Go vs. C++ – Which is faster? • Hacker News Board

Go vs. C++ – Which is faster?

#1

Post by sukeesh » Sun, May 24, 2020, 7:12 PM UTC

Go vs. C++ – Which is faster?
sukeesh.com

Re: Go vs. C++ – Which is faster?

#2

Post by Someone » Sun, May 24, 2020, 7:52 PM UTC

Likely explanation: using a byte for every flag takes a bit less CPU than packing 8 flags in a byte, but once the array outgrows the CPU’s cache, cache misses dominate running time.

Prediction: use std::bitset instead of an array of bool in the C++ code, and performance of the two programs will be (almost) identical.

Re: Go vs. C++ – Which is faster?

#3

Post by eesmith » Sun, May 24, 2020, 8:48 PM UTC

Someone wrote: post #2

Likely explanation: using a byte for every flag takes a bit less CPU than packing 8 flags in a byte, but once the array outgrows the CPU’s cache, cache misses dominate running time. Prediction: use std::bitset instead of an array of bool in the C++ code, and performance of the two programs will be (almost) identical.

Could be even simpler than that. I want to know how it was compiled.

I have a 2.3 GHz Intel Core i5 MacBook Pro, which is slower than the given hardware. The linked-to page reports "0.05s user".

I can reproduce that using g++ 9.1.0 using the default optimizations:

  % g++-9  sieve.cc
  % time ./a.out 4194304
  0.056u 0.003s 0:00.06 83.3% 0+0k 0+0io 99pf+0w

If I recompile with -O3, the result is significantly faster:

  % g++-9 -O3 sieve.cc
  % time ./a.out 4194304
  0.001u 0.003s 0:00.00 0.0% 0+0k 0+0io 98pf+0w