Simulation project leaders meeting Tuesday June 8 ================================================= ==== Attendees Paolo, John Harvey, John Apostolakis, Witek, Torre, Alfredo, Fabiola ==== Minutes There was a discussion of the simulation leader situation. It is still evolving. == Generator services - Paolo We have received the report of the generator services review. The review identifies the critical issues of manpower (librarian, project leader) and communication with generator authors. Steps are underway to address these. Arrangements are underway to bring new people into the generator library activity to strengthen both library support and communication with generator authors and users. A new 'PhenoGrid' project in the UK is being launched and will have .5-1 FTE working on generator infrastructure, particularly on ThePEG framework and its use with Herwig++. They have expressed interest in collaborating with the generator services subproject and Paolo is following up. EVTGEN is being added to GENSER. The most recent generator services meeting was dedicated to EVTGEN. There's a wide interest in both the LHC experimentalist and theoretical (Herwig++) communities to have soon a (B) decay plug-in amenable for the QCD models, in particular for what concerns the most recent MCs adopting OO GENSER is on a quarterly release cycle. New Pythia and Herwig versions will be in the next release, as well as a new minimum bias generator. No MCDB news. == Fluka - Alfredo Very busy with machine developments. DPMJET3 interface is now working, with 100k's of events produced. Can run Fe-Fe or p-p from DPMJET and shower in Fluka. Generator for em dissociation of heavy ions is running and benchmarked. There will be a paper in a few months. Comparisons with data are good. There will be a general Fluka meet this Friday in Milan. offer from the Nuclear Energy Agency (under OECD) to provide support for the code. particularly, they havve general code for generating and displaying geometries. adapted to a lot of radiation codes. would like to include fluka as well. Slowly progressing towards the public release, but busy with priority LHC issues. Public release still expected by end of year. Manual (350 pages) is more or less ready. Working out how to publish it. Elimination of the need for a preprocessing step to generate em cross section data is still planned for this year, but probably late this year (will be in the public release). Witek commented that this is important for FLUGG users. Alfredo commented that it is needed by ALICE. == Geant4 - John Apostolakis The minor release 6.2 will come out at the end of the month. It will include the LHCb requirement to optionally label as a new particle a primary if it has gone through a hard hadronic interaction . G4Allocator revisions needed by CMS to identify memory used and release it at the user's request (and responsibility) are in. Binary cascade optimizations have been done yielding ~10% improvement in performance for its use case. Revisions that enable dynamic loading support on Windows will be in the release. (CLHEP though cannot be loaded dynamically - we are communicating with its editors regarding this.) GDML (XML geometry description) output capability will be available in the GDML web page for use with the release. The coming release will be in the SPI external software library, either at the time of the release or shortly thereafter. (It will also be provided via the current standard Geant4 distribution mechanisms.) Witek mentioned the requirement (also heard from Rene) that default constructors be provided in order to support persistence of Geant4 classes. JohnA reported that the requirement was seen as important and was being analysed for its impact. Rafi Yaari has arrived for a 6 month visit to work on profiling. He is putting together an infrastructure to profile CMS and potentially other production simulations. In parallel Gabriele reported good preliminary performance results (~10% level) using O2 level optimization with gcc. This has now become the default optimisation level for the Linux/gcc setup for release 6.2. Alfredo commented that the Intel compiler gives factor 2 better performance than gcc for Fluka. JohnA commented that problems have been seen with gcc in obtaining consistent results between optimized and debug compiles: caused by precision changes depending on the degree of optimization (affected by what the compiler decides to keep in 80-bit registers as temporaries vs writing out as 64-bit doubles in memory). Test have shown that forcing the use of 64-bit only in arithmetic operations provides a much higher degree of agreement between debug and optimised. To date this is attained only in tests using compiler flags which are specific to Pentium4, and as such not portable and not suitable for production. The use of assembly instructions to put the floating point unit (FPU) into a mode where all operatios are done in 64 bits (used by Fluka as reported by Alfredo) is under investigation -- and will be discussed with experiments as it affects all FPU operations in an application. == 'Generic simulation framework' - Witek There have been several meetings with Radovan addressing the GDML work required in order for it to serve as a geometry exchange tool. Solutions have been found for problematic issues like parameterised volumes. There will be a meeting with GDML-interested people this Friday. Witek has been experimenting with python binding of Geant4 using the LCG dictionary and pyLCGDict to provide interaction with Geant4 classes from Python. He finds it a nice capability, providing a new Geant4 interface almost for free that can be used for building and configuring Geant4 geometries and applications in python. He reports interest from experiments. Witek noted that python may also be an alternative to XML as a means of writing a persistent description of a Geant4 geometry. He tried to persistify a Geant4 geometry with POOL and ran into the mentioned need for default constructors on Geant4 classes. It would be very nice to have the existing POOL persistency machinery able to save a geometry for reuse. JohnA noted that geometry persistency has been in the long term program of Geant4. == Simulation physics validation - Fabiola The pixel test beam study has been completed by Alberto. He has written a summary note which was circulated to the Geant4 and Fluka teams today. Should be released next week. We then need to discuss what the next physics validation exercise with Fluka/FLUGG and Geant4 should be. The hadronic physics note is almost complete; should be finished off this week by Fabiola and circulated for comment. The July meeting on physics validation will be mainly devoted to em physics. ATLAS em calorimeter results will be presented comparing Geant4 to test beam and looking at per mil effects in linearity of response. There will also be an LHCb ecal test beam - Geant4 comparison talk. Completion of Witek's simple benchmark study still requires some consultation with Fluka experts, and they have been very busy. Should be close to producing a summary note. Will then address if and how to continue with further simple benchmark studies.