https://github.com/vitay/ANNarchy

sort by:
Revision Author Date Message Commit Date
f438f03 Runge-Kutta 4 (rk4) is now available. 14 December 2022, 10:47:50 UTC
226063b Merge branch 'develop' 14 December 2022, 10:29:22 UTC
089d93a Call to warning() was not imported. 14 December 2022, 10:25:55 UTC
8474439 updated changelog. 14 December 2022, 10:11:47 UTC
b27ef35 Added a flag to explicitly disbale the mechanism introduced in a65cd0f (already did for debug in 488f97c). 14 December 2022, 09:52:20 UTC
a8f78d1 Bugfix in Convolution: the filters could not be changed after compile(). 14 December 2022, 08:39:38 UTC
c2393d9 In commit 77f3a3f we changed the implementation of the 'initialized' flag from only indicating the call of _connect to store if the c++ connectivity methods were correctly called. SpecificProjections override the _connect() method and consequently the 'initilized' flag was not initialized (i.e. None). 14 December 2022, 08:29:31 UTC
cd574fa Bugfix in code generation related to post-event and CSR on GPUs. 13 December 2022, 16:20:45 UTC
04c73ac removed unnecessary import introduced in commit e6bcde9. 13 December 2022, 11:20:53 UTC
e0b4daa Bugfix in code generation related to splitted CSRC/CSRC_T. 13 December 2022, 09:53:55 UTC
ed60b24 removed debug print 13 December 2022, 09:40:11 UTC
6613b97 Added sanity rule to prevent usage of CSRC_T and GPUs for non-default synapses. Improved readability of code (commit b66fdf5). 13 December 2022, 09:12:44 UTC
e3f2d82 Updated sanity rules to prevent the usage of continuous transmission on spiking synapses with pre-to-post matrix ordering. The continuous transmission kernel is derived from the rate-coded psp. Therefore, the inversed matrix view of the CSRC_T would introduce many misaligned memory accesses. In case of openMP one would also probably need atomic operations. 13 December 2022, 08:44:08 UTC
b66fdf5 Sanity rules for sparse matrix formats and rate-coded models should consider the hardware. 13 December 2022, 08:15:46 UTC
e16b13b Tentative: added implementation for spiking models using uniform delays and CSR on GPUs. 12 December 2022, 16:18:12 UTC
45588a6 minor change on unittest. 12 December 2022, 15:37:55 UTC
c5588c7 nonuniform delays are not supported yet on GPUs. 12 December 2022, 15:36:27 UTC
7bc8feb Bugfix in single item update using CSR. 12 December 2022, 11:19:40 UTC
8e507a1 ELLPACK-R: fixed single item update for CPU/GPU. 12 December 2022, 11:12:44 UTC
5d096dd Removed debug print left from commit 35b0de4. 12 December 2022, 11:07:08 UTC
82de4de Single item getter for dense matrix format was missing. 12 December 2022, 09:05:58 UTC
d3792c9 Implemented get/set for single matrix values for BSR (CPU/GPU) 12 December 2022, 08:49:55 UTC
d788821 Bugfix (commit 5b24fb5): the change performed in this commit did not adjust the code templates for min/max/mean. 11 December 2022, 17:31:59 UTC
064f68d Fixed code generation bugs related to non-uniform delays. 11 December 2022, 12:38:23 UTC
41d842d Addendum (commit 518b3a1): the custom connectivity classes did not considered the storage_format argument. 11 December 2022, 11:24:45 UTC
7f85445 Temporarily disabled pooling on GPU devices as the code templates are currently broken. The feature is anyways not useable right now, as the convolution is still missing on GPUs. 11 December 2022, 10:04:27 UTC
7396014 missing include statement for SpecificProjections on GPUs. 10 December 2022, 15:55:11 UTC
518b3a1 restructured unittests for custom connectivity and rate-coded psp. - no delay, uniform delay and non-uniform delay are now three classes - non-uniform delays are not supported on GPUs and therefore not tested 10 December 2022, 15:52:47 UTC
