a94ab94 | Julien Vitay | 06 April 2022, 12:22:08 UTC | Release 4.7.1.1 | 06 April 2022, 12:22:08 UTC |
65bdd10 | Helge Dinkelbach | 06 April 2022, 05:54:13 UTC | Merge branch 'master' of https://ai.informatik.tu-chemnitz.de/gogs/ANNarchy/ANNarchy | 06 April 2022, 05:54:13 UTC |
2af53a3 | Helge Dinkelbach | 06 April 2022, 05:53:39 UTC | Updated changelog. | 06 April 2022, 05:53:39 UTC |
1744754 | Julien Vitay | 05 April 2022, 14:23:22 UTC | Update README.md Zenodo badge for the latest release | 05 April 2022, 14:23:22 UTC |
1fda7fc | Helge Dinkelbach | 03 April 2022, 06:04:54 UTC | The device configuration for CUDA in annarchy.json was ignored. | 03 April 2022, 06:04:54 UTC |
470d592 | Helge Dinkelbach | 01 April 2022, 08:15:57 UTC | CPP-side connector methods should return a boolean (see commit 77f3a3f for details). | 01 April 2022, 08:15:57 UTC |
76fcbe5 | Helge Dinkelbach | 30 March 2022, 07:38:39 UTC | Tentative (CUDA): we test if the memory is available before allocation of the connectivity / variable. This is a tentative commit, as we do not adapt all available sparse matrix formats right now. | 30 March 2022, 07:38:39 UTC |
c002880 | Helge Dinkelbach | 29 March 2022, 14:16:55 UTC | Modification of the auto-tuning on CPUs: use CSR instead of LIL. | 29 March 2022, 14:16:55 UTC |
2067052 | Helge Dinkelbach | 18 March 2022, 13:30:34 UTC | Modification of commit 1cae984: the earlier build of LILConnectivity is only performed if required by the automatic format selection | 18 March 2022, 13:30:34 UTC |
88bc8f4 | Julien Vitay | 16 March 2022, 09:34:22 UTC | Monitor: visualization methods accessed the wrong array. | 16 March 2022, 09:34:22 UTC |
1cae984 | Helge Dinkelbach | 15 March 2022, 14:06:02 UTC | Updated heuristic for automatic format selection on GPUs. | 16 March 2022, 09:00:25 UTC |
fe86349 | Julien Vitay | 14 March 2022, 16:18:36 UTC | Fixed start and stop times of the Monitor. | 14 March 2022, 16:18:36 UTC |
802d765 | Julien Vitay | 14 March 2022, 12:25:31 UTC | TimedPoissonPopulation: rates accepts single values for the population. | 14 March 2022, 12:25:31 UTC |
e1378c3 | Julien Vitay | 14 March 2022, 11:16:37 UTC | Merge branch 'master' of https://ai.informatik.tu-chemnitz.de/gogs/annarchy/annarchy | 14 March 2022, 11:16:37 UTC |
68eb277 | Julien Vitay | 14 March 2022, 11:15:53 UTC | HomogeneousCorrelatedSpikeTrains: corr can now be a single value, even when using a schedule. | 14 March 2022, 11:15:53 UTC |
0e5dda8 | Helge Dinkelbach | 14 March 2022, 10:44:20 UTC | Modified default compiler flags. - cython modules build during setup.py use only -O3 - c++ (model compile) default flags are now: -O3 -ffast-math -fno-finite-math-only -march=native (following Brian2 default flags) further: - suppressed file copies of cudaCheck module (suppressed in commit 589b93f) | 14 March 2022, 10:44:20 UTC |
25be38d | Helge Dinkelbach | 14 March 2022, 10:38:59 UTC | Bugfix (performance-related): for CUDA devices wrong default flags in the Makefile were used, which reduce the achieved performance. | 14 March 2022, 10:39:50 UTC |
18c4baa | Helge Dinkelbach | 14 March 2022, 10:23:55 UTC | Bugfix (CUDA, dense): missing definition for vector variables (semi-global synapse attributes). | 14 March 2022, 10:23:55 UTC |
c80956d | Julien Vitay | 02 February 2022, 13:59:23 UTC | Removed warnings aout cpdef attributes in cython extensions. | 02 February 2022, 13:59:23 UTC |
c0bf65e | Julien Vitay | 02 February 2022, 12:35:58 UTC | Release 4.7.1. | 02 February 2022, 12:35:58 UTC |
3c765d3 | Julien Vitay | 02 February 2022, 11:24:02 UTC | parallel_run(): fixed max_processes for single networks and aadapted the example for osx. | 02 February 2022, 11:24:02 UTC |
