Re: [Qemu-devel] [PATCH 2/7] tcg: Optimize add2 + sub2

[Blue Swirl]

On Thu, Sep 27, 2012 at 5:19 PM, Richard Henderson <address@hidden> wrote:
> We can't do complete constant folding because we lack "mov2",
> or the ability to insert opcodes in the stream.  But we can
> at least canonicalize add2 operand ordering and simplify
> add2 to add when the lowpart adds a constant 0.

Couldn't we introduce add2_part1 and add2_part2, the latter being nop
for architectures that don't need it?

>
> Signed-off-by: Richard Henderson <address@hidden>
> ---
>  tcg/optimize.c | 31 +++++++++++++++++++++++++++++++
>  1 file changed, 31 insertions(+)
>
> diff --git a/tcg/optimize.c b/tcg/optimize.c
> index 55f2a24..004c336 100644
> --- a/tcg/optimize.c
> +++ b/tcg/optimize.c
> @@ -470,6 +470,11 @@ static TCGArg *tcg_constant_folding(TCGContext *s, 
> uint16_t *tcg_opc_ptr,
>              if (swap_commutative(args[0], &args[4], &args[3])) {
>                  args[5] = tcg_invert_cond(args[5]);
>              }
> +            break;
> +        case INDEX_op_add2_i32:
> +            swap_commutative(args[0], &args[2], &args[4]);
> +            swap_commutative(args[1], &args[3], &args[5]);
> +            break;
>          default:
>              break;
>          }
> @@ -522,6 +527,32 @@ static TCGArg *tcg_constant_folding(TCGContext *s, 
> uint16_t *tcg_opc_ptr,
>                  continue;
>              }
>              break;
> +        case INDEX_op_add2_i32:
> +        case INDEX_op_sub2_i32:
> +            /* Simplify op rl, rh, al, ah, 0, bh => op rh, ah, bh.
> +               The zero implies there will be no carry into the high part.
> +               But only when rl == al, since we can't insert the extra move
> +               that would be required.  */
> +            if (temps[args[4]].state == TCG_TEMP_CONST
> +                && temps[args[4]].val == 0
> +                && temps_are_copies(args[0], args[2])) {
> +                if (temps[args[5]].state == TCG_TEMP_CONST
> +                    && temps[args[5]].val == 0
> +                    && temps_are_copies(args[1], args[3])) {
> +                    gen_opc_buf[op_index] = INDEX_op_nop;
> +                } else {
> +                    gen_opc_buf[op_index] = (op == INDEX_op_add2_i32
> +                                             ? INDEX_op_add_i32
> +                                             : INDEX_op_sub_i32);
> +                    args[0] = args[1];
> +                    args[1] = args[3];
> +                    args[2] = args[5];
> +                    gen_args += 3;
> +                }
> +                args += 6;
> +                continue;
> +            }
> +            break;
>          default:
>              break;
>          }
> --
> 1.7.11.4
>
>