ARMv8.6-A Brings BFloat16, GEMM & Other Enhancements
Arm has outlined their architecture enhancements being introduced in ARMv8.6-A as their 2019 ARMv8 architecture update.
ARMv8.6 is adding General Matrix Multiply (GEMM) and BFloat16 as their big additions to this ARMv8 revision for boosting neural network / machine learning performance. There is also now support for fine grained traps for virtualization, wait-for-event instructions, extending pointer authentication, and a variety of other changes.
Arm acknowledges it will be "a few years time" before ARMv8.6-A hardware surfaces and they are now working to enable the open-source software support. Those wanting to learn more about ARMv8.6-A can do so via community.arm.com.
ARMv8.6 is adding General Matrix Multiply (GEMM) and BFloat16 as their big additions to this ARMv8 revision for boosting neural network / machine learning performance. There is also now support for fine grained traps for virtualization, wait-for-event instructions, extending pointer authentication, and a variety of other changes.
Arm acknowledges it will be "a few years time" before ARMv8.6-A hardware surfaces and they are now working to enable the open-source software support. Those wanting to learn more about ARMv8.6-A can do so via community.arm.com.
11 Comments