AVX-512: Fixed a bug in lowering saturated operations on KNL

[delena]

Summary

The code is not optimal, but it should be correct, at least.

[igorb]

LGTM

../lib/Target/X86/X86ISelLowering.cpp

if (N->getValueType(0) == MVT::i8) {
       Cond = X86::COND_O;
       break;
     }
     SDVTList VTs = DAG.getVTList(N->getValueType(0), N->getValueType(0),
                                  MVT::i32);
     SDValue Sum = DAG.getNode(X86ISD::UMUL, DL, VTs, LHS, RHS);
 
     SDValue SetCC =
-      DAG.getNode(X86ISD::SETCC, DL, N->getValueType(1),
+      DAG.getNode(X86ISD::SETCC, DL, MVT::i8,
                   DAG.getConstant(X86::COND_O, DL, MVT::i32),
                   SDValue(Sum.getNode(), 2));
 
+    if (N->getValueType(1) == MVT::i1)
+      SetCC = DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, SetCC);
+
     return DAG.getNode(ISD::MERGE_VALUES, DL, N->getVTList(), Sum, SetCC);
   }
   }
 
   // Also sets EFLAGS.
   SDVTList VTs = DAG.getVTList(N->getValueType(0), MVT::i32);
   SDValue Sum = DAG.getNode(BaseOp, DL, VTs, LHS, RHS);
 
   SDValue SetCC =
-    DAG.getNode(X86ISD::SETCC, DL, N->getValueType(1),
+    DAG.getNode(X86ISD::SETCC, DL, MVT::i8,
                 DAG.getConstant(Cond, DL, MVT::i32),
                 SDValue(Sum.getNode(), 1));
 
+  if (N->getValueType(1) == MVT::i1)
+    SetCC = DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, SetCC);
+
   return DAG.getNode(ISD::MERGE_VALUES, DL, N->getVTList(), Sum, SetCC);
 }
 
 /// Returns true if the operand type is exactly twice the native width, and
 /// the corresponding cmpxchg8b or cmpxchg16b instruction is available.
 /// Used to know whether to use cmpxchg8/16b when expanding atomic operations
 /// (otherwise we leave them alone to become __sync_fetch_and_... calls).
 bool X86TargetLowering::needsCmpXchgNb(Type *MemType) const {

../test/CodeGen/X86/xaluo.ll

 overflow:
   ret i1 false
 
 continue:
   ret i1 true
 }
 
 define i1 @bug27873(i64 %c1, i1 %c2) {
+; CHECK-LABEL: bug27873:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movl $160, %ecx
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    mulq %rcx
+; CHECK-NEXT:    seto %al
+; CHECK-NEXT:    orb %sil, %al
+; CHECK-NEXT:    retq
+;
 ; KNL-LABEL: bug27873:
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    andl $1, %esi
 ; KNL-NEXT:    kmovw %esi, %k0
 ; KNL-NEXT:    movl $160, %ecx
 ; KNL-NEXT:    movq %rdi, %rax
 ; KNL-NEXT:    mulq %rcx
 ; KNL-NEXT:    seto %al
+; KNL-NEXT:    andl $1, %eax
 ; KNL-NEXT:    kmovw %eax, %k1
 ; KNL-NEXT:    korw %k1, %k0, %k0
 ; KNL-NEXT:    kmovw %k0, %eax
 ; KNL-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
 ; KNL-NEXT:    retq
   %mul = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %c1, i64 160)
   %mul.overflow = extractvalue { i64, i1 } %mul, 1
   %x1 = or i1 %c2, %mul.overflow