Re: [PATCH 11/20] tcg/i386: Implement avx512 immediate rotate

[Richard Henderson]

On 2/3/22 01:05, Alex Bennée wrote:
> Is there any reason why gen_shift couldn't be pushed into a helper
> function so we just had:
>
>      static void tcg_out_vec_shift(s, vece, insn, sub, a0, a1, a2) {
>          tcg_debug_assert(vece != MO_8);
>          if (type == TCG_TYPE_V256) {
>              insn |= P_VEXL;
>          }
>          tcg_out_vex_modrm(s, insn, sub, a0, a1);
>          tcg_out8(s, a2);
>      }
>
>      ...
>
>      case INDEX_op_rotli_vec:
>          insn = OPC_PSHIFTD_Ib | P_EVEX;  /* VPROL[DQ] */
>          if (vece == MO_64) {
>              insn |= P_VEXW;
>          }
>          tcg_out_vec_shift(s, vece, insn, 1, a0, a1, a2);
>          break;
>
> Surely the compiler would inline if needed (and even if it didn't it the
> code generation that critical we care about a few cycles)?

Yes, I suppose I could pull out a helper or two.
Just one of those things where something used to be cleaner, and then the code 
grew.

r~