852b384 | Tim Haines | 23 October 2018, 23:21:25 UTC | Complete overhaul of build system Change-Id: Ic48c4ac92b03138cfdce71d0792caedf15a97d76 | 28 November 2018, 20:54:51 UTC |
bff00a3 | Tom Quinn | 24 November 2018, 20:35:13 UTC | HUGE has gone away; replace with FLT_MAX. Change-Id: I9fa87d419605c0f555ad26682b122828e67e6430 | 25 November 2018, 02:06:23 UTC |
45a3ed3 | Tim Haines | 10 December 2017, 10:35:00 UTC | Replace sprintf free buffers with formatted_string Compiling with gcc-7.2 shows calculated string sizes possibly larger than buffer size. These arbitrarily-sized buffers are replaced with the new formatted_string type that checks the format string length at runtime and allocates space either on the stack or the heap, accordingly. Change-Id: Ib57f3d868ce0e37251658e9778c88fc4dc66ac82 | 11 November 2018, 03:37:32 UTC |
41969a5 | Tim Haines | 02 November 2018, 22:18:19 UTC | Fix ci interface declaration mismatch for requestRemoteMoments In the ci file, we have void requestRemoteMoments(const Tree::NodeKey key, int sender) In TreePiece.cpp, we have void TreePiece::requestRemoteMoments(const Tree::NodeKey key, int sender) The mismatch of const on the first argument caused charmci to create a templated version which was never instantiated. This is only an issue when not using MERGE_REMOTE_REQUESTS as the [inline] modifier generates a non-template member function. Change-Id: I4ce600dbc66871d4f9eb6c0dc98ee86084c361e9 | 04 November 2018, 20:37:23 UTC |
16dde56 | Tom Quinn | 31 October 2018, 21:12:02 UTC | Found a missed CUDA to HAPI change. Change-Id: I81f2e67d4e7d54fbed387335e44789895321341b | 31 October 2018, 21:14:08 UTC |
5bd6256 | Tom Quinn | 07 August 2018, 23:09:30 UTC | allocatePinnedHostMemory(): use size_t. This allows larger than 2GB buffers to be allocated for the GPU. Also use size_t in GPU Transfer functions. Change-Id: I36bb3ec4156e4f7790ad24d0fc172b134a196c7e | 31 October 2018, 21:05:47 UTC |
9c94f4e | Michael Robson | 10 May 2017, 20:42:11 UTC | cuda: Update to new GPU Manager 6.9 API * Remove old commented out API vars * Change bufferInfo to buffers to match new API * Convert workRequests to new hapi_createWorkRequest API * Convert direct access to wr userData to get fn API * Update header file name * Various rewording changes to match new API * Convert CUDA_ defines to HAPI_ * Add hapi prefixes to HAPI structs * Guard setTraceName() with HAPI_TRACE flag * Use C++11 for NVCC and remove -lcuda from Makefile.in Change-Id: I8081e4044cf0bba6265d41c5a18895dfb23dcda8 | 22 October 2018, 17:20:37 UTC |
d6cd388 | Tim Haines | 02 October 2018, 06:13:56 UTC | Remove vestiges of callDummy in Compute.cpp This is a bit of cleanup from 04b13678ef64c56ca41ff48c2b3461578b389fb1 Change-Id: I4299fbb18f438a1647e1f2c2d62070c3156396fe | 04 October 2018, 03:06:18 UTC |
adfea1f | Tim Haines | 02 October 2018, 03:10:15 UTC | Fix compile error when using GPU local walk and not HEXADECAPOLE Change-Id: If3c9c1e4f640250a81895fe2dc0104164de51352 | 02 October 2018, 03:10:15 UTC |
27cbc83 | Tom Quinn | 26 July 2018, 23:01:57 UTC | GPU code: documentation and cleanup. Change-Id: Ia00263ea802ae044ef0421cdeff056fc8371c77d | 15 August 2018, 13:54:49 UTC |
04b1367 | Tom Quinn | 20 November 2017, 04:17:45 UTC | Clean up calls to sendXXXXInteractionsToGpu(). Change-Id: I99c2a5f84788a8276ef6582c7eb88feb707b3c21 | 14 August 2018, 16:55:20 UTC |
2437d32 | Tom Quinn | 12 November 2017, 02:21:46 UTC | GPU code cleanup/documenatation. Change-Id: Iaad82fbf3e19c7906ea39ac23bad20bfeb7b34d3 | 14 August 2018, 16:41:59 UTC |
f7d1a30 | Cambridge | 05 August 2018, 18:45:33 UTC | Remove two variables used for debugging/evaluation Change-Id: I11140097b4c9030ba9a815cb535fc7277bd1d153 | 05 August 2018, 19:47:24 UTC |
