diff options
| -rw-r--r-- | ravanan/propnet.scm | 50 |
1 files changed, 26 insertions, 24 deletions
diff --git a/ravanan/propnet.scm b/ravanan/propnet.scm index e6155d8..dfe61ac 100644 --- a/ravanan/propnet.scm +++ b/ravanan/propnet.scm @@ -1,5 +1,5 @@ ;;; ravanan --- High-reproducibility CWL runner powered by Guix -;;; Copyright © 2024 Arun Isaac <arunisaac@systemreboot.net> +;;; Copyright © 2024, 2025 Arun Isaac <arunisaac@systemreboot.net> ;;; ;;; This file is part of ravanan. ;;; @@ -204,28 +204,6 @@ add to the inbox." ;; cell-values-inbox are serviced. (() (match propagators-inbox - ;; Schedule one propagator in inbox. - ((propagator other-propagators ...) - (loop cells - cell-values-inbox - other-propagators - ;; We don't need to cancel or forget about previous runs of the - ;; same propagator because cells only *accumulate* information; - ;; they never remove it. Any previous runs of the same - ;; propagator will only *add to* the information in the output - ;; cells. Previous runs may be closer to completion and taking - ;; advantage of their output may allow later stages to start - ;; running sooner, thus improving throughput. In our CWL - ;; application of propnets, this will never result in the same - ;; step being recomputed; so this approach does not come at a - ;; higher computational cost. - (append (maybe-alist - (cons (propagator-name propagator) - (activate-propagator - scheduler - propagator - (propagator-input-values cells propagator)))) - propagators-in-flight))) ;; Poll propagators in flight and update cell values if any of them are ;; done. (() @@ -275,7 +253,31 @@ add to the inbox." (append-map propagator-state->cell-values finished-propagators)) propagators-inbox - propagators-still-in-flight)))))))))))) + propagators-still-in-flight))))))) + ;; Schedule propagators in inbox. + (_ + (loop cells + cell-values-inbox + (list) + ;; We don't need to cancel or forget about previous runs of the + ;; same propagator because cells only *accumulate* information; + ;; they never remove it. Any previous runs of the same + ;; propagator will only *add to* the information in the output + ;; cells. Previous runs may be closer to completion and taking + ;; advantage of their output may allow later stages to start + ;; running sooner, thus improving throughput. In our CWL + ;; application of propnets, this will never result in the same + ;; step being recomputed; so this approach does not come at a + ;; higher computational cost. + (append (append-map (lambda (propagator) + (maybe-alist + (cons (propagator-name propagator) + (activate-propagator + scheduler + propagator + (propagator-input-values cells propagator))))) + propagators-inbox) + propagators-in-flight)))))))) (define (capture-propnet-output state) "Return output of propagator network @var{state}." |