456351f Merge branch 'master' into develop 08 December 2022, 15:16:41 UTC
2a58a18 using consistant types. 08 December 2022, 10:36:16 UTC
8bc9663 Bug with midpoint. 05 December 2022, 16:25:45 UTC
8df6eb7 The parser made replacement errors when targets had overlapping names (e.g. sum(exc2) and sum(exc) used in that order). 05 December 2022, 16:14:04 UTC
73c3174 Version 4.7.1.6 25 November 2022, 15:56:42 UTC
439de48 updated changelog. 23 November 2022, 16:00:10 UTC
320af51 updated changelog. 23 November 2022, 15:48:43 UTC
ba6497d Bugfix: period_offset argument was ignored when creating object copy. 23 November 2022, 15:44:23 UTC
ff8b4f0 Tentative: modified test for save_and_load of projection parameters. The original tests fail, because the save function access the init attribute to gain the stored value. But the init value is only updated by access before compile(). The modifications which should be tested are performed after compile which let the test fail. Therefore I added another field in the test to store the initial states before compile(). 21 November 2022, 08:58:16 UTC
42fd859 Addendum (f2190e0): GPUs does not support CSRC_T, Dense_T for spiking models which are selected by automatic format selection on CPUs. Therefore, we fall back to CSRC on GPUs. 11 November 2022, 12:23:00 UTC
c5b7ac1 Continued work on fcba8bd. - the offsets need to be considered by get/set methods. 10 November 2022, 18:40:03 UTC
625954d Bugfix: related to Dense matrix representation and PopulationViews. If the matrix contains empty rows, the post_ranks need to be considered when constructing the mask which was not done. 10 November 2022, 17:20:41 UTC
aaa6223 Refactor (commit fcba8bd), same as commit d855701 but for openMP. Further, disabled the optimization for now until the feature is fully tested. 09 November 2022, 19:47:00 UTC
f92f460 Addendum (0491ef1): dense matrices. 09 November 2022, 19:30:04 UTC
d855701 Refactor (fcba8bd): instead of splitting the dictionary fields, I added a separate template file for this specialized implementation. 09 November 2022, 19:24:52 UTC
4052f6a updated changelog. 09 November 2022, 07:42:00 UTC
528eabb Bugfix (988b5e5): the psp argument provided to PoolingProjection() was not copied. 09 November 2022, 07:30:27 UTC
f1064f6 Suppressed xrange(). The usage of xrange is a Python2.x related optimization. In Python3.x the xrange() implementation was used to implement range(). 09 November 2022, 07:16:27 UTC
7e03c58 Changed ctype for boolean values on GPUs to char. For local/semiglobal variables we create a vector<ctype> instance. In the modern STL container implementation std::vector<bool> is a specialized container which (compiler dependent) may specialized as bitset container (see: https://en.cppreference.com/w/cpp/container/vector_bool). However, for GPUs this is not suitable and therefore transformation of the boolean value on GPUs into a char datatype. 09 November 2022, 06:57:21 UTC
f2190e0 Extended heuristic format selection. Spiking networks with high density should switch to Dense, pre-to-post representation. 08 November 2022, 07:18:55 UTC
099a162 ConnectorMethods functions should use None as default argument storage_format/storage_order. Using None we can clearly separate between explicit choices made by users and default arguments. 08 November 2022, 07:11:27 UTC
078ff35 updated changelog. 08 November 2022, 06:51:25 UTC
fe5b937 PopulationView expects a numpy.array since commit (5f9d58b). 08 November 2022, 06:47:40 UTC
35b0de4 Tentative: added new debug symbol (_DEBUG_ACCESSOR) to trace accessor function calls (test-wise implemented for LIL) 08 November 2022, 06:46:41 UTC
1bd613a Merge branch 'master' into develop 08 November 2022, 06:06:23 UTC
