Skip to content

[CodeGenPrepare] Filter out unrecreatable addresses from memory optimization #143566

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 9 commits into from
Jun 28, 2025
Merged

[CodeGenPrepare] Filter out unrecreatable addresses from memory optimization #143566

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
b7f4b17
[CodeGenPrepare] Handle address sinking obtained from invoke
e-kud Jun 10, 2025
8d928bc
Don't try to sink addresses that we can't recreate
e-kud Jun 11, 2025
0c804d3
Rename the test
e-kud Jun 11, 2025
a5b5b9a
Add tests for args and globals
e-kud Jun 13, 2025
81ddd17
Skip anything that is not an instruction from current basic block
e-kud Jun 13, 2025
360e261
Merge branch 'main' into cg-prepare-term-addr
e-kud Jun 17, 2025
ccf28b5
Skip only instructions from different basic blocks
e-kud Jun 26, 2025
85eb67b
Merge branch 'main' into cg-prepare-term-addr
e-kud Jun 26, 2025
5215657
Merge branch 'main' into cg-prepare-term-addr
e-kud Jun 28, 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
7 changes: 7 additions & 0 deletions llvm/lib/CodeGen/CodeGenPrepare.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6099,6 +6099,13 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
}

if (!ResultIndex) {
auto PtrInst = dyn_cast<Instruction>(ResultPtr);
// Here we know that we have just a pointer without any offsets. If
// this pointer comes from a different from the current basic block we
Copy link
Contributor

@weiguozhi weiguozhi Jun 13, 2025

Choose a reason for hiding this comment

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

different BB

// need to know how to recreate it in another basic block.
// Currently we don't support recreation of any of instruction.
if (PtrInst && PtrInst->getParent() != MemoryInst->getParent())
Copy link
Contributor Author

@e-kud e-kud Jun 11, 2025

Choose a reason for hiding this comment

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

Here we can change the predicate to

if (!PtrInst || PtrInst->getParent() != MemoryInst->getParent())

To skip arguments and global pointers. I'm not sure if it is completely correct.

Copy link
Contributor

@weiguozhi weiguozhi Jun 13, 2025

Choose a reason for hiding this comment

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

I prefer this condition. We should not modify a memory access instruction with pointer comes from argument or global value directly.

Copy link
Contributor Author

@e-kud e-kud Jun 13, 2025

Choose a reason for hiding this comment

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

Done. I also added tests for arguments and globals. @weiguozhi you may be interested that we don't reuse addresses in these cases: https://godbolt.org/z/q8fzce4ov

return Modified;
SunkAddr = ResultPtr;
} else {
if (ResultPtr->getType() != I8PtrTy)
Expand Down
42 changes: 42 additions & 0 deletions llvm/test/Transforms/CodeGenPrepare/X86/sink-addr-recreate.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,42 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -p 'require<profile-summary>,codegenprepare' -cgpp-huge-func=0 < %s | FileCheck %s

target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
target triple = "x86_64-grtev4-linux-gnu"

; Test a scenario when the address cannot be recreated in the current basic block

declare ptr @get_ptr(i64)
define void @addr_from_invoke() personality ptr null {
; CHECK-LABEL: define void @addr_from_invoke() personality ptr null {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[PTR:%.*]] = invoke ptr @get_ptr(i64 0)
; CHECK-NEXT: to label %[[BODY_1:.*]] unwind label %[[EHCLEANUP:.*]]
; CHECK: [[EHCLEANUP]]:
; CHECK-NEXT: [[PAD:%.*]] = cleanuppad within none []
; CHECK-NEXT: cleanupret from [[PAD]] unwind to caller
; CHECK: [[BODY_1]]:
; CHECK-NEXT: [[GEP1:%.*]] = bitcast ptr [[PTR]] to ptr
; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[GEP1]], align 4
; CHECK-NEXT: [[TMP0:%.*]] = bitcast ptr [[PTR]] to ptr
; CHECK-NEXT: [[UNUSED:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4
; CHECK-NEXT: store <4 x i32> zeroinitializer, ptr [[TMP0]], align 4
; CHECK-NEXT: ret void
;
entry:
%ptr = invoke ptr @get_ptr(i64 0) to label %body.1 unwind label %ehcleanup

body.2:
%unused = load <4 x i32>, ptr %gep, align 4
store <4 x i32> zeroinitializer, ptr %gep, align 4
ret void

ehcleanup:
%pad = cleanuppad within none []
cleanupret from %pad unwind to caller

body.1:
%gep = getelementptr { i32 }, ptr %ptr, i64 0, i32 0
store <4 x i32> zeroinitializer, ptr %gep, align 4
br label %body.2
}
Loading