SEGV on IR snippit reduced from test_simple_atomic() from pcsections.ll #160587
Description
While testing #160580 I noticed a test failure in llvm/test/CodeGen/AArch64/pcsections.ll's test_simple_atomic(). I tried to pull that function out to test it and I ended up with
@foo = dso_local global i64 0, align 8
@bar = dso_local global i64 0, align 8
define i64 @test_simple_atomic() {
; CHECK-LABEL: test_simple_atomic:
; CHECK: .Lpcsection1:
; CHECK-NEXT: ldr
; CHECK-NOT: .Lpcsection2
; CHECK: ldr
; CHECK: add
; CHECK-NEXT: ret
; CHECK: .section section_no_aux,"awo",@progbits,.text
; CHECK-NEXT: .Lpcsection_base3:
; DEFCM-NEXT: .word .Lpcsection1-.Lpcsection_base3
; LARGE-NEXT: .xword .Lpcsection1-.Lpcsection_base3
; CHECK-NEXT: .text
entry:
%0 = load atomic i64, ptr @foo monotonic, align 8, !pcsections !0
%1 = load i64, ptr @bar, align 8
%add = add nsw i64 %1, %0
ret i64 %add
}
!0 = !{!"section_no_aux"}
which SEGVs while generating assembly. Specifically:
$ ./build/bin/llc -march=aarch64 test.ll -o /dev/stdout
.build_version macos, 15, 0
.section __TEXT,__text,regular,pure_instructions
.globl _test_simple_atomic ; -- Begin function test_simple_atomic
.p2align 2
_test_simple_atomic: ; @test_simple_atomic
.cfi_startproc
; %bb.0: ; %entry
Lloh0:
adrp x8, _foo@PAGE
Lloh1:
adrp x9, _bar@PAGE
Lpcsection0:
Lloh2:
ldr x8, [x8, _foo@PAGEOFF]
Lloh3:
ldr x9, [x9, _bar@PAGEOFF]
add x0, x9, x8
ret
.loh AdrpLdr Lloh1, Lloh3
.loh AdrpLdr Lloh0, Lloh2
.cfi_endproc
.section PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace and instructions to reproduce the bug.
Stack dump:
0. Program arguments: ./build/bin/llc -march=aarch64 test.ll -o /dev/stdout
1. Running pass 'Function Pass Manager' on module 'test.ll'.
2. Running pass 'AArch64 Assembly Printer' on function '@test_simple_atomic'
#0 0x0000000103157408 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/Volumes/life/llvm-project/build/bin/llc+0x1022cb408)
#1 0x0000000103155344 llvm::sys::RunSignalHandlers() (/Volumes/life/llvm-project/build/bin/llc+0x1022c9344)
#2 0x0000000103157eb8 SignalHandler(int, __siginfo*, void*) (/Volumes/life/llvm-project/build/bin/llc+0x1022cbeb8)
#3 0x000000019c0296a4 (/usr/lib/system/libsystem_platform.dylib+0x1804ad6a4)
#4 0x0000000102b3af04 llvm::MCAsmInfoDarwin::printSwitchToSection(llvm::MCSection const&, unsigned int, llvm::Triple const&, llvm::raw_ostream&) const (/Volumes/life/llvm-project/build/bin/llc+0x101caef04)
#5 0x0000000102afbef0 (anonymous namespace)::MCAsmStreamer::switchSection(llvm::MCSection*, unsigned int) (/Volumes/life/llvm-project/build/bin/llc+0x101c6fef0)
#6 0x00000001024981e0 llvm::AsmPrinter::emitPCSections(llvm::MachineFunction const&)::$_1::operator()(llvm::MDNode const&, llvm::ArrayRef<llvm::MCSymbol const*>, bool) const (/Volumes/life/llvm-project/build/bin/llc+0x10160c1e0)
#7 0x0000000102497ec8 llvm::AsmPrinter::emitPCSections(llvm::MachineFunction const&) (/Volumes/life/llvm-project/build/bin/llc+0x10160bec8)
#8 0x000000010249c0bc llvm::AsmPrinter::emitFunctionBody() (/Volumes/life/llvm-project/build/bin/llc+0x1016100bc)
#9 0x0000000101059e88 (anonymous namespace)::AArch64AsmPrinter::runOnMachineFunction(llvm::MachineFunction&) (/Volumes/life/llvm-project/build/bin/llc+0x1001cde88)
#10 0x000000010269bad8 llvm::MachineFunctionPass::runOnFunction(llvm::Function&) (/Volumes/life/llvm-project/build/bin/llc+0x10180fad8)
#11 0x0000000102a67a50 llvm::FPPassManager::runOnFunction(llvm::Function&) (/Volumes/life/llvm-project/build/bin/llc+0x101bdba50)
#12 0x0000000102a6db58 llvm::FPPassManager::runOnModule(llvm::Module&) (/Volumes/life/llvm-project/build/bin/llc+0x101be1b58)
#13 0x0000000102a683c0 llvm::legacy::PassManagerImpl::run(llvm::Module&) (/Volumes/life/llvm-project/build/bin/llc+0x101bdc3c0)
#14 0x0000000100e8fbec main (/Volumes/life/llvm-project/build/bin/llc+0x100003bec)
#15 0x000000019bc4eb98
Segmentation fault: 11 ./build/bin/llc -march=aarch64 test.ll -o /dev/stdout
Note that I just started using both OSX and LLVM, so it's somewhat likely I've just done something stupid here. This is also all a little close to what I was just touching, but as far as I can tell it's still manifesting on a clean recent main -- specifically eb85899 ("[InstCombine] Fold selects into masked loads (#160522)"), but I don't have any reason to believe that's an issue.