3 ms·
Each is a multiply-accumulate instruction: https://developer.arm.com/documentation/dui0473/latest/neon-programming/fused-multiply-add-extension-for-neon https:/
by rayiner 6y ago
Each is a multiply-accumulate instruction: https://developer.arm.com/documentation/dui0473/latest/neon-programming/fused-multiply-add-extension-for-neon https://developer.arm.com/documentation/dui0473/latest/neon-...
- TinkersW 6y agoThey probably are FMA ports, but they didn't say that anywhere in the article, and it still doesn't explain the "quadruple the throughput", as both of Intels ports are FMA and 2/4 of AMDs are FMA.