b9278a2 | Leif Strand | 28 July 2009, 01:00:36 UTC | Experiments show that using a block size of 1 leads to poor performance. Using the recommended minimum of 64, gauss_seidel_0 and gauss_seidel_1 are about 3x to 4x faster. | 28 July 2009, 01:00:36 UTC |
b500001 | Leif Strand | 27 July 2009, 23:44:52 UTC | Fixed stupid reduce bug that affected several routines. (This is what I get for coping & pasting example code without thinking.) Now the only remaining broken routine is n_assemble_del2_u(). | 27 July 2009, 23:44:52 UTC |
9a97294 | Leif Strand | 25 July 2009, 02:48:38 UTC | Wrote two new versions of gauss_seidel_2() -- neither of which is very parallel. Both produce similar wrong answers. | 25 July 2009, 02:48:38 UTC |
fb1101a | Thorsten Becker | 24 July 2009, 01:03:47 UTC | This version of precise strain-rates may now be worth testing out. | 24 July 2009, 01:03:47 UTC |
1c4fd5d | Thorsten Becker | 23 July 2009, 18:40:41 UTC | Corrected call to get_ba in Topo_gravity for precise_strain_rate, still testing. Rest should be cosmetic. | 23 July 2009, 18:40:41 UTC |
b2a7cb3 | Leif Strand | 23 July 2009, 03:50:26 UTC | This junk is my effort to understand gauss_seidel_2(). Because of the way E->temp[] is used, it is quite difficult (impossible?) to parallelize. I even wrote code to generate a "dot" graph of the expression tree. Of course, this tree is too big to plot in its entirety. With neq == 2187, the tree has 87479 nodes and is 355 levels deep; each level can have anywhere from a handful to several hundred nodes, with over a thousand nodes in each of the bottommost layers. | 23 July 2009, 03:50:26 UTC |
02a3b02 | Thorsten Becker | 22 July 2009, 20:17:11 UTC | Tentative check in of stress computation modification such that Eh can take a look of what we're working on. | 22 July 2009, 20:17:11 UTC |
5bfdbd3 | Leif Strand | 17 July 2009, 21:22:38 UTC | Created drop-in, host-side replacements for each of the multigrid kernel functions. After swapping them in/out, it seems n_assemble_del2_u(), gauss_seidel_2(), and gauss_seidel_3() are all buggy :-( Both gauss_seidel_0() and gauss_seidel_1() appear to be OK... but that isn't worth writing home about, because they are both trivial. | 17 July 2009, 21:22:38 UTC |
9d02f10 | Leif Strand | 16 July 2009, 03:13:33 UTC | New scaffolding with annotations. | 16 July 2009, 03:13:33 UTC |
25fc342 | Leif Strand | 16 July 2009, 00:49:42 UTC | Bug fixes... still doesn't work right. | 16 July 2009, 00:49:42 UTC |
60515e6 | Leif Strand | 15 July 2009, 21:54:24 UTC | Wrote code to allocate & initialize device memory, and copy the result from the device. | 15 July 2009, 21:54:24 UTC |
0bef9e8 | Leif Strand | 15 July 2009, 01:58:53 UTC | Parallelized the last fragment of the gauss_seidel() main loop as gauss_seidel_3(). It turns out to be very similar to the second half of n_assemble_del2_u(). Tore down scaffolding. | 15 July 2009, 01:58:53 UTC |
362c2e3 | Leif Strand | 15 July 2009, 00:14:53 UTC | Parallel n_assemble_del2_u() takes shape. | 15 July 2009, 00:14:53 UTC |
3e27612 | Leif Strand | 10 July 2009, 22:15:02 UTC | Pardon my dust... tally_n_assemble_del2_u() is instrumention and printfs I wrote in order to understand n_assemble_del2_u(). As with e_assemble_del2_u(), it appears I need one block for each Au[i] element, with a reduction in shared memory. | 10 July 2009, 22:15:02 UTC |
5825a00 | Dan Bower | 10 July 2009, 21:08:14 UTC | include density (rho) scaling in convective heat flux (necessary for compressible flow) | 10 July 2009, 21:08:14 UTC |
