Troubles with arrayfire python

[Юрий Сидоров]

Hi! I've got some troubles

When I launch

import arrayfire as af

af.set_backend('cuda')

af.info()

ArrayFire v3.7.2 (CUDA, 64-bit Linux, build 218dd2c)
Platform: CUDA Runtime 10.0, Driver: 450.66
[0] GeForce GTX TITAN, 6083 MB, CUDA Compute 3.5

but when I try to run tests I've got error

python -m arrayfire.tests
ERROR:root:Traceback (most recent call last):
 File "/home/linuxboy/anaconda3/lib/python3.7/site-packages/arrayfire/tests/simple/_util.py", line 32, in run
   test(verbose)
 File "/home/linuxboy/anaconda3/lib/python3.7/site-packages/arrayfire/tests/simple/algorithm.py", line 20, in simple_algorithm
   af.eval(k)
 File "/home/linuxboy/anaconda3/lib/python3.7/site-packages/arrayfire/device.py", line 218, in eval
   __eval(*args)
 File "/home/linuxboy/anaconda3/lib/python3.7/site-packages/arrayfire/device.py", line 168, in __eval
   safe_call(backend.get().af_eval(args[0].arr))
 File "/home/linuxboy/anaconda3/lib/python3.7/site-packages/arrayfire/util.py", line 79, in safe_call
   raise RuntimeError(to_str(err_str))
RuntimeError: In function cuda::Module common::compileModule(const string&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, const std::vector<std::b
asic_string<char> >&, bool)
In file src/backend/cuda/compile_module.cpp:

Simple            arith: FAILED
arrayfire.Array()
Type: float

I run on Kubuntu 18.04.

If i run arrayfire examples all ok!

What should I do?

[Pradeep Garigipati]

Hi,

We have recently pushed some fixes to master to fix tests in the python wrapper project, but I haven't seen the error you posted during the debugging session though.

Are you using the latest master branch of arrayfire-python to run the tests ?

If examples are running fine, it could be something related to the python environment, I am not certain though. The line at which it fails is a very basic statement of creating an array of random numbers. Try copying this line to a working example and run the modified example and check what happens. Thay may help narrow down the problem further.

Regards,

Pradeep.

--
You received this message because you are subscribed to the Google Groups "ArrayFire Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to arrayfire-use...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/arrayfire-users/b471a628-781d-4e5c-9d48-b46ada2dd88en%40googlegroups.com.

--

Pradeep

https://www.linkedin.com/in/pradeepgarigipati

[PercentBoat4164]

Hey, I am encountering a similar issue. I am running Ubuntu 20.04 using python 3.8. Here are my results of af.info() with the different backends.

af.set_backend("cpu")

af.info()

ArrayFire v3.7.2 (CPU, 64-bit Linux, build 218dd2c)

[0] AMD: AMD Ryzen 5 3600 6-Core Processor

af.set_backend("cuda") or af.set_backend("opencl")

af.info():

ArrayFire v3.7.2 (CUDA, 64-bit Linux, build 218dd2c)

Platform: CUDA Runtime 10.0, Driver: 450.66

[0] GeForce RTX 2060 SUPER, 7981 MB, CUDA Compute 7.5

My problem arises when I try to run this function. It is part of a program that is suggested for testing by Arrayfire.

def calc_pi_device(samples):
    # Simple, array based API
    # Generate uniformly distributed random numbers
    x = af.randu(samples)
    y = af.randu(samples)
    # Supports Just In Time Compilation
    # The following line generates a single kernel
    within_unit_circle = (x * x + y * y) < 1
    # Intuitive function names
    return 4 * af.count(within_unit_circle) / samples

The error only arises when this function is run with the cuda, or opencl backends. They are as follows:

cuda:

Traceback (most recent call last):

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 19, in <module>

    print(calc_pi_device(100000000))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 14, in calc_pi_device

    return 4 * af.count(within_unit_circle) / samples

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 383, in count

    return _reduce_all(a, backend.get().af_count_all)

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 26, in _reduce_all

    safe_call(c_func(c_pointer(real), c_pointer(imag), a.arr))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/util.py", line 79, in safe_call

    raise RuntimeError(to_str(err_str))

RuntimeError: In function cuda::Module common::compileModule(const string&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, bool)

In file src/backend/cuda/compile_module.cpp:

opencl:

Traceback (most recent call last):

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 19, in <module>

    print(calc_pi_device(100000000))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 14, in calc_pi_device

    return 4 * af.count(within_unit_circle) / samples

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 383, in count

    return _reduce_all(a, backend.get().af_count_all)

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 26, in _reduce_all

    safe_call(c_func(c_pointer(real), c_pointer(imag), a.arr))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/util.py", line 79, in safe_call

    raise RuntimeError(to_str(err_str))

RuntimeError: In function cl::Program opencl::buildProgram(const std::vector<std::basic_string<char>, std::allocator<std::basic_string<char> > >&, const std::vector<std::basic_string<char>, std::allocator<std::basic_string<char> > >&)

In file src/backend/opencl/compile_module.cpp:

For a little more information here is another scenario with an error, and without an error. This section was all run with the opencl backend.

