Re: [PATCH v3 13/14] tcg/arm: Reserve a register for guest_base

[Peter Maydell]

On Wed, 18 Aug 2021 at 22:33, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Reserve a register for the guest_base using aarch64 for reference.
> By doing so, we do not have to recompute it for every memory load.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/arm/tcg-target.c.inc | 39 ++++++++++++++++++++++++++++-----------
>  1 file changed, 28 insertions(+), 11 deletions(-)
>
> diff --git a/tcg/arm/tcg-target.c.inc b/tcg/arm/tcg-target.c.inc
> index 35bd4c68d6..2728035177 100644
> --- a/tcg/arm/tcg-target.c.inc
> +++ b/tcg/arm/tcg-target.c.inc
> @@ -84,6 +84,9 @@ static const int tcg_target_call_oarg_regs[2] = {
>
>  #define TCG_REG_TMP  TCG_REG_R12
>  #define TCG_VEC_TMP  TCG_REG_Q15
> +#ifndef CONFIG_SOFTMMU
> +#define TCG_REG_GUEST_BASE  TCG_REG_R11
> +#endif
>
>  typedef enum {
>      COND_EQ = 0x0,
> @@ -1763,7 +1766,8 @@ static bool tcg_out_qemu_st_slow_path(TCGContext *s, 
> TCGLabelQemuLdst *lb)
>
>  static void tcg_out_qemu_ld_index(TCGContext *s, MemOp opc,
>                                    TCGReg datalo, TCGReg datahi,
> -                                  TCGReg addrlo, TCGReg addend)
> +                                  TCGReg addrlo, TCGReg addend,
> +                                  bool scratch_addend)
>  {
>      /* Byte swapping is left to middle-end expansion. */
>      tcg_debug_assert((opc & MO_BSWAP) == 0);
> @@ -1790,7 +1794,7 @@ static void tcg_out_qemu_ld_index(TCGContext *s, MemOp 
> opc,
>              && get_alignment_bits(opc) >= MO_64
>              && (datalo & 1) == 0 && datahi == datalo + 1) {
>              tcg_out_ldrd_r(s, COND_AL, datalo, addrlo, addend);
> -        } else if (datalo != addend) {
> +        } else if (scratch_addend) {
>              tcg_out_ld32_rwb(s, COND_AL, datalo, addend, addrlo);
>              tcg_out_ld32_12(s, COND_AL, datahi, addend, 4);
>          } else {

I don't understand this change. Yes, we can trash the addend
register, but if it's the same as 'datalo' then the second load
is not going to DTRT... Shouldn't this be
  if (scratch_addend && datalo != addend)
?

-- PMM