e0d4a44 | Dan Bower | 09 July 2009, 20:40:10 UTC | Second material element file only read in and processed when (pos_age) criteria met (i.e. consistent with vel b.c. approach). | 09 July 2009, 20:40:10 UTC |
0ee4a4c | Thorsten Becker | 09 July 2009, 17:27:41 UTC | Only one processor complains about removing rigid rotations while prescribing plate motions. | 09 July 2009, 17:27:41 UTC |
67d7472 | Leif Strand | 06 July 2009, 22:13:05 UTC | Make sure "-arch sm_13" is given -- for double-precision floating-point. Added missing #ifndef USE_CUDA. | 06 July 2009, 22:13:05 UTC |
735ae7c | Leif Strand | 02 July 2009, 18:41:00 UTC | Sketched data-parallel version of gauss_seidel(). I haven't figured out how to parallelize the following: Au[C[i]] += ... Ad[C[j]] += ... | 02 July 2009, 18:41:00 UTC |
43a1aa4 | Leif Strand | 01 July 2009, 19:09:31 UTC | Fixed a pair of mistakes I made: under cgrad, don't replace solve_del2_u(); under multigrid, 'level' is not a constant. | 01 July 2009, 19:09:31 UTC |
678e5fc | Leif Strand | 30 June 2009, 22:33:19 UTC | The 'temp1' array is unnecessary since nproc == 1. | 30 June 2009, 22:33:19 UTC |
9743b5e | Leif Strand | 29 June 2009, 22:10:40 UTC | For my own sanity, simplified code assuming the following: E->sphere.caps_per_proc == 1 E->parallel.nproc == 1 E->mesh.levmax == 0 E->mesh.nsd == 3 | 29 June 2009, 22:10:40 UTC |
87a1bca | Leif Strand | 29 June 2009, 20:18:25 UTC | Proposed CUDA kernel for multigrid solver. | 29 June 2009, 20:18:25 UTC |
d01a1c7 | Eh Tan | 25 June 2009, 16:51:10 UTC | forgot to add these files | 25 June 2009, 16:51:10 UTC |
f038904 | Eh Tan | 15 June 2009, 18:21:41 UTC | Update ChangeLog to r15180 | 15 June 2009, 18:21:41 UTC |
9a9cbd3 | Eh Tan | 10 June 2009, 20:42:43 UTC | Added COND_GGRD flags; removed redundant LIBS | 10 June 2009, 20:42:43 UTC |
c5915dc | Eh Tan | 10 June 2009, 20:41:31 UTC | Checking malloc.h earlier to initialize even when not using pyre | 10 June 2009, 20:41:31 UTC |
055dd50 | Eh Tan | 10 June 2009, 17:49:35 UTC | checking netcdf header and library | 10 June 2009, 17:49:35 UTC |
aa8a0ac | Eh Tan | 10 June 2009, 04:11:35 UTC | Append GMTHOME/lib to LD path | 10 June 2009, 04:11:35 UTC |
55d0112 | Eh Tan | 09 June 2009, 23:47:35 UTC | Adding ggrd compilation support | 09 June 2009, 23:47:35 UTC |
03dafa6 | Eh Tan | 08 June 2009, 21:03:32 UTC | One more feature and mention where to find the features in the cookbooks. | 08 June 2009, 21:03:32 UTC |
234f0ef | Eh Tan | 08 June 2009, 18:11:24 UTC | Added a safe guard | 08 June 2009, 18:11:24 UTC |
ba6a738 | Eh Tan | 04 June 2009, 21:04:33 UTC | Reordering parameter | 04 June 2009, 21:04:33 UTC |
bc9a15c | Eh Tan | 03 June 2009, 21:08:01 UTC | Update SVTInlet to inherit from BaseSVTInlet | 03 June 2009, 21:08:01 UTC |
56e2c06 | Eh Tan | 03 June 2009, 21:07:04 UTC | adding files to makefile | 03 June 2009, 21:07:04 UTC |
b3dfd01 | Eh Tan | 02 June 2009, 22:58:45 UTC | update to current coupler | 02 June 2009, 22:58:45 UTC |
7fd3100 | Eh Tan | 02 June 2009, 22:58:24 UTC | reducing the accuracy of solvers, otherwise, they will converge to the wrong result | 02 June 2009, 22:58:24 UTC |
748c305 | Eh Tan | 02 June 2009, 22:56:46 UTC | Added parameters 'amending_outflow' and 'exchange_pressure' to help the convergence of esolver When 'amending_outflow' is set to true, the imposed velocity BC will be amended slightly to be divergence-free. (The divergence, e.g. outflow, is caused by the combination of solver inaccuracy and interpolation inaccuracy.) When 'exchange_pressure' is set to true, the initial pressure (at element level) of the embedded solver is taken from the pressure solution of the containing solver. | 02 June 2009, 22:56:46 UTC |
