[Tarantool-patches] [PATCH v3 2/2] base64: improve decoder performance

[Leonid Vasiliev]

Hi!

On 25.12.2020 13:39, Sergey Nikiforov wrote:
> Hi!
> 
> On 24.12.2020 17:08, Leonid Vasiliev wrote:
>> Hi! Thank you for the patch.
>> AFAIU the status of the patch is follows:
>>
>>  > But I see we are not going anywhere here. You don't really need LGTM
>>  > from me on this patch, if you don't want to finish it. I am not
>>  > strictly against these changes, because *probably* they don't add new
>>  > bugs, and seem to be a tiny bit better for perf. I only don't like it
>>  > being not finished.
>>
>> I think the changes are ok, because they are good for perf (and we have
>> confirmation) and don't add degradation (our tests should guarantee
>> this).
>>
>> See some comments below:
>>
>> What about a benchmark. AFAIK A. Lyapunov propose saving all benchmarks
>> that we used. Did you have a conversation with him?
> 
> Yes, I had. "perf" in tarantool. Benchmark framework is not yet merged 
> (https://github.com/tarantool/tarantool/tree/i.kosarev/gh-5385-tiny-tuples-with-perf-test/perf) 
> so I plan to commit my microbenchmark a little later to avoid duplication.
> 

OK.

>> On 22.12.2020 13:41, Sergey Nikiforov wrote:
>>> Unnecessary checks were removed from internal loops.
>>> Benchmark shows that performance is now ~1.19 times higher
>>> (release build, Intel Core I7-9700K, only one thread).
>>> ---
>>>
>>> Branch: 
>>> https://github.com/tarantool/tarantool/tree/void234/gh-3069-fix-base64-memory-overrun-v3 
>>>
>>>
>>>   test/unit/base64.c      |  7 +++-
>>>   test/unit/base64.result | 84 +++++++++++++++++++++++++++--------------
>>>   third_party/base64.c    | 36 +++++++++++++-----
>>>   3 files changed, 89 insertions(+), 38 deletions(-)
>>>
>>
>> I left my questions about the test in the review of the previous patch.
> 
> Fixed.
> 
>>> diff --git a/third_party/base64.c b/third_party/base64.c
>>> index 3350a98ff..93442c04b 100644
>>> --- a/third_party/base64.c
>>> +++ b/third_party/base64.c
>>> @@ -257,10 +257,11 @@ base64_decode_block(const char *in_base64, int 
>>> in_len,
>>>           {
>>>       case step_a:
>>>               do {
>>> -                if (in_pos == in_end || out_pos >= out_end)
>>> +                if (in_pos >= in_end)
>>>                   {
>>>                       state->step = step_a;
>>> -                    state->result = curr_byte;
>>> +                    /* curr_byte is useless now */
>>> +                    /* state->result = curr_byte; */
>>
>> For multi-line comment, we use the following format:
>> /*
>>   * First line
>>   * Second line
>>   */
>>
>> And leaving a commented code is not best practice.
> 
> Fixed.