[llvm-commits] [llvm] r81814 - in /llvm/trunk: lib/Target/X86/X86Instr64bit.td test/CodeGen/X86/cmov-zext.ll

[Dan Gohman]

Author: djg
Date: Mon Sep 14 19:14:11 2009
New Revision: 81814

URL: http://llvm.org/viewvc/llvm-project?rev=81814&view=rev
Log:
On x86-64, the 32-bit cmov doesn't actually clear the high 32-bit of
its result if the condition is false.

Added:
    llvm/trunk/test/CodeGen/X86/cmov-zext.ll
Modified:
    llvm/trunk/lib/Target/X86/X86Instr64bit.td

Modified: llvm/trunk/lib/Target/X86/X86Instr64bit.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Instr64bit.td?rev=81814&r1=81813&r2=81814&view=diff

==============================================================================
--- llvm/trunk/lib/Target/X86/X86Instr64bit.td (original)
+++ llvm/trunk/lib/Target/X86/X86Instr64bit.td Mon Sep 14 19:14:11 2009
@@ -387,13 +387,15 @@
                     [(set GR64:$dst, (zextloadi64i32 addr:$src))]>;
 
 // Any instruction that defines a 32-bit result leaves the high half of the
-// register. Truncate can be lowered to EXTRACT_SUBREG, and CopyFromReg may
-// be copying from a truncate, but any other 32-bit operation will zero-extend
+// register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
+// be copying from a truncate. And x86's cmov doesn't do anything if the
+// condition is false. But any other 32-bit operation will zero-extend
 // up to 64 bits.
 def def32 : PatLeaf<(i32 GR32:$src), [{
   return N->getOpcode() != ISD::TRUNCATE &&
          N->getOpcode() != TargetInstrInfo::EXTRACT_SUBREG &&
-         N->getOpcode() != ISD::CopyFromReg;
+         N->getOpcode() != ISD::CopyFromReg &&
+         N->getOpcode() != X86ISD::CMOV;
 }]>;
 
 // In the case of a 32-bit def that is known to implicitly zero-extend,

Added: llvm/trunk/test/CodeGen/X86/cmov-zext.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/cmov-zext.ll?rev=81814&view=auto

==============================================================================
--- llvm/trunk/test/CodeGen/X86/cmov-zext.ll (added)
+++ llvm/trunk/test/CodeGen/X86/cmov-zext.ll Mon Sep 14 19:14:11 2009
@@ -0,0 +1,19 @@
+; RUN: llc < %s -march=x86-64 | FileCheck %s
+
+; x86's 32-bit cmov doesn't clobber the high 32 bits of the destination
+; if the condition is false. An explicit zero-extend (movl) is needed
+; after the cmov.
+
+; CHECK:      cmovne  %edi, %esi
+; CHECK-NEXT: movl    %esi, %edi
+
+declare void @bar(i64) nounwind
+
+define void @foo(i64 %a, i64 %b, i1 %p) nounwind {
+  %c = trunc i64 %a to i32
+  %d = trunc i64 %b to i32
+  %e = select i1 %p, i32 %c, i32 %d
+  %f = zext i32 %e to i64
+  call void @bar(i64 %f)
+  ret void
+}