050ac35 Addendum (a7123ed): suppressed the idx value of IndividualSynapse as it is not needed anymore. 08 November 2022, 05:57:54 UTC
a7123ed Bugfix related to the used index in IndividualSynapse When using calls like projection[post_rank][pre_rank] this leads to the creation of IndividualSynapse() objects. This object used the lil-idx (i.e. the position of pre_rank in the pre_ranks array) which was fine for earlier versions of ANNarchy which only used lil. For csrc, ellpack etc. I changed this accessors to expect the pre_rank not the index position. 08 November 2022, 05:48:36 UTC
e295a75 Merge branch 'master' into develop Notice: The conflict was on the configure_template_ids update tree. In earlier versions, we stored there the indices. Later I stored them together with the other template code. 03 November 2022, 11:01:14 UTC
407aae1 Bugfix in code generation: for sliced LIL, spiking and uniform delays the tid was forgotten when accessing the delay container. 03 November 2022, 10:50:14 UTC
0491ef1 Added inner-/outer-loop distinction for event-driven transmission. Our preliminary results and literature suggest that one need to distinguish between a parallelization on the inner- or outer-loop of the event-driven transmission kernel. For splitted matrices, one will always use outer-loop. 02 November 2022, 07:36:24 UTC
655dc5a Added partitioned implementation for CSRC + post_to_pre view. 02 November 2022, 06:51:39 UTC
eb2602b Bugfix (844c816): storage_order argument was missing in connect_from_matrix_market declaration 31 October 2022, 09:03:24 UTC
8d43915 Applied the changes made in commit fcba8bd also for dense matrix and post_to_pre ordering. 27 October 2022, 16:34:30 UTC
27d5d47 Merge branch 'master' into develop 27 October 2022, 08:36:57 UTC
d8b09de Bugfix: a simulate() after a Monitor.stop() call crashed. Monitor.stop() removes the C++ instance and set the corresponding position in the recorders array to nullptr. However, the simulation code need to check if a recorder was disabled in this way otherwise this leads to a segmentation fault. 27 October 2022, 08:29:18 UTC
c8569a6 CSRC_T and learning appears to work correctly, therefore I remove this sanity check. 27 October 2022, 06:11:45 UTC
85f53dd Profiling: added measurement of the post_event kernel. 26 October 2022, 16:38:13 UTC
4560eec Marked the profiling using PAPI-API as deprecated. 26 October 2022, 15:50:00 UTC
630fd77 The headers contained in thirdparty sub-folder need to be copied during setup. 26 October 2022, 15:35:25 UTC
5e7984f Extended the parser to compute the number of evaluated operations for equations. 26 October 2022, 15:32:32 UTC
d3f69a7 Tentative: for spiking neurons as target of convolution/pooling the psp-variable must be adjusted. 25 October 2022, 07:21:59 UTC
4db0fc1 Tentative: removed rate-coded neuron check for pooling/convolution to allow easier development on the ann_to_snn_conversion extension. 25 October 2022, 06:39:19 UTC
988b5e5 Pooling: Bugfix (fc64001) and allow to overwrite default psp (already allowed on convolution). 25 October 2022, 06:38:20 UTC
1a135ae Convolution: Bugfix (fc64001) and added same sanity check as in Pooling. 25 October 2022, 06:35:08 UTC
126575d Removed doubled CoupledEquations.c_code definition. 24 October 2022, 14:39:09 UTC
04ab383 Merge branch 'master' into develop 24 October 2022, 14:37:58 UTC
b1d79bc Extra tab in midpoint. 23 October 2022, 13:35:58 UTC
602cb5c removed trailing spaces. 20 October 2022, 08:05:23 UTC
d45ce63 Bugfix (844c816): storage_order argument was missing in connect_with_func declaration. 19 October 2022, 11:33:43 UTC
