Modify paper_2022 scripts for troubleshooting

brian2cuda/tools/benchmarking/paper_2022/run_all.sh

@@ -2,18 +2,18 @@
 # Any command line arguments are passed to `bash run_benchmark_suite.sh`
 
 SCRIPTS_NO_PROFILING="\
- run_benchmarks_fig5_figS1.py \
- run_benchmarks_fig6AB.py \
- run_benchmarks_fig6C.py \
- run_benchmarks_fig8.py"
+ run_benchmarks_fig5_figS1.py"
+ #run_benchmarks_fig6AB.py \
+ #run_benchmarks_fig6C.py \
+ #run_benchmarks_fig8.py"
 
 # XXX: fig9 simulations were executed on a different GPU (RTX2080)
 #run_benchmarks_fig9.py"
 
-SCRIPTS_WITH_PROFILING="\
- run_benchmarks_fig7A.py \
- run_benchmarks_fig7B.py \
- run_benchmarks_fig7CD.py"
+SCRIPTS_WITH_PROFILING=""
+ #run_benchmarks_fig7A.py \
+ #run_benchmarks_fig7B.py \
+ #run_benchmarks_fig7CD.py"
 
 cp ../run_benchmark_suite.py run_benchmark_suite.bak.py
 

brian2cuda/tools/benchmarking/paper_2022/run_benchmarks_fig5_figS1.py

@@ -133,28 +133,28 @@ def print_flushed(string, slack=True, new_reply=False, format_code=True):
 configurations = [
  # C++
  CPPStandaloneConfiguration,
- CPPStandaloneConfigurationSinglePrecision,
- CPPStandaloneConfigurationOpenMPMaxThreads,
- CPPStandaloneConfigurationOpenMPMaxThreadsSinglePrecision,
+ #CPPStandaloneConfigurationSinglePrecision,
+ #CPPStandaloneConfigurationOpenMPMaxThreads,
+ #CPPStandaloneConfigurationOpenMPMaxThreadsSinglePrecision,
 
- # Brian2CUDA
- ## number of partitions equal number of SMs on GPU (108 on A100 GPU)
- DynamicConfigCreator('CUDA standalone (max blocks, atomics)'),
- DynamicConfigCreator('CUDA standalone (single precision, max blocks, atomics)',
- prefs={'core.default_float_dtype': float32}),
+ ## Brian2CUDA
+ ### number of partitions equal number of SMs on GPU (108 on A100 GPU)
+ #DynamicConfigCreator('CUDA standalone (max blocks, atomics)'),
+ #DynamicConfigCreator('CUDA standalone (single precision, max blocks, atomics)',
+ # prefs={'core.default_float_dtype': float32}),
 
  ## 1 partition
  DynamicConfigCreator('CUDA standalone (1 block, atomics)',
  prefs={'devices.cuda_standalone.parallel_blocks': 1}),
- DynamicConfigCreator('CUDA standalone (single precision, 1 block, atomics)',
- prefs={'core.default_float_dtype': float32,
- 'devices.cuda_standalone.parallel_blocks': 1}),
+ #DynamicConfigCreator('CUDA standalone (single precision, 1 block, atomics)',
+ # prefs={'core.default_float_dtype': float32,
+ # 'devices.cuda_standalone.parallel_blocks': 1}),
 
  # Brian2GeNN
  GeNNConfiguration,
- GeNNConfigurationSinglePrecision,
- GeNNConfigurationSpanTypePre,
- GeNNConfigurationSinglePrecisionSpanTypePre,
+ #GeNNConfigurationSinglePrecision,
+ #GeNNConfigurationSpanTypePre,
+ #GeNNConfigurationSinglePrecisionSpanTypePre,
 ]
 
 # The `benchmark` classes are defined in brian2cuda/tests/features/speed.py. The
@@ -163,15 +163,15 @@ def print_flushed(string, slack=True, new_reply=False, format_code=True):
 speed_tests = [# benchmark n_slice
 
  # HH benchmark without synapses
- (COBAHHUncoupled, slice(None)),
+ (COBAHHUncoupled, slice(0, 1)),
  # HH benchmark with synapses
- (COBAHHPseudocoupled1000, slice(None)),
- # LIF benchmark with homogeneous delays
- (BrunelHakimHomogDelays, slice(None)),
- # STDP benchmark
- (STDPCUDARandomConnectivityHomogeneousDelays, slice(None)),
- # Mushroom body benchmark
- (MushroomBody, slice(None)),
+ #(COBAHHPseudocoupled1000, slice(None)),
+ ## LIF benchmark with homogeneous delays
+ #(BrunelHakimHomogDelays, slice(None)),
+ ## STDP benchmark
+ #(STDPCUDARandomConnectivityHomogeneousDelays, slice(None)),
+ ## Mushroom body benchmark
+ #(MushroomBody, slice(None)),
 ]
 
 slack_thread = None