On Sat, 18 May 2019 at 20:19, Richard Henderson
<richard.hender...@linaro.org> wrote:
>
> These instructions shift left or right depending on the sign
> of the input, and 7 bits are significant to the shift. This
> requires several masks and selects in addition to the actual
> shifts to form the complete answer.
>
> That said, the operation is still a small improvement even for
> two 64-bit elements -- 13 vector operations instead of 2 * 7
> integer operations.
>
> +void gen_ushl_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
> +{
> + TCGv_i32 lval = tcg_temp_new_i32();
> + TCGv_i32 rval = tcg_temp_new_i32();
> + TCGv_i32 lsh = tcg_temp_new_i32();
> + TCGv_i32 rsh = tcg_temp_new_i32();
> + TCGv_i32 zero = tcg_const_i32(0);
> + TCGv_i32 max = tcg_const_i32(32);
> +
> + /*
> + * Perform possibly out of range shifts, trusting that the operation
> + * does not trap. Discard unused results after the fact.
> + */
This comment reads to me like we're relying on a guarantee
that TCG doesn't make, but in fact the readme says it does:
out-of-range shifts are "unspecified behavior" which may give
bogus results but won't crash. Perhaps phrasing the comment
as "relying on the TCG guarantee that these are only
'unspecified behavior' and not 'undefined behavior' and so
won't crash" would be clearer ?
> + tcg_gen_ext8s_i32(lsh, b);
> + tcg_gen_neg_i32(rsh, lsh);
> + tcg_gen_shl_i32(lval, a, lsh);
> + tcg_gen_shr_i32(rval, a, rsh);
thanks
-- PMM
