X86: Emit Win64 SaveXMM opcodes at the right offset in the right order

[majnemer]

Summary

Walk the instructions marked FrameSetup and consider any stores of XMM
registers to the stack as needing a SaveXMM opcode.

This fixes PR22521.

[rnk]

Nice, lgtm

lib/Target/X86/X86FrameLowering.cpp

if (IsWinEH && HasFP) {
 
     if (NeedsWinEH)
       BuildMI(MBB, MBBI, DL, TII.get(X86::SEH_SetFrame))
           .addImm(FramePtr)
           .addImm(SEHFrameOffset)
           .setMIFlag(MachineInstr::FrameSetup);
   }
 
-  // Skip the rest of register spilling code
-  while (MBBI != MBB.end() && MBBI->getFlag(MachineInstr::FrameSetup))
+  while (MBBI != MBB.end() && MBBI->getFlag(MachineInstr::FrameSetup)) {
+    const MachineInstr *FrameInstr = &*MBBI;
     ++MBBI;
 
     if (NeedsWinEH) {
-    for (const CalleeSavedInfo &Info : MFI->getCalleeSavedInfo()) {
-      unsigned Reg = Info.getReg();
-      if (X86::GR64RegClass.contains(Reg) || X86::GR32RegClass.contains(Reg))
-        continue;
-      assert(X86::FR64RegClass.contains(Reg) && "Unexpected register class");
-
-      int Offset = getFrameIndexOffset(MF, Info.getFrameIdx());
+      int FI;
+      if (unsigned Reg = TII.isStoreToStackSlot(FrameInstr, FI)) {
+        if (X86::FR64RegClass.contains(Reg)) {
+          int Offset = getFrameIndexOffset(MF, FI);
           Offset += SEHFrameOffset;
 
           BuildMI(MBB, MBBI, DL, TII.get(X86::SEH_SaveXMM))
               .addImm(Reg)
               .addImm(Offset)
               .setMIFlag(MachineInstr::FrameSetup);
         }
+      }
+    }
+  }
 
+  if (NeedsWinEH)
     BuildMI(MBB, MBBI, DL, TII.get(X86::SEH_EndPrologue))
         .setMIFlag(MachineInstr::FrameSetup);
-  }
 
   // Realign stack after we spilled callee-saved registers (so that we'll be
   // able to calculate their offsets from the frame pointer).
   // Win64 requires aligning the stack after the prologue.
   if (IsWinEH && RegInfo->needsStackRealignment(MF)) {
     assert(HasFP && "There should be a frame pointer if stack is realigned.");
     uint64_t Val = -MaxAlign;
     MachineInstr *MI =
for (unsigned i = CSI.size(); i != 0; --i) {
     BuildMI(MBB, MI, DL, TII.get(Opc)).addReg(Reg, RegState::Kill)
       .setMIFlag(MachineInstr::FrameSetup);
   }
 
   // Make XMM regs spilled. X86 does not have ability of push/pop XMM.
   // It can be done by spilling XMMs to stack frame.
   for (unsigned i = CSI.size(); i != 0; --i) {
     unsigned Reg = CSI[i-1].getReg();
-    if (X86::GR64RegClass.contains(Reg) ||
-        X86::GR32RegClass.contains(Reg))
+    if (X86::GR64RegClass.contains(Reg) || X86::GR32RegClass.contains(Reg))
       continue;
     // Add the callee-saved register as live-in. It's killed at the spill.
     MBB.addLiveIn(Reg);
     const TargetRegisterClass *RC = TRI->getMinimalPhysRegClass(Reg);
 
     TII.storeRegToStackSlot(MBB, MI, Reg, true, CSI[i - 1].getFrameIdx(), RC,
                             TRI);
     --MI;

test/CodeGen/X86/win64_eh.ll

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