0f4637e | Helge Dinkelbach | 01 February 2022, 11:40:15 UTC | The intrinsics are limited to x86 / 64 CPUs. | 01 February 2022, 11:40:15 UTC |
2848e41 | Torsten Fietzek | 31 January 2022, 08:37:19 UTC | Merged in extensions_copy (pull request #5) Extensions: Adding the copy mechanism for convolution/pooling Compiler: make absolute change of working direktory | 31 January 2022, 08:37:19 UTC |
5e61b60 | Helge Dinkelbach | 31 January 2022, 08:12:18 UTC | Added missing del/sib templates for: CSR, splitted, pre_to_post | 31 January 2022, 08:12:18 UTC |
41d4e64 | Helge Dinkelbach | 24 January 2022, 11:05:24 UTC | updated changelog. | 24 January 2022, 11:05:24 UTC |
8607aff | Helge Dinkelbach | 24 January 2022, 10:59:27 UTC | Merge branch 'master' into develop | 24 January 2022, 10:59:27 UTC |
a301d96 | Helge Dinkelbach | 24 January 2022, 10:55:13 UTC | Improved code generation and solved memory leak issue. + added RNGs to size_in_bytes + detected and resolved memory leak in the record of spike events. | 24 January 2022, 10:55:13 UTC |
bb61289 | Helge Dinkelbach | 21 January 2022, 08:29:53 UTC | directly converting iterator to index should be done using std::distance (I think). | 21 January 2022, 08:29:53 UTC |
eb3d4c3 | Helge Dinkelbach | 20 January 2022, 10:17:11 UTC | Refactor: provide thread ID as function argument instead of calling omp_get_thread_num() in every step. | 20 January 2022, 10:17:11 UTC |
16ab4d6 | Julien Vitay | 19 January 2022, 09:07:18 UTC | Release 4.7.0.2. | 19 January 2022, 09:07:18 UTC |
72a325e | Helge Dinkelbach | 18 January 2022, 16:52:49 UTC | Merge branch 'master' into develop | 18 January 2022, 16:52:49 UTC |
2e2977d | Helge Dinkelbach | 18 January 2022, 15:49:32 UTC | Bugfix (openMP): the spike generation of SpikeSourceArray was not implemented correctly. | 18 January 2022, 15:49:32 UTC |
fc5a909 | Helge Dinkelbach | 18 January 2022, 15:48:04 UTC | Bugfix (CUDA): code generated for delays and spiking was not correct. | 18 January 2022, 15:48:04 UTC |
72db6f8 | Helge Dinkelbach | 18 January 2022, 15:13:11 UTC | code generation bugfix (commit 92a5ed8): dense and default spiking synapses (attention: still tentative feature) | 18 January 2022, 15:13:53 UTC |
5366d08 | Helge Dinkelbach | 18 January 2022, 10:38:45 UTC | minor refactoring in openMP generator. | 18 January 2022, 10:38:45 UTC |
2cf5263 | Helge Dinkelbach | 18 January 2022, 10:29:26 UTC | code generation bugfix (commit 92a5ed8): wrong include for specific projections. | 18 January 2022, 10:29:26 UTC |
0d48dd9 | Helge Dinkelbach | 17 January 2022, 12:51:17 UTC | changed parallel processing pattern: post_event, CSR | 17 January 2022, 12:51:17 UTC |
b3ce798 | Helge Dinkelbach | 17 January 2022, 11:54:38 UTC | Automatic format selection: LIL if structural plasticity is enabled. | 17 January 2022, 11:54:38 UTC |
60cee66 | Helge Dinkelbach | 17 January 2022, 11:01:26 UTC | added missing code templates - spiking (CSR, pre_to_post, openMP): delete data containers, missing link of post-event/synaptic updated. | 17 January 2022, 11:01:26 UTC |
92a5ed8 | Helge Dinkelbach | 17 January 2022, 10:34:00 UTC | Restructured include of SpMV template classes. | 17 January 2022, 10:34:00 UTC |
9269202 | Helge Dinkelbach | 16 January 2022, 11:17:00 UTC | Refactoring: - removed extern definitions of dt/t in "ANNarchy.h" as they should be only modified through the get/set methods. - removed function prototype in "ANNarchy.cpp" (single_step(), single_step(int, int)) by reordering cpp code. | 16 January 2022, 11:50:09 UTC |
0c1386e | Helge Dinkelbach | 11 January 2022, 12:41:26 UTC | fixed typos. | 11 January 2022, 12:41:26 UTC |
ab4598e | Helge Dinkelbach | 11 January 2022, 12:34:34 UTC | Added AVX to CSR implementation for default rate-coded synapses and single weights. | 11 January 2022, 12:34:34 UTC |
c47cb42 | Helge Dinkelbach | 11 January 2022, 11:27:10 UTC | Added warning: no code generation for plasticity for COO/Hybrid formats. | 11 January 2022, 11:27:10 UTC |
12eb01b | Helge Dinkelbach | 11 January 2022, 10:34:46 UTC | Merge branch 'master' into develop | 11 January 2022, 10:34:46 UTC |
8521886 | Helge Dinkelbach | 11 January 2022, 10:29:51 UTC | Small adjustment of commit 4ae3cc9: the compiler flags are only shown if verbose (--verbose/-v) enabled. | 11 January 2022, 10:29:51 UTC |
58a7aef | Julien Vitay | 10 January 2022, 11:07:01 UTC | Added check on the parameters of the random distributions. | 10 January 2022, 11:07:01 UTC |
0f3949a | Julien Vitay | 07 January 2022, 12:33:57 UTC | Reverted the cpdefs in Connector.pyx | 07 January 2022, 12:33:57 UTC |
4ae3cc9 | Julien Vitay | 07 January 2022, 12:15:49 UTC | MacOS: now supports openMP with gcc | 07 January 2022, 12:15:49 UTC |
a7498bb | Julien Vitay | 06 January 2022, 18:08:40 UTC | Merge branch 'master' of https://ai.informatik.tu-chemnitz.de/gogs/annarchy/annarchy | 06 January 2022, 18:08:40 UTC |
0b4ad83 | Julien Vitay | 06 January 2022, 18:07:02 UTC | MacOS: attempt at using openMP with gcc | 06 January 2022, 18:08:29 UTC |
6e5dfb4 | Helge Dinkelbach | 06 January 2022, 16:58:31 UTC | The <sched.h> import is only valid for linux systems. | 06 January 2022, 16:58:31 UTC |
5fd9f11 | Julien Vitay | 06 January 2022, 14:46:01 UTC | Merge branch 'master' of https://ai.informatik.tu-chemnitz.de/gogs/annarchy/annarchy | 06 January 2022, 14:46:01 UTC |
ab279af | Julien Vitay | 06 January 2022, 14:45:23 UTC | Adapted for mac m1. | 06 January 2022, 14:45:23 UTC |
2388bc6 | Helge Dinkelbach | 05 January 2022, 10:49:48 UTC | Replaced some occurances of hard-coded types by template parameters. | 05 January 2022, 10:49:48 UTC |
b627114 | Helge Dinkelbach | 16 December 2021, 15:16:29 UTC | Refactored CUDA code generation and added new tests to cover global/semiglobal synaptic variables. The different SpMV formats differ in two ways: first the number of arguments/participating variables and the algorithms to compute with. To ease the implementation in the future, the variables required for the connectivity representation and the indices now both located in the corresponding template file (instead of locating them in _configure_template_ids() for instance). | 16 December 2021, 15:28:50 UTC |
bf64e4f | Julien Vitay | 15 December 2021, 16:16:18 UTC | Script does not need to be executable. | 15 December 2021, 16:16:18 UTC |
9fa1982 | Helge Dinkelbach | 14 December 2021, 10:08:47 UTC | Unified structure of the conn_templates['synapse_update'] entry as done in commit f9de07e. | 14 December 2021, 10:08:47 UTC |
3b1f7e9 | Helge Dinkelbach | 13 December 2021, 13:40:30 UTC | minor bugfixes in non-default matrix formats. | 13 December 2021, 13:40:30 UTC |
87a3a7d | Helge Dinkelbach | 10 December 2021, 16:58:29 UTC | continued work on commit f9de07e Changed dense matrices to column_major for spiking models. Otherwise, the the event-driven transmission will perform not efficient. An open question (as for CSRC) remains wether a post_to_pre or pre_to_post order should be favored. | 10 December 2021, 16:58:29 UTC |
ccffe3f | Helge Dinkelbach | 10 December 2021, 08:17:46 UTC | Merge branch 'master' into develop resolved conflict: version strings | 10 December 2021, 08:17:46 UTC |
faa1653 | Helge Dinkelbach | 09 December 2021, 13:49:29 UTC | Bugfix/Enhancement (commit 281e272): constants (in this case the uniform delays) are stored as 0-dimensional ndarrays. To detect them, ndim check should be favoured instead of size. | 09 December 2021, 13:49:29 UTC |
