Source/Packages

Realization.Nvidia.SM86.GreedyArgmaxSM86

packages/realizations/cooperative/nvidia-sm86/src/Realization/Nvidia/SM86/GreedyArgmaxSM86.alpha

1,985 lines222 declarations73.3 KiBSHA-256 a4ae889c3756

def · lines 885–940

greedyArgmaxSM86ByteVocabularyLoopProgram

Full file
Bob uses one thread and one native backward branch to scan all 256 byte logits. The twelve-instruction loop advances R3 from 0 through 255, so the branch offset is -192 bytes from the instruction following BRA.

The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.