release/20.x: [RelLookupTableConverter] Drop unnamed_addr to avoid generating GOTPCREL relocations (#142304) #142311
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-llvm-transforms Author: dianqk (dianqk) ChangesBackport aa09dbb Requested by: @dianqk Full diff: https://github.com/llvm/llvm-project/pull/142311.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Utils/RelLookupTableConverter.cpp b/llvm/lib/Transforms/Utils/RelLookupTableConverter.cpp
index 2700b4307308c..4486cba2bf6c0 100644
--- a/llvm/lib/Transforms/Utils/RelLookupTableConverter.cpp
+++ b/llvm/lib/Transforms/Utils/RelLookupTableConverter.cpp
@@ -18,6 +18,7 @@
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Module.h"
+#include "llvm/TargetParser/Triple.h"
using namespace llvm;
@@ -108,8 +109,24 @@ static GlobalVariable *createRelLookupTable(Function &Func,
uint64_t Idx = 0;
SmallVector<Constant *, 64> RelLookupTableContents(NumElts);
+ Triple TT(M.getTargetTriple());
+ // FIXME: This should be removed in the future.
+ bool ShouldDropUnnamedAddr =
+ // Drop unnamed_addr to avoid matching pattern in
+ // `handleIndirectSymViaGOTPCRel`, which generates GOTPCREL relocations
+ // not supported by the GNU linker and LLD versions below 18 on aarch64.
+ TT.isAArch64()
+ // Apple's ld64 (and ld-prime on Xcode 15.2) miscompile something on
+ // x86_64-apple-darwin. See
+ // https://github.com/rust-lang/rust/issues/140686 and
+ // https://github.com/rust-lang/rust/issues/141306.
+ || (TT.isX86() && TT.isOSDarwin());
+
for (Use &Operand : LookupTableArr->operands()) {
Constant *Element = cast<Constant>(Operand);
+ if (ShouldDropUnnamedAddr)
+ if (auto *GlobalElement = dyn_cast<GlobalValue>(Element))
+ GlobalElement->setUnnamedAddr(GlobalValue::UnnamedAddr::None);
Type *IntPtrTy = M.getDataLayout().getIntPtrType(M.getContext());
Constant *Base = llvm::ConstantExpr::getPtrToInt(RelLookupTable, IntPtrTy);
Constant *Target = llvm::ConstantExpr::getPtrToInt(Element, IntPtrTy);
diff --git a/llvm/test/Transforms/RelLookupTableConverter/unnamed_addr.ll b/llvm/test/Transforms/RelLookupTableConverter/unnamed_addr.ll
new file mode 100644
index 0000000000000..78b8a4aa126c9
--- /dev/null
+++ b/llvm/test/Transforms/RelLookupTableConverter/unnamed_addr.ll
@@ -0,0 +1,119 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 5
+; REQUIRES: x86-registered-target
+; REQUIRES: aarch64-registered-target
+; RUN: opt < %s -passes=rel-lookup-table-converter -relocation-model=pic -mtriple=x86_64-apple-darwin -S | FileCheck -check-prefix=x86_64-apple-darwin %s
+; RUN: opt < %s -passes=rel-lookup-table-converter -relocation-model=pic -mtriple=aarch64 -S | FileCheck -check-prefix=aarch64 %s
+; RUN: opt < %s -passes=rel-lookup-table-converter -relocation-model=pic -mtriple=x86_64 -S | FileCheck -check-prefix=x86_64 %s
+
+@a0 = private unnamed_addr constant i32 0
+@a1 = private unnamed_addr constant i32 1
+@a2 = private unnamed_addr constant i32 2
+@load_relative_1.table = private unnamed_addr constant [3 x ptr] [ptr @a0, ptr @a1, ptr @a2]
+
+@x0 = internal unnamed_addr constant i64 0
+@x1 = internal unnamed_addr constant i64 1
+@x2 = internal unnamed_addr constant i64 2
+@x3 = internal unnamed_addr constant i64 3
+@y0 = internal unnamed_addr constant ptr @x3
+@y1 = internal unnamed_addr constant ptr @x2
+@y2 = internal unnamed_addr constant ptr @x1
+@y3 = internal unnamed_addr constant ptr @x0
+@load_relative_2.table = private unnamed_addr constant [4 x ptr] [ptr @y3, ptr @y2, ptr @y1, ptr @y0]
+
+;.