4825f44 | Tom Quinn | 30 June 2018, 21:49:04 UTC | Handle 1 or 2 pieces with Oct Decomp. Change-Id: I41b7a3f6c1bbdc3f510435272339ed1476ce679d | 31 July 2018, 03:27:19 UTC |
cbe1f43 | Thomas Quinn | 25 April 2018, 18:19:13 UTC | Clean up MultistepLB_notopo. Get rid of unused code, including greedy. Make the "migrate all" work correctly. Still needs work to avoid bias when the number of active piece is very small. Change-Id: I7bf8d26ac76f97930f5fbc1a0a6cf03e3c88a083 | 28 July 2018, 18:09:44 UTC |
163f370 | Tom Quinn | 22 June 2018, 22:04:59 UTC | Reserving lists allocates a lot of unused memory. Change-Id: Ie73ac34087b665d00aa561c751afcbaa8de278cf | 28 July 2018, 16:59:59 UTC |
9ec68a2 | Tom Quinn | 20 June 2018, 23:51:48 UTC | Refactor Main kick and updateuDot code. Change-Id: I0c6718576ceb48b8911d4c0ff5c4272d6fae5dda | 27 July 2018, 18:29:08 UTC |
23904fd | Tom Quinn | 04 May 2018, 18:09:59 UTC | Refactor Main externalGravity code. Change-Id: I4e70953fffad72475872e0e7787b8222318e44d9 | 27 July 2018, 18:28:47 UTC |
99bf1dc | Tom Quinn | 27 April 2018, 16:07:22 UTC | Refactor Main gravity. Change-Id: Ide635caa23ddf0eaa8dc055e9b9974152852c31d | 27 July 2018, 18:27:11 UTC |
ccb40fb | Tom Quinn | 16 March 2018, 23:40:48 UTC | Refactor buildTree in Main. Change-Id: Ia8c53781a4663450c864c73e38f256c64cb058c9 | 27 July 2018, 18:26:25 UTC |
ddfeb8d | Tom Quinn | 16 March 2018, 20:38:51 UTC | Refactor load balancing in Main. Change-Id: I531630b9935bad337371481e8d41d4a30e570c9a | 27 July 2018, 18:26:03 UTC |
fa1998e | Tom Quinn | 16 March 2018, 19:28:10 UTC | Refactor domain decomposition. Change-Id: Ib3938dd63adf40af01338f557e2c5afd913afd47 | 27 July 2018, 17:37:23 UTC |
5538f73 | Thomas Quinn | 14 July 2018, 16:50:00 UTC | sendNodeInteractionsToGpu(): don't send empty list. Skip sending an empty interaction list to the GPUManager. This avoids a lot of work that does nothing. Also added some memory corruption checks in cudaCallback() since the empty lists seem to be causing memory corruption issues, but it could be something else. Change-Id: Ie3397eb6d0cfbc36cd8c67b4b6a2438dccd6bf05 | 23 July 2018, 20:46:37 UTC |
e693f73 | Tom Quinn | 19 June 2018, 20:54:33 UTC | Check GPU kernel launches. Also check cudaMemcpyToSymbol(). These fail if the compile did not include the right device. Change-Id: I500f438d0f75b775d507da2abfd0fe4495169aa5 | 23 July 2018, 19:13:53 UTC |
a8b7965 | Cambridge | 30 April 2018, 06:07:45 UTC | All the changes for GPU local tree walk Change-Id: I1c58c774f3b21e3fdf085809bb49381f963ab52b | 13 July 2018, 19:11:06 UTC |
d84c8e1 | Tom Quinn | 29 June 2018, 19:18:18 UTC | assignKeys(): Always calculate boundingBox. The boundingBox should be calculated even if the TreePiece has no particles since it may get particles after domain decomposition. Change-Id: I0e885fc8ffbb8b5063b14862b9e8f529ca2fee0b | 29 June 2018, 19:18:18 UTC |
9e90a99 | Tom Quinn | 21 June 2018, 02:52:48 UTC | Handle cooling shutoff in cooling_grackle. Change-Id: I0e0bc10a85e5e9c2fd940a96f640f3d67e7adb39 | 21 June 2018, 02:56:00 UTC |
7e4383a | Evan Ramos | 06 November 2017, 23:05:21 UTC | Cast a void pointer to the proper non-void type Change-Id: I809ace0ae0ca7f25fe27b7e8b939164d4f977681 | 08 May 2018, 16:48:27 UTC |
08d3090 | Evan Ramos | 06 November 2017, 23:04:18 UTC | Declare Cha_CommitID as extern Change-Id: I503095a0a58efb3052011bcb1576dc2c20d7f9b1 | 08 May 2018, 16:48:18 UTC |
c1aac6e | Evan Ramos | 06 November 2017, 23:04:06 UTC | Ensure cosmo.h always includes pup_stl.h Change-Id: Ib4fdfce06497341833fbc4bdf0b1e148fac15be4 | 08 May 2018, 16:48:04 UTC |
