[PATCH] D105268: [X86] AVX512FP16 instructions enabling 5/6

Sat Aug 21 08:04:05 PDT 2021

pengfei added inline comments.

================
Comment at: clang/include/clang/Basic/BuiltinsX86.def:2010
+TARGET_BUILTIN(__builtin_ia32_vfmaddph, "V8xV8xV8xV8x", "ncV:128:", "avx512fp16,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_vfmaddph256, "V16xV16xV16xV16x", "ncV:256:", "avx512fp16,avx512vl")
+
----------------
LuoYuanke wrote:
> Can we arrange the vfmaddph variant together?  Move it to line 1997?
> Why there is no mask version for 128 and 256?
We followed what're ps/pd doing. As Craig explained, this is history's legacy. We should fix them in future.

================
Comment at: llvm/test/CodeGen/X86/vec-strict-128-fp16.ll:105

+define <8 x half> @f13(<8 x half> %a, <8 x half> %b, <8 x half> %c) #0 {
+; CHECK-LABEL: f13:
----------------
LuoYuanke wrote:
> Is it necessary to test 132, 231 version?
213 is the preferred version due to its order in `SelectCode` table. We can test others by using memory input, but they are covered in stack-folding-fp-avx512fp16vl-fma.ll. I don't think it's necessary to test here.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105268/new/

https://reviews.llvm.org/D105268