3rd March 2023, 3 min read

Float-parsing benchmark: Regular Visual Studio, ClangCL and Linux GCC

4 thoughts on “Float-parsing benchmark: Regular Visual Studio, ClangCL and Linux GCC”

1. Joseph says:
   March 6, 2023 at 2:26 pm

   Have you tried Intel’s compiler?

   1. Daniel Lemire says:
      March 6, 2023 at 7:55 pm

      I did not. Want to give it a try and report back?

2. JulioSAO says:
   March 7, 2023 at 9:56 am

   I think is not an issue with your metodology.
   I work as system programmer and in my experience, linux beats windows in every CPU, Disk and Memory benchmark. Even virtualized under windows many test still outperforms windows en several test. The only area where windows wins is in graphics.

   I dont have low level details about it but i suspect the cause is a combination of better optimization (ej, copy on write memory when fork a process), api exposed (system-v api vs win32) and complexity (linux boxes often run less process at idle state).

   And graphics stack in linux is getting better year at year. Now you even can run windows games in linux and some of them, run faster. Even “emulating” using proton.

   Regards.
   Regards.

   1. Aurélien RB says:
      March 11, 2023 at 9:18 pm

      The Linux version of his tests runs on a Windows kernel (WSL). Generally speaking I doubt the OS has a serious impact on such micro benchmark.
      It’s the compiler that is benckmarked here. And VC++ is known to not be the best in speed, both for backend code generation but also efficient STL implémentation.
      https://developercommunity.visualstudio.com/t/vs2019-generates-inefficient-code-compared-to-clan/457819
      Something interesting to test would be clang on WSL because I wonder how much clang-cl differs from regular clang.