Re: [PATCH 2/2] util/bufferiszero: Add simd acceleration for loongarch64

[Richard Henderson <richard.henderson@linaro.org>]

On 6/5/24 20:36, maobibo wrote:
>>> static biz_accel_fn const accel_table[] = {
>>>      buffer_is_zero_int_ge256,
>>> #ifdef __loongarch_sx
>>>      buffer_is_zero_lsx,
>>> #endif
>>> #ifdef __loongarch_asx
>>>      buffer_is_zero_lasx,
>>> #endif
>>> };
>>>
>>> static unsigned best_accel(void)
>>> {
>>> #ifdef __loongarch_asx
>>>      /* lasx may be index 1 or 2, but always last */
>>>      return ARRAY_SIZE(accel_table) - 1;
>>> #else
>>>      /* lsx is always index 1 */
>>>      return 1;
>>> #endif
>>> }
> size of accel_table is decided at compile-time, will it be better if runtime checking is 
> added also?  something like this:
> 
>   unsigned info = cpuinfo_init();
> 
>   #ifdef __loongarch_asx
>   if (info & CPUINFO_LASX) {
>        /* lasx may be index 1 or 2, but always last */
>        return ARRAY_SIZE(accel_table) - 1;
>   }
>   #endif

No, because the ifdef checks that the *compiler* is prepared to use LASX/LSX instructions 
itself without further checks.  There's no point in qemu checking further.


r~