X86: optimization for -(x != 0)

This patch will optimize -(x != 0) on X86
FROM
cmpl	$0x01,%edi
sbbl	%eax,%eax
notl	%eax
TO
negl %edi
sbbl %eax %eax

In order to generate negl, I added patterns in Target/X86/X86InstrCompiler.td:
def : Pat<(X86sub_flag 0, GR32:$src), (NEG32r GR32:$src)>;

rdar: 10961709

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156312 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 3e631382715f8f9ab00a7d1a75cadd46a57e4024..8404c7a3d5ed9b8b8d6b7d870accb3bd59d5a244 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8665,6 +8665,22 @@ SDValue X86TargetLowering::LowerSELECT(SDValue Op, SelectionDAG &DAG) const {
       SDValue Y = isAllOnes(Op2) ? Op1 : Op2;
 
       SDValue CmpOp0 = Cmp.getOperand(0);
+      // Apply further optimizations for special cases
+      // (select (x != 0), -1, 0) -> neg & sbb
+      // (select (x == 0), 0, -1) -> neg & sbb
+      if (ConstantSDNode *YC = dyn_cast<ConstantSDNode>(Y))
+        if (YC->isNullValue() && 
+            (isAllOnes(Op1) == (CondCode == X86::COND_NE))) {
+          SDVTList VTs = DAG.getVTList(CmpOp0.getValueType(), MVT::i32);
+          SDValue Neg = DAG.getNode(X86ISD::SUB, DL, VTs, 
+                                    DAG.getConstant(0, CmpOp0.getValueType()), 
+                                    CmpOp0);
+          SDValue Res = DAG.getNode(X86ISD::SETCC_CARRY, DL, Op.getValueType(),
+                                    DAG.getConstant(X86::COND_B, MVT::i8),
+                                    SDValue(Neg.getNode(), 1));
+          return Res;
+        }
+
       Cmp = DAG.getNode(X86ISD::CMP, DL, MVT::i32,
                         CmpOp0, DAG.getConstant(1, CmpOp0.getValueType()));
       Cmp = ConvertCmpIfNecessary(Cmp, DAG);

diff --git a/lib/Target/X86/X86InstrCompiler.td b/lib/Target/X86/X86InstrCompiler.td
index 6f9e8492613a74831068baff88ca65eaba7b5cb9..87634dde0b90358fe6255cb246e61240b96f25dc 100644 (file)
--- a/lib/Target/X86/X86InstrCompiler.td
+++ b/lib/Target/X86/X86InstrCompiler.td
@@ -1623,6 +1623,12 @@ def : Pat<(sub GR16:$src1, i16immSExt8:$src2),
 def : Pat<(sub GR32:$src1, i32immSExt8:$src2),
           (SUB32ri8 GR32:$src1, i32immSExt8:$src2)>;
 
+// sub 0, reg
+def : Pat<(X86sub_flag 0, GR8 :$src), (NEG8r  GR8 :$src)>;
+def : Pat<(X86sub_flag 0, GR16:$src), (NEG16r GR16:$src)>;
+def : Pat<(X86sub_flag 0, GR32:$src), (NEG32r GR32:$src)>;
+def : Pat<(X86sub_flag 0, GR64:$src), (NEG64r GR64:$src)>;
+
 // mul reg, reg
 def : Pat<(mul GR16:$src1, GR16:$src2),
           (IMUL16rr GR16:$src1, GR16:$src2)>;

diff --git a/test/CodeGen/X86/select.ll b/test/CodeGen/X86/select.ll
index 9adf4f9c050dd566fb193934479155ec709e85e0..c8d9345c400f6d9e4abfcf7ad19c54d6f03bb035 100644 (file)
--- a/test/CodeGen/X86/select.ll
+++ b/test/CodeGen/X86/select.ll
@@ -218,3 +218,33 @@ define i32 @test14(i32 %a, i32 %b) nounwind {
 ; CHECK-NEXT: ret
 }
 
+; rdar://10961709
+define i32 @test15(i32 %x) nounwind {
+entry:
+  %cmp = icmp ne i32 %x, 0
+  %sub = sext i1 %cmp to i32
+  ret i32 %sub
+; CHECK: test15:
+; CHECK: negl
+; CHECK: sbbl
+}
+
+define i64 @test16(i64 %x) nounwind uwtable readnone ssp {
+entry:
+  %cmp = icmp ne i64 %x, 0
+  %conv1 = sext i1 %cmp to i64
+  ret i64 %conv1
+; CHECK: test16:
+; CHECK: negq
+; CHECK: sbbq
+}
+
+define i16 @test17(i16 %x) nounwind {
+entry:
+  %cmp = icmp ne i16 %x, 0
+  %sub = sext i1 %cmp to i16
+  ret i16 %sub
+; CHECK: test17:
+; CHECK: negw
+; CHECK: sbbw
+}