Various bugs in ISGetNonlocal, MatGetSubMatricesParallel and GASM; GASM appears to work with straddling subdomains.Hg-commit: d2b62a60e18823706aa949d31d180dedbc6898d6
moved definition of PETSC_STDCALL to private/fortranimpl.h where it belongsremoved all DLL export/import stuff since it was never used and never will beremoved MatPermuteSparsify() since it sucked.
moved definition of PETSC_STDCALL to private/fortranimpl.h where it belongsremoved all DLL export/import stuff since it was never used and never will beremoved MatPermuteSparsify() since it sucked.Hg-commit: 72f7a5e60333f6c9780d56fcce45657940d16b13
show more ...
changed all EXTERN to extern the CAPS served no purposeHg-commit: cdf5985509992c2237dbbe07f14278cccee0306c
add SVD solver for testingfixed SNESMonitorVI() to include sign of f at endpointsHg-commit: 2720df5a8657cce461e62a85d7a63e401c4c853f
Logic in the GASM 2D grid partitioning now seems okay, but parallel submatrix extraction appears to be broken.Hg-commit: 30bf180ded81500d6d0f314913a0b25abc562b34
Merge.Hg-commit: e1fc2c73eb56030dc76b20b68b959231bf213deb
GASM: parallel subdomain definition based on a 2D regular grid. Might break ex8g.Hg-commit: 0b6767a66e5f65b42948ee0954ae7da568e81d5c
changed sacuda to use VecCUDAGetArray thingsHg-commit: ae26e2743e24e04e80b9ca8b3f49c2bbd8322e50
Bug fix in GASM. In uniproc case with mult. blks/rank ex8 and ex8g agree to tol, but not exactly!Hg-commit: c074ad75a2cdd1f38e3c79c3a45ce73fbee51126
Promote LocalToGlobalMapping and BlockSize from DA to DMHg-commit: adc5020243929ce40bd18f2655d78fe82dc79cc4
Minor fixes and restored PCCreate_GASM in pcregis.c.Hg-commit: d6d1f44f7147776a6fa4b2127ad79d6382c63fc4
added VecCUDAGet/Restore things (just framework, not yet used)commented out setting number of cycles for sacuda (does not work yet)Hg-commit: 442aa5908a6623001343cdab0b4b235874a717db
added vecset 0 to sacuda (just in case)Hg-commit: b8daaaa7666294ee625ef0376e958d1bdc05837f
stripped VecView for CUDA since it's not neededswitched sacuda to apply with multiplyHg-commit: 2fbeeabe7055701fef69f0d148ab87b9a6a3885c
experimental change to sacuda to apply with cusp::multiplyHg-commit: d3f0a354efdb04478871f4a4e62ba2c3279d8c57
Remove unused variablesHg-commit: 77d6469bd2a57fcebd12034a6ee69a8a737df696
increase object state in sacuda pcHg-commit: 54849ebd9603ebdf26784680b18b81e9fd1ebfc1
added richardson for sacuda (untested)Hg-commit: 351646d53d6e00e0f03453f4deb04d228e371f0f
added -pc_sacuda_cycles optionHg-commit: a6a7991be63fbbae8e9c4de8f5c0220ab293621a
removed unneccesary copyHg-commit: 04a9f34dd5d468851d4f4512bc4f298ae814fed2
Dmitry, you have to add the directory to the makefileif you are going to add it to the default build by referencing it in PCRegisterAll()Hg-commit: fb8128580aa2a5b61d20d773d672f57937634748
GASM ex8g works in the basic case on 2 procs (matches ex8).Hg-commit: 42bfe620a94d49788934218129aec2b1715a44fb
Debug and run targets for GASM's ex8gHg-commit: 2798e96707ddae77e3f150fef5a1d8843c475cdc
ex8g for GASM works in the simplest partitioning case and in serial, but broken with n > 1.Hg-commit: 65613701997382cc4b0bd324c40b5952ad6aeaa8
Minor fixes to GASM.Hg-commit: 7ab1fc104c49f34211f4baf26a62f75f2c0ebe9c
1...<<171172173174175176177178179180>>...210