(distribute-jobs) ( job-count per-job-shared max-shared-size max-block-size -- grid-size block-size per-block-shared )

Vocabulary

cuda.devices

Inputs

job-count	an object
per-job-shared	an object
max-shared-size	an object
max-block-size	an object

Outputs

grid-size	an object
block-size	an object
per-block-shared	an object

Definition

USING: kernel math math.order ;

IN: cuda.devices

:: (distribute-jobs)
    ( job-count per-job-shared max-shared-size max-block-size -- grid-size block-size per-block-shared )
    per-job-shared
    [ max-block-size ]
    [ max-shared-size swap /i max-block-size min ] if-zero
    job-count min
    :> job-max-block-size job-count job-max-block-size up/i
    :> grid-size job-count grid-size up/i
    :> block-size block-size per-job-shared *
    :> per-block-shared grid-size block-size per-block-shared ;
    inline