Announcement

Collapse
No announcement yet.

GCC 5 To Boast New x86 Optimizations

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

  • GCC 5 To Boast New x86 Optimizations

    Phoronix: GCC 5 To Boast New x86 Optimizations

    GCC 5 already boasts an incredible amount of new compiler features as laid out now over dozens of Phoronix articles, but there's even more abound for this major compiler update due out in 2015...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Three times faster? Wow, that's pretty impressive at face value.

    Evgeny Stupachenko of Intel
    hope they didn't manage to sneak any AMDcripple in

    Comment


    • #3
      Originally posted by peppercats View Post
      Three times faster? Wow, that's pretty impressive at face value.


      hope they didn't manage to sneak any AMDcripple in
      God I hope not......I had enough of that in the 80's and 90's.

      Comment


      • #4
        Originally posted by grndzro View Post
        God I hope not......I had enough of that in the 80's and 90's.
        You still have that in the intel compiler, which is used to compile benchmarks on windows.

        Comment


        • #5
          For one of the tests, with an Intel Core i7 Haswell CPU, the code is three times faster than GCC 4.9.
          Yes, it is called adding optimizing vectors for avx2 instruction sets. In short, 99% of people won't get squat out of it.

          Comment


          • #6
            Nope, they also tested Silvermont Atom, which doesn't have AVX2. Showed gains.

            Comment


            • #7
              Originally posted by Marc Driftmeyer View Post
              Yes, it is called adding optimizing vectors for avx2 instruction sets. In short, 99% of people won't get squat out of it.
              AVX2 has been supported since 4.8

              Comment


              • #8
                there is a solution to the AMDcripple problem it is called the AuthenticAMD patch, available on Linux but can also be used to patch windows EXE's with the -a option
                yes, the gains are real, the Intel compiler is revolutionary in terms of code performance, I have much experience with it, I tested many versions from 9.x to 15.x - it is worth, but the improvements between versions 10.x and 15.x are not that impressive, I only tested on Pentium 3 and Core 2 Duo, for Pentium 3 there is no difference between ICC and GCC, but for Core 2 Duo there is a big difference in performance compared to GCC
                also recently been testing ICC 15.0.1 and the good news is there is no need for a patch to compile the Linux kernel with it anymore - it works out of the box, the bad news: I also tested a kernel compiled by GCC5 (January 2015 version) and GCC5 wins in kernel performance (kernel tested was 3.15.1), but while there is no improvement in compiling the kernel with ICC, there is a great improvement compiling userspace with it
                GCC5 is the fastest GCC yet, hands down !

                Comment

                Working...
                X