fc64001 Extended sparse matrix format selection heuristics to select the storage_order for rate-coded and spiking models too. 18 October 2022, 17:41:12 UTC
5128dcd Added "annarchy_json" argument to parallel_run(). Normally, the user can adjust the compiler configuration in advance of calling this function. However, there is the option, that the compile is performed during parallel_run() and therefore this flag must be adjustable. 17 October 2022, 12:35:48 UTC
4ab147c Bugfix (6030556): row_nr_off needs to be computed in each iteration by the threads separatly, not incrementally add up. 15 October 2022, 07:51:46 UTC
2a14971 BSRMatrix: more fine-grained debug. The print-out of the tiled mask is now performed only if _DEBUG_CONN is enabled. Otherwise the normal _DEBUG is too noisy. 15 October 2022, 07:46:08 UTC
6030556 Tentative: openMP for block sparse row This commit adds a simple openMP parallelization for the BSR format. The parallelization is applied simply across the block_row. 15 October 2022, 06:49:57 UTC
a20665d cleaned up sliced ELLPACK template header. Added a hyper-parameter printout (as for BSR). 15 October 2022, 04:32:54 UTC
8ee024e Fixed sanity check for sliced ELLPACK - single thread. 15 October 2022, 04:32:14 UTC
ecf3f28 Merge branch 'develop' of https://ai.informatik.tu-chemnitz.de/gogs/annarchy/ANNarchy into develop 13 October 2022, 12:46:39 UTC
d18f9ee minor code change (TimedArray, CUDA) The NVCC compiler complained about the unnecessary first arguments in the for-loop statement. 13 October 2022, 08:38:54 UTC
5be9857 In verbose mode ANNarchy notifies the user which files were changed and had forced a recompile. 13 October 2022, 08:35:59 UTC
38963e1 Implemented two functions related to _DEBUG_CONN flag. 05 October 2022, 17:38:21 UTC
e31b53b updated changelog. 05 October 2022, 09:46:54 UTC
dc48293 Allowed thread placement using CPU_SET also for single-thread applications. 05 October 2022, 09:24:50 UTC
0eaf52a tentative feature: try to implement thread placement rules for parallel_run 05 October 2022, 07:02:09 UTC
df7d4a6 Updated changelog. 01 October 2022, 08:56:03 UTC
fcba8bd Added specialized implementation for dense matrices to reduce the memory requirement in context of PopulationViews 30 September 2022, 13:07:54 UTC
a4aab28 The mask type (default char) of dense matrices is now a template parameter. 29 September 2022, 05:47:59 UTC
f09ad6e Implemented two missing set methods for dense matrix format. 28 September 2022, 18:37:51 UTC
ec8f4ed Fixed class names in template header library. 28 September 2022, 13:39:07 UTC
4dd0aaa Bugfix: wrong index variable for synapse runinning index in the pre_spike event code (CSRC_T matrix). 28 September 2022, 13:29:15 UTC
ff9c934 QuickFix (commit 5f9d58b): Monitor crashes when accessing Population.ranks In commit 5f9d58b I replaced the Population.ranks generation from Python list to numpy.arange(). The monitors use the ranks and provide them to C++ core. This commit fixes this for now, by converting the ranks locally in the PopulationView object to a Python list. 26 September 2022, 16:46:49 UTC
2893199 Added new unittest related to record of PopulationViews. Next to slicing (e.g. Population[:,2]), one can unify two different populations. There are two cases: a) PopView + PopView and PopView + IndividualNeuron. The latter was not tested yet. 26 September 2022, 16:42:53 UTC
fa17a70 This commit relaxes the assert() in BSRMatrix::init_matrix_from_lil(). The previous assert condition on the theoretical number of allocated blocks was a bit strong. Now we check for allocated blocks if a problem arise. Additionally, we check the allocation of dense blocks if we have enough space in RAM. 25 September 2022, 06:14:04 UTC
back to top