[AArch64] [Windows] Share unwind codes between epilogues with identical unwind codes

[ssijaric]

Summary

There are cases where we have multiple epilogues that have the exact same unwind code sequence. This change merges such epilogues into a single epilogue.

This should get us past the assert "SEH unwind data splitting not yet implemented" in many cases. We still need to add support for generating multiple .pdata/.xdata sections for those functions that need to be split into fragments.

[efriedma]

It might be nice to merge together identical suffixes eventually, but that shouldn't block this change.

How hard would it be to add support for reusing opcodes from an identical prolog? The unwind format is designed to allow that in some cases.

lib/MC/MCWin64EH.cpp

459

Using a pair here seems weird... either use llvm:Optional, or just return a null pointer.

468

A bool would be sufficient here.

514

The DenseMaps here seem unnecessary; I mean, yes, it'll reduce the number of potential matching epilogs, but realistically functions have very few unique epilogs anyway.

529

It's not immediately obvious that you're mutating the EpilogMap here; probably worth calling that out explicitly in a comment. I guess it does the right thing because nothing else will use the FrameInfo after this.

[dmajor]

I've confirmed that I can build Firefox successfully with this patch on top of latest trunk. Thank you @ssijaric!

[ssijaric]

Address Eli's comments.

[efriedma]

LGTM

[ssijaric]

In D56813#1360762, @dmajor wrote:

I've confirmed that I can build Firefox successfully with this patch on top of latest trunk. Thank you @ssijaric!

Hi David,

Can you please try with the latest version of the patch? I am going to commit it shortly.

Thanks.

[TomTan]

In D56813#1360814, @ssijaric wrote:

In D56813#1360762, @dmajor wrote:

I've confirmed that I can build Firefox successfully with this patch on top of latest trunk. Thank you @ssijaric!

Hi David,

Can you please try with the latest version of the patch? I am going to commit it shortly.

Thanks.

FYI, confirmed that V8 builds successfully with the latest change which had the issue. But I am not sure about Firefox.

[dmajor]

The updated patch works for Firefox too.

lib/MC/MCWin64EH.cpp

case Win64EH::UOP_SaveFRegPX:
     break;
   case Win64EH::UOP_End:
     b = 0xE4;
     streamer.EmitIntValue(b, 1);
     break;
   }
 }
 
+// Returns the epilog symbol of an epilog with the exact same unwind code
+// sequence, if it exists.  Otherwise, returns nulltpr.
+// EpilogInstrs - Unwind codes for the current epilog.
+// Epilogs - Epilogs that potentialy match the current epilog.
+static MCSymbol*
+FindMatchingEpilog(const std::vector<WinEH::Instruction>& EpilogInstrs,
+                   const std::vector<MCSymbol *>& Epilogs,
+                   const WinEH::FrameInfo *info) {
+  for (auto *EpilogStart : Epilogs) {
+    auto InstrsIter = info->EpilogMap.find(EpilogStart);
+    assert(InstrsIter != info->EpilogMap.end() &&
+           "Epilog not found in EpilogMap");
+    const auto &Instrs = InstrsIter->second;
+
+    if (Instrs.size() != EpilogInstrs.size())
+      continue;
+
+    bool Match = true;
+    for (unsigned i = 0; i < Instrs.size(); ++i)
+      if (Instrs[i].Operation != EpilogInstrs[i].Operation ||
+          Instrs[i].Offset != EpilogInstrs[i].Offset ||
+          Instrs[i].Register != EpilogInstrs[i].Register) {
+         Match = false;
+         break;
+      }
+
+    if (Match)
+      return EpilogStart;
+  }
+  return nullptr;
+}
+
 // Populate the .xdata section.  The format of .xdata on ARM64 is documented at
 // https://docs.microsoft.com/en-us/cpp/build/arm64-exception-handling
 static void ARM64EmitUnwindInfo(MCStreamer &streamer, WinEH::FrameInfo *info) {
   // If this UNWIND_INFO already has a symbol, it's already been emitted.
   if (info->Symbol)
     return;
 