c34a147 | Thomas Quinn | 10 February 2018, 00:41:35 UTC | Change bool to int for LB params. The param module assumes "bool" the same size as an "int". Work around this by always using ints for boolean parameters. Change-Id: I5253018ba0c219163a73db89b2a89b53c1a9acfe | 07 May 2018, 18:15:05 UTC |
026bbe5 | Tom Quinn | 16 February 2018, 05:31:54 UTC | Fix issues with nSteps = 0 and timestep calculation. Change-Id: Ibf9b5ce825b14414bb765716f5361a80c79525fa | 01 May 2018, 19:04:34 UTC |
f9a522d | Tim Haines | 15 February 2018, 05:25:06 UTC | Remove vestigial COSMO_DEBUG logging in DataManager Change-Id: I312f6c559a2394a4e0d06c0a2f8e710b26f66b51 | 15 February 2018, 05:40:18 UTC |
eb4e8e5 | Tim Haines | 30 April 2017, 03:58:29 UTC | Use improved 128-bit hash for std::unordered_map This requires the Charm++ changeset https://charm.cs.illinois.edu/gerrit/#/c/2515/ that has the 128-bit template specialization of std::hash. Change-Id: I708d4703c0c85cdf0204049517885b38e02256ec | 08 February 2018, 03:33:36 UTC |
7c4d6f5 | Tim Haines | 31 January 2018, 18:37:01 UTC | Pass CkCallback by const& during CUDA instrumentation This one was missed by c49211. Change-Id: I769e550ebe0f2e1376293161a8d003d3ed5d6c0f | 08 February 2018, 03:24:07 UTC |
3780927 | Harshitha | 17 February 2016, 05:39:14 UTC | Reset the background load in HierarchOrbLB because we don't use that for LB Change-Id: Id35c291cc960d00ada65d81bfacd6ef5a902a8e7 | 01 February 2018, 20:32:22 UTC |
39ba07e | Tim Haines | 26 January 2018, 10:46:58 UTC | Remove usage of remoteBucketID This should have been removed in 68b9fbd199df1b34d5b5fb6f7e0ac2bd343583b3 Change-Id: I93eee10e45a8f02476f3bdb643146659e02c57b7 | 27 January 2018, 08:37:18 UTC |
acf1cad | Michael Robson | 06 December 2017, 20:41:51 UTC | cleanup: Rename C file extensions to cpp Use git log --follow path option to get full history after this commit. Change-Id: I20ad2475c72b407793fd2084cd42d2770fe5e703 | 25 January 2018, 16:28:15 UTC |
991ba99 | Tom Quinn | 15 December 2017, 01:15:23 UTC | Merge "Remove FastMultipole.C" | 15 December 2017, 01:15:23 UTC |
33dc323 | Tim Haines | 10 December 2017, 10:19:33 UTC | Remove FastMultipole.C This file does not compile against v3.2 Change-Id: I3ed992477b55010c1918436a3fca02ad25e25617 | 10 December 2017, 10:19:33 UTC |
30f002a | Tim Haines | 09 December 2017, 09:34:53 UTC | Remove deprecated CUDA device emulation support. Emulation support was removed in CUDA 3.0 in 2010. Change-Id: Id0754d05aa535de4543eb567f37b41a6d3b3ab00 | 09 December 2017, 22:43:24 UTC |
2eb2390 | Tim Haines | 09 December 2017, 22:39:14 UTC | Revert "Remove deprecated CUDA device emulation support." This reverts commit f534dfaad13ba2958ceb7f314dc558a71fcf0e2f. | 09 December 2017, 22:39:14 UTC |
f534dfa | Tim Haines | 09 December 2017, 09:34:53 UTC | Remove deprecated CUDA device emulation support. Emulation support was removed in CUDA 3.0 in 2010. Change-Id: Id0754d05aa535de4543eb567f37b41a6d3b3ab00 | 09 December 2017, 09:34:53 UTC |
be54ccf | Phil Miller | 01 December 2017, 17:10:18 UTC | Makefile: Fix parallel build failures for decl/def files Change-Id: Ida05530112bd068ad914fe4f8a3d4a8e5b0e09e5 | 05 December 2017, 00:44:43 UTC |
0304edf | Tom Quinn | 04 December 2017, 02:40:04 UTC | Undo last few changes and revert back to 9f57173 The last six changes appear to have been accidental. This reverts back to the "Fix memory issues" commit. Change-Id: I2cf25b8021cbce729a4970a7e69679bc89edec5d | 04 December 2017, 02:40:04 UTC |
57077f1 | Robel | 03 December 2017, 21:40:14 UTC | Revert "Remove test result" | 03 December 2017, 21:40:42 UTC |
