Re: [PATCH 5/5] target/ppc: Implement paddi and replace addi insns

[Richard Henderson]

On 4/14/21 12:11 PM, Richard Henderson wrote:
> This approach seems like it will work fine for MLS and MMIR prefixes.  For 8LS, 
> 8RR, and MRR prefixes, we'll need some extra help within ppc_tr_translate_insn. 
>   E.g.
>
>      insn = translator_ldl_swap(env, ctx->base.pc_next,
>                                 need_byteswap(ctx));
>      switch (ctx->prefix_type) {
>      case PREFIX_NONE:
>          ok = decode_opcode_space_0(ctx, insn) ||
>               decode_legacy(ctx, insn);
>          break;
>      case PREFIX_MLS:
>      case PREFIX_MMIRR:
>          ok = decode_opcode_space_0(ctx, insn);
>          break;

I played about with this last night, and there's an interesting trade-off:

(1) The thousands of 32-bit insns which do not allow prefixes
    now each require 3 lines to assert that no prefix is present,

(2) There are only 12 MLS and 29 MMIRR prefixed insns.

I think it may well be that eliminating boiler-plate from thousands of insns it 
a good trade-off to special-casing 41 insns.

At which point, considering the multiple variations on 8RR and MMIRR prefixes, 
seems to indicate that we should decode all 64 bits all at once.


r~