   MCContext &context = streamer.getContext();
   MCSymbol *Label = context.createTempSymbol();
 
   streamer.EmitValueToAlignment(4);
   streamer.EmitLabel(Label);
   info->Symbol = Label;
 
   uint32_t FuncLength = 0x0;
   if (info->FuncletOrFuncEnd)
     FuncLength = (uint32_t)GetAbsDifference(streamer, info->FuncletOrFuncEnd,
                                             info->Begin);
   FuncLength /= 4;
   uint32_t PrologCodeBytes = ARM64CountOfUnwindCodes(info->Instructions);
   uint32_t TotalCodeBytes = PrologCodeBytes;
 
   // Process epilogs.
   MapVector<MCSymbol *, uint32_t> EpilogInfo;
+  // Epilogs processed so far.
+  std::vector<MCSymbol *> AddedEpilogs;
+
   for (auto &I : info->EpilogMap) {
     MCSymbol *EpilogStart = I.first;
     auto &EpilogInstrs = I.second;
     uint32_t CodeBytes = ARM64CountOfUnwindCodes(EpilogInstrs);
+
+    uint32_t NumUnwindCodes = EpilogInstrs.size();
+    MCSymbol* MatchingEpilog =
+      FindMatchingEpilog(EpilogInstrs, AddedEpilogs, info);
+    if (MatchingEpilog) {
+      assert(EpilogInfo.find(MatchingEpilog) != EpilogInfo.end() &&
+             "Duplicate epilog not found");
+      EpilogInfo[EpilogStart] = EpilogInfo[MatchingEpilog];
+      // Clear the unwind codes in the EpilogMap, so that they don't get output
+      // in the logic below.
+      EpilogInstrs.clear();
+    } else {
       EpilogInfo[EpilogStart] = TotalCodeBytes;
       TotalCodeBytes += CodeBytes;
+      AddedEpilogs.push_back(EpilogStart);
+    }
   }
 
   // Code Words, Epilog count, E, X, Vers, Function Length
   uint32_t row1 = 0x0;
   uint32_t CodeWords = TotalCodeBytes / 4;
   uint32_t CodeWordsMod = TotalCodeBytes % 4;
   if (CodeWordsMod)
     CodeWords++;

test/CodeGen/AArch64/wineh4.mir

 # RUN: llc -o - %s -mtriple=aarch64-windows -start-after=prologepilog \
 # RUN:   -disable-branch-fold  -filetype=obj \
 # RUN: | llvm-readobj -unwind | FileCheck %s
-# Check that multiple epilgoues are correctly placed in .xdata.
+# Check that identical multiple epilgoues are correctly shared in .xdata.
 
 # CHECK:        ExceptionData {
 # CHECK-NEXT:      FunctionLength: 164
 # CHECK-NEXT:      Version: 0
 # CHECK-NEXT:      ExceptionData: No
 # CHECK-NEXT:      EpiloguePacked: No
 # CHECK-NEXT:      EpilogueScopes: 2
-# CHECK-NEXT:      ByteCodeLength: 48
+# CHECK-NEXT:      ByteCodeLength: 32
 # CHECK-NEXT:      Prologue [
 # CHECK-NEXT:        0xc80c              ; stp x19, x20, [sp, #96]
 # CHECK-NEXT:        0xc88a              ; stp x21, x22, [sp, #80]
 # CHECK-NEXT:        0xc908              ; stp x23, x24, [sp, #64]
 # CHECK-NEXT:        0xc986              ; stp x25, x26, [sp, #48]
 # CHECK-NEXT:        0xca04              ; stp x27, x28, [sp, #32]
 # CHECK-NEXT:        0xd802              ; stp d8, d9, [sp, #16]
 # CHECK-NEXT:        0xda8d              ; stp d10, d11, [sp, #-112]!

 # CHECK-NEXT:            0xca04              ; ldp x27, x28, [sp, #32]
 # CHECK-NEXT:            0xd802              ; ldp d8, d9, [sp, #16]
 # CHECK-NEXT:            0xda8d              ; ldp d10, d11, [sp], #112
 # CHECK-NEXT:            0xe4                ; end
 # CHECK-NEXT:          ]
 # CHECK-NEXT:        }
 # CHECK-NEXT:        EpilogueScope {
 # CHECK-NEXT:          StartOffset: 33
-# CHECK-NEXT:          EpilogueStartIndex: 30
+# CHECK-NEXT:          EpilogueStartIndex: 15
 # CHECK-NEXT:          Opcodes [
 # CHECK-NEXT:            0xc80c              ; ldp x19, x20, [sp, #96]
 # CHECK-NEXT:            0xc88a              ; ldp x21, x22, [sp, #80]
 # CHECK-NEXT:            0xc908              ; ldp x23, x24, [sp, #64]
 # CHECK-NEXT:            0xc986              ; ldp x25, x26, [sp, #48]
 # CHECK-NEXT:            0xca04              ; ldp x27, x28, [sp, #32]
 # CHECK-NEXT:            0xd802              ; ldp d8, d9, [sp, #16]
 # CHECK-NEXT:            0xda8d              ; ldp d10, d11, [sp], #112