If I execute

y = af.randu(100)
af.display(y)

I get

y

[

I thought this might be due to the just in time compilation. So I had it multiply all the results by 1.

y = af.randu(100) * 1
af.display(y)

I got the following error though.

y

[Traceback (most recent call last):

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 19, in <module>

    af.display(y)

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/array.py", line 1426, in display

    safe_call(backend.get().af_print_array_gen(name.encode('utf-8'),

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/util.py", line 79, in safe_call

    raise RuntimeError(to_str(err_str))

RuntimeError: In function void print(const char*, af_array, int, std::ostream&, bool) [with T = float; af_array = void*; std::ostream = std::basic_ostream<char>]

In file src/api/c/print.cpp:

If anyone has a suggestion please let me know. Thank you for your time.

[Pradeep Garigipati]

Please run the code with environment variables AF_PRINT_ERRORS, AF_TRACE set to 1 and all respectively, and share the full log.

You can do that by running "export AF_PRINT_ERRORS=1 && export AF_TRACE=all" from command line before running the python script.

To view this discussion on the web visit https://groups.google.com/d/msgid/arrayfire-users/bf81996a-1358-4fb2-9b0f-6257fbea4ea8n%40googlegroups.com.

--

Pradeep

https://www.linkedin.com/in/pradeepgarigipati

[PercentBoat4164]

I ran the following with the environment variables set as you requested

import arrayfire as af

def calc_pi_device(samples):
    # Simple, array based API
    # Generate uniformly distributed random numbers
    x = af.randu(samples)
    y = af.randu(samples)
    # Supports Just In Time Compilation
    # The following line generates a single kernel
    within_unit_circle = (x * x + y * y) < 1
    # Intuitive function names
    return 4 * af.count(within_unit_circle) / samples

af.set_backend('cuda')
print(calc_pi_device(100))

and received this as output

In function cuda::Module common::compileModule(const string&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, bool)

In file src/backend/cuda/compile_module.cpp:

Traceback (most recent call last):

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 18, in <module>

    print(calc_pi_device(100))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/GPUtesting.py", line 14, in calc_pi_device

    return 4 * af.count(within_unit_circle) / samples

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 383, in count

    return _reduce_all(a, backend.get().af_count_all)

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/algorithm.py", line 26, in _reduce_all

    safe_call(c_func(c_pointer(real), c_pointer(imag), a.arr))

  File "/home/thaddeus/Documents/Programming/Python/VideoStreaming/venv/lib/python3.8/site-packages/arrayfire/util.py", line 79, in safe_call

    raise RuntimeError(to_str(err_str))

RuntimeError: In function cuda::Module common::compileModule(const string&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, const std::vector<std::basic_string<char> >&, bool)

In file src/backend/cuda/compile_module.cpp:

I hope this helps. Thank you for taking the time to help me out.

[Pradeep Garigipati]

The error message still seems minimal, not sure what that is the case. Usually, with trace and print environment variables set, it should print out a whole lot more to the console.

Let me try to reproduce this on my end. Right now, I have neither the Ryzen nor RTX card. I shall try to access a similar compute version card.

Sorry about the inconvenience.

Pradeep.

To view this discussion on the web visit https://groups.google.com/d/msgid/arrayfire-users/bb3e1303-7120-4a27-aafa-b1474c9bd2cbn%40googlegroups.com.

--

Pradeep

https://www.linkedin.com/in/pradeepgarigipati

[PercentBoat4164]

Alright. If there is any other information I can give don't hesitate to ask.

Thanks

[Barbara F]

I have similar problem Running latest AF libraries and latest python wrapper. Running on Linux, python 3.8. I have created new environment and installed only ArrayFire.

LD_LIBRARY_PATH is set to /path/to/arrayfire/lib64

running test:

python -m arrayfire.tests

Results:

Simple        algorithm: FAILED

ERROR:root:Traceback (most recent call last):

  File "/home/xxx/yyy/anaconda3/envs/aft/lib/python3.8/site-packages/arrayfire/tests/simple/_util.py", line 32, in run

    test(verbose)

  File "/home/xxx/yyy/anaconda3/envs/aft/lib/python3.8/site-packages/arrayfire/tests/simple/algorithm.py", line 20, in simple_algorithm

    af.eval(k)

  File "/home/xxx/yyy/anaconda3/envs/aft/lib/python3.8/site-packages/arrayfire/device.py", line 218, in eval

    __eval(*args)

  File "/home/xxx/yyy/anaconda3/envs/aft/lib/python3.8/site-packages/arrayfire/device.py", line 168, in __eval

    safe_call(backend.get().af_eval(args[0].arr))

  File "/home/xxx/yyy/anaconda3/envs/aft/lib/python3.8/site-packages/arrayfire/util.py", line 79, in safe_call

    raise RuntimeError(to_str(err_str))

RuntimeError: In function cl::Program opencl::buildProgram(const std::vector<std::basic_string<char>, std::allocator<std::basic_string<char> > >&, const std::vector<std::basic_string<char>, std::allocator<std::basic_string<char> > >&)

In file src/backend/opencl/compile_module.cpp:

Segmentation fault (core dumped)