[PATCH] D152350: [X86][FP16] Do not generate VBROADCAST for fp16

[Phoebe Wang via Phabricator via llvm-commits]

pengfei added inline comments.


================
Comment at: llvm/test/CodeGen/X86/vector-half-conversions.ll:942
+; AVX1-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; AVX1-NEXT:    vbroadcastss 8(%rdi), %xmm0
+; AVX1-NEXT:    callq __extendhfsf2 at PLT
----------------
LuoYuanke wrote:
> vbroadcastss would access 32-bit memory and vpinsrw only access 16-bit memory?
The `vbroadcastss` is an optimization because the high 16-bit is undefined.
`ptr %a0` is a `<8 x i16>` pointer, so we are free to access 32-bit memory without worrying about the size.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152350/new/

https://reviews.llvm.org/D152350