test/CodeGen/AArch64/wineh8.mir

+# RUN: llc -o - %s -mtriple=aarch64-windows -start-after=prologepilog \
+# RUN:   -disable-branch-fold  -filetype=obj \
+# RUN: | llvm-readobj -unwind | FileCheck %s
+# Check that non-identical multiple epilgoues are correctly shared in .xdata.
+
+# CHECK:        ExceptionData {
+# CHECK-NEXT:      FunctionLength: 160
+# CHECK-NEXT:      Version: 0
+# CHECK-NEXT:      ExceptionData: No
+# CHECK-NEXT:      EpiloguePacked: No
+# CHECK-NEXT:      EpilogueScopes: 2
+# CHECK-NEXT:      ByteCodeLength: 44
+# CHECK-NEXT:      Prologue [
+# CHECK-NEXT:        0xc80c              ; stp x19, x20, [sp, #96]
+# CHECK-NEXT:        0xc88a              ; stp x21, x22, [sp, #80]
+# CHECK-NEXT:        0xc908              ; stp x23, x24, [sp, #64]
+# CHECK-NEXT:        0xc986              ; stp x25, x26, [sp, #48]
+# CHECK-NEXT:        0xca04              ; stp x27, x28, [sp, #32]
+# CHECK-NEXT:        0xd802              ; stp d8, d9, [sp, #16]
+# CHECK-NEXT:        0xda8d              ; stp d10, d11, [sp, #-112]!
+# CHECK-NEXT:        0xe4                ; end
+# CHECK-NEXT:      ]
+# CHECK-NEXT:      EpilogueScopes [
+# CHECK-NEXT:        EpilogueScope {
+# CHECK-NEXT:          StartOffset: 16
+# CHECK-NEXT:          EpilogueStartIndex: 15
+# CHECK-NEXT:          Opcodes [
+# CHECK-NEXT:            0xc80c              ; ldp x19, x20, [sp, #96]
+# CHECK-NEXT:            0xc88a              ; ldp x21, x22, [sp, #80]
+# CHECK-NEXT:            0xc908              ; ldp x23, x24, [sp, #64]
+# CHECK-NEXT:            0xc986              ; ldp x25, x26, [sp, #48]
+# CHECK-NEXT:            0xd802              ; ldp d8, d9, [sp, #16]
+# CHECK-NEXT:            0xda8d              ; ldp d10, d11, [sp], #112
+# CHECK-NEXT:            0xe4                ; end
+# CHECK-NEXT:          ]
+# CHECK-NEXT:        }
+# CHECK-NEXT:        EpilogueScope {
+# CHECK-NEXT:          StartOffset: 32
+# CHECK-NEXT:          EpilogueStartIndex: 28
+# CHECK-NEXT:          Opcodes [
+# CHECK-NEXT:            0xc80c              ; ldp x19, x20, [sp, #96]
+# CHECK-NEXT:            0xc88a              ; ldp x21, x22, [sp, #80]
+# CHECK-NEXT:            0xc908              ; ldp x23, x24, [sp, #64]
+# CHECK-NEXT:            0xc986              ; ldp x25, x26, [sp, #48]
+# CHECK-NEXT:            0xca04              ; ldp x27, x28, [sp, #32]
+# CHECK-NEXT:            0xd802              ; ldp d8, d9, [sp, #16]
+# CHECK-NEXT:            0xda8d              ; ldp d10, d11, [sp], #112
+# CHECK-NEXT:            0xe4                ; end
+# CHECK-NEXT:          ]
+# CHECK-NEXT:        }
+# CHECK-NEXT:      ]
+# CHECK-NEXT:    }
+...
+---
+name:            test
+alignment:       2
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       true
+registers:
+liveins:
+  - { reg: '$w0', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       112
+  offsetAdjustment: 0
+  maxAlignment:    8
+  adjustsStack:    false
+  hasCalls:        false
+  stackProtector:  ''
+  maxCallFrameSize: 0
+  hasOpaqueSPAdjustment: true
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:
+stack:
