CUDA Shared Memory Swizzling (leimao.github.io)

54 points by jxmorris12 6 days ago

2 comments:

by johndough 2 hours ago

I was wondering why the author was using braced initialization like

    size_t i{0};
instead of the more common

    size_t i = 0;
Apparently, braced initialization does not allow narrowing conversion, so you'd get a compiler error for e.g. casting double to float

    size_t i{0.0};
and a warning for

    double d = 0.0;
    size_t i{d};
which might silently overflow size_t otherwise, so this is a bit safer.

In C++, you can get the same effect without the unusual syntax by passing -Wfloat-conversion to gcc/clang, but not sure how to do that with CUDA:

by KeplerBoy 3 hours ago

Lei Mao's blog is such an amazing resource for GPU performance engineering. I am stunned by the sheer amount of insight he puts out on his blog.

Data from: Hacker News, provided by Hacker News (unofficial) API