[lld] [DO NOT MERGE][WIP] -- [lld][ICF] Don't merge symbols with different addends (PR #139493)
via llvm-commits
llvm-commits at lists.llvm.org
Sun May 11 21:26:04 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-lld
Author: Pranav Kant (pranavk)
<details>
<summary>Changes</summary>
Fixes scenario reported in https://github.com/llvm/llvm-project/pull/134342#issuecomment-2867252474
---
Full diff: https://github.com/llvm/llvm-project/pull/139493.diff
2 Files Affected:
- (modified) lld/ELF/ICF.cpp (+15-14)
- (added) lld/test/ELF/icf-addend.s (+33)
``````````diff
diff --git a/lld/ELF/ICF.cpp b/lld/ELF/ICF.cpp
index 1882116d4d5f0..797d0968da6c2 100644
--- a/lld/ELF/ICF.cpp
+++ b/lld/ELF/ICF.cpp
@@ -334,26 +334,27 @@ bool ICF<ELFT>::equalsConstant(const InputSection *a, const InputSection *b) {
: constantEq(a, ra.relas, b, rb.relas);
}
-template <class RelTy>
-static SmallVector<Symbol *> getReloc(const InputSection *sec,
+template <class ELFT, class RelTy>
+static SmallVector<std::pair<Symbol *, uint64_t>> getReloc(const InputSection *sec,
Relocs<RelTy> relocs) {
- SmallVector<Symbol *> syms;
+ SmallVector<std::pair<Symbol *, uint64_t>> syms;
for (auto ri = relocs.begin(), re = relocs.end(); ri != re; ++ri) {
Symbol &sym = sec->file->getRelocTargetSym(*ri);
- syms.push_back(&sym);
+ uint64_t addend = getAddend<ELFT>(*ri);
+ syms.push_back(std::make_pair(&sym, addend));
}
return syms;
}
template <class ELFT>
-static SmallVector<Symbol *> getRelocTargetSyms(const InputSection *sec) {
+static SmallVector<std::pair<Symbol *, uint64_t>> getRelocTargets(const InputSection *sec) {
const RelsOrRelas<ELFT> rel = sec->template relsOrRelas<ELFT>();
if (rel.areRelocsCrel())
- return getReloc(sec, rel.crels);
+ return getReloc<ELFT>(sec, rel.crels);
if (rel.areRelocsRel())
- return getReloc(sec, rel.rels);
+ return getReloc<ELFT>(sec, rel.rels);
- return getReloc(sec, rel.relas);
+ return getReloc<ELFT>(sec, rel.relas);
}
// Compare two lists of relocations. Returns true if all pairs of
@@ -568,19 +569,19 @@ template <class ELFT> void ICF<ELFT>::run() {
if (end - begin == 1)
return;
print() << "selected section " << sections[begin];
- SmallVector<Symbol *> syms = getRelocTargetSyms<ELFT>(sections[begin]);
+ SmallVector<std::pair<Symbol *, uint64_t>> syms = getRelocTargets<ELFT>(sections[begin]);
for (size_t i = begin + 1; i < end; ++i) {
print() << " removing identical section " << sections[i];
sections[begin]->replace(sections[i]);
- SmallVector<Symbol *> replacedSyms =
- getRelocTargetSyms<ELFT>(sections[i]);
+ SmallVector<std::pair<Symbol *, uint64_t>> replacedSyms =
+ getRelocTargets<ELFT>(sections[i]);
assert(syms.size() == replacedSyms.size() &&
"Should have same number of syms!");
for (size_t i = 0; i < syms.size(); i++) {
- if (syms[i] == replacedSyms[i] || !syms[i]->isGlobal() ||
- !replacedSyms[i]->isGlobal())
+ if (syms[i].first == replacedSyms[i].first || !syms[i].first->isGlobal() ||
+ !replacedSyms[i].first->isGlobal() || syms[i].second != replacedSyms[i].second)
continue;
- symbolEquivalence.unionSets(syms[i], replacedSyms[i]);
+ symbolEquivalence.unionSets(syms[i].first, replacedSyms[i].first);
}
// At this point we know sections merged are fully identical and hence
diff --git a/lld/test/ELF/icf-addend.s b/lld/test/ELF/icf-addend.s
new file mode 100644
index 0000000000000..0023bbf2ff421
--- /dev/null
+++ b/lld/test/ELF/icf-addend.s
@@ -0,0 +1,33 @@
+# REQUIRES: x86
+# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t.o
+# RUN: ld.lld %t.o -o /dev/null --icf=all --print-icf-sections | FileCheck %s
+
+# Check that ICF doesn't fold different symbols when functions referencing them
+# can be folded because they are pointing to the same address.
+
+# CHECK: selected section {{.*}}:(.text.f1)
+# CHECK: removing identical section {{.*}}:(.text.f2)
+# CHECK-NOT: redirecting 'y' in symtab to 'x'
+# CHECK-NOT: redirecting 'y' to 'x'
+
+.globl x, y
+
+.section .rodata,"a", at progbits
+x:
+.long 11
+y:
+.long 12
+
+.section .text.f1,"ax", at progbits
+f1:
+movq x+4(%rip), %rax
+
+.section .text.f2,"ax", at progbits
+f2:
+movq y(%rip), %rax
+
+.section .text._start,"ax", at progbits
+.globl _start
+_start:
+call f1
+call f2
``````````
</details>
https://github.com/llvm/llvm-project/pull/139493
More information about the llvm-commits
mailing list