Hi there, today I upgraded af_python via pip. I uninstalled the devel version previously installed with setup.py install. Then installed with pip. Afterwards I tried to do simple_tests.py from the devel package. I got 'Intel MKL FATAL ERROR: Cannot load mkl_intel_thread.dll.' My af-python based application executes without any result, no error message, no sign of doing calculations, no nothing. I also reinstalled the arrayfire library from the website, without any success. af.info() on the other hand works and gives the version and device message. What's going on here? Is there a way to explicitly point af_python to the arrayfire path?

Hello, I'm sorry I know this issue has been around for a while but I couldn't solve it based on the posted solutions. I'm trying to install it for Mac OSx with cpu and opencl, without cuda. So I write this in the terminal: export DYLD_LIBRARY_PATH=/usr/local/lib:$DYLD_LIBRARY_PATH as this is where my libaf*.dylib files are located. Still, when trying to import the module I get this message:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/ivangusachenko/anaconda/lib/python3.6/site-packages/arrayfire/__init__.py", line 56, in <module>
    from .library    import *
  File "/Users/ivangusachenko/anaconda/lib/python3.6/site-packages/arrayfire/library.py", line 600, in <module>
    backend = _clibrary()
  File "/Users/ivangusachenko/anaconda/lib/python3.6/site-packages/arrayfire/library.py", line 576, in __init__
    raise RuntimeError("Could not load any ArrayFire libraries.\n" + more_info_str)
RuntimeError: Could not load any ArrayFire libraries.
Please look at https://github.com/arrayfire/arrayfire-python/wiki for more information.

hello, we are currently working on a finite difference code using arrayfire. First benchmarks showed that for small systems the C/C++ versions seems to be significantly faster than the Python version. Therefor at least the core routines should be implemented in C/C++. On the other hand the final interface should be available in Python. We typically use Cython for that purpose and we already archive to pass arrays from C to Python and call some low-level routines using cython wrappers.

Unfortunately we are not yet able to combine the cython interface with the ctypes wrappers. Finally we would like to be able to create an array from arrayfire-python. Somehow determine the address of the C/C++ array and pass this to our cython wrappers. We tried several methods to determine the address of the underlying array, and we are quite sure that one of these versions should give the correct address, but we still get the following error:

terminate called after throwing an instance of 'af::exception'
  what():  ArrayFire Exception (Unknown error:208):
In function const ArrayInfo& getInfo(af_array, bool, bool)
In file src/api/c/array.cpp:30
Input Array not created on current device

In function void af::print(const char*, const af::array&)
In file src/api/cpp/util.cpp:21
Aborted

It looks like that there are two "instances" of arrayfire which do not work together!? Does anybody have experience with arrayfire and cython? any ideas how to debug such a problem?

thanks for any suggestion Florian

I am using a function which uses af.broadcast inside it.

When I use the program in windows PC, I do not see any slowdown of the processing using CPU. However, when I use my linux PC which has both CPU and AMD RX 480 GPU, I see the code slow down drastically.

In Issue#123 @pavanky mentioned that to use af.eval and af.sync when using broadcast function.

Can any one given me a simple explanation of this if possible.

Thanks

EDIT(June 30 2017): The problem was I did not update my arrayfire to newest version on linux.

I built ArrayFire from source, and installed it in ${HOME}/arrayfire I am using anaconda3 on Linux Ubuntu 16.04 (64-bit). I installed arrayfire python bindings from pip : pip install arrayfire.