306f79c | Robel Geda | 03 December 2017, 20:48:11 UTC | Revert "Remove test result" This reverts commit 599e81ad9b415b124c0e8ceaa3a04ca3ad6b4f90. Change-Id: If2462c496faa7145ca6dfd7772826e9b88f41f4e | 03 December 2017, 20:48:11 UTC |
174437b | Robel | 03 December 2017, 20:45:53 UTC | Merge branch 'master' of https://charm.cs.illinois.edu/gerrit/cosmo/changa | 03 December 2017, 20:45:53 UTC |
86a2f27 | Robel | 03 December 2017, 20:44:48 UTC | Found the bug and added a fix. Not meant to merge Change-Id: Icfa509729edf6223caa1ee6e0a593a7c41e66049 | 03 December 2017, 20:44:48 UTC |
42f8641 | Robel | 03 December 2017, 20:43:10 UTC | Found the bug and added a fix. Not meant to merge Change-Id: If94c709ef672c1d1e0ff4d9415c382989bbd347b | 03 December 2017, 20:43:10 UTC |
599e81a | Tom Quinn | 18 November 2017, 05:54:02 UTC | Remove test result Change-Id: I1b3f2d2ec3b5ea469adfcf57177be62d51e6e262 | 03 December 2017, 04:19:31 UTC |
9f57173 | Tom Quinn | 18 November 2017, 05:54:02 UTC | Fix memory issues: a leak and unitialized dTimeFB Change-Id: I588be5297e2bedd0a6b075b64b6beb6a19f67b79 | 18 November 2017, 05:54:02 UTC |
6a59a43 | Phil Miller | 07 November 2017, 17:00:14 UTC | Feedback: const safety Change-Id: I20c38bfb2b8803931e747911c4ebe86f61f0ea6c | 17 November 2017, 16:23:29 UTC |
9659552 | Phil Miller | 07 November 2017, 16:45:13 UTC | Improve const safety around Romberg integrators Change-Id: I8809fcf9a5914c78c3d74dfc884ee906a0aefafb | 16 November 2017, 06:03:55 UTC |
c49211a | Phil Miller | 05 November 2017, 16:35:15 UTC | Pass all CkCallback objects around by const& Change-Id: I756fcbcb92dbfed15a70e1e10efc6191331728e3 | 16 November 2017, 05:49:02 UTC |
543a188 | Phil Miller | 07 November 2017, 16:22:32 UTC | NodeType and GenericTrees enums: move PUP into namespace for proper ADL resolution Change-Id: I447a5f99e167643487bc6b40103572a4b6ecd2b2 | 16 November 2017, 05:02:31 UTC |
740e68d | Tim Haines | 30 April 2017, 03:57:36 UTC | Only initialize partMap in DoubleWalkState ctor if CUDA is enabled Change-Id: I300403315cc146cae1ad8a80305bcdc24f5bf77e | 16 November 2017, 04:48:54 UTC |
e7dadad | Tom Quinn | 22 October 2017, 00:57:01 UTC | Eliminate unused load balancing variables. Change-Id: Ia4e1f2243e9050afb9802cdd8c5109a7a4e3af14 | 16 November 2017, 04:34:54 UTC |
9229669 | Tom Quinn | 22 October 2017, 00:26:19 UTC | Only store LB user data in balancers expecting it. The standard load balancers that come with Charm (e.g. OrbLB) are not expecting centroid information, so storing user data caused memory corruption. Change-Id: I4a68151b63c41ca3dc685c490fc5ede17b98b1bf | 16 November 2017, 04:14:56 UTC |
7cc0b9c | Tom Quinn | 16 November 2017, 04:04:45 UTC | GPU callbacks change did not compile on non-GPU. Change-Id: I672f5a97b8a515713e0c5414814e3719acb5d8f4 | 16 November 2017, 04:04:45 UTC |
7f2bc87 | Thomas Quinn | 20 October 2017, 01:26:32 UTC | Change GPU transfer callbacks to DataManager methods. This also gets rid of the now vestigial DataManagerHelper group. Change-Id: I878a8c5c759af1fa0647c4bb48b7bef51d8fe0ce | 14 November 2017, 22:17:38 UTC |
1450b79 | Thomas Quinn | 12 November 2017, 02:55:37 UTC | Merge pull request #197 from N-BodyShop/sinks Sinks | 12 November 2017, 02:55:37 UTC |
a9209de | Thomas Quinn | 01 July 2015, 18:32:09 UTC | Main::advanceBigStep(): doSinks for all sinks, not just black holes. | 30 October 2017, 17:27:16 UTC |
787771e | Tom Quinn | 30 October 2017, 04:11:04 UTC | Merge branch 'public' into sinks Conflicts: GravityParticle.h Makefile.dep Makefile.in ParallelGravity.cpp feedback.C parameters.h smoothparams.h | 30 October 2017, 17:01:32 UTC |
