AlphaGitHub
Source/Packages

Realization.Nvidia.SM86.ExactCrossEntropySM86

packages/realizations/cooperative/nvidia-sm86/src/Realization/Nvidia/SM86/ExactCrossEntropySM86.alpha

1,471 lines228 declarations53.2 KiBSHA-256 411308c1414a
Download exact fileView at revisionf5034773ad4bb52133

def · lines 992–993

exactCrossEntropyMeanPartialSlots

Full file
Both reductions read a complete CTA, including lanes masked after their load. The workspace therefore needs 64 slots beyond the loss rows even when fewer row groups contain useful partials.
992def exactCrossEntropyMeanPartialSlots : Nat =
993  reductionSM86WarpSum64RequiredThreads

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.