Re: [boost] [beast] Formal review

Subject: Re: [boost] [beast] Formal review
From: Vinnie Falco (vinnie.falco_at_[hidden])
Date: 2017-07-10 18:00:29

On Mon, Jul 10, 2017 at 9:57 AM, Phil Endecott via Boost
<boost_at_[hidden]> wrote:
> So you're reinterpret_casting a char* to a size_t* and then dereferencing
> it.  Isn't that undefined behaviour?

Which platform do you think this doesn't work on?

I have to apologize again for the earlier misleading comments. The
benchmarks were measuring the wrong thing. The actual figure is more
in line with 20,000%.

The reinterpret_cast<> can be trivially changed to std::memcpy:

    std::size_t temp;
    std::memcpy(&temp, in, sizeof(temp));
    if((temp & mask) != 0)

This hurts MSVC a bit, gcc/clang/MIPS not at all, but ARM takes quite a hit:

<https://godbolt.org/g/HFxxub>

(Thanks to Peter for working up the CE ARM example)

Updated benchmarks (run on MSVC with optimizations)

With reinterpret_cast
beast.benchmarks.utf8_checker
beast:  1,977,629,044 char/s
beast:  1,474,907,121 char/s
beast:  1,930,979,173 char/s
beast:  1,899,078,149 char/s
beast:  1,893,740,588 char/s
locale: 81,966,125 char/s
locale: 82,200,658 char/s
locale: 81,802,070 char/s
locale: 82,103,369 char/s
locale: 81,802,149 char/s
Longest suite times:
    2.7s beast.benchmarks.utf8_checker
2.7s, 1 suite, 1 case, 1 test total, 0 failures
The program '[79364] benchmarks.exe' has exited with code 0 (0x0).

With std::memcpy
beast.benchmarks.utf8_checker
beast:  1,124,515,969 char/s
beast:  1,336,074,093 char/s
beast:  1,494,183,562 char/s
beast:  1,506,365,044 char/s
beast:  1,533,419,187 char/s
locale: 75,457,683 char/s
locale: 81,358,140 char/s
locale: 80,413,657 char/s
locale: 81,635,114 char/s
locale: 67,234,619 char/s
Longest suite times:
    3.0s beast.benchmarks.utf8_checker
3.0s, 1 suite, 1 case, 1 test total, 0 failures
The program '[82220] benchmarks.exe' has exited with code 0 (0x0).