developpement:activites:integration:cuda4jessie

Différences

Ci-dessous, les différences entre deux révisions de la page.

--- developpement:activites:integration:cuda4jessie [2015/11/17 13:31] – equemene
+++ developpement:activites:integration:cuda4jessie [2016/04/21 17:10] (Version actuelle) – equemene
@@ Ligne 1: / Ligne 1: @@
 ====== Rétroportage de pilote Nvidia & Cuda sur une Debian Jessie ======
+<note warning>Attention ! Cassé au 21 avril 2016 pour la partie PyCUDA et PyOpenCL</note>
 ===== Rétroportage de CUDA 6.5 sous Jessie =====
@@ Ligne 7: / Ligne 8: @@
-Voici la méthode permettant d'installer à partir d'un rétro-portage les paquets Nvidia ainsi que tout l'environnement associé. Au 28 mai 2015, il n'est pas possible de récupérer la version 346.59 se trouvant dans l'archive **experimental** : le module NVIDIA casse à l'installation par DKMS. Il est donc nécessaire d'installer le dernier pilote opérationnel issu des snaphots Debian, le pilote 343.36. L'environnement de développement, le 6.5.14, sera rétroporté directement de l'archive **experimental**.
+Voici la méthode permettant d'installer à partir d'un rétro-portage les paquets Nvidia ainsi que tout l'environnement associé. Au 17 novembre 2015, il est à nouveau possible de récupérer la version 352.55 se trouvant dans l'archive **experimental** : le module NVIDIA se construit correctement à l'installation par DKMS. L'environnement de développement, le 6.5.19, sera rétroporté directement de l'archive **experimental**.
 <note warning>Pour une compilation dans SIDUS, ne pas oublier de monter le ''/proc''</note>
@@ Ligne 81: / Ligne 82: @@
 Les paquets créés sont les suivants :
 <code>
-libcuda1_352.55-3_amd64.deb
+libcuda1_355.11-2_amd64.deb
-libegl1-nvidia_352.55-3_amd64.deb
+libegl1-nvidia_355.11-2_amd64.deb
-libgl1-nvidia-glx_352.55-3_amd64.deb
+libegl-nvidia0_355.11-2_amd64.deb
-libgles1-nvidia_352.55-3_amd64.deb
+libgl1-nvidia-glx_355.11-2_amd64.deb
-libgles2-nvidia_352.55-3_amd64.deb
+libgles1-nvidia_355.11-2_amd64.deb
-libnvcuvid1_352.55-3_amd64.deb
+libgles2-nvidia_355.11-2_amd64.deb
-libnvidia-compiler_352.55-3_amd64.deb
+libglvnd-nvidia_355.11-2_amd64.deb
-libnvidia-eglcore_352.55-3_amd64.deb
+libnvcuvid1_355.11-2_amd64.deb
-libnvidia-encode1_352.55-3_amd64.deb
+libnvidia-compiler_355.11-2_amd64.deb
-libnvidia-fbc1_352.55-3_amd64.deb
+libnvidia-eglcore_355.11-2_amd64.deb
-libnvidia-ifr1_352.55-3_amd64.deb
+libnvidia-encode1_355.11-2_amd64.deb
-libnvidia-ml1_352.55-3_amd64.deb
+libnvidia-fbc1_355.11-2_amd64.deb
-nvidia-alternative_352.55-3_amd64.deb
+libnvidia-ifr1_355.11-2_amd64.deb
-nvidia-cuda-mps_352.55-3_amd64.deb
+libnvidia-ml1_355.11-2_amd64.deb
-nvidia-detect_352.55-3_amd64.deb
+nvidia-alternative_355.11-2_amd64.deb
-nvidia-driver_352.55-3_amd64.deb
+nvidia-cuda-mps_355.11-2_amd64.deb
-nvidia-driver-bin_352.55-3_amd64.deb
+nvidia-detect_355.11-2_amd64.deb
-nvidia-kernel-dkms_352.55-3_amd64.deb
+nvidia-driver_355.11-2_amd64.deb
-nvidia-kernel-source_352.55-3_amd64.deb
+nvidia-driver-bin_355.11-2_amd64.deb
-nvidia-kernel-support_352.55-3_amd64.deb
+nvidia-kernel-dkms_355.11-2_amd64.deb
