[X86] Don't promote i16 compares to i32 if the immediate will fit in 8 bits.

The comments in this code say we were trying to avoid 16-bit immediates, but if the immediate fits in 8-bits this isn't an issue. This avoids creating a zero extend that probably won't go away.

The movmskb related changes are interesting. The movmskb instruction writes a 32-bit result, but fills the upper bits with 0. So the zero_extend we were previously emitting was free, but we turned a -1 immediate that would fit in 8-bits into a 32-bit immediate so it was still bad.

llvm-svn: 343871

1 files changed, 5 insertions, 2 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 91fa216c053..f63fd3e0463 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -18435,8 +18435,11 @@ SDValue X86TargetLowering::EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
        Op0.getValueType() == MVT::i32 || Op0.getValueType() == MVT::i64)) {
     // Only promote the compare up to I32 if it is a 16 bit operation
     // with an immediate.  16 bit immediates are to be avoided.
-    if ((Op0.getValueType() == MVT::i16 &&
-         (isa<ConstantSDNode>(Op0) || isa<ConstantSDNode>(Op1))) &&
+    if (Op0.getValueType() == MVT::i16 &&
+        ((isa<ConstantSDNode>(Op0) &&
+          !cast<ConstantSDNode>(Op0)->getAPIntValue().isSignedIntN(8)) ||
+         (isa<ConstantSDNode>(Op1) &&
+          !cast<ConstantSDNode>(Op1)->getAPIntValue().isSignedIntN(8))) &&
         !DAG.getMachineFunction().getFunction().optForMinSize() &&
         !Subtarget.isAtom()) {
       unsigned ExtendOp =