something wrong in function `kernel_bucketing` in `GPUStreamGeneral.cu`
Closed this issue · 1 comments
Summwer commented
Hello, developer, when I implement the command
python lwe_challenge.py 40 --lwe/alpha 0.035 --threads 2 --gpus 1
with the pnj-bkz blocksize = 89, it returns the error:
File "lwe_challenge_last_pump.py", line 322, in <module>
lwe()
File "lwe_challenge_last_pump.py", line 312, in lwe
stats = run_all(lwe_kernel, list(all_params.values()), # noqa
File "/home/cryptothesis/summer/G6K-GPU-Tensor/g6k/utils/cli.py", line 109, in run_all
res = f(copy.deepcopy(job))
File "lwe_challenge_last_pump.py", line 202, in lwe_kernel
pump_n_jump_bkz_tour(g6k, tracer, blocksize, jump=jump,
File "/home/cryptothesis/summer/G6K-GPU-Tensor/g6k/algorithms/bkz.py", line 118, in pump_n_jump_bkz_tour
pump(g6k, tracer, kappa, beta, f, **pump_params)
File "/home/cryptothesis/summer/G6K-GPU-Tensor/g6k/algorithms/pump.py", line 159, in pump
if not wrapped_sieve(pump):
File "/home/cryptothesis/summer/G6K-GPU-Tensor/g6k/algorithms/pump.py", line 37, in wrapped_sieve
pump.g6k(alg=alg, tracer=pump.tracer)
File "g6k/siever.pyx", line 1263, in g6k.siever.Siever.__call__
with tracer.context("gpu"):
File "g6k/siever.pyx", line 1264, in g6k.siever.Siever.__call__
self.gpu_sieve(reset_stats=reset_stats)
File "g6k/siever.pyx", line 1172, in g6k.siever.Siever.gpu_sieve
sig_on()
cysignals.signals.SignalError: Segmentation fault
Besides, while I debug on this error, I just found out that there's something wrong in function kernel_bucketing
in GPUStreamGeneral.cu
, maybe there's a RAM allocate error or the threads/warps/fragment allocate error. But I cannot find out the concrete error. Hope your help, thanks!
WvanWoerden commented
Fixed in branch python3lwe.