Re: [PATCH v2 08/14] tcg/riscv: Implement vector cmp ops

[Richard Henderson]

On 9/2/24 23:45, Richard Henderson wrote:
> I think the first implementation should be simpler:
>
> CONST('C', TCG_CT_CONST_CMP_VI)
>
> tcg_target_const_match()
> {
>      ...
>      if ((ct & TCG_CT_CONST_CMP_VI) &&
>          val >= tcg_cmpcond_to_rvv_vi[cond].min &&
>          val <= tcg_cmpcond_to_rvv_vi[cond].max) {
>          return true;
>      }
> }
>
>      case INDEX_op_cmp_vec:
>          riscv_set_vec_config_vl_vece(s, type, vece);
>          cond = args[3];
>          if (c2) {
>              tcg_out_opc_vi(s, tcg_cmpcond_to_rvv_vi[cond].op, a0, a1,
>                             a2 - tcg_cmpcond_to_rvv_vi[cond].adjust);
>          } else if (tcg_cmpcond_to_rvv_vv[cond].swap) {
>              tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op, a0, a2, a1);
>          } else {
>              tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op, a0, a1, a2);
>          }
>          break;
>
> This appears to not require any expansion in tcg_expand_vec_op at all.

I knew I should have slept on that answer.
Of course you need expansion, because riscv cmp_vv produces a mask.

However, I think we should simply model this as INDEX_op_cmpsel_vec:

    case INDEX_op_cmpsel_vec:
          riscv_set_vec_config_vl_vece(s, type, vece);
          a3 = args[3];
          a4 = args[4];
          cond = args[5];
          /* Use only vmerge_vim if possible, by inverting the test. */
          if (const_args[4] && !const_args[3]) {
              cond = tcg_cond_inv(cond);
              a3 = a4;
              a4 = args[3];
              const_args[3] = true;
              const_args[4] = false;
          }
          /* Perform the comparison into V0 mask. */
          if (const_args[2]) {
              tcg_out_opc_vi(s, tcg_cmpcond_to_rvv_vi[cond].op,
                             TCG_REG_V0, a1,
                             a2 - tcg_cmpcond_to_rvv_vi[cond].adjust);
          } else if (tcg_cmpcond_to_rvv_vv[cond].swap) {
              tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
                             TCG_REG_V0, a2, a1);
          } else {
              tcg_out_opc_vv(s, tcg_cmpcond_to_rvv_vv[cond].op,
                             TCG_REG_V0, a1, a2);
          }
          if (const_args[3]) {
              if (const_args[4]) {
                  tcg_out_opc_vi(s, OPC_VMV_V_I, a0, TCG_REG_V0, a4, true);
                  a4 = a0;
              }
              tcg_out_opc_vim_mask(s, OPC_VMERGE_VIM, a0, a3, a4);
          } else {
              tcg_out_opc_vvm_mask(s, OPC_VMERGE_VVM, a0, a3, a4);
          }
          break;

Then INDEX_op_cmp_vec should be expanded to

    INDEX_op_cmpsel_vec a0, a1, a2, -1, 0, a3


r~