[InstrRef] Preserve debug instr num in aarch64-expand-pseudo LOADgot expansion #128081

rastogishubham · 2025-02-20T22:39:41Z

The aarch64-expand-pseudo pass expands the LOADgot instruction to an ADRP instruction and a LDRXui instruction. If the LOADgot had a debug-instr-number, the pass doesn't preserve this to the new expansion.

This patch fixes the issue by making sure the debug-instr-number is correctly applied to the LDRXui instruction generated.

llvmbot · 2025-02-20T22:40:12Z

@llvm/pr-subscribers-backend-aarch64

Author: Shubham Sandeep Rastogi (rastogishubham)

Changes

The aarch64-expand-pseudo pass expands the LOADgot instruction to an ADRP instruction and a LDRXui instruction. If the LOADgot had a debug-instr-number, the pass doesn't preserve this to the new expansion.

This patch fixes the issue by making sure the debug-instr-number is correctly applied to the LDRXui instruction generated.

Full diff: https://github.com/llvm/llvm-project/pull/128081.diff

2 Files Affected:

(modified) llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp (+5)
(added) llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir (+283)

diff --git a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
index b44c48afe705b..55a441b7d22b6 100644
--- a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
@@ -1382,6 +1382,11 @@ bool AArch64ExpandPseudo::expandMI(MachineBasicBlock &MBB,
                                       AArch64II::MO_NC);
       }
 
