GCC 10 Adds ARMv8.6-A Targeting, BFloat16 + i8MM Options
Building on earlier GCC commits for Arm's BFloat16 (BF16) support and other new extensions, a late change landing for GCC 10 is the command line options for targeting the ARMv8.6-A architecture and optionally toggling i8mm and BF16 extensions.
ARMv8.6-A brings BFloat16 for helping the performance of neural network performance running on Arm-based systems. There has been compiler support worked on already for Arm BFloat16 while now the CLI switches are there for toggling it with +bf16.
Another new CLI option is +i8mm for enabling Arm's new 8-bit Integer Matrix Multiply instructions.
Most importantly is the CLI switch for allowing -march=armv8.6-a targeting for the eventual ARMv8.6-A processors, per this Git activity on Thursday.
No Arm processors based on ARMv8.6-A have yet to be announced, so there's still plenty of time for the open-source toolchain and other software support to mature for it. GCC 10.1 should be the first stable release of this annual compiler update due out around April.
ARMv8.6-A brings BFloat16 for helping the performance of neural network performance running on Arm-based systems. There has been compiler support worked on already for Arm BFloat16 while now the CLI switches are there for toggling it with +bf16.
Another new CLI option is +i8mm for enabling Arm's new 8-bit Integer Matrix Multiply instructions.
Most importantly is the CLI switch for allowing -march=armv8.6-a targeting for the eventual ARMv8.6-A processors, per this Git activity on Thursday.
No Arm processors based on ARMv8.6-A have yet to be announced, so there's still plenty of time for the open-source toolchain and other software support to mature for it. GCC 10.1 should be the first stable release of this annual compiler update due out around April.
1 Comment