Re: [PATCH 4/7] x86: Add SSSE3 optimized chacha20

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 14/04/2022 16:25, Noah Goldstein wrote:
> On Wed, Apr 13, 2022 at 3:27 PM Adhemerval Zanella via Libc-alpha
> <libc-alpha@sourceware.org> wrote:
>>
>> +
>> +static inline void
>> +chacha20_crypt (struct chacha20_state *state, uint8_t *dst, const uint8_t *src,
>> +               size_t bytes)
>> +{
>> +  if (CPU_FEATURE_USABLE_P (cpu_features, SSSE3) && bytes >= CHACHA20_BLOCK_SIZE * 4)
>> +    {
>> +      size_t nblocks = bytes / CHACHA20_BLOCK_SIZE;
>> +      nblocks -= nblocks % 4;
> 
> Are we locking ourselves into the api of __chacha_* expecting
> this precomputation? I imagine we might want to move this to
> assembly unless `nblock` is a compile time constant.
> 
>> +      __chacha20_ssse3_blocks8 (state->ctx, dst, src, nblocks);
>> +      bytes -= nblocks * CHACHA20_BLOCK_SIZE;
>> +      dst += nblocks * CHACHA20_BLOCK_SIZE;
>> +      src += nblocks * CHACHA20_BLOCK_SIZE;
>> +    }
>> +
>> +  if (bytes > 0)
>> +    chacha20_crypt_generic (state, dst, src, bytes);
>> +}
>> --
>> 2.32.0
>>

I think should be ok to _Static_assert that CHACHA20_BUFSIZE is a multiple of
the expected nblocks used by the optimized version and just call it without
the need to handle nblocks.

I will change it to v2.