179463a | Eh Tan | 02 June 2009, 22:09:25 UTC | Boundary for global mesh, not sure whether it works with interpolator or not | 02 June 2009, 22:09:25 UTC |
2ef0c3d | Eh Tan | 02 June 2009, 21:58:40 UTC | init'd E->output.seismic and E->output.coord_bin | 02 June 2009, 21:58:40 UTC |
101123f | Eh Tan | 02 June 2009, 21:58:23 UTC | refining cookbook10 | 02 June 2009, 21:58:23 UTC |
e1af994 | Leif Strand | 28 May 2009, 01:50:40 UTC | CUDA-ized e_assemble_del2_u(), and ran it successfully -- and slowly! -- under the device emulator. Added a new 'configure' option: "--with-cuda". | 28 May 2009, 01:50:40 UTC |
500e3f5 | Leif Strand | 27 May 2009, 02:32:03 UTC | Parallelized strip_bcs_from_residual(). | 27 May 2009, 02:32:03 UTC |
511d3f3 | Leif Strand | 27 May 2009, 01:28:55 UTC | Sketched data-parallel version of e_assemble_del2_u(). | 27 May 2009, 01:28:55 UTC |
060b73c | Thorsten Becker | 26 May 2009, 15:21:42 UTC | Added blob_bc_persist parameter (default = off) to allow assigning blob-like temperature anomalies to BCs. | 26 May 2009, 15:21:42 UTC |
5a6accf | Thorsten Becker | 26 May 2009, 15:03:36 UTC | Fixed typo in blob temperature assignment. Changed conversion of coordinates, now assignment works. | 26 May 2009, 15:03:36 UTC |
accc9b4 | Eh Tan | 21 May 2009, 23:16:27 UTC | add changes in r14999 | 21 May 2009, 23:16:27 UTC |
f99c960 | Eh Tan | 15 May 2009, 22:32:48 UTC | ignoring the degree-0 and degree-1 components of geoid | 15 May 2009, 22:32:48 UTC |
1670e55 | Eh Tan | 10 May 2009, 23:18:05 UTC | add more chaanges in v3.1.0 | 10 May 2009, 23:18:05 UTC |
a622619 | Eh Tan | 10 May 2009, 16:37:31 UTC | Change version number for v3.1.0 alpha release | 10 May 2009, 16:37:31 UTC |
658d80a | Eh Tan | 10 May 2009, 16:35:07 UTC | Summary of changes in this release | 10 May 2009, 16:35:07 UTC |
573f5d0 | Eh Tan | 01 May 2009, 22:19:35 UTC | Adding cookbook for synthetic seismogram, mostly copied from cookbook7 | 01 May 2009, 22:19:35 UTC |
7a399ef | Eh Tan | 01 May 2009, 22:18:00 UTC | Depth is measured from r=1.0, not outer radius, because in coupled model, the outer radius can be < 1.0. | 01 May 2009, 22:18:00 UTC |
c9c08bb | Eh Tan | 01 May 2009, 22:17:27 UTC | Printing a big warning when plate velocity is imposed but remove_angular_momentum or remove_rigid_rotation is on. | 01 May 2009, 22:17:27 UTC |
24392fd | Eh Tan | 01 May 2009, 22:15:05 UTC | Removing angular momentum in pseudo free surface case | 01 May 2009, 22:15:05 UTC |
5ab9dda | Eh Tan | 01 May 2009, 22:14:29 UTC | ignore anything after ';' | 01 May 2009, 22:14:29 UTC |
b215c86 | Eh Tan | 01 May 2009, 22:13:26 UTC | explicitly add Python.h to requirement | 01 May 2009, 22:13:26 UTC |
05da830 | Leif Strand | 28 April 2009, 22:53:38 UTC | For my own sanity, simplified code assuming the following: E->sphere.caps_per_proc == 1 E->parallel.nproc == 1 E->mesh.levmax == 0 E->mesh.nsd == 3 | 28 April 2009, 22:53:38 UTC |
2064543 | Leif Strand | 27 April 2009, 20:15:10 UTC | Moved malloc/free calls outside of kernel. (The amount of memory allocated appears to be invariant.) | 27 April 2009, 20:15:10 UTC |
fa871b7 | Leif Strand | 27 April 2009, 18:39:15 UTC | Consolidated calls to malloc/free. | 27 April 2009, 18:39:15 UTC |
2d995dd | Thorsten Becker | 24 April 2009, 18:37:49 UTC | Added flag to allow using the more precise method of computing strain-rates throughout the globe, and not just close to the poles. precise_strain_rate=on will select this, default is "odd", i.e. poles only as before. | 24 April 2009, 18:37:49 UTC |
