frasercrmck added a comment.
Is "destination operand" the terminology we want? I'd have thought "passthru"
was more conventional.
================
Comment at: llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp:1194
Operands.push_back(Node->getOperand(CurOp++));
+ if (Node->getOperand(CurOp).isUndef())
+ CurOp++;
----------------
If I'm reading this right, isn't it possible that we could have a masked/tu
intrinsic, so push back the passthru operand, increment `CurOp` then check
whether the //pointer// operand is Undef, then wrongly increment `CurOp` a
second time? So basically an undef base pointer has the potential to
(presumably) crash the compiler somewhere later?
Don't we unconditionally want to increment `CurOp` (a single time) since we now
always have a passthru operand that we want to skip over?
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D117647/new/
https://reviews.llvm.org/D117647
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
