[clang] [compiler-rt] [llvm] [X86][AVX10.2] Support AVX10.2 option and VMPSADBW/VADDP[D,H,S] new instructions (PR #101452)

[Phoebe Wang via cfe-commits]

================
@@ -0,0 +1,33 @@
+//===-- X86InstrAVX10.td - AVX10 Instruction Set -----------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file describes the X86 AVX10 instruction set, defining the
+// instructions, and properties of the instructions which are needed for code
+// generation, machine code emission, and analysis.
+//
+//===----------------------------------------------------------------------===//
+
+// VMPSADBW
+defm VMPSADBW : avx512_common_3Op_rm_imm8<0x42, X86Vmpsadbw, "vmpsadbw", SchedWritePSADBW,
+                                          avx512vl_i16_info, avx512vl_i8_info,
+                                          HasAVX10_2>,
+                    XS, EVEX_CD8<32, CD8VF>;
+
+// YMM Rounding
+multiclass avx256_fp_binop_p_round<bits<8> opc, string OpcodeStr, SDNode OpNodeRnd,
----------------
phoebewang wrote:

I think for brand new instructions, it's better to use avx10 prefix. Here I want to highlight they are for YMM rounding only. So avx256 should be better.

https://github.com/llvm/llvm-project/pull/101452