Internal:GPU Usage: Difference between revisions

From crtc.cs.odu.edu
Jump to navigation Jump to search
← Older edit
Spyridon97 (talk | contribs)
CNF, external, internal
2,385 edits
No edit summary
Spyridon97 (talk | contribs)
CNF, external, internal
2,385 edits
 
(15 intermediate revisions by the same user not shown)
Line 1: Line 1:
===PODM Refactoring===
=Explore Usage of GPU=
# Find a Predicates library to replace CGAL library. A possible solution is [https://hal.inria.fr/hal-01225202 Predicate Construction Kit] and its [https://gforge.inria.fr/frs/?group_id=5833 code].
# For “itkSignedMaurerParallelDistanceMapImageFilter” code:
# Remove Boost library and replace with std::thread, ask Christos for Barrier.
## Understand the code by reading the [https://cepm.cs.odu.edu/pub/papers/journal_23.pdf Parallel N-Dimensional Exact Signed Euclidean Distance Transform] paper
# Remove VTK library
## Study if it’s beneficial to convert it in order to use GPUs. See the papers:
# Update ITK
### [https://crtc.cs.odu.edu/pub/papers/journal_18.pdf A Multigrain Delaunay Mesh Generation Method for Multicore SMT-based Architectures], (TODO)
# Choose a specific C++ coding style✔
### [https://crtc.cs.odu.edu/pub/papers/journal_19.pdf Algorithm, Software, and Hardware Optimizations for Delaunay Mesh Generation on Simultaneous Multithreaded Architectures] (TODO)
# Use C++17 standard✔
# For Predicates:
# Create a general Mesh Format which supports both 3D and 4D (and higher dimensions). This will be helpful for Linear interpolation.
## Based on [https://imr.sandia.gov/_assets/documents/2019_IMR_Papers/6B.1-de%20Matos%20Menezes.pdf ACCELERATING THE EXACT EVALUATION OF GEOMETRIC PREDICATES WITH GPUS] (TODO)
# Combine 3D and 4D PODM versions into one abstract program, in which the user will choose what he wants to use.
## Based on [https://ieeexplore.ieee.org/abstract/document/8692354 GPredicates - GPU Implementation of Robust and Adaptive Floating-Point Predicates for Computational Geometry] (TODO)
## Create namespaces for each PODM which will help creating a templated PODM wrapper✔
## Decouple Args from PODM✔
## Create setArguments function✔
## Remove setImage* functions and create setInputImage(ImageType::Pointer, string)✔
## Remove GetShortFileName✔
## Create a Mesh Unstructured Grid to facilitate Linear Interpolation in N-Dimensions
## Template the common code of each PODM for 3-4 dimensions (or even more or less)
## Keep only useful Defines
# Use GPUs:
## For “itkSignedMaurerParallelDistanceMapImageFilter” code:
### Understand the code by reading the [https://cepm.cs.odu.edu/pub/papers/journal_23.pdf Parallel N-Dimensional Exact Signed Euclidean Distance Transform] paper
### Study if it’s beneficial to convert it in order to use GPUs. See the papers:
#### [https://cepm.cs.odu.edu/pub/papers/journal_18.pdf A Multigrain Delaunay Mesh Generation Method for Multicore SMT-based Architectures],  
#### [https://cepm.cs.odu.edu/pub/papers/journal_19.pdf Algorithm, Software, and Hardware Optimizations for Delaunay Mesh Generation on Simultaneous Multithreaded Architectures]
## For Predicates:
### Based on [https://imr.sandia.gov/_assets/documents/2019_IMR_Papers/6B.1-de%20Matos%20Menezes.pdf ACCELERATING THE EXACT EVALUATION OF GEOMETRIC PREDICATES WITH GPUS]
### Based on [https://ieeexplore.ieee.org/abstract/document/8692354 GPredicates - GPU Implementation of Robust and Adaptive Floating-Point Predicates for Computational Geometry]
# Decouple Contentions managers and Load Balancing from the Meshing Procedure.
# Add PDR
# Add Smoothing

Latest revision as of 16:20, 30 April 2020

Explore Usage of GPU

  1. For “itkSignedMaurerParallelDistanceMapImageFilter” code:
    1. Understand the code by reading the Parallel N-Dimensional Exact Signed Euclidean Distance Transform paper
    2. Study if it’s beneficial to convert it in order to use GPUs. See the papers:
      1. A Multigrain Delaunay Mesh Generation Method for Multicore SMT-based Architectures, (TODO)
      2. Algorithm, Software, and Hardware Optimizations for Delaunay Mesh Generation on Simultaneous Multithreaded Architectures (TODO)
  2. For Predicates:
    1. Based on ACCELERATING THE EXACT EVALUATION OF GEOMETRIC PREDICATES WITH GPUS (TODO)
    2. Based on GPredicates - GPU Implementation of Robust and Adaptive Floating-Point Predicates for Computational Geometry (TODO)
Retrieved from "https://crtc.cs.odu.edu/index.php?title=Internal:GPU_Usage&oldid=5806"