+  - { id: 0, name: '', type: spill-slot, offset: -8, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x19', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 1, name: '', type: spill-slot, offset: -16, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x20', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 2, name: '', type: spill-slot, offset: -24, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x21', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 3, name: '', type: spill-slot, offset: -32, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x22', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 4, name: '', type: spill-slot, offset: -40, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x23', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 5, name: '', type: spill-slot, offset: -48, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x24', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 6, name: '', type: spill-slot, offset: -56, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x25', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 7, name: '', type: spill-slot, offset: -64, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x26', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 8, name: '', type: spill-slot, offset: -72, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x27', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 9, name: '', type: spill-slot, offset: -80, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$x28', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 10, name: '', type: spill-slot, offset: -88, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$d8', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 11, name: '', type: spill-slot, offset: -96, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$d9', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 12, name: '', type: spill-slot, offset: -104, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$d10', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 13, name: '', type: spill-slot, offset: -112, size: 8, alignment: 8,
+      stack-id: 0, callee-saved-register: '$d11', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+constants:
+body:             |
+  bb.0.entry:
+    successors: %bb.2(0x40000000), %bb.1(0x40000000)
+    liveins: $x0, $x1, $d0, $d1, $d10, $d11, $d8, $d9, $x27, $x28, $x25, $x26, $x23, $x24, $x21, $x22, $x19, $x20
+
+    early-clobber $sp = frame-setup STPDpre killed $d10, killed $d11, $sp, -14 :: (store 8 into %stack.12), (store 8 into %stack.13)
+    frame-setup SEH_SaveFRegP_X 10, 11, -112
+    frame-setup STPDi killed $d8, killed $d9, $sp, 2 :: (store 8 into %stack.10), (store 8 into %stack.11)
+    frame-setup SEH_SaveFRegP 8, 9, 16
+    frame-setup STPXi killed $x27, killed $x28, $sp, 4 :: (store 8 into %stack.8), (store 8 into %stack.9)
+    frame-setup SEH_SaveRegP 27, 28, 32
+    frame-setup STPXi killed $x25, killed $x26, $sp, 6 :: (store 8 into %stack.6), (store 8 into %stack.7)
+    frame-setup SEH_SaveRegP 25, 26, 48
+    frame-setup STPXi killed $x23, killed $x24, $sp, 8 :: (store 8 into %stack.4), (store 8 into %stack.5)
+    frame-setup SEH_SaveRegP 23, 24, 64
+    frame-setup STPXi killed $x21, killed $x22, $sp, 10 :: (store 8 into %stack.2), (store 8 into %stack.3)
+    frame-setup SEH_SaveRegP 21, 22, 80
+    frame-setup STPXi killed $x19, killed $x20, $sp, 12 :: (store 8 into %stack.0), (store 8 into %stack.1)
+    frame-setup SEH_SaveRegP 19, 20, 96
+    frame-setup SEH_PrologEnd
+    frame-setup CFI_INSTRUCTION def_cfa_offset 112
+    frame-setup CFI_INSTRUCTION offset $w19, -8
+    frame-setup CFI_INSTRUCTION offset $w20, -16
