Florian Rathgeber

lp:~florian-rathgeber/+junk/dolfin-gpu-lma-spmv

Created by Florian Rathgeber and last modified
Get this branch:
bzr branch lp:~florian-rathgeber/+junk/dolfin-gpu-lma-spmv
Only Florian Rathgeber can upload to this branch. If you are Florian Rathgeber please log in for upload directions.

Related bugs

Related blueprints

Branch information

Owner:
Florian Rathgeber
Status:
Development

Recent revisions

83. By Florian Rathgeber <florian@thehoff>

experimental: fused stages 1,2,3 of spmv for LMA in GPUSolver to single stage (uses atomic adds for addto to destination -> 3x slower)

82. By Florian Rathgeber <florian@thehoff>

added build with memory monitoring to regression test

81. By Florian Rathgeber <florian@thehoff>

.bzrignore

80. By Florian Rathgeber <florian@thehoff>

demos: removed output of file for final timestep

79. By Florian Rathgeber <florian@thehoff>

Assemble-solve benchmark:
  * added 2 tables for time fraction spent in assemble and in solve resp.
  * compute speedup of different implementations against baseline (PETSc assembled) for timing tables
  * catch exceptions occuring in the run and write out the tables so that results are not lost

78. By Florian Rathgeber <florian@thehoff>

optimized PETSc parameters for demos

77. By Florian Rathgeber <florian@thehoff>

added memory monitoring (enabled with enableMemoryMonitoring build parameter) for GPU memory; seems not to work currently (Cuda driver error 3)

76. By Florian Rathgeber <florian@thehoff>

Restructured build system (now hierarchical), remove CPU reference variant, added memory monitoring variant

75. By Florian Rathgeber <florian@thehoff>

improved benchmark: added form without coefficients, PETSc with optimized parameters and with defaults

74. By Florian Rathgeber <florian@thehoff>

fixed debug mode build errors

Branch metadata

Branch format:
Branch format 7
Repository format:
Bazaar repository format 2a (needs bzr 1.16 or later)
This branch contains Public information 
Everyone can see this information.