[Statepoint] Handle 'undef' operand tied to def

[skatkov]

Summary

FixupStatepoints pass does not take into account the undef use
it skips may have a tied def. So when defs are handled pass
considers that tied-use should be spilled and triggers an assert.

FixupStatepoints should skip undef def as well.

[dantrushin]

LGTM.

BTW, we can completely delete tied dead,undef pair. But additional complexity might not worth it.

llvm/lib/CodeGen/FixupStatepointCallerSaved.cpp

MachineInstr *rewriteStatepoint() {
 
     // New indices for the remaining defs.
     SmallVector<unsigned, 8> NewIndices;
     unsigned NumDefs = MI.getNumDefs();
     for (unsigned I = 0; I < NumDefs; ++I) {
       MachineOperand &DefMO = MI.getOperand(I);
       assert(DefMO.isReg() && DefMO.isDef() && "Expected Reg Def operand");
       Register Reg = DefMO.getReg();
+      assert(DefMO.isTied() && "Def is expected to be tied");
+      // We skipped undef uses and did not spill them, so we should not
+      // proceed with defs here.
+      if (MI.getOperand(MI.findTiedOperandIdx(I)).isUndef()) {
+        if (AllowGCPtrInCSR) {
+          NewIndices.push_back(NewMI->getNumOperands());
+          MIB.addReg(Reg, RegState::Define);
+        }
+        continue;
+      }
       if (!AllowGCPtrInCSR) {
         assert(is_contained(RegsToSpill, Reg));
         RegsToReload.push_back(Reg);
       } else {
         if (isCalleeSaved(Reg)) {
           NewIndices.push_back(NewMI->getNumOperands());
           MIB.addReg(Reg, RegState::Define);
         } else {