+      // If the LOADgot instruction has a debug-instr-number, annotate the
+      // LDRWui instruction that it is expanded to with the same
+      // debug-instr-number to preserve debug information.
+      if (MI.peekDebugInstrNum() != 0)
+        MIB2->setDebugInstrNum(MI.peekDebugInstrNum());
       transferImpOps(MI, MIB1, MIB2);
     }
     MI.eraseFromParent();
diff --git a/llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir b/llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir
new file mode 100644
index 0000000000000..f5273d345d216
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir
@@ -0,0 +1,283 @@
+# RUN: llc -run-pass=aarch64-expand-pseudo -mtriple=aarch64-unknown-linux-gnu -o - %s | FileCheck %s
+
+# This testcase was obtained by looking at FileCheck.cpp and reducing it down via llvm-reduce 
+
+# Check that the LDRXui preserves the debug info by retaining the debug-instr-number 1 in _ZN4llvm12handleErrorsIJZNS_12consumeErrorENS_5ErrorEEUlRKNS_13ErrorInfoBaseEE_EEES1_S1_DpOT_
+# CHECK: $x8 = ADRP target-flags(aarch64-page, aarch64-got) @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev
+# CHECK-NEXT: renamable $x8 = LDRXui killed $x8, target-flags(aarch64-pageoff, aarch64-got, aarch64-nc) @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev, debug-instr-number 1
+# CHECK-NEXT: DBG_INSTR_REF !9, !DIExpression(DW_OP_LLVM_arg, 0), dbg-instr-ref(1, 0), debug-location
+
+--- |
+  ; ModuleID = '/Users/shubhamrastogi/Development/test105791650/FileCheck-extract-reduced.ll'
+  source_filename = "/Users/shubhamrastogi/Development/test105791650/FileCheck-extract-reduced.ll"
+  target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-n32:64-S128-Fn32"
+  target triple = "arm64-apple-macosx15.0.0"
+  
+  declare i64 @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev()
+  
+  ; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: read)
+  define i1 @_ZNSt3__1neB8nn180100IPNS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEEEEbRKNS_11__wrap_iterIT_EESC_(ptr readnone captures(address) %__x, ptr readonly captures(none) %__y) local_unnamed_addr #0 {
+    %1 = load ptr, ptr %__y, align 8
+    %cmp = icmp eq ptr %__x, %1
+    ret i1 %cmp
+  }
+  
+  ; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: read)
+  define ptr @_ZNKSt3__111__wrap_iterIPNS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEEE4baseB8nn180100Ev(ptr readonly captures(none) %this) local_unnamed_addr #0 {
+  entry:
+    %0 = load ptr, ptr %this, align 8
+    ret ptr %0
+  }
+  
+  ; Function Attrs: mustprogress nofree norecurse nounwind willreturn
+  define void @_ZN4llvm12handleErrorsIJZNS_12consumeErrorENS_5ErrorEEUlRKNS_13ErrorInfoBaseEE_EEES1_S1_DpOT_(ptr readnone captures(address) %call6) local_unnamed_addr #1 !dbg !4 {
+    %call8 = load volatile i64, ptr null, align 4294967296, !dbg !24
+      #dbg_value(ptr @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev, !9, !DIExpression(), !25)
+    %cmp.i = icmp eq ptr %call6, @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev
+    br i1 %cmp.i, label %for.body, label %common.ret
+  
+  common.ret:                                       ; preds = %0
+    ret void
+  
+  for.body:                                         ; preds = %0
+    tail call void @llvm.lifetime.start.p0(i64 0, ptr null)
+    ret void
+  }
+  
+  ; Function Attrs: mustprogress nocallback nofree nosync nounwind willreturn memory(argmem: readwrite)
+  declare void @llvm.lifetime.start.p0(i64 immarg, ptr captures(none)) #2
+  
+  attributes #0 = { mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: read) }
+  attributes #1 = { mustprogress nofree norecurse nounwind willreturn }
+  attributes #2 = { mustprogress nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
+  
+  !llvm.module.flags = !{!0}
+  !llvm.dbg.cu = !{!1}
+  
+  !0 = !{i32 2, !"Debug Info Version", i32 3}
+  !1 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_14, file: !2, producer: "clang version 21.0.0git (\0A\0A\0A\[email protected]:llvm/llvm-project.git 6deee0d5b36c8b4b83209759df8d4933e4922bc8\0A\0A\0A\0A)", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !3, retainedTypes: !3, globals: !3, imports: !3, splitDebugInlining: false, nameTableKind: Apple, sysroot: "/Library/Developer/CommandLineTools/SDKs/MacOSX15.3.sdk", sdk: "MacOSX15.3.sdk")
+  !2 = !DIFile(filename: "/Users/shubhamrastogi/Development/llvm-project-instr-ref/llvm-project/llvm/lib/FileCheck/FileCheck.cpp", directory: "/Users/shubhamrastogi/Development/llvm-project-instr-ref/llvm-project/build-instr-ref-stage2", checksumkind: CSK_MD5, checksum: "e718c2a5b2d3baab240a5e370113901e")
+  !3 = !{}
+  !4 = distinct !DISubprogram(name: "handleErrors<(\0A\0A\0A\0Alambda at /Users/shubhamrastogi/Development/llvm-project-instr-ref/llvm-project/llvm/include/llvm/Support/Error.h:1070:35\0A\0A\0A\0A)>", linkageName: "_ZN4llvm12handleErrorsIJZNS_12consumeErrorENS_5ErrorEEUlRKNS_13ErrorInfoBaseEE_EEES1_S1_DpOT_", scope: !6, file: !5, line: 954, type: !7, scopeLine: 954, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !1, templateParams: !3, retainedNodes: !8)
+  !5 = !DIFile(filename: "llvm/include/llvm/Support/Error.h", directory: "/Users/shubhamrastogi/Development/llvm-project-instr-ref/llvm-project", checksumkind: CSK_MD5, checksum: "0a75676311531ba2140b3f0d994b3c33")
+  !6 = !DINamespace(name: "llvm", scope: null)
+  !7 = distinct !DISubroutineType(types: !3)
+  !8 = !{!9}
+  !9 = !DILocalVariable(name: "__end3", scope: !10, type: !13, flags: DIFlagArtificial)
+  !10 = distinct !DILexicalBlock(scope: !11, file: !5, line: 963, column: 5)
+  !11 = distinct !DILexicalBlock(scope: !12, file: !5, line: 960, column: 34)
+  !12 = distinct !DILexicalBlock(scope: !4, file: !5, line: 960, column: 7)
+  !13 = !DIDerivedType(tag: DW_TAG_typedef, name: "iterator", scope: !15, file: !14, line: 403, baseType: !18, flags: DIFlagPublic)
+  !14 = !DIFile(filename: "/Library/Developer/CommandLineTools/SDKs/MacOSX15.3.sdk/usr/include/c++/v1/vector", directory: "")
+  !15 = distinct !DICompositeType(tag: DW_TAG_class_type, name: "vector<std::__1::unique_ptr<llvm::ErrorInfoBase,\0A\0A\0A\0Astd::__1::default_delete<llvm::ErrorInfoBase> >,\0A\0A\0A\0Astd::__1::allocator<std::__1::unique_ptr<llvm::ErrorInfoBase,\0A\0A\0A\0Astd::__1::default_delete<llvm::ErrorInfoBase> > > >", scope: !16, file: !14, line: 387, size: 192, flags: DIFlagTypePassByReference | DIFlagNonTrivial, elements: !3, templateParams: !3, identifier: "_ZTSNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEEE")
+  !16 = !DINamespace(name: "__1", scope: !17, exportSymbols: true)
+  !17 = !DINamespace(name: "std", scope: null)
+  !18 = distinct !DICompositeType(tag: DW_TAG_class_type, name: "__wrap_iter<std::__1::unique_ptr<llvm::ErrorInfoBase,\0A\0A\0A\0Astd::__1::default_delete<llvm::ErrorInfoBase> > *>", scope: !16, file: !19, line: 41, baseType: !20, size: 64)
+  !19 = !DIFile(filename: "/Library/Developer/CommandLineTools/SDKs/MacOSX15.3.sdk/usr/include/c++/v1/__iterator/wrap_iter.h", directory: "")
+  !20 = !DIDerivedType(tag: DW_TAG_typedef, name: "iterator_type", scope: !18, file: !19, line: 30, baseType: !21, flags: DIFlagPublic)
+  !21 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !22, size: 64)
+  !22 = distinct !DICompositeType(tag: DW_TAG_class_type, name: "unique_ptr<llvm::ErrorInfoBase,\0A\0A\0A\0Astd::__1::default_delete<llvm::ErrorInfoBase> >", scope: !16, file: !23, line: 124, size: 64, flags: DIFlagTypePassByReference | DIFlagNonTrivial, elements: !3, templateParams: !3, identifier: "_ZTSNSt3__110unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS2_EEEE")
+  !23 = !DIFile(filename: "/Library/Developer/CommandLineTools/SDKs/MacOSX15.3.sdk/usr/include/c++/v1/__memory/unique_ptr.h", directory: "")
+  !24 = !DILocation(line: 963, column: 18, scope: !10)
+  !25 = !DILocation(line: 0, scope: !10)
+
+...
+---
+name:            _ZNSt3__1neB8nn180100IPNS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEEEEbRKNS_11__wrap_iterIT_EESC_
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+noPhis:          true
+isSSA:           false
+noVRegs:         true
+hasFakeUses:     false
+callsEHReturn:   false
+callsUnwindInit: false
+hasEHCatchret:   false
+hasEHScopes:     false
+hasEHFunclets:   false
+isOutlined:      false
+debugInstrRef:   true
+failsVerification: false
+tracksDebugUserValues: true
+registers:       []
+liveins:
+  - { reg: '$x0', virtual-reg: '' }
+  - { reg: '$x1', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    1
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  functionContext: ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  hasTailCall:     false
+  isCalleeSavedInfoValid: true
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+entry_values:    []
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo:
+  hasRedZone:      false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: $x0, $x1
+  
+    renamable $x8 = LDRXui killed renamable $x1, 0 :: (load (s64) from %ir.__y)
+    dead $xzr = SUBSXrr killed renamable $x0, killed renamable $x8, implicit-def $nzcv
+    renamable $w0 = CSINCWr $wzr, $wzr, 1, implicit killed $nzcv
+    RET_ReallyLR implicit $w0
+
+...
+---
+name:            _ZNKSt3__111__wrap_iterIPNS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEEE4baseB8nn180100Ev
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+noPhis:          true
+isSSA:           false
+noVRegs:         true
+hasFakeUses:     false
+callsEHReturn:   false
+callsUnwindInit: false
+hasEHCatchret:   false
+hasEHScopes:     false
+hasEHFunclets:   false
+isOutlined:      false
+debugInstrRef:   true
+failsVerification: false
+tracksDebugUserValues: true
+registers:       []
+liveins:
+  - { reg: '$x0', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    1
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  functionContext: ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  hasTailCall:     false
+  isCalleeSavedInfoValid: true
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+entry_values:    []
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo:
+  hasRedZone:      false
+body:             |
+  bb.0.entry:
+    liveins: $x0
+  
+    renamable $x0 = LDRXui killed renamable $x0, 0 :: (load (s64) from %ir.this)
+    RET_ReallyLR implicit $x0
+
+...
+---
+name:            _ZN4llvm12handleErrorsIJZNS_12consumeErrorENS_5ErrorEEUlRKNS_13ErrorInfoBaseEE_EEES1_S1_DpOT_
+alignment:       4
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+noPhis:          true
+isSSA:           false
+noVRegs:         true
+hasFakeUses:     false
+callsEHReturn:   false
+callsUnwindInit: false
+hasEHCatchret:   false
+hasEHScopes:     false
+hasEHFunclets:   false
+isOutlined:      false
+debugInstrRef:   true
+failsVerification: false
+tracksDebugUserValues: true
+registers:       []
+liveins:
+  - { reg: '$x0', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    1
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  functionContext: ''
+  maxCallFrameSize: 0
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  hasTailCall:     false
+  isCalleeSavedInfoValid: true
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:           []
+entry_values:    []
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo:
+  hasRedZone:      false
+body:             |
+  bb.0 (%ir-block.0):
+    liveins: $x0
+  
+    $x8 = MOVZXi 0, 0
+    dead $xzr = LDRXui killed renamable $x8, 0, debug-location !24 :: (volatile load (s64) from `ptr null`, align 4294967296)
+    renamable $x8 = LOADgot target-flags(aarch64-got) @_ZNSt3__16vectorINS_10unique_ptrIN4llvm13ErrorInfoBaseENS_14default_deleteIS3_EEEENS_9allocatorIS6_EEE3endB8nn180100Ev, debug-instr-number 1
+    DBG_INSTR_REF !9, !DIExpression(DW_OP_LLVM_arg, 0), dbg-instr-ref(1, 0), debug-location !25
+    dead $xzr = SUBSXrr killed renamable $x0, killed renamable $x8, implicit-def $nzcv
+    RET_ReallyLR
+
+...

