[PATCH] D106376: [SVE] Add support for folding for select + masked loads

[Steven Johnson via Phabricator via llvm-commits]

srj added a comment.

This appears to have inserted an obscure failure into Halide code generation, but only for certain vector architectures (e.g. x86-64 with AVX or AVX2, but not SSE4.1). We end up assert-failing:

  InstructionSimplify.cpp:1981: llvm::Value* simplifyLogicOfAddSub(llvm::Value*, llvm::Value*, llvm::Instruction::BinaryOps): Assertion `Op0->getType() == Op1->getType() && "Mismatched binop types"' failed.

where the two binops are

  %483 = icmp slt i32 %482, 0
  %481 = icmp sle <8 x i32> %478, %480

(I'm working on a repro case to provide, but so far I'm hampered by the fact that the assertion failure occurs when trying to produce a useful .ll file -- I'm posting this as a heads-up in case the short description rings any bells.)


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106376/new/

https://reviews.llvm.org/D106376