fun trick to do f32 -> bf16 conversion with PTX: prmt.b32 combined_val, val_0, val_1, 12918; previously I had been using CPU brain and had been taking val_0 and val_1 and >>16'ing them and…

Posters: cis_female. cis_female

2 likes recorded by the source corpus.

Topics: Software

View Post: https://x.com/cis_female/status/1569850593728135168