From 5e74c97ad77f097f3363c930cad1c2a42ba80014 Mon Sep 17 00:00:00 2001
From: David Majnemer <david.majnemer@gmail.com>
Date: Thu, 5 May 2016 06:00:56 +0000
Subject: [PATCH] [X86] Use the right type when folding xor (truncate (shift))
 -> setcc

The result type of setcc is dependent on whether or not AVX512 is
present.
We had an X86-specific DAG-combine which assumed that the result type
should be i8 when it could be i1.
This meant that we would generate illegal setccs which LowerSETCC did
not like.

Instead, use an appropriate type and zero extend to i8.

Also, there were some scenarios where the fold should have fired but
didn't because we were overly cautious about the types.  This meant that
we generated:

        shrl    $31, %edi
        andl    $1, %edi
        kmovw   %edi, %k0
        kxnorw  %k0, %k0, %k1
        kshiftrw        $15, %k1, %k1
        kxorw   %k1, %k0, %k0
        kmovw   %k0, %eax

instead of:

        testl   %edi, %edi
        setns   %al

This fixes PR27638.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@268609 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp | 12 +++++++++---
 test/CodeGen/X86/setcc.ll          | 24 ++++++++++++++++++++++++
 2 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 531a0a15341..d3457ed841f 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -27307,8 +27307,9 @@ static SDValue combineIntegerAbs(SDNode *N, SelectionDAG &DAG) {
 /// into:
 ///   SETGT(X, -1)
 static SDValue foldXorTruncShiftIntoCmp(SDNode *N, SelectionDAG &DAG) {
-  // This is only worth doing if the output type is i8.
-  if (N->getValueType(0) != MVT::i8)
+  // This is only worth doing if the output type is i8 or i1.
+  EVT ResultType = N->getValueType(0);
+  if (ResultType != MVT::i8 && ResultType != MVT::i1)
     return SDValue();
 
   SDValue N0 = N->getOperand(0);
@@ -27343,8 +27344,13 @@ static SDValue foldXorTruncShiftIntoCmp(SDNode *N, SelectionDAG &DAG) {
   SDLoc DL(N);
   SDValue ShiftOp = Shift.getOperand(0);
   EVT ShiftOpTy = ShiftOp.getValueType();
-  SDValue Cond = DAG.getSetCC(DL, MVT::i8, ShiftOp,
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  EVT SetCCResultType = TLI.getSetCCResultType(DAG.getDataLayout(),
+                                               *DAG.getContext(), ResultType);
+  SDValue Cond = DAG.getSetCC(DL, SetCCResultType, ShiftOp,
                               DAG.getConstant(-1, DL, ShiftOpTy), ISD::SETGT);
+  if (SetCCResultType != ResultType)
+    Cond = DAG.getNode(ISD::ZERO_EXTEND, DL, ResultType, Cond);
   return Cond;
 }
 
diff --git a/test/CodeGen/X86/setcc.ll b/test/CodeGen/X86/setcc.ll
index b4847c54ffa..d5874344fd1 100644
--- a/test/CodeGen/X86/setcc.ll
+++ b/test/CodeGen/X86/setcc.ll
@@ -54,3 +54,27 @@ entry:
   %add = shl nuw nsw i32 %conv4.2, 16
   ret i32 %add
 }
+
+define i8 @t5(i32 %a) #0 {
+entry:
+; CHECK-LABEL: t5:
+; CHECK:  testl   %edi, %edi
+; CHECK:  setns   %al
+  %.lobit = lshr i32 %a, 31
+  %trunc = trunc i32 %.lobit to i8
+  %.not = xor i8 %trunc, 1
+  ret i8 %.not
+}
+
+define zeroext i1 @t6(i32 %a) #0 {
+entry:
+; CHECK-LABEL: t6:
+; CHECK:  testl   %edi, %edi
+; CHECK:  setns   %al
+  %.lobit = lshr i32 %a, 31
+  %trunc = trunc i32 %.lobit to i1
+  %.not = xor i1 %trunc, 1
+  ret i1 %.not
+}
+
+attributes #0 = { "target-cpu"="skylake-avx512" }
-- 
2.50.1