-nvidia-libopencl1_352.55-3_amd64.deb
+nvidia-kernel-source_355.11-2_amd64.deb
-nvidia-opencl-common_352.55-3_amd64.deb
+nvidia-kernel-support_355.11-2_amd64.deb
-nvidia-opencl-icd_352.55-3_amd64.deb
+nvidia-legacy-check_355.11-2_amd64.deb
-nvidia-smi_352.55-3_amd64.deb
+nvidia-libopencl1_355.11-2_amd64.deb
-nvidia-vdpau-driver_352.55-3_amd64.deb
+nvidia-opencl-common_355.11-2_amd64.deb
-xserver-xorg-video-nvidia_352.55-3_amd64.deb
+nvidia-opencl-icd_355.11-2_amd64.deb
+nvidia-smi_355.11-2_amd64.deb
+nvidia-vdpau-driver_355.11-2_amd64.deb
+xserver-xorg-video-nvidia_355.11-2_amd64.deb
 </code>
@@ Ligne 119: / Ligne 123: @@
 Les paquets créés sont les suivants :
 <code>
-nvidia-modprobe_358.09-1_amd64.deb
+nvidia-modprobe_361.28-1_amd64.deb
 </code>
@@ Ligne 162: / Ligne 166: @@
 Les paquets créés sont les suivants :
 <code>
-nvidia-installer-cleanup_20151021+1_amd64.deb
+nvidia-installer-cleanup_20151021+4_amd64.deb
-nvidia-kernel-common_20151021+1_amd64.deb
+nvidia-kernel-common_20151021+4_amd64.deb
-nvidia-support_20151021+1_amd64.deb
+nvidia-support_20151021+4_amd64.deb
 </code>
@@ Ligne 183: / Ligne 187: @@
 dpkg -i nvidia-modprobe_*_amd64.deb
+dpkg -i nvidia-legacy-check-*_amd64.deb
 dpkg -i nvidia-alternative_*_amd64.deb
@@ Ligne 195: / Ligne 199: @@
 dpkg -i nvidia-kernel-dkms_*_amd64.deb
-ls -1 lib*352.55*deb | xargs -I '{}' dpkg -i '{}'
+ls -1 lib*$(dpkg -l | grep  nvidia-kernel-dkms | awk '{ print $3 }')*deb | xargs -I '{}' dpkg -i '{}'
 apt-get -f install
@@ Ligne 203: / Ligne 207: @@
 dpkg -i nvidia-vdpau-driver_*_amd64.deb nvidia-driver_*_amd64.deb nvidia-driver-bin_*_amd64.deb
-ls *352.55*deb  | grep -v ^lib | grep -v nvidia-kernel | grep -v libopencl | xargs -I '{}' dpkg -i '{}'
+ls *$(dpkg -l | grep  nvidia-kernel-dkms | awk '{ print $3 }')*deb  | grep -v ^lib | grep -v nvidia-kernel | grep -v libopencl | xargs -I '{}' dpkg -i '{}'
 dpkg -i nvidia-driver-bin_*deb nvidia-driver_*deb nvidia-xconfig* nvidia-settings* libxnvctrl* nvidia-smi_*
@@ Ligne 226: / Ligne 230: @@
 Les paquets créés sont les suivants :
 <code>
+libcublas7.0_7.0.28-4_amd64.deb
+libcudart7.0_7.0.28-4_amd64.deb
+libcufft7.0_7.0.28-4_amd64.deb
+libcufftw7.0_7.0.28-4_amd64.deb
+libcuinj64-7.0_7.0.28-4_amd64.deb
+libcupti7.0_7.0.28-4_amd64.deb
+libcupti-dev_7.0.28-4_amd64.deb
+libcupti-doc_7.0.28-4_all.deb
+libcurand7.0_7.0.28-4_amd64.deb
+libcusolver7.0_7.0.28-4_amd64.deb
+libcusparse7.0_7.0.28-4_amd64.deb
+libnppc7.0_7.0.28-4_amd64.deb
+libnppi7.0_7.0.28-4_amd64.deb
+libnpps7.0_7.0.28-4_amd64.deb
+libnvblas7.0_7.0.28-4_amd64.deb
+libnvrtc7.0_7.0.28-4_amd64.deb
+libnvtoolsext1_7.0.28-4_amd64.deb
+libnvvm3_7.0.28-4_amd64.deb
+nvidia-cuda-dev_7.0.28-4_amd64.deb
+nvidia-cuda-doc_7.0.28-4_all.deb
+nvidia-cuda-gdb_7.0.28-4_amd64.deb
+nvidia-cuda-toolkit_7.0.28-4_amd64.deb
+nvidia-nsight_7.0.28-4_amd64.deb
+nvidia-opencl-dev_7.0.28-4_amd64.deb
+nvidia-profiler_7.0.28-4_amd64.deb
+nvidia-visual-profiler_7.0.28-4_amd64.deb
 </code>
@@ Ligne 234: / Ligne 263: @@
 apt-get install -y opencl-headers ocl-icd-opencl-dev
 cd /root/nvidia/debian
-ls *6.5.19*deb | grep -v opencl | xargs -I '{}' dpkg -i '{}'
