Call push kernel only with max bundle size threads

brian-team · Feb 15, 2022 · e7ad71b · e7ad71b
1 parent dba8942
commit e7ad71b
Showing 1 changed file with 6 additions and 2 deletions.
diff --git a/brian2cuda/templates/synapses_push_spikes.cu b/brian2cuda/templates/synapses_push_spikes.cu
@@ -1054,12 +1054,16 @@ void _run_{{codeobj_name}}()
  */
  needed_shared_memory = (2 * {{owner.name}}_max_num_unique_delays + 1) * sizeof(int);
  assert (needed_shared_memory <= max_shared_mem_size);
+ // We don't need more then max(num_synapses) threads per block.
+ num_threads = {{owner.name}}_max_size;
  {% else %}{# bundle_mode #}
  needed_shared_memory = 0;
+ // We don't need more then max(num_delays) threads per block.
+ num_threads = {{owner.name}}_bundle_size_max;
  {% endif %}{# not bundle_mode #}
 
- // We don't need more then max(num_synapses) threads per block.
- num_threads = {{owner.name}}_max_size;
+ // TODO: warp size multiple for num_threads?
+
  if (num_threads > max_threads_per_block)
  {
  num_threads = max_threads_per_block;