[clang] [llvm] [AArch64] Add intrinsic support for Fdot instr. (PR #189987)

[via cfe-commits]

================
@@ -6699,6 +6699,13 @@ multiclass SIMDThreeSameVectorFDot<string asm, SDPatternOperator OpNode = null_f
                                          v2f32, v4f16, OpNode>;
   def v8f16_v4f32 : BaseSIMDThreeSameVectorDot<1, 0, 0b10, 0b1111, asm, ".4s", ".8h", V128,
                                          v4f32, v8f16, OpNode>;
+
+  def : Pat<(v2f32 (int_aarch64_neon_fdot (v2f32 V64:$Rd),
----------------
CarolineConcatto wrote:

Can we follow what was done for UDOT and SDOT. They did some work on the dot instructions to be able to optimize:
commit f51b3de4e851812b5f7d7c307ddb7b6ec61c05ab
So maybe we should create ISD node for FDOT as well.

https://github.com/llvm/llvm-project/pull/189987