+ls *7.0.28*deb | grep -v opencl | xargs -I '{}' dpkg -i '{}'
 # Il arrive que l'installeur rale sur nvidia-cuda-toolkit. Si c'est le cas, la commande suivante s'impose
 apt-get -f install
@@ Ligne 243: / Ligne 272: @@
 <code>
 cd /root/nvidia
-apt-get -y source python-pyopencl
+apt-get -y build-dep pycuda
-apt-get -y build-dep python-pyopencl
+apt-get -y install python-setuptools python3-setuptools
-cd pyopencl-*
+wget http://snapshot.debian.org/archive/debian/20150617T043723Z/pool/main/p/pyopencl/pyopencl_2015.1-2.debian.tar.xz
+wget http://snapshot.debian.org/archive/debian/20150617T043723Z/pool/main/p/pyopencl/pyopencl_2015.1-2.dsc
+wget http://snapshot.debian.org/archive/debian/20150610T042543Z/pool/main/p/pyopencl/pyopencl_2015.1.orig.tar.xz
+tar Jxf pyopencl_2015.1.orig.tar.xz
+cd pyopencl-*/
+tar Jxf ../pyopencl_2015.1-2.debian.tar.xz
 debuild
 cd ..
@@ Ligne 254: / Ligne 288: @@
 Les paquets créés sont les suivants :
 <code>
-python3-pyopencl_2015.1-1_amd64.deb
+python3-pyopencl_2015.1-2_amd64.deb
-python3-pyopencl-dbg_2015.1-1_amd64.deb
+python3-pyopencl-dbg_2015.1-2_amd64.deb
-python-pyopencl_2015.1-1_amd64.deb
+python-pyopencl_2015.1-2_amd64.deb
-python-pyopencl-dbg_2015.1-1_amd64.deb
+python-pyopencl-dbg_2015.1-2_amd64.deb
-python-pyopencl-doc_2015.1-1_all.deb
+python-pyopencl-doc_2015.1-2_all.deb
 </code>
@@ Ligne 269: / Ligne 303: @@
 </code>
+Il n'est pas possible de rétroporter la dernière version de pycuda en Jessie à cause des dépendances entre PyCUDA et d'autres librairies récentes n'acceptant pas le rétroportage.
 <code>
 cd /root/nvidia
-apt-get source -y pycuda
 apt-get -y build-dep pycuda
+apt-get -y install python-setuptools python3-setuptools
+wget http://snapshot.debian.org/archive/debian/20150710T034220Z/pool/contrib/p/pycuda/pycuda_2015.1.2-1.debian.tar.xz
+wget http://snapshot.debian.org/archive/debian/20150710T034220Z/pool/contrib/p/pycuda/pycuda_2015.1.2-1.dsc
+wget http://snapshot.debian.org/archive/debian/20150710T034220Z/pool/contrib/p/pycuda/pycuda_2015.1.2.orig.tar.xz
+tar Jxf pycuda_2015.1.2.orig.tar.xz
 cd pycuda-*/
+tar Jxf ../pycuda_2015.1.2-1.debian.tar.xz
 debuild
 cd ..
@@ Ligne 302: / Ligne 342: @@
 <code>
 cd /root/nvidia/debian
-apt-get install -y python-pytest python3-pytest
+apt-get install -y python-pytest python3-pytest python3-appdirs python-appdirs
 dpkg -i python-pycuda*deb python3-pycuda*deb
 </code>
@@ Ligne 308: / Ligne 348: @@
 ==== Exécution des exemples ====
-<code>
+Sur une machine équipée de 3 cartes vidéo et 3 implémentations d'OpenCL pour CPU (AMD, Intel, PortableCL)
+<code>
 python /usr/share/doc/python-pyopencl-doc/examples/benchmark.py
 </code>
-Sur des cartes Tesla C1060 & Quadro FX 580 (et un Dell Precision 3500):<code>
+<code>
-('Execution time of test without OpenCL: ', 7.415176868438721, 's')
 ===============================================================
-('Platform name:', 'NVIDIA CUDA')
+Platform name: AMD Accelerated Parallel Processing
-('Platform profile:', 'FULL_PROFILE')
+Platform profile: FULL_PROFILE
-('Platform vendor:', 'NVIDIA Corporation')
+Platform vendor: Advanced Micro Devices, Inc.