0eb5c37 | Leif Strand | 23 April 2009, 02:32:01 UTC | Don't call through function pointers -- otherwise, 'nvcc' crashes! | 23 April 2009, 02:32:01 UTC |
185800d | Leif Strand | 23 April 2009, 02:22:19 UTC | Proposed CUDA kernel for conjugate gradient solver. | 23 April 2009, 02:22:19 UTC |
a6bc514 | Leif Strand | 22 April 2009, 00:59:20 UTC | Avoid core dump in record() when E->fp hasn't been set yet. | 22 April 2009, 00:59:20 UTC |
e45eab4 | Leif Strand | 21 April 2009, 22:29:26 UTC | The new Python.h header check reorders this script somewhat. CIT_CHECK_PYTHON_HEADER must follow AC_PROG_CC, and precede CIT_PYTHON_EGG_SETUP. | 21 April 2009, 22:29:26 UTC |
c46afd8 | Leif Strand | 21 April 2009, 19:33:59 UTC | Issue45: Check for Python.h. | 21 April 2009, 19:33:59 UTC |
756b97b | Thorsten Becker | 17 April 2009, 03:37:28 UTC | Added viscosity options 9 and 10 for TDPEPV which are identical to 3 and 8 but temperature is not limited to [0;1]. This is cumbersome, but perhaps needed for backward compatibility. | 17 April 2009, 03:37:28 UTC |
9ea8a44 | Eh Tan | 16 April 2009, 18:24:28 UTC | Move parameter "mantle_temp" from CitcomS.solver.param to CitcomS.solver.ic. Renamed E->control.lith_age_mantle_temp to E->control.mantle_temp. | 16 April 2009, 18:24:28 UTC |
0591133 | Eh Tan | 16 April 2009, 18:23:06 UTC | update cookbook8 with the manual | 16 April 2009, 18:23:06 UTC |
487eeea | Eh Tan | 16 April 2009, 18:22:22 UTC | Add -DUSE_HDF5 to CPPFLAGS in module/Exchanger | 16 April 2009, 18:22:22 UTC |
075b734 | Eh Tan | 16 April 2009, 18:21:50 UTC | change the viscosity section to be the same as that in cookbook3. | 16 April 2009, 18:21:50 UTC |
2c35d25 | Eh Tan | 16 April 2009, 18:20:08 UTC | Removed domain extents, since they are determined by coor_file. | 16 April 2009, 18:20:08 UTC |
a7389af | Eh Tan | 16 April 2009, 18:07:53 UTC | Removed instructions to 'make install', 'make clean' and 'make distclean'. Added dependencies. | 16 April 2009, 18:07:53 UTC |
a2a39ef | Eh Tan | 09 April 2009, 20:53:58 UTC | Fixed a bug in cgrad solver introduced in r13270. | 09 April 2009, 20:53:58 UTC |
7dbc76c | Thorsten Becker | 09 April 2009, 20:39:24 UTC | Left debugging output in, now cleaned up | 09 April 2009, 20:39:24 UTC |
d5b89ca | Thorsten Becker | 09 April 2009, 20:37:59 UTC | Fixed file output issue which caused vtk log file I/O to bomb during Citcom finalize. | 09 April 2009, 20:37:59 UTC |
9d6cfb4 | Eh Tan | 08 April 2009, 23:39:49 UTC | Sync'ing C and Pyre and adding comments | 08 April 2009, 23:39:49 UTC |
971856b | Eh Tan | 08 April 2009, 23:38:51 UTC | Added a function citcom_finalize() to close files and MPI before exit. Return 0 when the code is finished normally. Rearrang post_processing and stokes_flow_only. | 08 April 2009, 23:38:51 UTC |
6e42502 | Eh Tan | 08 April 2009, 23:34:43 UTC | Clean up. | 08 April 2009, 23:34:43 UTC |
811838a | Eh Tan | 08 April 2009, 23:31:11 UTC | Removed multigrid-el option of stokes solver, which is broken for a long time and not used by anyone. | 08 April 2009, 23:31:11 UTC |
f482a84 | Eh Tan | 08 April 2009, 23:27:51 UTC | Moved where to call read_rayleigh_from_file() earlier. I hope this doesn't break the ray_control stuff. | 08 April 2009, 23:27:51 UTC |
5ab6970 | Thorsten Becker | 08 April 2009, 20:24:59 UTC | Fixed bug in vtop assignment. | 08 April 2009, 20:24:59 UTC |
c6a6fd7 | Eh Tan | 07 April 2009, 22:50:02 UTC | Revert the unintended modification of examples/Makefile.am | 07 April 2009, 22:50:02 UTC |