cff4d8d | Tom Quinn | 23 October 2017, 16:57:28 UTC | GPU Race condition fix didn't compile on non-GPU. Change-Id: I620e52bc716b57d97a03e7fb2776f6cb833184bd | 23 October 2017, 16:57:28 UTC |
be63d23 | Tom Quinn | 21 October 2017, 02:09:26 UTC | Add LICENSE file. Change-Id: Ie37cd0735d60fb81343919af86b5ec274b60f3da | 23 October 2017, 16:47:49 UTC |
6aedff9 | Thomas R. Quinn | 24 August 2017, 04:13:52 UTC | Fixes to handle empty TreePieces. This includes forwarding node requests in requestRemoteMoments() and marking buckets as "initialized". Change-Id: I2fa46fc532429f1a1fabbc5284dd900754c436dd | 23 October 2017, 16:46:11 UTC |
d5c944d | Thomas R. Quinn | 11 October 2017, 22:28:50 UTC | Fix SPH/GPU race on acceleration. Change-Id: I21e0101bc50432191adfa6b249f286c30e3c4a6d | 23 October 2017, 16:44:56 UTC |
2be7c1d | Tom Quinn | 07 October 2017, 22:54:07 UTC | Fix AGORApreCheckSmoothParams::isSmoothActive() bug. Change-Id: I57b4afd2778a21da22538bdcc4a026b0db9513de | 07 October 2017, 22:54:07 UTC |
464fb95 | Tom Quinn | 14 August 2017, 22:41:05 UTC | Update to v3.3 Change-Id: Icb82cbd8a7e18f918a512d4385ed6412415d5500 | 11 September 2017, 16:37:31 UTC |
70c2568 | Tom Quinn | 06 September 2017, 19:48:02 UTC | Report SPH time when no self gravity. Change-Id: Ibb80be81e6640856562d53047a1097d63cb73500 | 06 September 2017, 19:48:02 UTC |
b1a69a4 | Thomas R. Quinn | 31 August 2017, 21:11:04 UTC | ExternalGravity::applyPotential(): be sure idt2 is positive. Change-Id: Ida5307ba166c63ba6b82eb8f0a59403d88085777 | 31 August 2017, 21:11:04 UTC |
2c05d51 | Tom Quinn | 30 August 2017, 20:38:09 UTC | Make unknown cooling complain a little louder. Change-Id: I6838a4a7f3673df2df9ff665181224021a349fa5 | 30 August 2017, 22:32:31 UTC |
0a42aea | Thomas R. Quinn | 02 August 2017, 04:42:06 UTC | GPU fix: clear registered TreePieces on the DataManager. If the tree is build, but gravity isn't done. Then the registeredTreePieces on the DataManager don't get cleared. Add code to clear them in this case. Note: probably have not caught all the cases. Change-Id: I6264eb80d1ce9a2df1ae8ef9d6628dcfb5a7c0f0 | 23 August 2017, 21:56:49 UTC |
85da1f3 | Tom Quinn | 12 August 2017, 23:39:00 UTC | More documentation for makeOctChildren and getChunks. Change-Id: Ic538a6f75ef846eb29562b16eea5f89e2f723fb0 | 16 August 2017, 23:06:12 UTC |
b870f07 | Tom Quinn | 21 July 2017, 03:12:13 UTC | testcollapse: update gasoline results. The gasoline results in test collapse are now from a version of gasoline based on the Wadsley et al 2017 paper. Change-Id: I653e6c770d6bfac5e5698ce6f3753d751f3ad819 | 21 July 2017, 03:12:13 UTC |
1ba0135 | Tom Quinn | 20 July 2017, 20:27:27 UTC | Merge branch 'ppl_master' into public | 20 July 2017, 20:27:27 UTC |
edefcc1 | Thomas Quinn | 20 July 2017, 20:25:09 UTC | Merge pull request #182 from ibackus/quintic-kernel Quintic kernel and de-macroed kernels | 20 July 2017, 20:25:09 UTC |
0d1f3da | Tom Quinn | 12 July 2017, 15:26:32 UTC | NodeKey: downgrade from "using" to "typedef" for older compilers. Change-Id: Ie5caa6b5e0b221688169f4ce1d4f0ddfa953e84d | 20 July 2017, 13:58:09 UTC |
6d6acf2 | Robel Geda | 19 July 2017, 21:46:41 UTC | Combined Ewald Kernels Change-Id: I9e893d883f9616669113dafe048710162e60e882 | 20 July 2017, 03:37:34 UTC |
3823065 | Thomas Quinn | 18 July 2017, 04:25:50 UTC | Merge pull request #181 from N-BodyShop/dtadjust AGORA feedback | 18 July 2017, 04:25:50 UTC |
7ce0541 | Isaac Backus | 13 July 2017, 01:50:24 UTC | Added configure option to allow user to override default compiler options. (#177) Added the flag --enable-opts="options...". Example usage: Defaults: ./configure No options: ./configure --enable-opts -or- ./configure --enable-opts="" Change optimization and still use lpthread: ./configure --enable-opts="-O2 -lpthread" | 13 July 2017, 01:50:24 UTC |
