Re: [Qemu-devel] [PATCH v3 10/16] tcg-arm: Define TCG_TARGET_INSN_UNIT_SIZE

[Peter Maydell <peter.maydell@linaro.org>]

On 28 April 2014 20:28, Richard Henderson <rth@twiddle.net> wrote:
> And use tcg pointer differencing functions as appropriate.
>
> Cc: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
> @@ -1670,51 +1651,28 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
>
>      switch (opc) {
>      case INDEX_op_exit_tb:
> -        if (use_armv7_instructions || check_fit_imm(args[0])) {
> -            tcg_out_movi32(s, COND_AL, TCG_REG_R0, args[0]);
> -            tcg_out_goto(s, COND_AL, (tcg_target_ulong) tb_ret_addr);
> -        } else {
> -            uint8_t *ld_ptr = s->code_ptr;
> -            tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
> -            tcg_out_goto(s, COND_AL, (tcg_target_ulong) tb_ret_addr);
> -            *ld_ptr = (uint8_t) (s->code_ptr - ld_ptr) - 8;
> -            tcg_out32(s, args[0]);
> -        }
> +        tcg_out_movi32(s, COND_AL, TCG_REG_R0, args[0]);
> +        tcg_out_goto(s, COND_AL, tb_ret_addr);

Why is it OK not to have the pre-v7/non-fitting-immediate
code now? I guess movi32 will handle it all though probably
in a less efficient way.

>          break;
>      case INDEX_op_goto_tb:
>          if (s->tb_jmp_offset) {
>              /* Direct jump method */
> -#if defined(USE_DIRECT_JUMP)
> -            s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
> +            s->tb_jmp_offset[args[0]] = tcg_current_code_size(s);
>              tcg_out_b_noaddr(s, COND_AL);
> -#else
> -            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, -4);
> -            s->tb_jmp_offset[args[0]] = s->code_ptr - s->code_buf;
> -            tcg_out32(s, 0);
> -#endif
>          } else {
>              /* Indirect jump method */
> -#if 1
> -            c = (int) (s->tb_next + args[0]) - ((int) s->code_ptr + 8);
> -            if (c > 0xfff || c < -0xfff) {
> -                tcg_out_movi32(s, COND_AL, TCG_REG_R0,
> -                                (tcg_target_long) (s->tb_next + args[0]));
> -                tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
> -            } else
> -                tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_PC, c);
> -#else
> -            tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_PC, 0);
> -            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, 0);
> -            tcg_out32(s, (tcg_target_long) (s->tb_next + args[0]));
> -#endif
> +            intptr_t ptr = (intptr_t)(s->tb_next + args[0]);
> +            tcg_out_movi32(s, COND_AL, TCG_REG_R0, ptr & ~0xfff);
> +            tcg_out_ld32_12(s, COND_AL, TCG_REG_PC, TCG_REG_R0, ptr & 0xfff);
>          }

This change also confused me but we're again relying on movi32
generating correct-but-inefficient code now, right?

thanks
-- PMM