Error when using af.af.lapack.norm
Linuxboy2007 opened this issue · comments
When I try to compute norm of matrix (vector) I got error:
ArrayFire v3.6.4 (CUDA, 64-bit Linux, build 1b8030c)
Platform: CUDA Toolkit 10.0, Driver: 418.87.00
[0] GeForce GTX TITAN, 6081 MB, CUDA Compute 3.5
(100, 100)
Traceback (most recent call last):
File "/home/linuxboy/PycharmProjects/test/main.py", line 8, in <module>
nrma = af.lapack.norm(dA,norm_type = af.NORM.MATRIX_INF)
File "/home/linuxboy/PycharmProjects/test/venv/lib/python3.7/site-packages/arrayfire/lapack.py", line 341, in norm
c_double_t(p), c_double_t(q)))
File "/home/linuxboy/PycharmProjects/test/venv/lib/python3.7/site-packages/arrayfire/util.py", line 79, in safe_call
raise RuntimeError(to_str(err_str))
RuntimeError: In function cuda::Kernel cuda::buildKernel(int, const string&, const string&, const std::vector<std::basic_string<char> >&, bool) In file src/backend/cuda/nvrtc/cache.cpp:
Cuda examples working, other af functions, for example, det, lu, matmul working good. Example bench_blas.py working.
I've got Ubuntu 19.04, ArrayFire 3.6.4, CudaToolkit 10.1.
import arrayfire as af
af.set_backend('cuda')
af.info()
dA = af.randn(100,100,1,1,af.Dtype.f64)
print(dA.shape)
nrma = af.lapack.norm(dA,af.NORM.MATRIX_INF)
print(nrma)
I solved the problem - I downgraded version of ArrayFire to 3.6.2.