You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
An ongoing problem in PSyclone is (esp. in the context of kernel extraction) that it has to know the source file for a given module name. Compilers solve this by creating .mod files, but they are proprietary and therefore cannot be used.
Given that Fab already analyses the files and creates the analyse files, I wonder if it would be feasible to store the information which source file contains a given module (the .an files have besides other information the reverse information: which modules are contained in a file). Either one file per module (similar to the .mod files), or maybe even one file (properly synchronised to allow the analysis run in parallel). PSyclone could then use this file (optionally) to reduce the time required to find a source file for a module.
The only issue would be that the _alg.f90 and _psy.f90 files do not exist at the time the analysis runs. But the analysis could perhaps be done in two steps: first run on all .f90/.f files, then run PSyclone (all kernel files will have been processed at this stage)? The devil is in the detail obviously :)
The text was updated successfully, but these errors were encountered:
An ongoing problem in PSyclone is (esp. in the context of kernel extraction) that it has to know the source file for a given module name. Compilers solve this by creating .mod files, but they are proprietary and therefore cannot be used.
Given that Fab already analyses the files and creates the analyse files, I wonder if it would be feasible to store the information which source file contains a given module (the .an files have besides other information the reverse information: which modules are contained in a file). Either one file per module (similar to the .mod files), or maybe even one file (properly synchronised to allow the analysis run in parallel). PSyclone could then use this file (optionally) to reduce the time required to find a source file for a module.
The only issue would be that the _alg.f90 and _psy.f90 files do not exist at the time the analysis runs. But the analysis could perhaps be done in two steps: first run on all .f90/.f files, then run PSyclone (all kernel files will have been processed at this stage)? The devil is in the detail obviously :)
The text was updated successfully, but these errors were encountered: