[PATCH] D95890: [RISCV] Alternate attempt to optimize sign-extended EXTRACT_VECTOR_ELT nodes.

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: llvm/test/CodeGen/Mips/atomic.ll:4300
 ; MIPS32EB-NEXT:    jr $ra
 ; MIPS32EB-NEXT:    sra $2, $1, 24
 entry:
----------------
jrtc27 wrote:
> Strange that this one didn't change
AtomicCmpSwap8 uses ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS which is expanded by SelectionDAGLegalize::ExpandNode to a sequence that includes an AssertSExt.

AtomicSwap8 just uses ISD::ATOMIC_SWAP and does not go through any expanding. A custom inserter creates ATOMIC_SWAP_I8_POSTRA and then that is expanded by MipsExpandPseudo.cpp. Only after that do the earlier shifts appear.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D95890/new/

https://reviews.llvm.org/D95890