+    frame-setup CFI_INSTRUCTION offset $w21, -24
+    frame-setup CFI_INSTRUCTION offset $w22, -32
+    frame-setup CFI_INSTRUCTION offset $w23, -40
+    frame-setup CFI_INSTRUCTION offset $w24, -48
+    frame-setup CFI_INSTRUCTION offset $w25, -56
+    frame-setup CFI_INSTRUCTION offset $w26, -64
+    frame-setup CFI_INSTRUCTION offset $w27, -72
+    frame-setup CFI_INSTRUCTION offset $w28, -80
+    frame-setup CFI_INSTRUCTION offset $b8, -88
+    frame-setup CFI_INSTRUCTION offset $b9, -96
+    frame-setup CFI_INSTRUCTION offset $b10, -104
+    frame-setup CFI_INSTRUCTION offset $b11, -112
+    $x19 = ADDXrr $x0, killed $x1
+    $d8 = FADDDrr killed $d0, $d1
+    $d9 = FADDDrr $d8, $d1
+    $d10 = FADDDrr $d9, $d8
+    $d11 = FADDDrr killed $d9, $d10
+    $x20 = SUBSXrr $x19, killed $x0, implicit-def $nzcv
+    Bcc 1, %bb.2, implicit killed $nzcv
+    B %bb.1
+
+  bb.1:
+    liveins: $x19, $x20
+
+    $x21 = ADDXrr $x20, killed $x19
+    $x22 = ADDXrr $x21, killed $x20
+    $x23 = ADDXrr $x22, killed $x21
+    $x24 = ADDXrr $x23, killed $x22
+    $x25 = ADDXrr $x24, killed $x23
+    $x26 = ADDXrr $x25, killed $x24
+    $x27 = ADDXrr $x26, killed $x25
+    $x28 = ADDXrr $x27, killed $x26
+    $x0 = COPY $x28
+    frame-destroy SEH_EpilogStart
+    $x19, $x20 = frame-destroy LDPXi $sp, 12 :: (load 8 from %stack.0), (load 8 from %stack.1)
+    frame-destroy SEH_SaveRegP 19, 20, 96
+    $x21, $x22 = frame-destroy LDPXi $sp, 10 :: (load 8 from %stack.2), (load 8 from %stack.3)
+    frame-destroy SEH_SaveRegP 21, 22, 80
+    $x23, $x24 = frame-destroy LDPXi $sp, 8 :: (load 8 from %stack.4), (load 8 from %stack.5)
+    frame-destroy SEH_SaveRegP 23, 24, 64
+    $x25, $x26 = frame-destroy LDPXi $sp, 6 :: (load 8 from %stack.6), (load 8 from %stack.7)
+    frame-destroy SEH_SaveRegP 25, 26, 48
+    $x27, $x28 = frame-destroy LDPXi $sp, 4 :: (load 8 from %stack.8), (load 8 from %stack.9)
+    frame-destroy SEH_SaveRegP 27, 28, 32
+    $d8, $d9 = frame-destroy LDPDi $sp, 2 :: (load 8 from %stack.10), (load 8 from %stack.11)
+    frame-destroy SEH_SaveFRegP 8, 9, 16
+    early-clobber $sp, $d10, $d11 = frame-destroy LDPDpost $sp, 14 :: (load 8 from %stack.12), (load 8 from %stack.13)
+    frame-destroy SEH_SaveFRegP_X 10, 11, -112
+    frame-destroy SEH_EpilogEnd
+    RET_ReallyLR implicit $x0
+
+  bb.2:
+    liveins: $x28, $d11
+
+    $x0 = COPY $d11
+    $x0 = ADDXrr $x0, killed $x28
+    frame-destroy SEH_EpilogStart
+    $x19, $x20 = frame-destroy LDPXi $sp, 12 :: (load 8 from %stack.0), (load 8 from %stack.1)
+    frame-destroy SEH_SaveRegP 19, 20, 96
+    $x21, $x22 = frame-destroy LDPXi $sp, 10 :: (load 8 from %stack.2), (load 8 from %stack.3)
+    frame-destroy SEH_SaveRegP 21, 22, 80
+    $x23, $x24 = frame-destroy LDPXi $sp, 8 :: (load 8 from %stack.4), (load 8 from %stack.5)
+    frame-destroy SEH_SaveRegP 23, 24, 64
+    $x25, $x26 = frame-destroy LDPXi $sp, 6 :: (load 8 from %stack.6), (load 8 from %stack.7)
+    frame-destroy SEH_SaveRegP 25, 26, 48
+    $d8, $d9 = frame-destroy LDPDi $sp, 2 :: (load 8 from %stack.10), (load 8 from %stack.11)
+    frame-destroy SEH_SaveFRegP 8, 9, 16
+    early-clobber $sp, $d10, $d11 = frame-destroy LDPDpost $sp, 14 :: (load 8 from %stack.12), (load 8 from %stack.13)
+    frame-destroy SEH_SaveFRegP_X 10, 11, -112
+    frame-destroy SEH_EpilogEnd
+    RET_ReallyLR implicit $x0
+
+...