Tags give the ability to mark specific points in history as being important
-
-
-
-
-
-
0.6.0-dev2506111
Release: 0.6.0-dev2506111c25002c1 · ·Checks for compatibility for the workaround of device identification. Work in progress, development release
-
0.5.6-dev2506101
Release: 0.5.6-dev250610111053cc7 · ·Development release to test new gpuId assignation of values.
-
-
-
0.5.4
Release: 0.5.48493686b · ·Multiple bug fixes related to MPI information: - Now tasks are sorted by Rank when MPI is available - Node synchronization for multi-node report merging, only for MPI - Solves a bug that crashed application when UVA memory calls were found - Solves a bug that did not include the names of not predefined MPI calls - Solves a bug that added duplicate copies of the same records for MPI_Wait and calls of "other" category
-
-
0.5.2
Release: 0.5.2d81f2ee3 · ·Patch release 0.5.2 This release solves a software regression bug that did't allow to translate GPU metrics never. This regression started from dev release 0.5.0-dev2503171. It also solves a bug where traces that included the GPU metrics of devices that were not in the context of the application were incorrectly translated, writing malformed records.
-
-
-
0.5.0-dev2503271
Release: 0.5.0-dev2503271aef55474 · ·Mainly adds the different type for memory copies kernel events.
-
0.5.0-dev2503171
Release: 0.5.0-dev2503171308cfc0b · ·Solves some small fixes Adds graceful disable of GPU metrics when not found Adds values of memory usage in kernels
-
-
-
0.5.0-dev2503071
Release: 0.5.0-dev25030711062a5b7 · ·Hotfix release Solves a bug when adding number of normal threads to cuda streams task number, in the case of having processes with no device associated.
-
0.5.0-dev0503251
Release: 0.5.0-dev0503251f3800a74 · ·Development version 0.5.0-dev0503251 Includes release candidate support of CUDA Graphs Solves many small bugs Makes output of files more stable