llvm/test/CodeGen/X86/statepoint-fixup-undef-def.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -x mir -run-pass fixup-statepoint-caller-saved -verify-machineinstrs < %s | FileCheck %s
+
+--- |
+  ; ModuleID = 'undef.ll'
+  source_filename = "test_undef.ll"
+  target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+  target triple = "x86_64-unknown-linux-gnu"
+
+  define void @test_undef(i8 addrspace(1)* %arg1, i8 addrspace(1)* %arg2, i8 addrspace(1)* %arg3, i8 addrspace(1)* %arg4) #0 gc "statepoint-example" {
+  bb:
+    %tmp1 = lshr i32 0, undef
+    %tmp2 = load atomic i32, i32 addrspace(1)* undef unordered, align 8
+    %tmp3 = load atomic i32, i32 addrspace(1)* undef unordered, align 8
+    %tmp19 = inttoptr i64 undef to i8 addrspace(1)*
+    br label %bb7
+
+  bb7:                                              ; preds = %bb
+    %tmp4 = icmp slt i32 %tmp3, undef
+    %tmp5 = select i1 %tmp4, i32 6, i32 undef
+    %tmp6 = add i32 %tmp5, %tmp2
+    %tmp7 = call i8 addrspace(1)* @wombat()
+    %tmp20 = call token (i64, i32, void (i8 addrspace(1)*, i32, i32, i8 addrspace(1)*, i32)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidp1i8i32i32p1i8i32f(i64 2, i32 5, void (i8 addrspace(1)*, i32, i32, i8 addrspace(1)*, i32)* nonnull @hoge, i32 5, i32 0, i8 addrspace(1)* %tmp19, i32 %tmp2, i32 %tmp6, i8 addrspace(1)* %tmp7, i32 0, i32 0, i32 0) [ "deopt"(i8 addrspace(1)* %tmp19, i8 addrspace(1)* %arg1, i8 addrspace(1)* %arg3, i8 addrspace(1)* %arg4, i32 %tmp2, i32 %tmp1, i32 %tmp5), "gc-live"(i8 addrspace(1)* %tmp19, i8 addrspace(1)* %arg2) ]
+    %tmp21 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tmp20, i32 0, i32 0) ; (%tmp19, %tmp19)
+    %tmp22 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tmp20, i32 1, i32 1) ; (%arg2, %arg2)
+    ret void
+  }
+
+  declare void @hoge(i8 addrspace(1)*, i32, i32, i8 addrspace(1)*, i32) #0
+
+  declare i8 addrspace(1)* @wombat() #0
+
+  ; Function Attrs: nounwind readonly
+  declare i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token, i32 immarg, i32 immarg) #1
+
+  declare token @llvm.experimental.gc.statepoint.p0f_isVoidp1i8i32i32p1i8i32f(i64 immarg, i32 immarg, void (i8 addrspace(1)*, i32, i32, i8 addrspace(1)*, i32)*, i32 immarg, i32 immarg, ...)
+
+  attributes #0 = { "target-cpu"="broadwell" }
+  attributes #1 = { nounwind readonly }
+
+...
+---
+name:            test_undef
+alignment:       16
+exposesReturnsTwice: false
+legalized:       false
+regBankSelected: false
+selected:        false
+failedISel:      false
+tracksRegLiveness: true
+hasWinCFI:       false
+registers:       []
+liveins:
+  - { reg: '$rdi', virtual-reg: '' }
+  - { reg: '$rsi', virtual-reg: '' }
+  - { reg: '$rdx', virtual-reg: '' }
+  - { reg: '$rcx', virtual-reg: '' }
+frameInfo:
+  isFrameAddressTaken: false
+  isReturnAddressTaken: false
+  hasStackMap:     false
+  hasPatchPoint:   false
+  stackSize:       0
+  offsetAdjustment: 0
+  maxAlignment:    8
+  adjustsStack:    false
+  hasCalls:        true
+  stackProtector:  ''
+  maxCallFrameSize: 4294967295
+  cvBytesOfCalleeSavedRegisters: 0
+  hasOpaqueSPAdjustment: false
+  hasVAStart:      false
+  hasMustTailInVarArgFunc: false
+  localFrameSize:  0
+  savePoint:       ''
+  restorePoint:    ''
+fixedStack:      []
+stack:
+  - { id: 0, name: '', type: default, offset: 0, size: 8, alignment: 8,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 1, name: '', type: default, offset: 0, size: 8, alignment: 8,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+  - { id: 2, name: '', type: default, offset: 0, size: 8, alignment: 8,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      debug-info-variable: '', debug-info-expression: '', debug-info-location: '' }
+callSites:       []
+debugValueSubstitutions: []
+constants:       []
+machineFunctionInfo: {}
+body:             |
+  ; CHECK-LABEL: name: test_undef
+  ; CHECK: bb.0.bb:
+  ; CHECK:   successors: %bb.1(0x80000000)
