Difference: GPUStudies (1 vs. 9)

Revision 92019-08-05 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Changed:
<
<
To be able to process data at 30 MHz, the possibility of running (part of) the algorithms of HLT1 on GPUs is currently being studied.
>
>
To be able to process data at 30 MHz, the possibility of running the full HLT1 on GPUs is currently being studied.
  The repository of the project is Allen.
Added:
>
>
A full HLT1 implementation on GPUs is available and was documented in detail for the accelerator's review in June 2019:

review document

Link to review Indico page

 If you want to join the effort, you are more than welcome. Please contact Daniel (dcampora@cernNOSPAMPLEASE.ch) and/or Dorothea (dorothea.vom.bruch@cernNOSPAMPLEASE.ch).
Changed:
<
<
The mailing list is lhcb-parallelization.
>
>
The mailing list is lhcb-rta-accelerators.
 

To do list

The list of items to be worked on, currently being worked on, and finished can be found here together with the people to whom they are assigned: https://gitlab.cern.ch/lhcb-parallelization/Allen/boards?=

Past meetings

Added:
>
>
Past meetings since Allen is part of WP6 of RTA:https://indico.cern.ch/category/10878/
 
Changed:
<
<
The archive of past meetings can be found here: https://indico.cern.ch/category/7616/
>
>
Past meetings before Allen became part of WP6 of RTA: https://indico.cern.ch/category/7616/
  Minutes from the meetings:

Revision 82019-01-07 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Line: 13 to 13
 

To do list

The list of items to be worked on, currently being worked on, and finished can be found here together with the people to whom they are assigned:
Changed:
<
<
https://gitlab.cern.ch/lhcb-parallelization/cuda_hlt/boards?=
>
>
https://gitlab.cern.ch/lhcb-parallelization/Allen/boards?=
 

Past meetings

Revision 72018-12-13 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

To be able to process data at 30 MHz, the possibility of running (part of) the algorithms of HLT1 on GPUs is currently being studied.

Changed:
<
<
The repository in which the code is developed and growing is cuda_hlt.
>
>
The repository of the project is Allen.
  If you want to join the effort, you are more than welcome. Please contact Daniel (dcampora@cernNOSPAMPLEASE.ch) and/or Dorothea (dorothea.vom.bruch@cernNOSPAMPLEASE.ch).

Revision 62018-11-27 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Line: 24 to 24
 
Changed:
<
<
>
>
 

Revision 52018-10-23 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Line: 15 to 15
 The list of items to be worked on, currently being worked on, and finished can be found here together with the people to whom they are assigned: https://gitlab.cern.ch/lhcb-parallelization/cuda_hlt/boards?=
Deleted:
<
<

Sub-projects

Possible Scenarios in the DAQ

 

Past meetings

The archive of past meetings can be found here: https://indico.cern.ch/category/7616/

Line: 32 to 23
 
Added:
>
>
 

Revision 42018-07-23 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Line: 12 to 12
 The mailing list is lhcb-parallelization.

To do list

Changed:
<
<
  • improvements of current algorithms (mainly the SearchByTriplet)
  • Decoding and clustering of the various detectors
  • add VeloUT and forward tracking
  • add the Kalman filter
  • reconstruct vertices
  • validation: physics performance (efficiencies), use Manuel's c++ implementation of standalone PrChecker
  • reproducibility: compare CPU and GPU results
  • back of the envelope calculation of data rates that need to be shipped around to have the information available for the whole tracking chain

Current distribution of tasks

  • Daniel Hugo Campora Perez: goal for the next 1-2 weeks: finish first demonstrator including clustering and decoding afterwards, Daniel will focus mainly on other projects
  • Daniel's summer student (2-3 months): to be decided
  • Placido : currently working on information formalism for Kalman filter (~until the summer), then he is interested in joining the GPU effort (for ~6 months)
  • Roel Aaij: framework integration (starting point is the repository he set up as interface to the current framework), maybe we can create an input from the standalone programs running on the GPU to be fed into and checked by the PrChecker?; (Velo decoding and clustering)?
  • Julius Roeder: Cellular automaton; could be used as input for the SearchByTriplet or possibly for other tracking algorithms along the chain (?)
  • Da Yu Tou: neural networks
  • Florian Reiss: vertex reconstruction or UT tracking
  • Arthur Hennequin: fast clustering algorithm (light speed labeling)
  • Dorothea vom Bruch: continue work on SearchByTriplet (apply change to Kalman filter code developed with Pierre to use floats safely), coordination
>
>
The list of items to be worked on, currently being worked on, and finished can be found here together with the people to whom they are assigned: https://gitlab.cern.ch/lhcb-parallelization/cuda_hlt/boards?=
 

Sub-projects

Added:
>
>
 

Possible Scenarios in the DAQ

Changed:
<
<

Minutes from the meetings

>
>

Past meetings

The archive of past meetings can be found here: https://indico.cern.ch/category/7616/

Minutes from the meetings:

 
Added:
>
>
 

Revision 32018-04-24 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

To be able to process data at 30 MHz, the possibility of running (part of) the algorithms of HLT1 on GPUs is currently being studied.

Changed:
<
<
At present, the SearchByTriplet project by Daniel Hugo Campora Perez exists, with which Velo tracks are reconstructed on a GPU, using the PrPixelTracking as starting point. In the future, more components of the tracking chain and also the decoding and clustering will be ported to GPUs.
>
>
The repository in which the code is developed and growing is cuda_hlt.
  If you want to join the effort, you are more than welcome. Please contact Daniel (dcampora@cernNOSPAMPLEASE.ch) and/or Dorothea (dorothea.vom.bruch@cernNOSPAMPLEASE.ch).
Line: 34 to 34
 
  • Arthur Hennequin: fast clustering algorithm (light speed labeling)
  • Dorothea vom Bruch: continue work on SearchByTriplet (apply change to Kalman filter code developed with Pierre to use floats safely), coordination
Added:
>
>

Sub-projects

Possible Scenarios in the DAQ

Minutes from the meetings

 

-- DorotheaVomBruch - 2018-03-13 \ No newline at end of file

Revision 22018-03-19 - DorotheaVomBruch

Line: 1 to 1
 
META TOPICPARENT name="LHCbTrigger"

Idea

Line: 28 to 28
 
  • Daniel's summer student (2-3 months): to be decided
  • Placido : currently working on information formalism for Kalman filter (~until the summer), then he is interested in joining the GPU effort (for ~6 months)
  • Roel Aaij: framework integration (starting point is the repository he set up as interface to the current framework), maybe we can create an input from the standalone programs running on the GPU to be fed into and checked by the PrChecker?; (Velo decoding and clustering)?
Changed:
<
<
  • Julian Roeder: Cellular automaton; could be used as input for the SearchByTriplet or possibly for other tracking algorithms along the chain (?)
>
>
  • Julius Roeder: Cellular automaton; could be used as input for the SearchByTriplet or possibly for other tracking algorithms along the chain (?)
 
  • Da Yu Tou: neural networks
  • Florian Reiss: vertex reconstruction or UT tracking
  • Arthur Hennequin: fast clustering algorithm (light speed labeling)

Revision 12018-03-13 - DorotheaVomBruch

Line: 1 to 1
Added:
>
>
META TOPICPARENT name="LHCbTrigger"

Idea

To be able to process data at 30 MHz, the possibility of running (part of) the algorithms of HLT1 on GPUs is currently being studied. At present, the SearchByTriplet project by Daniel Hugo Campora Perez exists, with which Velo tracks are reconstructed on a GPU, using the PrPixelTracking as starting point. In the future, more components of the tracking chain and also the decoding and clustering will be ported to GPUs.

If you want to join the effort, you are more than welcome. Please contact Daniel (dcampora@cernNOSPAMPLEASE.ch) and/or Dorothea (dorothea.vom.bruch@cernNOSPAMPLEASE.ch).

The mailing list is lhcb-parallelization.

To do list

  • improvements of current algorithms (mainly the SearchByTriplet)
  • Decoding and clustering of the various detectors
  • add VeloUT and forward tracking
  • add the Kalman filter
  • reconstruct vertices
  • validation: physics performance (efficiencies), use Manuel's c++ implementation of standalone PrChecker
  • reproducibility: compare CPU and GPU results
  • back of the envelope calculation of data rates that need to be shipped around to have the information available for the whole tracking chain

Current distribution of tasks

  • Daniel Hugo Campora Perez: goal for the next 1-2 weeks: finish first demonstrator including clustering and decoding afterwards, Daniel will focus mainly on other projects
  • Daniel's summer student (2-3 months): to be decided
  • Placido : currently working on information formalism for Kalman filter (~until the summer), then he is interested in joining the GPU effort (for ~6 months)
  • Roel Aaij: framework integration (starting point is the repository he set up as interface to the current framework), maybe we can create an input from the standalone programs running on the GPU to be fed into and checked by the PrChecker?; (Velo decoding and clustering)?
  • Julian Roeder: Cellular automaton; could be used as input for the SearchByTriplet or possibly for other tracking algorithms along the chain (?)
  • Da Yu Tou: neural networks
  • Florian Reiss: vertex reconstruction or UT tracking
  • Arthur Hennequin: fast clustering algorithm (light speed labeling)
  • Dorothea vom Bruch: continue work on SearchByTriplet (apply change to Kalman filter code developed with Pierre to use floats safely), coordination

-- DorotheaVomBruch - 2018-03-13

 
This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback