-
legion-master-final-pre-control-replication
This is the final commit of the master branch of Legion prior to the merge of the control replication branch. Prior to this there is no support for control replication. The primary method for physical analysis in this branch is still based on Warnock's algorithm for decomposing equivalence sets. After the control replication merge, the primary method for physical analysis switches to ray casting.
-
legion-control-replication-final
This is the final commit of the long-lived control replication branch. Control replication adds support for replicating logical tasks into implicit shards which can run in parallel but still behave as one logical task. To make control replication possible a number of fundamental changes had to be made to the internal Legion architecture and required significant modifications to how Legion works internally. The version of Legion after this branch lands in the master branch bears little resemblance to the version of Legion that preceded it.
-
legion-control-replication-final-pre-shardrefine
This is the last commit of the long-running control replication branch before the shardrefine branch was merged into it. It contains an older version of the equivalence set algorithm which used heuristics to select disjoint and complete partitions to serve as the bounding volume hierarchy to use for creating equivalence sets at the leaves. After this commit we switch to the new algorithm implemented in the shardrefine branch.
-
legion-shardrefine-final
This is the final commit of the long-running shardrefine branch which introduced a new mechanism and heuristics for creating and tracking equivalence sets
-
papers/equiv-ppopp23/paint
PPoPP 2023 equivalence set paper, paint version.
-
papers/equiv-ppopp23/oldeqcr
PPoPP 2023 equivalence set paper, oldeqcr version.
-
papers/equiv-ppopp23/neweqcr
PPoPP 2023 equivalence set paper, neweqcr version.
-
papers/dcr-ppopp21/regent
Software artifact for Regent experiments from the PPoPP 2021 paper. Scaling Implicit Parallelism via Dynamic Control Replication. Michael Bauer, Wonchan Lee, Elliott Slaughter, Zhihao Jia, Mario Di Renzo, Manolis Papadakis, Galen Shipman, Patrick McCormick, Michael Garland, and Alex Aiken. In Principles and Practices of Parallel Programming (PPoPP), February 2021.
-
papers/index-launch-sc21
Final code artifacts for SC21 paper.