1 files changed, 18 insertions, 14 deletions

diff --git a/lib/Target/X86/X86VZeroUpper.cpp b/lib/Target/X86/X86VZeroUpper.cpp
index d93baeb..99ba4c0 100644
--- a/lib/Target/X86/X86VZeroUpper.cpp
+++ b/lib/Target/X86/X86VZeroUpper.cpp
@@ -9,7 +9,7 @@
 //
 // This file defines the pass which inserts x86 AVX vzeroupper instructions
 // before calls to SSE encoded functions. This avoids transition latency
-// penalty when tranfering control between AVX encoded instructions and old
+// penalty when transferring control between AVX encoded instructions and old
 // SSE encoding mode.
 //
 //===----------------------------------------------------------------------===//
@@ -171,7 +171,7 @@ void VZeroUpperInserter::addDirtySuccessor(MachineBasicBlock &MBB) {
 }
 
 /// processBasicBlock - Loop over all of the instructions in the basic block,
-/// inserting vzero upper instructions before function calls.
+/// inserting vzeroupper instructions before function calls.
 void VZeroUpperInserter::processBasicBlock(MachineBasicBlock &MBB) {
 
   // Start by assuming that the block PASS_THROUGH, which implies no unguarded
@@ -202,7 +202,7 @@ void VZeroUpperInserter::processBasicBlock(MachineBasicBlock &MBB) {
     // If the call won't clobber any YMM register, skip it as well. It usually
     // happens on helper function calls (such as '_chkstk', '_ftol2') where
     // standard calling convention is not used (RegMask is not used to mark
-    // register clobbered and register usage (def/imp-def/use) is well-dfined
+    // register clobbered and register usage (def/imp-def/use) is well-defined
     // and explicitly specified.
     if (MI->isCall() && !callClobbersAnyYmmReg(MI))
       continue;
@@ -245,25 +245,29 @@ void VZeroUpperInserter::processBasicBlock(MachineBasicBlock &MBB) {
 }
 
 /// runOnMachineFunction - Loop over all of the basic blocks, inserting
-/// vzero upper instructions before function calls.
+/// vzeroupper instructions before function calls.
 bool VZeroUpperInserter::runOnMachineFunction(MachineFunction &MF) {
-  const X86Subtarget &ST = MF.getTarget().getSubtarget<X86Subtarget>();
+  const X86Subtarget &ST = MF.getSubtarget<X86Subtarget>();
   if (!ST.hasAVX() || ST.hasAVX512())
     return false;
-  TII = MF.getSubtarget().getInstrInfo();
+  TII = ST.getInstrInfo();
   MachineRegisterInfo &MRI = MF.getRegInfo();
   EverMadeChange = false;
 
+  bool FnHasLiveInYmm = checkFnHasLiveInYmm(MRI);
+
   // Fast check: if the function doesn't use any ymm registers, we don't need
   // to insert any VZEROUPPER instructions.  This is constant-time, so it is
   // cheap in the common case of no ymm use.
-  bool YMMUsed = false;
-  const TargetRegisterClass *RC = &X86::VR256RegClass;
-  for (TargetRegisterClass::iterator i = RC->begin(), e = RC->end();
-       i != e; i++) {
-    if (!MRI.reg_nodbg_empty(*i)) {
-      YMMUsed = true;
-      break;
+  bool YMMUsed = FnHasLiveInYmm;
+  if (!YMMUsed) {
+    const TargetRegisterClass *RC = &X86::VR256RegClass;
+    for (TargetRegisterClass::iterator i = RC->begin(), e = RC->end(); i != e;
+         i++) {
+      if (!MRI.reg_nodbg_empty(*i)) {
+        YMMUsed = true;
+        break;
+      }
     }
   }
   if (!YMMUsed) {
@@ -282,7 +286,7 @@ bool VZeroUpperInserter::runOnMachineFunction(MachineFunction &MF) {
 
   // If any YMM regs are live in to this function, add the entry block to the
   // DirtySuccessors list
-  if (checkFnHasLiveInYmm(MRI))
+  if (FnHasLiveInYmm)
     addDirtySuccessor(MF.front());
 
   // Re-visit all blocks that are successors of EXITS_DIRTY bsocks. Add