adrian-prantl

That looks plausible.

llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir

jmorse

LGTM; realistically we can drop the bodies of most of the functions in the IR block, and the MIR can be generated with --simplify-mir to cut down the number of redundant lines expressing default values. To be absolutely minimal you could probably drop all but the final MIR blocks (I don't think llc will need them for the LOADgot instruction, only a declare in the LLVM-IR block?).

…expansion The aarch64-expand-pseudo pass expands the LOADgot instruction to an ADRP instruction and a LDRXui instruction. If the LOADgot had a debug-instr-number, the pass doesn't preserve this to the new expansion. This patch fixes the issue by making sure the debug-instr-number is correctly applied to the LDRXui instruction generated.

rastogishubham · 2025-02-25T05:41:57Z

I have reduced the test down to 45 lines from 283 lines, thanks for the suggestions!

…expansion (llvm#128081) The aarch64-expand-pseudo pass expands the LOADgot instruction to an ADRP instruction and a LDRXui instruction. If the LOADgot had a debug-instr-number, the pass doesn't preserve this to the new expansion. This patch fixes the issue by making sure the debug-instr-number is correctly applied to the LDRXui instruction generated. (cherry picked from commit b3c51db)

llvmbot added the backend:AArch64 label Feb 20, 2025

rastogishubham requested review from jmorse and adrian-prantl February 20, 2025 22:39

adrian-prantl approved these changes Feb 22, 2025

View reviewed changes

adrian-prantl reviewed Feb 22, 2025

View reviewed changes

llvm/test/CodeGen/AArch64/expand-load-got-pseudo.mir Show resolved Hide resolved

jmorse approved these changes Feb 24, 2025

View reviewed changes

rastogishubham force-pushed the FixAArch64ExpandPseudo branch from a369e49 to d113900 Compare February 25, 2025 03:23

rastogishubham merged commit b3c51db into llvm:main Feb 25, 2025
11 checks passed

rastogishubham deleted the FixAArch64ExpandPseudo branch February 25, 2025 05:42

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[InstrRef] Preserve debug instr num in aarch64-expand-pseudo LOADgot expansion #128081

[InstrRef] Preserve debug instr num in aarch64-expand-pseudo LOADgot expansion #128081

Uh oh!

rastogishubham commented Feb 20, 2025

Uh oh!

llvmbot commented Feb 20, 2025

Uh oh!

adrian-prantl left a comment

Uh oh!

Uh oh!

jmorse left a comment

Uh oh!

rastogishubham commented Feb 25, 2025

Uh oh!

Uh oh!

Uh oh!

[InstrRef] Preserve debug instr num in aarch64-expand-pseudo LOADgot expansion #128081

[InstrRef] Preserve debug instr num in aarch64-expand-pseudo LOADgot expansion #128081

Uh oh!

Conversation

rastogishubham commented Feb 20, 2025

Uh oh!

llvmbot commented Feb 20, 2025

Uh oh!

adrian-prantl left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

jmorse left a comment

Choose a reason for hiding this comment

Uh oh!

rastogishubham commented Feb 25, 2025

Uh oh!

Uh oh!

Uh oh!