Re: [Qemu-devel] [PATCH v11 13/20] target/arm: Use vector infrastructure for aa64 mov/not/neg

["Alex Bennée" <alex.bennee@linaro.org>]

Richard Henderson <richard.henderson@linaro.org> writes:

> Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>

> ---
>  target/arm/translate-a64.c | 42 ++++++++++++++++++++++++++++++++++++------
>  1 file changed, 36 insertions(+), 6 deletions(-)
>
> diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
> index 5a4e62ae0f..11310f1a7a 100644
> --- a/target/arm/translate-a64.c
> +++ b/target/arm/translate-a64.c
> @@ -86,6 +86,7 @@ typedef void CryptoThreeOpIntFn(TCGv_ptr, TCGv_ptr, TCGv_i32);
>  typedef void CryptoThreeOpFn(TCGv_ptr, TCGv_ptr, TCGv_ptr);
>
>  /* Note that the gvec expanders operate on offsets + sizes.  */
> +typedef void GVecGen2Fn(unsigned, uint32_t, uint32_t, uint32_t, uint32_t);
>  typedef void GVecGen3Fn(unsigned, uint32_t, uint32_t,
>                          uint32_t, uint32_t, uint32_t);
>
> @@ -631,6 +632,14 @@ static TCGv_ptr get_fpstatus_ptr(void)
>      return statusptr;
>  }
>
> +/* Expand a 2-operand AdvSIMD vector operation using an expander function.  */
> +static void gen_gvec_fn2(DisasContext *s, bool is_q, int rd, int rn,
> +                         GVecGen2Fn *gvec_fn, int vece)
> +{
> +    gvec_fn(vece, vec_full_reg_offset(s, rd), vec_full_reg_offset(s, rn),
> +            is_q ? 16 : 8, vec_full_reg_size(s));
> +}
> +
>  /* Expand a 3-operand AdvSIMD vector operation using an expander function.  */
>  static void gen_gvec_fn3(DisasContext *s, bool is_q, int rd, int rn, int rm,
>                           GVecGen3Fn *gvec_fn, int vece)
> @@ -4596,14 +4605,17 @@ static void handle_fp_1src_double(DisasContext *s, int opcode, int rd, int rn)
>      TCGv_i64 tcg_op;
>      TCGv_i64 tcg_res;
>
> +    switch (opcode) {
> +    case 0x0: /* FMOV */
> +        gen_gvec_fn2(s, false, rd, rn, tcg_gen_gvec_mov, 0);
> +        return;
> +    }
> +
>      fpst = get_fpstatus_ptr();
>      tcg_op = read_fp_dreg(s, rn);
>      tcg_res = tcg_temp_new_i64();
>
>      switch (opcode) {
> -    case 0x0: /* FMOV */
> -        tcg_gen_mov_i64(tcg_res, tcg_op);
> -        break;
>      case 0x1: /* FABS */
>          gen_helper_vfp_absd(tcg_res, tcg_op);
>          break;
> @@ -9185,7 +9197,11 @@ static void disas_simd_3same_logic(DisasContext *s, uint32_t insn)
>          gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_andc, 0);
>          return;
>      case 2: /* ORR */
> -        gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_or, 0);
> +        if (rn == rm) { /* MOV */
> +            gen_gvec_fn2(s, is_q, rd, rn, tcg_gen_gvec_mov, 0);
> +        } else {
> +            gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_or, 0);
> +        }
>          return;
>      case 3: /* ORN */
>          gen_gvec_fn3(s, is_q, rd, rn, rm, tcg_gen_gvec_orc, 0);
> @@ -10059,8 +10075,7 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
>          return;
>      case 0x5: /* CNT, NOT, RBIT */
>          if (u && size == 0) {
> -            /* NOT: adjust size so we can use the 64-bits-at-a-time loop. */
> -            size = 3;
> +            /* NOT */
>              break;
>          } else if (u && size == 1) {
>              /* RBIT */
> @@ -10312,6 +10327,21 @@ static void disas_simd_two_reg_misc(DisasContext *s, uint32_t insn)
>          tcg_rmode = NULL;
>      }
>
> +    switch (opcode) {
> +    case 0x5:
> +        if (u && size == 0) { /* NOT */
> +            gen_gvec_fn2(s, is_q, rd, rn, tcg_gen_gvec_not, 0);
> +            return;
> +        }
> +        break;
> +    case 0xb:
> +        if (u) { /* NEG */
> +            gen_gvec_fn2(s, is_q, rd, rn, tcg_gen_gvec_neg, size);
> +            return;
> +        }
> +        break;
> +    }
> +
>      if (size == 3) {
>          /* All 64-bit element operations can be shared with scalar 2misc */
>          int pass;


--
Alex Bennée