+; x86_64-apple-darwin: @a0 = private constant i32 0
+; x86_64-apple-darwin: @a1 = private constant i32 1
+; x86_64-apple-darwin: @a2 = private constant i32 2
+; x86_64-apple-darwin: @load_relative_1.table.rel = private unnamed_addr constant [3 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @a0 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a1 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a2 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32)], align 4
+; x86_64-apple-darwin: @x0 = internal unnamed_addr constant i64 0
+; x86_64-apple-darwin: @x1 = internal unnamed_addr constant i64 1
+; x86_64-apple-darwin: @x2 = internal unnamed_addr constant i64 2
+; x86_64-apple-darwin: @x3 = internal unnamed_addr constant i64 3
+; x86_64-apple-darwin: @y0 = internal constant ptr @x3
+; x86_64-apple-darwin: @y1 = internal constant ptr @x2
+; x86_64-apple-darwin: @y2 = internal constant ptr @x1
+; x86_64-apple-darwin: @y3 = internal constant ptr @x0
+; x86_64-apple-darwin: @load_relative_2.table.rel = private unnamed_addr constant [4 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @y3 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y2 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y1 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y0 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32)], align 4
+;.
+; aarch64: @a0 = private constant i32 0
+; aarch64: @a1 = private constant i32 1
+; aarch64: @a2 = private constant i32 2
+; aarch64: @load_relative_1.table.rel = private unnamed_addr constant [3 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @a0 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a1 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a2 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32)], align 4
+; aarch64: @x0 = internal unnamed_addr constant i64 0
+; aarch64: @x1 = internal unnamed_addr constant i64 1
+; aarch64: @x2 = internal unnamed_addr constant i64 2
+; aarch64: @x3 = internal unnamed_addr constant i64 3
+; aarch64: @y0 = internal constant ptr @x3
+; aarch64: @y1 = internal constant ptr @x2
+; aarch64: @y2 = internal constant ptr @x1
+; aarch64: @y3 = internal constant ptr @x0
+; aarch64: @load_relative_2.table.rel = private unnamed_addr constant [4 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @y3 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y2 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y1 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y0 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32)], align 4
+;.
+; x86_64: @a0 = private unnamed_addr constant i32 0
+; x86_64: @a1 = private unnamed_addr constant i32 1
+; x86_64: @a2 = private unnamed_addr constant i32 2
+; x86_64: @load_relative_1.table.rel = private unnamed_addr constant [3 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @a0 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a1 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @a2 to i64), i64 ptrtoint (ptr @load_relative_1.table.rel to i64)) to i32)], align 4
+; x86_64: @x0 = internal unnamed_addr constant i64 0
+; x86_64: @x1 = internal unnamed_addr constant i64 1
+; x86_64: @x2 = internal unnamed_addr constant i64 2
+; x86_64: @x3 = internal unnamed_addr constant i64 3
+; x86_64: @y0 = internal unnamed_addr constant ptr @x3
+; x86_64: @y1 = internal unnamed_addr constant ptr @x2
+; x86_64: @y2 = internal unnamed_addr constant ptr @x1
+; x86_64: @y3 = internal unnamed_addr constant ptr @x0
+; x86_64: @load_relative_2.table.rel = private unnamed_addr constant [4 x i32] [i32 trunc (i64 sub (i64 ptrtoint (ptr @y3 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y2 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y1 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32), i32 trunc (i64 sub (i64 ptrtoint (ptr @y0 to i64), i64 ptrtoint (ptr @load_relative_2.table.rel to i64)) to i32)], align 4
+;.
