From 9401b2fcabb264c0397a6650c7696c44c236be5b Mon Sep 17 00:00:00 2001
From: Dan Gohman <gohman@apple.com>
Date: Tue, 15 Sep 2009 00:14:11 +0000
Subject: [PATCH] On x86-64, the 32-bit cmov doesn't actually clear the high
 32-bit of its result if the condition is false.

llvm-svn: 81814
---
 lib/Target/X86/X86Instr64bit.td |  8 +++++---
 test/CodeGen/X86/cmov-zext.ll   | 19 +++++++++++++++++++
 2 files changed, 24 insertions(+), 3 deletions(-)
 create mode 100644 test/CodeGen/X86/cmov-zext.ll

diff --git a/lib/Target/X86/X86Instr64bit.td b/lib/Target/X86/X86Instr64bit.td
index 353868a407a..ac90357038b 100644
--- a/lib/Target/X86/X86Instr64bit.td
+++ b/lib/Target/X86/X86Instr64bit.td
@@ -387,13 +387,15 @@ def MOVZX64rm32 : I<0x8B, MRMSrcMem, (outs GR64:$dst), (ins i32mem:$src),
                     [(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,
diff --git a/test/CodeGen/X86/cmov-zext.ll b/test/CodeGen/X86/cmov-zext.ll
new file mode 100644
index 00000000000..8df228a00aa
--- /dev/null
+++ b/test/CodeGen/X86/cmov-zext.ll
@@ -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
+}