[DAG] Handle missing transform in fold of value extension case.

[niravd]

Summary

When pushing an extension of a constant bitwise operator on a load
into the load, change other uses of the load value if they exist to
prevent the old load from persisting.

[efriedma]

Do we need to check isTruncateFree or something here?

[niravd]

There's a check hidden in ExtendUsesToFormExtLoad.

In D35030#802717, @efriedma wrote:

Do we need to check isTruncateFree or something here?

[niravd]

Ping.

[niravd]

Rebase past PR34095 change

[RKSimon]

LGTM

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (LN0->getExtensionType() != ISD::ZEXTLOAD && LN0->isUnindexed()) {
         APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
         Mask = Mask.sext(VT.getSizeInBits());
         SDValue And = DAG.getNode(N0.getOpcode(), DL, VT,
                                   ExtLoad, DAG.getConstant(Mask, DL, VT));
         SDValue Trunc = DAG.getNode(ISD::TRUNCATE,
                                     SDLoc(N0.getOperand(0)),
                                     N0.getOperand(0).getValueType(), ExtLoad);
         ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL, ISD::SIGN_EXTEND);
+        bool NoReplaceTruncAnd = !N0.hasOneUse();
         bool NoReplaceTrunc = SDValue(LN0, 0).hasOneUse();
         CombineTo(N, And);
+        // If N0 has multiple uses, change other uses as well.
+        if (NoReplaceTruncAnd) {
+          SDValue TruncAnd =
+              DAG.getNode(ISD::TRUNCATE, DL, N0.getValueType(), And);
+          CombineTo(N0.getNode(), TruncAnd);
+        }
         if (NoReplaceTrunc)
           DAG.ReplaceAllUsesOfValueWith(SDValue(LN0, 1), ExtLoad.getValue(1));
         else
           CombineTo(LN0, Trunc, ExtLoad.getValue(1));
         return SDValue(N,0); // Return N so it doesn't get rechecked!
       }
     }
   }
if (LN0->getExtensionType() != ISD::SEXTLOAD && LN0->isUnindexed()) {
         Mask = Mask.zext(VT.getSizeInBits());
         SDLoc DL(N);
         SDValue And = DAG.getNode(N0.getOpcode(), DL, VT,
                                   ExtLoad, DAG.getConstant(Mask, DL, VT));
         SDValue Trunc = DAG.getNode(ISD::TRUNCATE,
                                     SDLoc(N0.getOperand(0)),
                                     N0.getOperand(0).getValueType(), ExtLoad);
         ExtendSetCCUses(SetCCs, Trunc, ExtLoad, DL, ISD::ZERO_EXTEND);
+        bool NoReplaceTruncAnd = !N0.hasOneUse();
         bool NoReplaceTrunc = SDValue(LN0, 0).hasOneUse();
         CombineTo(N, And);
+        // If N0 has multiple uses, change other uses as well.
+        if (NoReplaceTruncAnd) {
+          SDValue TruncAnd =
+              DAG.getNode(ISD::TRUNCATE, DL, N0.getValueType(), And);
+          CombineTo(N0.getNode(), TruncAnd);
+        }
         if (NoReplaceTrunc)
           DAG.ReplaceAllUsesOfValueWith(SDValue(LN0, 1), ExtLoad.getValue(1));
         else
           CombineTo(LN0, Trunc, ExtLoad.getValue(1));
         return SDValue(N,0); // Return N so it doesn't get rechecked!
       }
     }
   }

llvm/trunk/test/CodeGen/X86/avx512-mask-op.ll

 ;}
 
 @f1.v = internal unnamed_addr global i1 false, align 4
 
 define void @f1(i32 %c) {
 ; CHECK-LABEL: f1:
 ; CHECK:       ## BB#0: ## %entry
 ; CHECK-NEXT:    movzbl {{.*}}(%rip), %edi
-; CHECK-NEXT:    movb {{.*}}(%rip), %al
-; CHECK-NEXT:    notb %al
-; CHECK-NEXT:    andb $1, %al
-; CHECK-NEXT:    movb %al, {{.*}}(%rip)
 ; CHECK-NEXT:    xorl $1, %edi
+; CHECK-NEXT:    movb %dil, {{.*}}(%rip)
 ; CHECK-NEXT:    jmp _f2 ## TAILCALL
 entry:
   %.b1 = load i1, i1* @f1.v, align 4
   %not..b1 = xor i1 %.b1, true
   store i1 %not..b1, i1* @f1.v, align 4
   %0 = zext i1 %not..b1 to i32
   tail call void @f2(i32 %0) #2
   ret void