• voir http://wiki.tiker.net/PyCuda/Installation/Linux/Ubuntu
• me suis aussi inspiré de https://wiki.calculquebec.ca/w/Python/fr

tar xzf Downloads/pycuda-2015.1.3.tar.gz
cd pycuda-2015.1.3/
export PATH=/local/apps/cuda-7.5/bin:$PATH
export LD_LIBRARY_PATH=/local/apps/cuda-7.5/lib64:$LD_LIBRARY_PATH
./configure.py --python-exe=/usr/bin/python3 --cuda-root=/local/apps/cuda-7.5 --cudadrv-lib-dir=/usr/lib/x86_64-linux-gnu --boost-inc-dir=/usr/include --boost-lib-dir=/usr/lib --boost-python-libname=boost_python-py34 --boost-thread-libname=boost_thread --no-use-shipped-boost

erreur:

ImportError: No module named 'setuptools'
<Code>
résolu en installant:
<Code>
apt-get install python3-setuptools

ppuis:

ImportError: No module named 'numpy'

# apt-get install python3-scipy
Reading package lists... Done
Building dependency tree
Reading state information... Done
The following extra packages will be installed:
  python3-decorator python3-numpy
  

plus

apt-get install libpython3.4-dev

une fois fait, on va installer dans un virtualenv pour ne pas toucher aux fichiers système:

• on ajoute un alias dans .bashrc pour que python lance python3

• on crée le venv en python3

pyvenv-3.4 env_pycuda

qui crée le dossier env_pycuda, et on l'active

source ~/env_pycuda/bin/activate

dans ce venv, il faut ajouter numpy:

pip install numpy

et enfin:

export PATH=/local/apps/cuda-7.5/bin:$PATH
export LD_LIBRARY_PATH=/local/apps/cuda-7.5/lib64:$LD_LIBRARY_PATH
python setup.py install

ce qui donne:

$ pip list
appdirs (1.4.0)
decorator (4.0.6)
numpy (1.10.4)
pip (1.5.4)
py (1.4.31)
pycuda (2015.1.3)
pytest (2.8.7)
pytools (2016.1)
setuptools (3.3)
six (1.10.0)

on récupère les examples dans le dossier:

./pycuda-2015.1.3/examples/download-examples-from-wiki.py

les exemples sont dans le dossier wiki-examples/

virtualenv env_pycuda_py2
source env_pycuda_py2/bin/activate
pip install numpy
cd pycuda-2015.1.3/
rm siteconf.py

./configure.py --cuda-root=/local/apps/cuda-7.5/ --cudadrv-lib-dir=/usr/lib/x86_64-linux-gnu --boost-inc-dir=/usr/include --boost-lib-dir=/usr/lib --boost-python-libname=boost_python --boost-thread-libname=boost_thread --no-use-shipped-boost

python setup.py install

pip install .

et on vérifie:

pip list
appdirs (1.4.0)
argparse (1.2.1)
decorator (4.0.6)
numpy (1.10.4)
pip (1.5.4)
py (1.4.31)
pycuda (2015.1.3)
pytest (2.8.7)
pytools (2016.1)
setuptools (2.2)
six (1.10.0)
wsgiref (0.1.2)

export PATH=/local/apps/cuda-7.5/bin:$PATH
export LD_LIBRARY_PATH=/local/apps/cuda-7.5/lib64:$LD_LIBRARY_PATH

les exemples sont dans /local/admin1/NVIDIA_CUDA-7.5_Samples/

• pour les recopier: sh cuda-install-samples-7.5.sh ~
• Exemple avec nbody:

./nbody -bench


GPU Device 0: "Quadro K620" with compute capability 5.0

> Compute 5.0 CUDA device: [Quadro K620]
3072 bodies, total time for 10 iterations: 4.661 ms
= 20.246 billion interactions per second
= 404.929 single-precision GFLOP/s at 20 flops per interaction

dans le dossier ~/NVIDIA_CUDA-7.5_Samples/1_Utilities/

./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "Quadro K620"
  CUDA Driver Version / Runtime Version          7.5 / 7.5
  CUDA Capability Major/Minor version number:    5.0
  Total amount of global memory:                 2047 MBytes (2146762752 bytes)
  ( 3) Multiprocessors, (128) CUDA Cores/MP:     384 CUDA Cores
  GPU Max Clock rate:                            1124 MHz (1.12 GHz)
  Memory Clock rate:                             900 Mhz
  Memory Bus Width:                              128-bit
  L2 Cache Size:                                 2097152 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096)
  Maximum Layered 1D Texture Size, (num) layers  1D=(16384), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(16384, 16384), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 1 copy engine(s)
  Run time limit on kernels:                     Yes
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Disabled
  Device supports Unified Addressing (UVA):      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 1 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 7.5, CUDA Runtime Version = 7.5, NumDevs = 1, Device0 = Quadro K620
Result = PASS

[CUDA Bandwidth Test] - Starting...
Running on...

 Device 0: Quadro K620
 Quick Mode

 Host to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)        Bandwidth(MB/s)
   33554432                     6417.3

 Device to Host Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)        Bandwidth(MB/s)
   33554432                     6471.0

 Device to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)        Bandwidth(MB/s)
   33554432                     26349.3

Result = PASS

NOTE: The CUDA Samples are not meant for performance measurements. Results may vary when GPU Boost is enabled.