Win64: Stack alignment constraints aren't applied during SET_FPREG

Stack realignment occurs after the prolog, not during, for Win64.
Because of this, don't factor in the maximum stack alignment when
establishing a frame pointer.

This fixes PR22572.

llvm-svn: 230113

llvm/lib/Target/X86/X86FrameLowering.cpp

@@ -458,7 +458,7 @@ static unsigned calculateSetFPREG(uint64_t SPAdjust) {
   const uint64_t Win64MaxSEHOffset = 128;
   uint64_t SEHFrameOffset = std::min(SPAdjust, Win64MaxSEHOffset);
   // Win64 ABI requires 16-byte alignment for the UWOP_SET_FPREG opcode.
-  return static_cast<unsigned>(RoundUpToAlignment(SEHFrameOffset, 16));
+  return SEHFrameOffset & -16;
 }
 
 // If we're forcing a stack realignment we can't rely on just the frame
@@ -672,14 +672,12 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
     // If required, include space for extra hidden slot for stashing base pointer.
     if (X86FI->getRestoreBasePointer())
       FrameSize += SlotSize;
-    if (RegInfo->needsStackRealignment(MF)) {
+
-      // Callee-saved registers are pushed on stack before the stack
+    NumBytes = FrameSize - X86FI->getCalleeSavedFrameSize();
-      // is realigned.
+
-      FrameSize -= X86FI->getCalleeSavedFrameSize();
+    // Callee-saved registers are pushed on stack before the stack is realigned.
-      NumBytes = RoundUpToAlignment(FrameSize, MaxAlign);
+    if (RegInfo->needsStackRealignment(MF) && !IsWinEH)
-    } else {
+      NumBytes = RoundUpToAlignment(NumBytes, MaxAlign);
-      NumBytes = FrameSize - X86FI->getCalleeSavedFrameSize();
-    }
 
     // Get the offset of the stack slot for the EBP register, which is
     // guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
@@ -801,7 +799,10 @@ void X86FrameLowering::emitPrologue(MachineFunction &MF) const {
   // responsible for adjusting the stack pointer.  Touching the stack at 4K
   // increments is necessary to ensure that the guard pages used by the OS
   // virtual memory manager are allocated in correct sequence.
-  if (NumBytes >= StackProbeSize && UseStackProbe) {
+  uint64_t AlignedNumBytes = NumBytes;
+  if (IsWinEH && RegInfo->needsStackRealignment(MF))
+    AlignedNumBytes = RoundUpToAlignment(AlignedNumBytes, MaxAlign);
+  if (AlignedNumBytes >= StackProbeSize && UseStackProbe) {
     // Check whether EAX is livein for this function.
     bool isEAXAlive = isEAXLiveIn(MF);
 
@@ -1005,14 +1006,12 @@ void X86FrameLowering::emitEpilogue(MachineFunction &MF,
   if (hasFP(MF)) {
     // Calculate required stack adjustment.
     uint64_t FrameSize = StackSize - SlotSize;
-    if (RegInfo->needsStackRealignment(MF)) {
+    NumBytes = FrameSize - CSSize;
-      // Callee-saved registers were pushed on stack before the stack
+
-      // was realigned.
+    // Callee-saved registers were pushed on stack before the stack was
-      FrameSize -= CSSize;
+    // realigned.
-      NumBytes = (FrameSize + MaxAlign - 1) / MaxAlign * MaxAlign;
+    if (RegInfo->needsStackRealignment(MF) && !IsWinEH)
-    } else {
+      NumBytes = RoundUpToAlignment(FrameSize, MaxAlign);
-      NumBytes = FrameSize - CSSize;
-    }
 
     // Pop EBP.
     BuildMI(MBB, MBBI, DL,
@@ -1185,25 +1184,15 @@ int X86FrameLowering::getFrameIndexOffset(const MachineFunction &MF,
   int64_t FPDelta = 0;
 
   if (IsWinEH) {
-    uint64_t NumBytes = 0;
+    assert(!MFI->hasCalls() || (StackSize % 16) == 8);
+
     // Calculate required stack adjustment.
     uint64_t FrameSize = StackSize - SlotSize;
     // If required, include space for extra hidden slot for stashing base pointer.
     if (X86FI->getRestoreBasePointer())
       FrameSize += SlotSize;
-    uint64_t SEHStackAllocAmt = StackSize;
+    uint64_t NumBytes = FrameSize - CSSize;
-    if (RegInfo->needsStackRealignment(MF)) {
-      // Callee-saved registers are pushed on stack before the stack
-      // is realigned.
-      FrameSize -= CSSize;
 
-      uint64_t MaxAlign =
-          calculateMaxStackAlign(MF); // Desired stack alignment.
-      NumBytes = RoundUpToAlignment(FrameSize, MaxAlign);
-      SEHStackAllocAmt = RoundUpToAlignment(SEHStackAllocAmt, 16);
-    } else {
-      NumBytes = FrameSize - CSSize;
-    }
     uint64_t SEHFrameOffset = calculateSetFPREG(NumBytes);
     if (FI && FI == X86FI->getFAIndex())
       return -SEHFrameOffset;
@@ -1212,7 +1201,9 @@ int X86FrameLowering::getFrameIndexOffset(const MachineFunction &MF,
     // pointer followed by return address and the location required by the
     // restricted Win64 prologue.
     // Add FPDelta to all offsets below that go through the frame pointer.
-    FPDelta = SEHStackAllocAmt - SEHFrameOffset;
+    FPDelta = FrameSize - SEHFrameOffset;
+    assert((!MFI->hasCalls() || (FPDelta % 16) == 0) &&
+           "FPDelta isn't aligned per the Win64 ABI!");
   }
 