c5e1db6 | Eh Tan | 07 April 2009, 22:45:32 UTC | Change the viscosity option in cookbook3 and cookbook4 | 07 April 2009, 22:45:32 UTC |
ce1c4ce | Eh Tan | 07 April 2009, 22:44:24 UTC | Change the mesh size in cookbook2 | 07 April 2009, 22:44:24 UTC |
02125c8 | Eh Tan | 04 April 2009, 00:59:32 UTC | Renamed Seismic_model.c to Mineral_physics_models.c | 04 April 2009, 00:59:32 UTC |
1ff1990 | Eh Tan | 04 April 2009, 00:57:47 UTC | Fixed a few bugs in seismic output * fixed errors in the coefficient table * fixed typos * normalized drho by reference density profile * more digits in the prem radius table * added two specfem flags in PREM calculation. * "dv" output for debugging purpose, disabled. | 04 April 2009, 00:57:47 UTC |
795f249 | Eh Tan | 04 April 2009, 00:52:15 UTC | Inconsisent parameters with the manual | 04 April 2009, 00:52:15 UTC |
9de0f51 | Thorsten Becker | 28 March 2009, 02:29:00 UTC | Modified the way the compositional init with ggrd tracer can now select either a range of layers (>0) or a specific layer (<0). | 28 March 2009, 02:29:00 UTC |
dc9672a | Eh Tan | 20 March 2009, 23:36:56 UTC | limiting the number of multigrid cycles in solve_del2_u(). A new input parameter solver.vsolver.max_mg_cycles (default to 50) controls the max. number of multigrid cycles when solving the velocity. Without this parameter, when the stiffness matrix is near singular, the the multigrid solver may become an infinite loop. | 20 March 2009, 23:36:56 UTC |
25a7493 | Eh Tan | 18 March 2009, 19:39:54 UTC | Remove mgunitx etc from pyre input. This restores the behavior in v3.0 and earlier version. In pyre version, the mesh size is always specified by nodex etc. In C version, the mesh size is specified by nodex if Solver=cgrad, and by mgunitx and levels if Solver=multigrid. | 18 March 2009, 19:39:54 UTC |
04c4125 | Eh Tan | 17 March 2009, 23:42:01 UTC | Rollback the unintended change in velo output. | 17 March 2009, 23:42:01 UTC |
6e50c43 | Eh Tan | 17 March 2009, 00:18:09 UTC | Check return values of sscanf/fscanf. This fixs issue265. | 17 March 2009, 00:18:09 UTC |
213f078 | Eh Tan | 17 March 2009, 00:17:04 UTC | Removed obsolete function read_previous_field() | 17 March 2009, 00:17:04 UTC |
c1fbc1d | Eh Tan | 17 March 2009, 00:16:11 UTC | remove angular momentum from stokes solution by setting solver.vsolver.remove_angular_mementum=1 | 17 March 2009, 00:16:11 UTC |
e8525af | Eh Tan | 16 March 2009, 23:17:30 UTC | Converting temperature/composition fields to seismic velocities, in a format that specfem3d portal and v4.1 can understand. * solver.param.mineral_physics_mode: default to 3, the model of Tramper, Vacher and Vlaar's PEPI 2001. * output_optional=seismic will write coordinates and seismic velocities in binary files for specfem comsumption. * domain bounds file "datafile.domain": binary file with nproc * 10 doubles. The 10 doubles are (rmin, rmax) and four (theta, phi) pairs for the four bottom corner nodes. This file is written by rank-0 processor only. | 16 March 2009, 23:17:30 UTC |
ebff88c | Eh Tan | 16 March 2009, 22:52:08 UTC | add test cases for the stokes solver under bousinessq approximation and truncated anelastic liquid approximation | 16 March 2009, 22:52:08 UTC |
e1f9320 | Eh Tan | 16 March 2009, 22:50:06 UTC | seperate initialization and parameter input | 16 March 2009, 22:50:06 UTC |
ac6cb95 | Eh Tan | 16 March 2009, 22:44:42 UTC | change the default of mgunit | 16 March 2009, 22:44:42 UTC |
8c586f0 | Eh Tan | 16 March 2009, 22:43:37 UTC | Make the initialization step in pyre conforms to C version. This fixs a bug in r13270. | 16 March 2009, 22:43:37 UTC |