6f75ad5 | Helge Dinkelbach | 09 December 2021, 11:10:06 UTC | continued work on commit f9de07e Dense matrices and synaptic plasticity (rate-coded) should be fully functional now. | 09 December 2021, 11:10:06 UTC |
41ee1ae | Helge Dinkelbach | 09 December 2021, 11:07:29 UTC | Bugfix (commit 145fc8d): the CUDA device code was not compiled with debug symbols. | 09 December 2021, 11:08:13 UTC |
488f97c | Helge Dinkelbach | 09 December 2021, 11:01:30 UTC | Enhancement (commit a65cd0f, f3e4d9c): disable the introduced mechanism for debug purposes. The introduced mechanism disturbs the usage of gdb/cuda-gdb as the ANNarchyCoreXX.so library is always deleted/moved. Therefore I decided to disable this mechanism if ANNarchy is in debug mode (either by --debug on command line or by compile() argument). | 09 December 2021, 11:01:30 UTC |
7ebdb9e | Helge Dinkelbach | 09 December 2021, 10:50:05 UTC | Modified FNP/FP pattern to sort the pre-synaptic ranks in ascending order after RNG. | 09 December 2021, 10:50:05 UTC |
f9de07e | Helge Dinkelbach | 03 December 2021, 08:59:39 UTC | Added synaptic plasticity (rate-coded) for dense formats on GPU. Experimental: dense matrix format for event-driven transmission (psp-only) | 03 December 2021, 08:59:39 UTC |
20db1a5 | Helge Dinkelbach | 24 November 2021, 09:05:54 UTC | Merge branch 'tuning' into develop | 24 November 2021, 09:05:54 UTC |
a9ef1af | Helge Dinkelbach | 24 November 2021, 08:40:50 UTC | Tentative: added unrolled versions for 2x2 or 3x3 kernel of the BSR matrix format (single thread) | 24 November 2021, 08:40:50 UTC |
285526d | Helge Dinkelbach | 22 November 2021, 10:06:48 UTC | Bugfix: code generation error in dense matrix code. | 22 November 2021, 10:07:15 UTC |
6350748 | Helge Dinkelbach | 22 November 2021, 10:03:08 UTC | Added dense matrix heuristic. | 22 November 2021, 10:03:08 UTC |
a46d305 | Julien Vitay | 19 November 2021, 08:20:39 UTC | Release 4.7.0.1 | 19 November 2021, 08:20:39 UTC |
f6c0468 | Helge Dinkelbach | 18 November 2021, 08:50:19 UTC | Merge branch 'develop' into tuning | 18 November 2021, 08:50:19 UTC |
462bdb2 | Helge Dinkelbach | 18 November 2021, 08:44:52 UTC | Tried to optimize the performance of the dense matrix implementation on GPUs. Instead of 1-dimensional indexing and only global memory, we use a 2d kernel and a fixed size shared memory. | 18 November 2021, 08:44:52 UTC |
002a2d1 | Helge Dinkelbach | 18 November 2021, 08:40:14 UTC | Added dense matrices to GPU heuristics. | 18 November 2021, 08:40:14 UTC |
e247c04 | Helge Dinkelbach | 15 November 2021, 14:58:42 UTC | ELLPACK-R: clearer distinction between dense and sparse dimensions. | 16 November 2021, 08:27:24 UTC |
c7f9b42 | Helge Dinkelbach | 15 November 2021, 14:59:39 UTC | Merge branch 'develop' into tuning | 15 November 2021, 15:52:56 UTC |
84fd1bd | Helge Dinkelbach | 15 November 2021, 14:58:42 UTC | ELLPACK-R: need to distinguish between dense and sparse dimensions. | 15 November 2021, 14:58:42 UTC |
68f9a7a | Helge Dinkelbach | 15 November 2021, 13:59:03 UTC | played around with some heuristics. | 15 November 2021, 13:59:03 UTC |
9c8061d | Helge Dinkelbach | 12 November 2021, 10:36:47 UTC | Merge branch 'master' into develop | 12 November 2021, 10:36:47 UTC |
5d62a1e | Helge Dinkelbach | 11 November 2021, 13:46:12 UTC | BoldModel: allow strings as argument for *inputs* if only one variable should be considered. | 11 November 2021, 13:46:12 UTC |
943cc1c | Helge Dinkelbach | 11 November 2021, 10:49:28 UTC | Added ANNarchy version tag to some generated files. | 11 November 2021, 10:49:28 UTC |
aedfece | Helge Dinkelbach | 11 November 2021, 10:41:10 UTC | Merge branch 'master' into develop | 11 November 2021, 10:41:10 UTC |
796c473 | Helge Dinkelbach | 11 November 2021, 09:36:16 UTC | Updated changelog. | 11 November 2021, 09:36:16 UTC |