 
@@ -1237,8 +1228,6 @@ int X86FrameLowering::getFrameIndexOffset(const MachineFunction &MF,
   } else {
     if (!HasFP)
       return Offset + StackSize;
-    if (IsWinEH)
-      return Offset + FPDelta;
 
     // Skip the saved EBP.
     Offset += SlotSize;
@@ -1249,7 +1238,7 @@ int X86FrameLowering::getFrameIndexOffset(const MachineFunction &MF,
       Offset -= TailCallReturnAddrDelta;
   }
 
-  return Offset;
+  return Offset + FPDelta;
 }
 
 int X86FrameLowering::getFrameIndexReference(const MachineFunction &MF, int FI,

llvm/test/CodeGen/X86/win64_eh.ll

@@ -150,18 +150,18 @@ entry:
 ; WIN64: .seh_pushreg 7
 ; WIN64: pushq %rbx
 ; WIN64: .seh_pushreg 3
-; WIN64: subq  $128, %rsp
+; WIN64: subq  $96, %rsp
-; WIN64: .seh_stackalloc 128
+; WIN64: .seh_stackalloc 96
-; WIN64: leaq  128(%rsp), %rbp
+; WIN64: leaq  96(%rsp), %rbp
-; WIN64: .seh_setframe 5, 128
+; WIN64: .seh_setframe 5, 96
-; WIN64: movaps  %xmm7, -32(%rbp)        # 16-byte Spill
+; WIN64: movaps  %xmm7, -16(%rbp)        # 16-byte Spill
-; WIN64: .seh_savexmm 7, 96
+; WIN64: .seh_savexmm 7, 80
-; WIN64: movaps  %xmm6, -48(%rbp)        # 16-byte Spill
+; WIN64: movaps  %xmm6, -32(%rbp)        # 16-byte Spill
-; WIN64: .seh_savexmm 6, 80
+; WIN64: .seh_savexmm 6, 64
 ; WIN64: .seh_endprologue
 ; WIN64: andq  $-64, %rsp
-; WIN64: movaps  -48(%rbp), %xmm6        # 16-byte Reload
+; WIN64: movaps  -32(%rbp), %xmm6        # 16-byte Reload
-; WIN64: movaps  -32(%rbp), %xmm7        # 16-byte Reload
+; WIN64: movaps  -16(%rbp), %xmm7        # 16-byte Reload
 ; WIN64: leaq  (%rbp), %rsp
 ; WIN64: popq  %rbx
 ; WIN64: popq  %rdi

llvm/test/CodeGen/X86/win64_frame.ll

@@ -9,10 +9,10 @@ define i32 @f1(i32 %p1, i32 %p2, i32 %p3, i32 %p4, i32 %p5) "no-frame-pointer-el
 define void @f2(i32 %p, ...) "no-frame-pointer-elim"="true" {
   ; CHECK-LABEL: f2:
   ; CHECK:      .seh_stackalloc 8
-  ; CHECK:      leaq    16(%rsp), %rbp
+  ; CHECK:      leaq    (%rsp), %rbp
-  ; CHECK:      .seh_setframe 5, 16
+  ; CHECK:      .seh_setframe 5, 0
-  ; CHECK:      movq    %rdx, 16(%rbp)
+  ; CHECK:      movq    %rdx, 32(%rbp)
-  ; CHECK:      leaq    16(%rbp), %rax
+  ; CHECK:      leaq    32(%rbp), %rax
   %ap = alloca i8, align 8
   call void @llvm.va_start(i8* %ap)
   ret void
@@ -76,20 +76,21 @@ define i32 @f7(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) "no-frame-pointer-elim"="
   ; CHECK-LABEL: f7:
   ; CHECK:      pushq   %rbp
   ; CHECK:      .seh_pushreg 5
-  ; CHECK:      subq    $320, %rsp
+  ; CHECK:      subq    $304, %rsp
-  ; CHECK:      .seh_stackalloc 320
+  ; CHECK:      .seh_stackalloc 304
   ; CHECK:      leaq    128(%rsp), %rbp
   ; CHECK:      .seh_setframe 5, 128
-  ; CHECK:      movl    240(%rbp), %eax
+  ; CHECK:      andq    $-64, %rsp
-  ; CHECK:      leaq    192(%rbp), %rsp
+  ; CHECK:      movl    224(%rbp), %eax
+  ; CHECK:      leaq    176(%rbp), %rsp
   alloca [300 x i8], align 64
   ret i32 %e
 }
 
 define i32 @f8(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) "no-frame-pointer-elim"="true" {
   ; CHECK-LABEL: f8:
-  ; CHECK:        subq    $384, %rsp
+  ; CHECK:        subq    $352, %rsp
-  ; CHECK:        .seh_stackalloc 384
+  ; CHECK:        .seh_stackalloc 352
   ; CHECK:        leaq    128(%rsp), %rbp
   ; CHECK:        .seh_setframe 5, 128
 
@@ -113,7 +114,7 @@ define i32 @f8(i32 %a, i32 %b, i32 %c, i32 %d, i32 %e) "no-frame-pointer-elim"="
 
   ret i32 %e
   ; CHECK:        movl    %esi, %eax
-  ; CHECK:        leaq    256(%rbp), %rsp
+  ; CHECK:        leaq    224(%rbp), %rsp
 }
 
 declare i8* @llvm.returnaddress(i32) nounwind readnone