27dfad5 | Isaac Backus | 11 July 2017, 21:34:11 UTC | Switched SPH kernels from macros to inline functions. The SPH kernels were formerly macros. Different kernels implicitly required different arguments to be defined, but these were not visibile in the function call. This commit addresses that. Tests indicate that this approach is just as fast. I ran the shocktube test on a 4-core laptop with default arguments. The results were exactly identical and the walltime per step were the same. | 11 July 2017, 21:34:11 UTC |
8a81f0e | Isaac Backus | 11 July 2017, 18:29:24 UTC | Merge branch 'public' into quintic-kernel | 11 July 2017, 18:54:37 UTC |
c0089f2 | Tom Quinn | 05 July 2017, 17:48:22 UTC | Particle attribute documentation. Change-Id: I87b0c0300c6e92cc99b8996323aca4b300cfb46e | 05 July 2017, 17:48:22 UTC |
b3420df | Tom Quinn | 01 July 2017, 02:02:02 UTC | Merge branch 'public' into dtadjust | 01 July 2017, 02:02:02 UTC |
aefff5b | Tom Quinn | 16 June 2017, 17:29:35 UTC | Merge branch 'ppl_master' into public | 16 June 2017, 17:29:35 UTC |
a884b9f | Tom Quinn | 16 June 2017, 17:29:00 UTC | More documentation for "dumpframe". | 16 June 2017, 17:29:00 UTC |
5390a74 | Thomas Quinn | 16 June 2017, 15:58:40 UTC | DumpFrameInit(): also adjust frame number for dDumpFrameTime. (#174) This is a fix to issue #172 : dumpframe overwriting at restart. | 16 June 2017, 15:58:40 UTC |
dbac1eb | spencerw | 16 June 2017, 15:55:18 UTC | Frame Transformation (#173) * Allow particles with soft=0 and m=0 * Potential reacts to gravitational force of particles * Code updates according to Tom's comments | 16 June 2017, 15:55:18 UTC |
4fe6bc9 | Isaac Backus | 16 June 2017, 15:37:01 UTC | Refactor SphPressureTerms (#146) Refactor SphPressureTerms * Tab indented SphPressureTerms.h so that the nested #ifdefs can be understood * Added /* compile-time flag */ comments to compile-time macros * Added comments separating the 2 main sections of code * Replaced DRHODTACTIVE macro with an #ifdef * Flattened artificial viscosity into the code calculation * Wrapped the diffusion calls in an #ifdef DIFFUSION These diffusion calls are #defined such that if DIFFUSION is not #defined, they are empty macros. Therefore, wrapping them in an #ifdef DIFFUSION will not affect their execution. * Replaced DIFFUSIONBase() Macro with code execution * Replaced DIFFUSIONThermal(dt) macro with code execution * Replaced DIFFUSIONMetalsBase() macro with code execution * Replaced DIFFUSIONMetals() macro with code execution * Replaced DIFFUSIONMetalsOxygen() macro with code execution * Replaced DIFFUSIONMetalsIron() macro with code execution * Replaced DIFFUSIONMass() macro with code execution * Replaced DIFFUSIONVelocity() macro with code execution * Replaced SETDTNEW_PQ(dt) macro with code execution * Moved the Accp and Accq calculation to after the viscosity calculation. Nothing in the viscosity calculation depends on Accp or Accq, and nothing in those depends on the viscosity. The only substantial change is adding visc = 0.0 for dvdotdr >=0 and adding visc to Acc always. * Shorted Accp and Accq calculations * Moved PdV() calculations to after the viscosity calculation. * Moved the DRHODT calculations to after viscosity. These calculations are independent of the viscosity calc * Moved //DIFFUSIONBase(); code to before particle updating since it is independent of the particle updates * Added comments that detail what the DIFFUSION code snippets depend on and what they update/change * Mapped out macro-function dependencies Note that the 