[llvm] [X86] Promote approximate v8f16 FMA to v8f32 (PR #228900)

[via llvm-commits]

================
@@ -0,0 +1,1060 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=x86-64-v2 | FileCheck %s --check-prefix=NOFMA
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=x86-64-v3 | FileCheck %s --check-prefix=FMA
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=x86-64-v4 -mattr=+avx512fp16 | FileCheck %s --check-prefix=FP16
+
+; Check v8f16 FMA lowering with approximate, exact, and native FP16 paths.
+
+define <8 x half> @afn_fma(<8 x half> %a, <8 x half> %b, <8 x half> %c) {
+; NOFMA-LABEL: afn_fma:
+; NOFMA:       # %bb.0:
+; NOFMA-NEXT:    subq $120, %rsp
+; NOFMA-NEXT:    .cfi_def_cfa_offset 128
----------------
Andarwinux wrote:

add nounwind to silence cfi noise

https://github.com/llvm/llvm-project/pull/228900