for hiding this comment

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

I was able to get around the pre-SelectionDAG CSE by using the stackrestore which causes a addrspacecast to be added during DAG legalization, so that isn't a problem. However it seems like there is some post-ISel (perhaps DAG-CSE on the machine nodes?) which is preventing the impact of this change from being easily observed in the final output. That's why I'm using the debug output.

Copy link
Contributor

Choose a reason for hiding this comment

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

-stop-after=finalize-isel might work for avoiding machine cse

define void @foo(ptr %p) {
; CHECK-LABEL: Initial selection DAG
;
; CHECK: [[ASC:t[0-9]+]]{{.*}} = addrspacecast
; CHECK: store{{.*}} [[ASC]]
; CHECK: store{{.*}} [[ASC]]
;
; CHECK-LABEL: Optimized lowered selection
;
%a1 = addrspacecast ptr %p to ptr addrspace(5)
%a2 = addrspacecast ptr %p to ptr addrspace(5)
store i32 0, ptr addrspace(5) %a1
store i32 0, ptr addrspace(5) %a2
ret void
}
Loading