'Children' are just macro 'functions' that depend on a variable which the function updates. Nothing about the order of execution is considered. Being a child doesn't mean you depend on the other function. * Added dependency documentation for DIFFUSIONBase() * Added dependency comments for PRESSUREUpdateTerms() * double checked/updated dependencies and Keep track of which functions also update objects * Moved PRESSUREUpdateTerms() to end of calculation * Merged DIFFUSIONBase into the same #ifdef DIFFUSION as the following DIFFUSION code * Moved DIFFUSIONMetalsBase() to match the new execution order suggested by https://github.com/ibackus/changa_uw/wiki/ * Moved the particle updating of DIFFUSIONThermal(dt) to the end of the code. This depends only on diffu, which required moving double diffu outside the scope of DIFFUSIONThermal. * Removed scope around DIFFUSIONMetals() and renamed diff to diffMetals * Moved particle updating of DIFFUSIONMetals() to the end * Moved particle updating of DIFFUSIONMetalsOxygen() to the end * Moved particle updating of DIFFUSIONMetalsIron() to the end * Moved particle updating of DIFFUSIONMass() to the end * Moved particle updating of DIFFUSIONVelocity() to the end * Removed DIFFUSIONThermaluNoncool() to push its particle updates to the end of the code * Updating comments for compile-time flags. Now they include a description in the file comment * Replaced PRES_PDV(a,b) with inline function presPdv(a,b) * Replaced PRES_ACC(a,b) with inline function presAcc(a,b) * Replaced SWITCHCOMBINE(a,b) with inline function switchCombine(a,b) * Replaced RHO_DIVV(a,b) with inline function rhoDivv(a,b) * Replaced ALPHA and BETA macros with inline functions varAlpha and varBeta * Replaced DIFFUSIONLimitTest() macro with inline function diffusionLimitTest(...) * Replaced MASSDIFFFAC(p) macro with inline function massDiffFac(p) * Flattened the DIFFUSIONThermalCondBase(dt_) macro into the code. That macro was only called once. * Removed dependencies comments and updated other comments to match the new state of this code * Moved all SphPressureTerms prefactor term calculations to Sph.C:PressureSmoothParams::fcnSmooth( ) * Replaced mixed indentation in PressureSmoothParams::fcnSmooth(...) with spaces * Added PressSmoothUpdate struct to store variables used to update particles in PressureSmoothParams::fcnSmooth( ) * Replaced updateParams->pDensity uses with just p->fDensity (they are equal) * Split off particle-specific update properties (such as Accp, Accq or rp, rq) into a separate struc PressSmoothParticle * Renamed updateParams to params * Converted 3D velocity and position quantities to Vector3D * Simplified particle updating if statements * Adjusted SETDTNEW_PQ formatting * Moved DTADJUST>SETDTNEW_PQ to Sph.C from SphPressureTerms.h * Created updateParticle(...) to handle updating the particles p and q. * Moved rp, rq to updateParams * Moved dvdotdr to updateParams * Moved pDensity to updateParams * Moved pPoverRho2, qPoverRho2 to updateParams * Moved visc to updateParams * Moved pPoverRho2f, qPoverRho2f to updateParams * Moved aFac to updateParams * Moved Accq, Accp to updateParams * Moved dx, dy, dz to updateParams * Moved diffu to updateParams * Moved diffuNc to updateParams * Moved diffMetals, diffMetalsOxygen, diffMetalsIron to updateParams * Moved diffMass to updateParams * Moved diffVelocity0, diffVelocity1, diffVelocity2 to updateParams * small cleanups * Suggested optimization for divvi: factor the p->fDensity normalization out of the loop * Basic cleanup: removed unused hav, removed fNorm (used only once), removed unnecessary sqrt * Commented out unimplemented compile-time