Announcement

Collapse
No announcement yet.

GCC 10.1 Compiler Optimization Benchmarks

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • GCC 10.1 Compiler Optimization Benchmarks

    Phoronix: GCC 10.1 Compiler Optimization Benchmarks

    With the recently minted GCC 10 compiler there was a request to see some fresh benchmarks at different compiler optimization levels and flags like LTO...

    http://www.phoronix.com/scan.php?pag...-Optimizations

  • #2
    -Ofast -flto?

    Comment


    • #3
      Thanks for this.

      -Os would have been interesting as well.

      Comment


      • #4
        Another interesting metric would be to compare the memory usage against performance.

        Comment


        • #5
          Originally posted by Raka555 View Post
          Another interesting metric would be to compare the memory usage against performance.
          Why exactly would this be useful? The compiler isn't going to magically be able to optimise the underlying algorithms that lead to increased memory usage.

          Comment


          • #6
            Originally posted by AHSauge View Post

            Why exactly would this be useful? The compiler isn't going to magically be able to optimise the underlying algorithms that lead to increased memory usage.
            To see for instance how much speed you gain versus extra code generated for example by unrolling loops and inline functions.

            Comment


            • #7
              -flto has been rock solid for quite some time now and it looks like it's worth enabling this option by default at all times. It never slows down and sometimes gives nice boosts.

              Originally posted by Raka555 View Post
              Thanks for this.

              -Os would have been interesting as well.
              It's been irrelevant for ages now. Hardly faster than -O1.

              Comment


              • #8
                Originally posted by Raka555 View Post

                To see for instance how much speed you gain versus extra code generated for example by unrolling loops and inline functions.
                Yeah, again, why is memory consumption a useful metrics here? That's only going to be a tiny blip in comparison to all the dynamically allocated data. Of course, the final binary size vs. speed is interesting, but I really don't understand why memory consumption is relevant.

                Comment


                • #9
                  Originally posted by birdie View Post
                  -flto has been rock solid for quite some time now and it looks like it's worth enabling this option by default at all times. It never slows down and sometimes gives nice boosts.
                  Except in the Crypto++ bench for some strange reason.

                  Comment


                  • #10
                    Originally posted by AHSauge View Post

                    Why exactly would this be useful? The compiler isn't going to magically be able to optimise the underlying algorithms that lead to increased memory usage.
                    Ok, so I am not allowed to be interested in that. Got it.

                    Comment

                    Working...
                    X