+define ptr @load_relative_1(i64 %offset) {
+; x86_64-apple-darwin-LABEL: define ptr @load_relative_1(
+; x86_64-apple-darwin-SAME: i64 [[OFFSET:%.*]]) {
+; x86_64-apple-darwin-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; x86_64-apple-darwin-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_1.table.rel, i64 [[RELTABLE_SHIFT]])
+; x86_64-apple-darwin-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+; aarch64-LABEL: define ptr @load_relative_1(
+; aarch64-SAME: i64 [[OFFSET:%.*]]) {
+; aarch64-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; aarch64-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_1.table.rel, i64 [[RELTABLE_SHIFT]])
+; aarch64-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+; x86_64-LABEL: define ptr @load_relative_1(
+; x86_64-SAME: i64 [[OFFSET:%.*]]) {
+; x86_64-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; x86_64-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_1.table.rel, i64 [[RELTABLE_SHIFT]])
+; x86_64-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+ %gep = getelementptr inbounds [3 x ptr], ptr @load_relative_1.table, i64 0, i64 %offset
+ %load = load ptr, ptr %gep
+ ret ptr %load
+}
+
+define ptr @load_relative_2(i64 %offset) {
+; x86_64-apple-darwin-LABEL: define ptr @load_relative_2(
+; x86_64-apple-darwin-SAME: i64 [[OFFSET:%.*]]) {
+; x86_64-apple-darwin-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; x86_64-apple-darwin-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_2.table.rel, i64 [[RELTABLE_SHIFT]])
+; x86_64-apple-darwin-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+; aarch64-LABEL: define ptr @load_relative_2(
+; aarch64-SAME: i64 [[OFFSET:%.*]]) {
+; aarch64-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; aarch64-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_2.table.rel, i64 [[RELTABLE_SHIFT]])
+; aarch64-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+; x86_64-LABEL: define ptr @load_relative_2(
+; x86_64-SAME: i64 [[OFFSET:%.*]]) {
+; x86_64-NEXT: [[RELTABLE_SHIFT:%.*]] = shl i64 [[OFFSET]], 2
+; x86_64-NEXT: [[RELTABLE_INTRINSIC:%.*]] = call ptr @llvm.load.relative.i64(ptr @load_relative_2.table.rel, i64 [[RELTABLE_SHIFT]])
+; x86_64-NEXT: ret ptr [[RELTABLE_INTRINSIC]]
+;
+ %gep = getelementptr inbounds [4 x ptr], ptr @load_relative_2.table, i64 0, i64 %offset
+ %load = load ptr, ptr %gep
+ ret ptr %load
+}
+;.
+; x86_64-apple-darwin: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(argmem: read) }
+;.
+; aarch64: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(argmem: read) }
+;.
+; x86_64: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(argmem: read) }
+;.
|
MaskRay
approved these changes
Jun 1, 2025
nikic
approved these changes
Jun 2, 2025
github-project-automation
bot
moved this from Needs Triage
to Needs Merge
in LLVM Release Status
…REL relocations (llvm#142304) Follow llvm#72584 (comment), the patch will drop the `unnamed_addr` attribute when generating relative lookup tables. I'm not very confident about this patch, but it does resolve rust-lang/rust#140686, rust-lang/rust#141306 and rust-lang/rust#141737. But I don't think this will result in worse problems. > LLVM provides that the calculation of such a constant initializer will not overflow at link time under the medium code model if x is an unnamed_addr function. However, it does not provide this guarantee for a constant initializer folded into a function body. This intrinsic can be used to avoid the possibility of overflows when loading from such a constant. ([‘llvm.load.relative’ Intrinsic](https://llvm.org/docs/LangRef.html#id2592)) This is my concern. I'm not sure how unnamed_addr provides this guarantee, and I haven't found any test cases. (cherry picked from commit aa09dbb)
github-project-automation
bot
moved this from Needs Merge
to Done
in LLVM Release Status
|
@dianqk (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR. |
Member
Author
Add a workaround for relocations in the data sections for Apple's ld64, ld-prime, and GNU linker, which will be removed in the future. |
Member
Author
|
Note to me: I propose to keep a major version around until the gnu linker release supports it and GitHub Actions defaults to Xcode 15.3+. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Backport aa09dbb
Requested by: @dianqk