flags. A large fraction of the compile-time modules present in PressureSmoothParams::fcnSmooth( ) and updateParticles() are not implemented. To make this clear, they have been commented out. In the future when they are implemented and tested they can be uncommented. * Deleted now deprecated SphPressureTerms.h * Stop initializing params with new(). This makes params get initialized on the stack * Stop initializing qParams and pParams with new(). This makes them get initialized on the stack * Replaced PressSmoothParticleStruct.acc with a local variable in updateParticle PressSmoothParticleStruct.acc is only used as a local variable in the updatePartice(), so it has been removed from the struct and placed as a local variable. * Removed unnecessary temporary variables pc and pMass * Moved inline functions, function declarations, and struct definitions specific to PressureSmoothParams::fcnSmooth() to their own header file SphUtils.h There are a large number of functions, incline functions, and structs which are only needed by Sph.C and don't match the standard usage of Sph.h which therefore should be placed in their own header file. Note: at this point there is a macro definition in Sph.h which is required by SphUtils.h, so SphUtils.h depends on Sph.h. * Suggested optimization: move divvi, divvj calculation inside RTFORCE flag. The calculation of divvi, divvj is only required for RTFORCE, specifically to calculate the fDivv_Corrector. There is a for loop over neighbors which is unnecessary when RTFORCE is not compiled. | 16 June 2017, 15:37:01 UTC |
45252c1 | Isaac Backus | 06 June 2017, 18:34:31 UTC | Merge branch 'public' into quintic-kernel | 06 June 2017, 18:34:31 UTC |
6900c93 | Tom Quinn | 11 May 2017, 22:56:28 UTC | Merge "Log COSMO_FLOAT." | 11 May 2017, 22:56:28 UTC |
6b14c06 | Thomas R. Quinn | 11 May 2017, 04:03:36 UTC | Log COSMO_FLOAT. Change-Id: Ie32fa52ebc4d11de159e3111cd59cb5aad039e75 | 11 May 2017, 04:03:36 UTC |
f7f3c46 | spencerw | 03 May 2017, 15:12:40 UTC | Central force fix (#171) * Theta force not set correctly | 03 May 2017, 15:12:40 UTC |
3d4ed18 | Tim Haines | 30 April 2017, 03:54:12 UTC | Use KeyType from keytype.h for NodeKey Change-Id: Id3ef5042f7ba59da69fa95a720d9a5a9c000b0c9 | 03 May 2017, 02:49:49 UTC |
5fefcda | Thomas R. Quinn | 01 May 2017, 16:24:04 UTC | GPU: free device buffers for remote interactions in TranserVarsBack. TransferParticleVarsBack() frees the local particle and moment device buffers. This patch adds freeing the remote data, instead of a separate call at the end of the remote walk. Previously, this was not done reliably, resulting in a CUDA memory error when an old buffer got reused. Change-Id: I12aba021fdf278377324eca16c4eee05f7dd974f | 01 May 2017, 23:47:00 UTC |
bab215e | Tom Quinn | 30 April 2017, 04:13:27 UTC | Merge branch 'ppl_master' into public Conflicts: GravityParticle.h Makefile.dep Makefile.in ParallelGravity.cpp configure | 30 April 2017, 04:13:27 UTC |
42d8edb | Thomas Quinn | 27 April 2017, 04:11:01 UTC | Remove dead GPU code. Change-Id: I35de38cc61d604fcff2b44db84014af903507a0f | 29 April 2017, 05:06:20 UTC |
fd8a405 | Thomas Quinn | 26 April 2017, 19:18:59 UTC | Add separate callback to signal transfering local data to GPU. The local walk can only be started after the local/particle data has been transferred. This could be delayed until after the remote data transfer, so it needs its own callback. Also fixed an Ewald bug that caused hangs with multistepping, and added a check for too many treepieces on a node due to a limitation in hybridAPI. Change-Id: I7939e2e3aecab92d5760e46c1c70ba9fd7cf17ae | 29 April 2017, 05:04:32 UTC |