I have followed the instructions mentioned in the arrayfire-python/wiki export LD_LIBRARY_PATH=${HOME}/arrayfire/lib:$LD_LIBRARY_PATH

When I run python -m arrayfire.tests, I get

Traceback (most recent call last):
  File "/home/gitansh95/anaconda3/lib/python3.6/runpy.py", line 183, in _run_module_as_main
    mod_name, mod_spec, code = _get_module_details(mod_name, _Error)
  File "/home/gitansh95/anaconda3/lib/python3.6/runpy.py", line 109, in _get_module_details
    __import__(pkg_name)
  File "/home/gitansh95/anaconda3/lib/python3.6/site-packages/arrayfire/__init__.py", line 56, in <module>
    from .library    import *
  File "/home/gitansh95/anaconda3/lib/python3.6/site-packages/arrayfire/library.py", line 601, in <module>
    backend = _clibrary()
  File "/home/gitansh95/anaconda3/lib/python3.6/site-packages/arrayfire/library.py", line 577, in __init__
    raise RuntimeError("Could not load any ArrayFire libraries.\n" + more_info_str)
RuntimeError: Could not load any ArrayFire libraries.
Please look at https://github.com/arrayfire/arrayfire-python/wiki for more information.

These are the libraries in ${HOME}/arrayfire/lib:

[email protected]:~/arrayfire/lib$ ls -l
total 113172
lrwxrwxrwx 1 gitansh95 gitansh95       13 Aug  8 23:41 libafcpu.so -> libafcpu.so.3
lrwxrwxrwx 1 gitansh95 gitansh95       17 Aug  8 23:41 libafcpu.so.3 -> libafcpu.so.3.6.0
-rw-r--r-- 1 gitansh95 gitansh95 37576264 Aug  8 19:37 libafcpu.so.3.6.0
lrwxrwxrwx 1 gitansh95 gitansh95       16 Aug  8 23:41 libafopencl.so -> libafopencl.so.3
lrwxrwxrwx 1 gitansh95 gitansh95       20 Aug  8 23:41 libafopencl.so.3 -> libafopencl.so.3.6.0
-rw-r--r-- 1 gitansh95 gitansh95 67891256 Aug  8 19:19 libafopencl.so.3.6.0
lrwxrwxrwx 1 gitansh95 gitansh95       10 Aug  8 23:41 libaf.so -> libaf.so.3
lrwxrwxrwx 1 gitansh95 gitansh95       14 Aug  8 23:41 libaf.so.3 -> libaf.so.3.6.0
-rw-r--r-- 1 gitansh95 gitansh95  3576680 Aug  8 19:45 libaf.so.3.6.0
lrwxrwxrwx 1 gitansh95 gitansh95       13 Aug  8 23:41 libforge.so -> libforge.so.1
lrwxrwxrwx 1 gitansh95 gitansh95       17 Aug  8 23:41 libforge.so.1 -> libforge.so.1.0.2
-rw-r--r-- 1 gitansh95 gitansh95  6834384 Aug  8 18:24 libforge.so.1.0.2

I also tried export AF_PATH=${HOME}/arrayfire/lib Did not work. What am I missing? Please help.

• cast numpy scalars to af scalars, so that things like np.sin(scalar)+af_array=np_array don't happen. It's confusing and people are bound to use math. or cmath. functions. Pyopencl does the casting silently. funny: af_array+np.sin(scalar)=af_array
• change array layout to row-major as in numpy and C/C++ so that arrays copied from the host and arrays created on the device match (or is there a reason for column-major arrays?))
• make the api a little more like numpy (I know, afnumpy is there) e.g. implement more array methods (.abs(), .min(), .max(), .sum(), .imag, .real, .conj(), .transpose() and so on) and rename the array attributes (.elements() to .size, .dims() to .shape, af.moddims() to .reshape()...)

data=np.random.random(size=(100,100)) af.np_to_af_array(data)

gets error: af.np_to_af_array(data) File "C:\Python27\lib\site-packages\arrayfire\interop.py", line 53, in np_to_af_array res = Array(in_ptr, shape, in_dtype) File "C:\Python27\lib\site-packages\arrayfire\array.py", line 428, in init raise TypeError("src is an object of unsupported class") TypeError: src is an object of unsupported class

My Environment: Python27 64 bit Windows 7 64 AF version: 3.3.2 or sys.version_info(major=2, minor=7, micro=10, releaselevel='final', serial=0) CUDA kit 7.5

More background: I just installed the package so likely a setup issue on my side... The examples work for me, both cpu & cuda.

