[DAG] Avoid Node Update assertion due to AND simplification

[niravd]

Summary

Check for construction-time folding for incomplete AND nodes in BackwardsPropagateMask. Fixes PR38185.

[samparker]

This looks like an odd solution to me, I haven't seen TokenFactors used like that before. Isn't it okay for the AND to be folded? Why not just check that the AND hasn't be folded into a constant before trying to update its, now non-existent, operands?

[niravd]

Using a temporary node avoids the intermediate loop issue, though a better temp node structure would be preferable.
But you're correct that any simplification to the AND obviates the need for fixup so let's us that.

[samparker]

LGTM, thanks!

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

if (SearchForAndLoads(N, Loads, NodesWithConsts, Mask, FixupNode)) {
     // If it exists, fixup the single node we allow in the tree that needs
     // masking.
     if (FixupNode) {
       LLVM_DEBUG(dbgs() << "First, need to fix up: "; FixupNode->dump());
       SDValue And = DAG.getNode(ISD::AND, SDLoc(FixupNode),
                                 FixupNode->getValueType(0),
                                 SDValue(FixupNode, 0), MaskOp);
       DAG.ReplaceAllUsesOfValueWith(SDValue(FixupNode, 0), And);
-      DAG.UpdateNodeOperands(And.getNode(), SDValue(FixupNode, 0),
-                             MaskOp);
+      if (And.getOpcode() == ISD ::AND)
+        DAG.UpdateNodeOperands(And.getNode(), SDValue(FixupNode, 0), MaskOp);
     }
 
     // Narrow any constants that need it.
     for (auto *LogicN : NodesWithConsts) {
       SDValue Op0 = LogicN->getOperand(0);
       SDValue Op1 = LogicN->getOperand(1);
 
       if (isa<ConstantSDNode>(Op0))
           std::swap(Op0, Op1);
 
       SDValue And = DAG.getNode(ISD::AND, SDLoc(Op1), Op1.getValueType(),
                                 Op1, MaskOp);
 
       DAG.UpdateNodeOperands(LogicN, Op0, And);
     }
 
     // Create narrow loads.
     for (auto *Load : Loads) {
       LLVM_DEBUG(dbgs() << "Propagate AND back to: "; Load->dump());
       SDValue And = DAG.getNode(ISD::AND, SDLoc(Load), Load->getValueType(0),
                                 SDValue(Load, 0), MaskOp);
       DAG.ReplaceAllUsesOfValueWith(SDValue(Load, 0), And);
-      DAG.UpdateNodeOperands(And.getNode(), SDValue(Load, 0), MaskOp);
+      if (And.getOpcode() == ISD ::AND)
+        And = SDValue(
+            DAG.UpdateNodeOperands(And.getNode(), SDValue(Load, 0), MaskOp), 0);
       SDValue NewLoad = ReduceLoadWidth(And.getNode());
       assert(NewLoad &&
              "Shouldn't be masking the load if it can't be narrowed");
       CombineTo(Load, NewLoad, NewLoad.getValue(1));
     }
     DAG.ReplaceAllUsesWith(N, N->getOperand(0).getNode());
     return true;
   }

llvm/trunk/test/CodeGen/X86/pr38185.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -o - %s -mtriple=x86_64--unknown-linux-gnu | FileCheck %s
+
+define void @foo(i32* %a, i32* %b, i32* noalias %c, i64 %s) {
+; CHECK-LABEL: foo:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movq $0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    jmp .LBB0_1
+; CHECK-NEXT:    .p2align 4, 0x90
+; CHECK-NEXT:  .LBB0_2: # %body
+; CHECK-NEXT:    # in Loop: Header=BB0_1 Depth=1
+; CHECK-NEXT:    movl $1, (%rdx,%r9,4)
+; CHECK-NEXT:    movzbl (%rdi,%r9,4), %r8d
+; CHECK-NEXT:    movzbl (%rsi,%r9,4), %eax
+; CHECK-NEXT:    andl %r8d, %eax
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    movl %eax, (%rdi,%r9,4)
+; CHECK-NEXT:    incq %r9
+; CHECK-NEXT:    movq %r9, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:  .LBB0_1: # %loop
+; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT:    movq -{{[0-9]+}}(%rsp), %r9
+; CHECK-NEXT:    cmpq %rcx, %r9
+; CHECK-NEXT:    jne .LBB0_2
+; CHECK-NEXT:  # %bb.3: # %endloop
+; CHECK-NEXT:    retq
+%i = alloca i64
+store i64 0, i64* %i
+br label %loop
+
+loop:
+%ct = load i64, i64* %i
+%comp = icmp eq i64 %ct, %s
+br i1 %comp, label %endloop, label %body
+
+body:
+%var0 = getelementptr i32, i32* %c, i64 %ct
+store i32 1, i32* %var0
+%var1 = getelementptr i32, i32* %c, i64 %ct
+%var2 = load i32, i32* %var1
+%var3 = add i32 %var2, 1
+%var4 = getelementptr i32, i32* %a, i64 %ct
+%var5 = load i32, i32* %var4
+%var6 = and i32 %var3, %var5
+%var7 = add i32 %var6, 1
+%var8 = getelementptr i32, i32* %a, i64 %ct
+%var9 = load i32, i32* %var8
+%var10 = and i32 %var7, %var9
+%var11 = getelementptr i32, i32* %c, i64 %ct
+%var12 = load i32, i32* %var11
+%var13 = and i32 %var10, %var12
+%var14 = getelementptr i32, i32* %b, i64 %ct
+%var15 = load i32, i32* %var14
+%var16 = and i32 %var15, 63
+%var17 = and i32 %var13, %var16
+%var18 = getelementptr i32, i32* %a, i64 %ct
+store i32 %var17, i32* %var18
+%z = add i64 1, %ct
+store i64 %z, i64* %i
+br label %loop
+
+endloop:
+ret void
+}