[clang] [llvm] [Clang][AArch64] Implement widening FMMLA intrinsics (PR #165282)

[via llvm-commits]

================
@@ -4745,10 +4745,14 @@ defm FMLALLTT_ZZZ : sve2_fp8_mla<0b011, ZPR32, "fmlalltt", nxv4f32, int_aarch64_
 
 let Predicates = [HasSVE2, HasF8F32MM] in {
   def FMMLA_ZZZ_BtoS :  sve2_fp8_mmla<0b0, ZPR32, "fmmla">;
+  def : Pat<(nxv4f32 (int_aarch64_sve_fmmla_mf8f32 nxv4f32:$acc, nxv16i8:$zn, nxv16i8:$zm)), 
----------------
Lukacma wrote:

It will probably be better to copy what was done for sve_fp_matrix_mla and create a multiclass for it, where you bundle pattern and instruction definition together.

https://github.com/llvm/llvm-project/pull/165282