+  ; CHECK:   liveins: $rcx, $rdi, $rdx, $rsi
+  ; CHECK:   renamable $r15 = COPY $rcx
+  ; CHECK:   renamable $r12 = COPY $rdx
+  ; CHECK:   renamable $r14 = COPY $rsi
+  ; CHECK:   renamable $r13 = COPY $rdi
+  ; CHECK:   renamable $ebx = MOV32rm undef renamable $rax, 1, $noreg, 0, $noreg, implicit-def $rbx :: (load unordered 4 from `i32 addrspace(1)* undef`, align 8, addrspace 1)
+  ; CHECK: bb.1.bb7:
+  ; CHECK:   liveins: $rbx, $r12, $r13, $r14, $r15
+  ; CHECK:   renamable $ebp = LEA64_32r renamable $rbx, 1, $noreg, 6, $noreg
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   CALL64pcrel32 target-flags(x86-plt) @wombat, csr_64, implicit $rsp, implicit $ssp, implicit-def $rsp, implicit-def $ssp, implicit-def $rax
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $r13 :: (store 8 into %stack.0)
+  ; CHECK:   MOV64mr %stack.1, 1, $noreg, 0, $noreg, killed renamable $r12 :: (store 8 into %stack.1)
+  ; CHECK:   MOV64mr %stack.2, 1, $noreg, 0, $noreg, killed renamable $r15 :: (store 8 into %stack.2)
+  ; CHECK:   ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   $esi = COPY renamable $ebx
+  ; CHECK:   $edx = COPY killed renamable $ebp
+  ; CHECK:   $rcx = COPY killed renamable $rax
+  ; CHECK:   $r8d = MOV32r0 implicit-def dead $eflags
+  ; CHECK:   MOV64mr %stack.3, 1, $noreg, 0, $noreg, killed $r14 :: (store 8 into %stack.3)
+  ; CHECK:   STATEPOINT 2, 5, 5, undef renamable $rax, undef $rdi, $esi, $edx, $rcx, killed $r8d, 2, 0, 2, 0, 2, 7, undef renamable $rax, 1, 8, %stack.0, 0, 1, 8, %stack.1, 0, 1, 8, %stack.2, 0, renamable $ebx, undef renamable $eax, 2, 6, 2, 2, 1, 8, %stack.3, 0, undef renamable $rbx, 2, 0, 2, 2, 0, 0, 1, 1, csr_64, implicit-def $rsp, implicit-def $ssp, implicit killed $rbx :: (volatile load store 8 on %stack.0), (volatile load store 8 on %stack.1), (volatile load store 8 on %stack.2), (load store 8 on %stack.3)
+  ; CHECK:   $r14 = MOV64rm %stack.3, 1, $noreg, 0, $noreg :: (load 8 from %stack.3)
+  ; CHECK:   ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+  ; CHECK:   RET 0
+  bb.0.bb:
+    successors: %bb.1(0x80000000)
+    liveins: $rcx, $rdi, $rdx, $rsi
+
+    renamable $r15 = COPY $rcx
+    renamable $r12 = COPY $rdx
+    renamable $r14 = COPY $rsi
+    renamable $r13 = COPY $rdi
+    renamable $ebx = MOV32rm undef renamable $rax, 1, $noreg, 0, $noreg, implicit-def $rbx :: (load unordered 4 from `i32 addrspace(1)* undef`, align 8, addrspace 1)
+
+  bb.1.bb7:
+    liveins: $rbx, $r12, $r13, $r14, $r15
+
+    renamable $ebp = LEA64_32r renamable $rbx, 1, $noreg, 6, $noreg
+    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+    CALL64pcrel32 target-flags(x86-plt) @wombat, csr_64, implicit $rsp, implicit $ssp, implicit-def $rsp, implicit-def $ssp, implicit-def $rax
+    ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+    MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $r13 :: (store 8 into %stack.0)
+    MOV64mr %stack.1, 1, $noreg, 0, $noreg, killed renamable $r12 :: (store 8 into %stack.1)
+    MOV64mr %stack.2, 1, $noreg, 0, $noreg, killed renamable $r15 :: (store 8 into %stack.2)
+    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+    $esi = COPY renamable $ebx
+    $edx = COPY killed renamable $ebp
+    $rcx = COPY killed renamable $rax
+    $r8d = MOV32r0 implicit-def dead $eflags
+    dead renamable $r14, dead renamable $rbx = STATEPOINT 2, 5, 5, undef renamable $rax, undef $rdi, $esi, $edx, $rcx, killed $r8d, 2, 0, 2, 0, 2, 7, undef renamable $rax, 1, 8, %stack.0, 0, 1, 8, %stack.1, 0, 1, 8, %stack.2, 0, renamable $ebx, undef renamable $eax, 2, 6, 2, 2, killed renamable $r14(tied-def 0), undef renamable $rbx(tied-def 1), 2, 0, 2, 2, 0, 0, 1, 1, csr_64, implicit-def $rsp, implicit-def $ssp, implicit killed $rbx :: (volatile load store 8 on %stack.0), (volatile load store 8 on %stack.1), (volatile load store 8 on %stack.2)
+    ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+    RET 0
+
+...