Intel® ISA Extensions
Use hardware-based isolation and memory encryption to provide more code protection in your solutions.
1110 Discussions

mnemonic / description of AVX10.2 VSCALEFPBF16

Beulich__Jan
Beginner
153 Views

VSCALEFP{S,D,H} permit (embedded) rounding. For other BF16 insns with such counterparts the mnemonics include NE and the descriptions explicitly say the round-to-nearest is always used. If the behavior is similar as for other BF16 insns, shouldn't the mnemonic include NE? In any event it would be helpful if behavior was described explicitly, no matter whether it's similar to or different from other insns.

0 Kudos
1 Reply
Beulich__Jan
Beginner
145 Views

Hmm, is this perhaps because FTZ is unconditionally in effect, and hence no actual rounding would ever occur?

0 Kudos
Reply