Any ideas?

Would it be possible to support interoperability with numpy?

import numpy as np
import arrayfire as af

A = np.zeros((2,3), dtype=np.complex64)
Ad = af.array(A) # <-- load data from a numpy array
af.display(Ad)
print(Ad.elements(), Ad.type(), Ad.dims(), Ad.numdims())

Currently:

Using opencl backend
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-3-c8c7947108da> in <module>()
      3 
      4 A = np.zeros((2,3), dtype=np.complex64)
----> 5 Ad = af.array(A) # <-- load data from a numpy array
      6 af.display(Ad)
      7 print(Ad.elements(), Ad.type(), Ad.dims(), Ad.numdims())

/Users/cmey/Code/GitHub/arrayfire-python/arrayfire/array.pyc in __init__(self, src, dims)
    133                 type_char = tmp.typecode
    134             else:
--> 135                 raise TypeError("src is an object of unsupported class")
    136 
    137             elements = 1

TypeError: src is an object of unsupported class

I'm a bit confused about the use of the interpolation function arrayfire.signal.approx1(signal, pos0). The documentation states that it accepts 2 arguments: the signal value, and the interpolation points. How can I specify the array which defines the points where the signal has been defined? I'll take an example using the numpy function np.interp:

x = np.linspace(0, 2*np.pi, 100)
y = np.sin(x)

x_interp = np.linspace(0, 2*np.pi, 50)
y_interp = np.interp(x_interp, x, y) # generates the values at the interpolated points

How would I approach this using the arrayfire functions?

I've installed the ArrayFire libraries, pip-installed the python wrapper, and followed the directions on the wiki, but I haven't been able to run any code yet.

When I run af.info(), I get the following:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.7/site-packages/arrayfire/device.py", line 32, in info
    safe_call(backend.get().af_info())
  File "/usr/local/lib/python2.7/site-packages/arrayfire/util.py", line 75, in safe_call
    raise RuntimeError(to_str(err_str), af_error)
RuntimeError: ('', 501)

I'm on an AWS g2.2xlarge, with the NVIDIA/Amazon Linux image.

Any debugging advice?

I've implemented an interp1d and interp2d function similar to the one found in scipy. Rather than having the user manually transform the input points from [0, N-1] when using approx1 and approx2, this function takes care of the transformation by taking the input data points in addition to the interpolation points.

However, there is one issue: I'm using sum to get a scalar value:

dx   = sum(x_input[1, 0, 0, 0] - x_input[0, 0, 0, 0])
pos0 = (x_interpolated - sum(x_input[0, 0, 0, 0]))/dx

Do you recommend using something else?

Inside a new thread, Arrayfire uses the default backend rather than the backend set by the parent thread:

>>> import arrayfire as af
>>> af.get_available_backends()
('cpu', 'cuda', 'opencl')
>>> af.get_active_backend()
'cuda'
>>> af.set_backend('opencl')
>>> af.get_active_backend()
'opencl'
>>> from threading import Thread
>>> Thread(target=lambda: print(af.get_active_backend())).start()
cuda

Is this expected behavior and/or documented anywhere? I would think that the active backend should be process local rather than thread local.

If it matters, here's my version info: OS: Ubuntu 20.04 arrayfire: 3.9.0 (commit 955152b6) arrayfire-python: 3.8.10

cuda: 11.4.0-1 ocl-icd-libopencl1: 2.2.11-1 intel-mkl: 2020.0.166-1

Using this arrayfire release (https://arrayfire.s3.amazonaws.com/3.8.0/ArrayFire-v3.8.0_Linux_x86_64.sh) on the opencl backend with the python api I get the following error.

import arrayfire as af
af.set_backend('opencl')
random_signal = af.random.randn(100)
test = af.signal.fft(random_signal)

Error

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
<ipython-input-7-a744aa37e707> in <module>
----> 1 test = af.signal.fft(random_signal)

