Announcement

Collapse
No announcement yet.

Clang Fights GCC On AMD's Athlon AM1 APU With Jaguar Cores

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

  • Clang Fights GCC On AMD's Athlon AM1 APU With Jaguar Cores

    Phoronix: Clang Fights GCC On AMD's Athlon AM1 APU With Jaguar Cores

    A few days ago I did my latest benchmarks of GCC vs. LLVM/Clang and that was using an Intel Core i7 4770K "Haswell" processor. The tables have now turned and in this article are GCC vs. LLVM Clang benchmarks of the AMD Athlon 5350 APU with four Jaguar CPU cores.

    http://www.phoronix.com/vr.php?view=20248

  • #2


    How I see compiler benchmarks these days

    Comment


    • #3
      It would be a better idea to compile full-featured programs and games instead of synthetic benchmarks (compilation time + run-time performance).

      Comment


      • #4
        Originally posted by Calinou View Post
        It would be a better idea to compile full-featured programs and games instead of synthetic benchmarks (compilation time + run-time performance).
        Timed compilation of the common subset of the GCC & Clang compilable-software of the whole Debian archive, followed by latency (latencytop anyone?) and performance tests

        Comment


        • #5
          Originally posted by doom_Oo7 View Post
          Timed compilation of the common subset of the GCC & Clang compilable-software of the whole Debian archive, followed by latency (latencytop anyone?) and performance tests
          Don't forget the comparison against an infinite group of monkeys with keyboards simultaneously writing hamlet, and the binary versions of the compiled code for the debian archive. Whether the monkeys target -march=native, or just -O2 is an exercise left to the reader.

          Comment


          • #6
            Originally posted by Veerappan View Post
            Don't forget the comparison against an infinite group of monkeys with keyboards simultaneously writing hamlet, and the binary versions of the compiled code for the debian archive. Whether the monkeys target -march=native, or just -O2 is an exercise left to the reader.
            You guys are so silly. Monkeys can't code you need to get a Siberian Husky to do that.

            Comment


            • #7
              Originally posted by Calinou View Post
              It would be a better idea to compile full-featured programs and games instead of synthetic benchmarks (compilation time + run-time performance).
              I recently ran Firefox benchmarks on AMD hardware http://hubicka.blogspot.ca/2014/04/l...2-firefox.html

              Comment


              • #8
                Originally posted by Veerappan View Post
                Don't forget the comparison against an infinite group of monkeys with keyboards simultaneously writing hamlet, and the binary versions of the compiled code for the debian archive. Whether the monkeys target -march=native, or just -O2 is an exercise left to the reader.
                I guess an architecture might be conceivable (by the same monkeys, at the same time), where -O2 == -march=native

                Comment


                • #9
                  Originally posted by doom_Oo7 View Post
                  I guess an architecture might be conceivable (by the same monkeys, at the same time), where -O2 == -march=native
                  You want the monkeys to start doing CPU design? Wait a minute...

                  Comment


                  • #10
                    Micheal, the compilation times seem to be out if the norm here

                    I know CLang is fast but this seems to be excessively fast in the compiling tests. This makes me wonder why, is Clang making better use if all of the cores in the machine? Or maybe GCC wasn't using the cores at all. I expect better compiling times but these are a blow out.

                    Comment


                    • #11
                      I won't switch to LLVM/Clang even if it's the absolute winner on all benchmarks by a big margin.

                      And some compilation benchmarks really show a suspecting difference. The compilation on LLVM/Clang is parallelized by default? I know that on GCC that's not true.

                      Comment

                      Working...
                      X