On Tuesday, 18 June 2019 09:01:31 PDT Thiago Macieira wrote: > They're the only ones we couldn't fall back to Linux's AF_ALG or OpenSSL's > support, which often contain more optimised code that ours.
If we remove Keccak, then we could remove all the hash implementations from QtCore, on Linux, and simply fall back to AF_ALG. See prototype at https://codereview.qt-project.org/c/qt/qtbase/+/215796 I haven't benchmarked whether the kernel/userspace transitions make a perceptible difference. But for any non-trivial content, we should expect a performance gain if there's an implementation in https://code.woboq.org/linux/linux/arch/x86/crypto/ (basically, SHA1, SHA256 and SHA512). Better than that, only if we used OpenSSL, but there's also a set-up cost because EVP_MD will try to allocate memory. -- Thiago Macieira - thiago.macieira (AT) intel.com Software Architect - Intel System Software Products _______________________________________________ Development mailing list [email protected] https://lists.qt-project.org/listinfo/development
