There is a problem with max and min

[wjr-z]

There is a problem with max and min

[bernhardmgruber]

No worries, I could translate! I will answer in English, so you can pick your favorite translation engine!

I can confirm, this is a bug in the implementation of uchar Vc::Detail::max(__m128i a, uchar). It should use uchar instead of schar in it's implementation. The bug also occurs when the function is called via the public API:

    uint8_t a[32]{0, 10, 250};
    std::cout << (int)Vc::schar_v((signed char*)a).max() << '\n'; // correctly returns 10
    std::cout << (int)Vc::uchar_v(a).max() << '\n'; // returns 10, should be 250

See: https://godbolt.org/z/8K6Yb3oYv.

Would you like to provide us with a pull request? Otherwise, I can fix it myself.

[wjr-z]

不用担心，我可以翻译！我会用英语回答，所以你可以选择你最喜欢的翻译引擎！

我可以确认，这是实现中的错误。它应该使用而不是在其实现中使用。当通过公共 API 调用函数时，也会发生此错误：uchar Vc::Detail::max(__m128i a, uchar)``uchar``schar

    uint8_t a[32]{0, 10, 250};
    std::cout << (int)Vc::schar_v((signed char*)a).max() << '\n'; // correctly returns 10
    std::cout << (int)Vc::uchar_v(a).max() << '\n'; // returns 10, should be 250

请参阅：https://godbolt.org/z/8K6Yb3oYv。

您想向我们提供拉取请求吗？否则，我可以自己修复它。

我不太会pull请求，您可以自己修复它。

[wjr-z]

No worries, I could translate! I will answer in English, so you can pick your favorite translation engine!

I can confirm, this is a bug in the implementation of uchar Vc::Detail::max(__m128i a, uchar). It should use uchar instead of schar in it's implementation. The bug also occurs when the function is called via the public API:

    uint8_t a[32]{0, 10, 250};
    std::cout << (int)Vc::schar_v((signed char*)a).max() << '\n'; // correctly returns 10
    std::cout << (int)Vc::uchar_v(a).max() << '\n'; // returns 10, should be 250

See: https://godbolt.org/z/8K6Yb3oYv.

Would you like to provide us with a pull request? Otherwise, I can fix it myself.

实际上min也有问题，不过我当时只列出了max的测试，抱歉！
对于min的测试如下
https://godbolt.org/z/d36c8YPsY
显然第二个的min应该是0而不是1。
对于a[2+16]=250，如果没有这一句，很容易发现不出错误。
https://godbolt.org/z/q9dnqx4v8
造成这个的原因大概是使用Vc_AVX_TO_SSE_2_NEW，导致先使用了AVX2中的unsigned函数，后使用SSE2中的signed函数，使得250先与1进行min_epu8，导致发现不出错误。

[bernhardmgruber]

Reopen, because of remaining issue with clang.