[FastISel][X86] Fix INC/DEC optimization (r215230)

I accidentally also used INC/DEC for unsigned arithmetic which doesn't work,
because INC/DEC don't set the required flag which is used for the overflow
check.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215237 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 9563db7e45049f3a047110bf015d705080d4bc52..db3a4fdc9cc7a098608efeeda3a0be6239da108b 100644 (file)
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -2410,7 +2410,7 @@ bool X86FastISel::FastLowerIntrinsicCall(const IntrinsicInst *II) {
           { X86::DEC8r, X86::DEC64_16r, X86::DEC64_32r, X86::DEC64r }  }
       };
 
           { X86::DEC8r, X86::DEC64_16r, X86::DEC64_32r, X86::DEC64r }  }
       };
 

-      if (UseIncDec) {
+      if (BaseOpc == X86ISD::INC || BaseOpc == X86ISD::DEC) {

         ResultReg = createResultReg(TLI.getRegClassFor(VT));
         bool Is64Bit = Subtarget->is64Bit();
         bool IsDec = BaseOpc == X86ISD::DEC;
         ResultReg = createResultReg(TLI.getRegClassFor(VT));
         bool Is64Bit = Subtarget->is64Bit();
         bool IsDec = BaseOpc == X86ISD::DEC;

diff --git a/test/CodeGen/X86/xaluo.ll b/test/CodeGen/X86/xaluo.ll
index 5e6cd25adedea307db314697341b65ec0427197f..bc1d0ce77c6f2c5e5fbf8727d642dde4e728b900 100644 (file)
--- a/test/CodeGen/X86/xaluo.ll
+++ b/test/CodeGen/X86/xaluo.ll
@@ -231,6 +231,51 @@ entry:
   ret i1 %obit
 }
 
   ret i1 %obit
 }
 

+; UADDO reg, 1 | NOT INC
+define zeroext i1 @uaddo.inc.i8(i8 %v1, i8* %res) {
+entry:
+; CHECK-LABEL: uaddo.inc.i8
+; CHECK-NOT:   incb %dil
+  %t = call {i8, i1} @llvm.uadd.with.overflow.i8(i8 %v1, i8 1)
+  %val = extractvalue {i8, i1} %t, 0
+  %obit = extractvalue {i8, i1} %t, 1
+  store i8 %val, i8* %res
+  ret i1 %obit
+}
+
+define zeroext i1 @uaddo.inc.i16(i16 %v1, i16* %res) {
+entry:
+; CHECK-LABEL: uaddo.inc.i16
+; CHECK-NOT:   incw %di
+  %t = call {i16, i1} @llvm.uadd.with.overflow.i16(i16 %v1, i16 1)
+  %val = extractvalue {i16, i1} %t, 0
+  %obit = extractvalue {i16, i1} %t, 1
+  store i16 %val, i16* %res
+  ret i1 %obit
+}
+
+define zeroext i1 @uaddo.inc.i32(i32 %v1, i32* %res) {
+entry:
+; CHECK-LABEL: uaddo.inc.i32
+; CHECK-NOT:   incl %edi
+  %t = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %v1, i32 1)
+  %val = extractvalue {i32, i1} %t, 0
+  %obit = extractvalue {i32, i1} %t, 1
+  store i32 %val, i32* %res
+  ret i1 %obit
+}
+
+define zeroext i1 @uaddo.inc.i64(i64 %v1, i64* %res) {
+entry:
+; CHECK-LABEL: uaddo.inc.i64
+; CHECK-NOT:   incq %rdi
+  %t = call {i64, i1} @llvm.uadd.with.overflow.i64(i64 %v1, i64 1)
+  %val = extractvalue {i64, i1} %t, 0
+  %obit = extractvalue {i64, i1} %t, 1
+  store i64 %val, i64* %res
+  ret i1 %obit
+}
+

 ; SSUBO
 define zeroext i1 @ssubo.i32(i32 %v1, i32 %v2, i32* %res) {
 entry:
 ; SSUBO
 define zeroext i1 @ssubo.i32(i32 %v1, i32 %v2, i32* %res) {
 entry:


@@ -758,6 +803,8 @@ declare {i8,  i1} @llvm.sadd.with.overflow.i8 (i8,  i8 ) nounwind readnone
 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16, i16) nounwind readnone
 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32, i32) nounwind readnone
 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64, i64) nounwind readnone
 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16, i16) nounwind readnone
 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32, i32) nounwind readnone
 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64, i64) nounwind readnone

+declare {i8,  i1} @llvm.uadd.with.overflow.i8 (i8,  i8 ) nounwind readnone
+declare {i16, i1} @llvm.uadd.with.overflow.i16(i16, i16) nounwind readnone

 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone
 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64, i64) nounwind readnone
 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone
 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone
 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64, i64) nounwind readnone
 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone