Skip to content

CallPromotionUtils: Correctly use IndexSize when determining the bit width of pointer offsets. #119138

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 1 commit into from
Dec 10, 2024
Merged

CallPromotionUtils: Correctly use IndexSize when determining the bit width of pointer offsets. #119138

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 3 additions & 3 deletions llvm/lib/Transforms/Utils/CallPromotionUtils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -692,14 +692,14 @@ bool llvm::tryPromoteCall(CallBase &CB) {
if (!VTableEntryLoad)
return false; // Not a vtable entry load.
Value *VTableEntryPtr = VTableEntryLoad->getPointerOperand();
APInt VTableOffset(DL.getTypeSizeInBits(VTableEntryPtr->getType()), 0);
APInt VTableOffset(DL.getIndexTypeSizeInBits(VTableEntryPtr->getType()), 0);
Value *VTableBasePtr = VTableEntryPtr->stripAndAccumulateConstantOffsets(
DL, VTableOffset, /* AllowNonInbounds */ true);
LoadInst *VTablePtrLoad = dyn_cast<LoadInst>(VTableBasePtr);
if (!VTablePtrLoad)
return false; // Not a vtable load.
Value *Object = VTablePtrLoad->getPointerOperand();
APInt ObjectOffset(DL.getTypeSizeInBits(Object->getType()), 0);
APInt ObjectOffset(DL.getIndexTypeSizeInBits(Object->getType()), 0);
Value *ObjectBase = Object->stripAndAccumulateConstantOffsets(
DL, ObjectOffset, /* AllowNonInbounds */ true);
if (!(isa<AllocaInst>(ObjectBase) && ObjectOffset == 0))
Expand All @@ -712,7 +712,7 @@ bool llvm::tryPromoteCall(CallBase &CB) {
VTablePtrLoad, VTablePtrLoad->getParent(), BBI, 0, nullptr, nullptr);
if (!VTablePtr)
return false; // No vtable found.
APInt VTableOffsetGVBase(DL.getTypeSizeInBits(VTablePtr->getType()), 0);
Copy link
Collaborator

@jrtc27 jrtc27 Dec 11, 2024

Choose a reason for hiding this comment

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

Presumably FindAvailableLoadedValue can return a pointer-sized integer (really, anything that CastInst::isBitOrNoopPointerCastable) if you end up storing a uintptr_t that you then load and call. If so, stripAndAccumulateConstantOffsets returns its input for non-pointers and then the dyn_cast will definitely give null.

(Obviously not true on CHERI for various reasons)

The other instances are fine because they're using the pointer operands of loads so by definition are pointers, whereas this one is some random other Value that is known to be the same bit pattern as the value loaded here.

So probably line 713 needs to gain a type check. Maybe you could cast back instead, though I'd worry about provenance due to peeking through inttoptr(ptrtoint(...)), and AFAICT the former would make the new code behave like the old code.

Copy link
Collaborator Author

@resistor resistor Dec 11, 2024

Choose a reason for hiding this comment

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

Thanks. I'm doing a bootstrap build now to try to extract a test case. I'll put up a new PR with the test and the fix once I have that.

APInt VTableOffsetGVBase(DL.getIndexTypeSizeInBits(VTablePtr->getType()), 0);
Value *VTableGVBase = VTablePtr->stripAndAccumulateConstantOffsets(
DL, VTableOffsetGVBase, /* AllowNonInbounds */ true);
GlobalVariable *GV = dyn_cast<GlobalVariable>(VTableGVBase);
Expand Down
48 changes: 48 additions & 0 deletions llvm/test/Transforms/Inline/promote-call-bitwidth.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,48 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --scrub-attributes
; RUN: opt -S -passes=inline < %s | FileCheck %s

;; Check that we correctly use the index size when accumulating offsets during CallPromotion

target datalayout = "p200:128:128:128:64-A200-P200-G200"

define void @test(ptr addrspace(200) %arg1, ptr addrspace(200) %arg2) local_unnamed_addr addrspace(200) {
; CHECK-LABEL: define {{[^@]+}}@test
; CHECK-SAME: (ptr addrspace(200) [[ARG1:%.*]], ptr addrspace(200) [[ARG2:%.*]]) local_unnamed_addr addrspace(200) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = load ptr addrspace(200), ptr addrspace(200) [[ARG2]], align 16
; CHECK-NEXT: call addrspace(200) void [[TMP0]](ptr addrspace(200) [[ARG1]])
; CHECK-NEXT: ret void
;
entry:
call void @call_fnptr(ptr addrspace(200) %arg1, ptr addrspace(200) %arg2)
ret void
}

define internal void @call_fnptr(ptr addrspace(200) %this, ptr addrspace(200) %arg) unnamed_addr addrspace(200) align 2 {
entry:
%0 = load ptr addrspace(200), ptr addrspace(200) %arg, align 16
call void %0(ptr addrspace(200) %this)
ret void
}

define void @test2(ptr addrspace(200) %this) local_unnamed_addr addrspace(200) {
; CHECK-LABEL: define {{[^@]+}}@test2
; CHECK-SAME: (ptr addrspace(200) [[THIS:%.*]]) local_unnamed_addr addrspace(200) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[VTABLE_I:%.*]] = load ptr addrspace(200), ptr addrspace(200) [[THIS]], align 16
; CHECK-NEXT: [[FN_I:%.*]] = load ptr addrspace(200), ptr addrspace(200) [[VTABLE_I]], align 16
; CHECK-NEXT: call addrspace(200) void [[FN_I]](ptr addrspace(200) [[THIS]])
; CHECK-NEXT: ret void
;
entry:
call void @call_via_vtable(ptr addrspace(200) %this)
ret void
}

define internal void @call_via_vtable(ptr addrspace(200) %this) unnamed_addr addrspace(200) {
entry:
%vtable = load ptr addrspace(200), ptr addrspace(200) %this, align 16
%fn = load ptr addrspace(200), ptr addrspace(200) %vtable, align 16
call void %fn(ptr addrspace(200) %this)
ret void
}
Loading