I agree, I’ve had really good results cross plaform code (Linux, Mac, Windows) with MingW64 w/ pthreads on Windows. I code to the C++11 standard though, not certain about support for newer standards. One other thimg that is important to me is ease of integration with the nvcc compiler for linking CUDA kernels - only certain compilers and versions are supported. The author’s use of the word “best” doesn’t seem to address that aspect.
Agner’s test results are on page 76 of Optimizing software in C++. He covers mostly optimizations from the 1986 Dragon Book and heavily emphasizes algebraic optimizations. Loop vectorization is not covered in detail. Loop nests are not mentioned.
No mention of msys2? They have a windows native clang http://repo.msys2.org/mingw/x86_64/
Also cygwin has a native version too http://mirrors.kernel.org/sourceware/cygwin/x86_64/release/clang/mingw64-x86_64-clang/
I agree, I’ve had really good results cross plaform code (Linux, Mac, Windows) with MingW64 w/ pthreads on Windows. I code to the C++11 standard though, not certain about support for newer standards. One other thimg that is important to me is ease of integration with the nvcc compiler for linking CUDA kernels - only certain compilers and versions are supported. The author’s use of the word “best” doesn’t seem to address that aspect.
After a multi-year effort chromium switched to Clang on Windows last year. They wrote about it and then differences in build size, build time, performance, etc: http://blog.llvm.org/2018/03/clang-is-now-used-to-build-chrome-for.html?m=1
Agner’s test results are on page 76 of Optimizing software in C++. He covers mostly optimizations from the 1986 Dragon Book and heavily emphasizes algebraic optimizations. Loop vectorization is not covered in detail. Loop nests are not mentioned.