Announcement

Collapse
No announcement yet.

Arm Cortex-A710 Support Merged Into GCC 12 Compiler

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

  • Arm Cortex-A710 Support Merged Into GCC 12 Compiler

    Phoronix: Arm Cortex-A710 Support Merged Into GCC 12 Compiler

    Announced back in May was the Cortex-A710 as the first-generation Armv9 "big" core and successor to the Cortex-A78. The initial Cortex-A710 support is now present in the GCC 12 code compiler...

    https://www.phoronix.com/scan.php?pa...0-GCC-12-Lands

  • #2
    Perhaps Phoronix could run some ARMv8 vs ARMv9 benchmarks, if possible?

    There's clearly not a lot of hand written SVE2/TME code floating around in released stuff, but Phoronix's benchmarks have shown some significant gains from autovectorization and specific architecture optimizations before.

    Comment


    • #3
      Originally posted by brucethemoose View Post
      Perhaps Phoronix could run some ARMv8 vs ARMv9 benchmarks, if possible?

      There's clearly not a lot of hand written SVE2/TME code floating around in released stuff, but Phoronix's benchmarks have shown some significant gains from autovectorization and specific architecture optimizations before.
      I don't think Michael has the hardware for that.
      Anyway iirc, ARMv9 is mostly about additional instructions for learning and such. Code that uses that will fly, compared to ARMv8.x, code that doesn't will not be significantly faster. Waiting for benchmarks, of course, but that's what I expect to see.

      Comment


      • #4
        Originally posted by bug77 View Post

        I don't think Michael has the hardware for that.
        Anyway iirc, ARMv9 is mostly about additional instructions for learning and such. Code that uses that will fly, compared to ARMv8.x, code that doesn't will not be significantly faster. Waiting for benchmarks, of course, but that's what I expect to see.
        Of course, I meant when ARMv9 hardware is out in the wild. Which shouldn't be *that* far away.

        Comment


        • #5
          Originally posted by bug77 View Post
          Anyway iirc, ARMv9 is mostly about additional instructions for learning and such.
          It is actually designed for the opposite. Indeed, you need to have some large inner loops, but SVE is made for speculative vectorization. However it is not enough to check if SVE is enabled, it also depends a lot on the cleverness of the compiler.

          Comment

          Working...
          X