3b22830 | Helge Dinkelbach | 11 November 2021, 08:51:17 UTC | Added openMP support for the BOLD monitor extension. | 11 November 2021, 09:30:08 UTC |
17015bb | Helge Dinkelbach | 09 November 2021, 08:39:15 UTC | bugfix on BSR matrix for matrices with blocks of empty rows. | 09 November 2021, 08:39:15 UTC |
0fbd8bf | Helge Dinkelbach | 02 November 2021, 12:59:42 UTC | Commit 77f3a3f10 didn't considered openMP. | 02 November 2021, 12:59:42 UTC |
2b4c2b5 | Helge Dinkelbach | 01 November 2021, 16:42:14 UTC | Tentative: "blocked sparse row" (BSR) format for GPUs. This implementation is for now intended solely as rate-coded benchmark (e. g. learning is not working). | 01 November 2021, 16:42:14 UTC |
615ce4d | Helge Dinkelbach | 25 October 2021, 08:59:04 UTC | Tentative: moved index definitions (old position from Projection.[..]Generator._configure_template_ids() ) to the specific template files. | 25 October 2021, 08:59:04 UTC |
d4f6970 | Helge Dinkelbach | 22 October 2021, 07:05:54 UTC | Corrected size_in_bytes (cuda, host side) computation for coo, csr, ell, hyb. | 22 October 2021, 07:05:54 UTC |
145fc8d | Helge Dinkelbach | 21 October 2021, 08:58:16 UTC | Restructured CUDA code. Moved all device specific functions to ANNarchyDevice.cu. Added new field to template files which can be used to define common functions for this specific template. | 21 October 2021, 09:01:15 UTC |
7be3f79 | Helge Dinkelbach | 21 October 2021, 08:04:41 UTC | Merge branch 'master' into develop | 21 October 2021, 08:04:41 UTC |
88975af | Helge Dinkelbach | 21 October 2021, 08:00:38 UTC | addition to commit 281e272 | 21 October 2021, 08:00:38 UTC |
bb35b3c | Helge Dinkelbach | 21 October 2021, 07:37:11 UTC | Improved performance of Coordinate / in particular Hybrid format on GPUs. The current implementation uses atomicAdd to update the target array. To improve the performance of that we want to use shared memory. This requires a guarantee that the threads of one block will not access array elements outside the block. To ensure this, we group the rows in fixed segments. This is done by a helper array built on top and will not effect the ordering of the coordinate format itself. | 21 October 2021, 07:37:11 UTC |
77f3a3f | Helge Dinkelbach | 19 October 2021, 08:32:00 UTC | Added memory checks for some sparse matrix formats (GPU as well as CPU). If this checks fail, the init_from_lil method returns false and Projection.initialized will be false. TODO: the CPP-side connectors (currently disabled) need to be adapted. | 19 October 2021, 08:32:00 UTC |
ff30d7a | Helge Dinkelbach | 14 October 2021, 10:20:33 UTC | Added ELLPACK format on GPUs (required as part of the hybrid ELL/COO format). Code generation fixes for hybrid format (missing ST template parameter). | 14 October 2021, 10:37:08 UTC |
e68c82b | Helge Dinkelbach | 12 October 2021, 17:03:55 UTC | Added missing deletion of GPU variables in ProjStruct::clear(). Added SIMD implementation for CSR and non-constant weights. Bugfix on dense implementation (spmv code and dendrite_size()). Added CPP side implementation for fixed_probability pattern on dense formats. | 12 October 2021, 17:03:55 UTC |
06e6697 | Helge Dinkelbach | 11 October 2021, 16:26:17 UTC | Merge branch 'develop' of https://github.com/ANNarchy/ANNarchy into develop | 11 October 2021, 16:26:17 UTC |
4ef956c | Helge Dinkelbach | 11 October 2021, 16:22:42 UTC | Added additional debug messages and sanity checks (e. g. check available memory before allocations on GPUs). Added CPP-side implementation of fixed_number_pre and fixed_probability pattern for ELLPACK-R. | 11 October 2021, 16:22:42 UTC |
ddc4580 | Julien Vitay | 11 October 2021, 14:34:06 UTC | Updated travis. | 11 October 2021, 14:34:06 UTC |