~/miniconda3/envs/af/lib/python3.7/site-packages/arrayfire/signal.py in fft(signal, dim0, scale)
    177 
    178     output = Array()
--> 179     safe_call(backend.get().af_fft(c_pointer(output.arr), signal.arr, c_double_t(scale), c_dim_t(dim0)))
    180     return output
    181 

~/miniconda3/envs/af/lib/python3.7/site-packages/arrayfire/util.py in safe_call(af_error)
     77         err_len = c_dim_t(0)
     78         backend.get().af_get_last_error(c_pointer(err_str), c_pointer(err_len))
---> 79         raise RuntimeError(to_str(err_str))
     80 
     81 def get_version():

RuntimeError: In function void opencl::fft_inplace(opencl::Array<T>&, int, bool) [with T = cl_float2]
In file src/backend/opencl/fft.cpp:

I'm running Ubuntu 20.04 with Nvidia OpenCL Drivers. I have arrayfire debugging enabled and got the following output when the error occurred.

                        BUILD LOG
************************************************
<kernel>:19:10: error: expected expression
(float2)(};
         ^
<kernel>:891:1: error: expected '}'
^
<kernel>:18:32: note: to match this '{'
__constant float2 twiddles[] = {
                               ^
<kernel>:890:2: error: expected ';' after top level declarator
}
 ^
 ;

************************************************
FFTGeneratedStockhamAction::compileKernels failed

This line has a comment # Iterate in reverse order of preference. However, it loads them with some random order, because the object to iterate is a set, which doesn't keep the order. In my Linux computer, when I run import arrayfire as af, the function will be called and it loads "" first, and due to err!=0, it directly returns and not loads anything.

By the way, I simply change {"cpu", "opencl", "cuda", ""} to ["cuda"] and I can import arrayfire with no error.

• Remove custom test runner tool
• Add pytest runner (there is no need to install it locally from requirements because it is used as setup test requirement) These tests are not checking the correctness of the executed method and they are nothing more than smoke tests. So, they are put in the appropriate tests subfolder.

The tests can be run with the python setup.py test command.

Minimal coverage was set to fit the current coverage value -- 70% (was 72.3%, actual 71.86% after removing some prints)

• Remove the check for enum import. Motivation: check for enum's import error is unnecessary on modern python versions because its built-in library is available with a basic python build version. It is pointless to cover very-very-very situational python builds that could not contain enum in it.
• Remove the C++ style formatting for constants declaration. Motivation: FYI, @syurkevi : https://www.python.org/dev/peps/pep-0008/#other-recommendations. Especially example before link that is under "More than one space around an assignment (or other) operator to align it with another" and highlighted "never use more than one space" :middle_finger: :sunglasses:
• Remove wrap for Enum constants with _Enum_Type function. Motivation: There is no point to use such a method like _Enum_Type that returns the arg as it is passed into it.
• to_str is moved to the "library" module to avoid circular import with the "util" module. Note: This change does not break the public API because it is still declared in __init__.

Hi, I am using MACOS, I have 1. Download and install binaries 2. pip install arrayfire 3. export DYLD_LIBRARY_PATH=/opt/arrayfire/lib:$DYLD_LIBRARY_PATH

It can be used with command line

But when I use the same conda environment and python VeM.py which is a script containing import arrayfire, it shows

Traceback (most recent call last):
  File "/Users/zetao/Documents/code/edu/python_code/first_try/VEM_deblur.py", line 3, in <module>
    import arrayfire as af 
  File "/Users/zetao/opt/anaconda3/lib/python3.7/site-packages/arrayfire/__init__.py", line 56, in <module>
    from .library    import *
  File "/Users/zetao/opt/anaconda3/lib/python3.7/site-packages/arrayfire/library.py", line 643, in <module>
    backend = _clibrary()
  File "/Users/zetao/opt/anaconda3/lib/python3.7/site-packages/arrayfire/library.py", line 619, in __init__
    raise RuntimeError("Could not load any ArrayFire libraries.\n" + more_info_str)
RuntimeError: Could not load any ArrayFire libraries.

Can anyone kindly explain why?