-('Platform version:', 'OpenCL 1.1 CUDA 6.0.1')
+Platform version: OpenCL 2.0 AMD-APP (1800.11)
 ---------------------------------------------------------------
-('Device name:', 'Tesla C1060')
+Device name: Fiji
-('Device type:', 'GPU')
+Device type: GPU
-('Device memory: ', 4095, 'MB')
+Device memory:  4045 MB
-('Device max clock speed:', 1296, 'MHz')
+Device max clock speed: 1000 MHz
-('Device compute units:', 30)
+Device compute units: 64
-Execution time of test: 0.00188525 s
+Device max work group size: 256
+Device max work item sizes: [256, 256, 256]
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 64
+Execution time of test: 0.00037168 s
 Results OK
 ===============================================================
-('Platform name:', 'NVIDIA CUDA')
+Platform name: AMD Accelerated Parallel Processing
-('Platform profile:', 'FULL_PROFILE')
+Platform profile: FULL_PROFILE
-('Platform vendor:', 'NVIDIA Corporation')
+Platform vendor: Advanced Micro Devices, Inc.
-('Platform version:', 'OpenCL 1.1 CUDA 6.0.1')
+Platform version: OpenCL 2.0 AMD-APP (1800.11)
 ---------------------------------------------------------------
-('Device name:', 'Quadro FX 580')
+Device name: Intel(R) Xeon(R) CPU E5-2665 0 @ 2.40GHz
-('Device type:', 'GPU')
+Device type: CPU
-('Device memory: ', 511, 'MB')
+Device memory:  128966 MB
-('Device max clock speed:', 1125, 'MHz')
+Device max clock speed: 2309 MHz
-('Device compute units:', 4)
+Device compute units: 32
-Execution time of test: 0.0126466 s
+Device max work group size: 1024
+Device max work item sizes: [1024, 1024, 1024]
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 1
+Execution time of test: 0.0192504 s
 Results OK
 ===============================================================
-('Platform name:', 'AMD Accelerated Parallel Processing')
+Platform name: Intel(R) OpenCL
-('Platform profile:', 'FULL_PROFILE')
+Platform profile: FULL_PROFILE
-('Platform vendor:', 'Advanced Micro Devices, Inc.')
+Platform vendor: Intel(R) Corporation
-('Platform version:', 'OpenCL 1.2 AMD-APP (938.2)')
+Platform version: OpenCL 1.2 LINUX
 ---------------------------------------------------------------
-('Device name:', 'Intel(R) Xeon(R) CPU           W3565  @ 3.20GHz')
+Device name:        Intel(R) Xeon(R) CPU E5-2665 0 @ 2.40GHz
-('Device type:', 'CPU')
+Device type: CPU
-('Device memory: ', 12041, 'MB')
+Device memory:  128966 MB
-('Device max clock speed:', 3199, 'MHz')
+Device max clock speed: 2400 MHz
-('Device compute units:', 4)
+Device compute units: 32
-Execution time of test: 0.00191834 s
+Device max work group size: 8192
+Device max work item sizes: [8192, 8192, 8192]
+/usr/lib/python2.7/dist-packages/pyopencl/__init__.py:63: CompilerWarning: Non-empty compiler output encountered. Set the environment variable PYOPENCL_COMPILER_OUTPUT=1 to see more.
+  "to see more.", CompilerWarning)
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 128
+Execution time of test: 0.00310517 s
+Results OK
+===============================================================
+Platform name: Portable Computing Language
+Platform profile: FULL_PROFILE
+Platform vendor: The pocl project
+Platform version: OpenCL 1.2 pocl 0.10
+---------------------------------------------------------------
+Device name: pthread-Intel(R) Xeon(R) CPU E5-2665 0 @ 2.40GHz
+Device type: CPU
+Device memory:  128966 MB
+Device max clock speed: 3100 MHz
+Device compute units: 32
+Device max work group size: 1024
+Device max work item sizes: [1024, 1024, 1024]
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 8
+Execution time of test: 0.007638 s
 Results OK
-</code>
-Sur une carte GT650M (et un Mac Book Pro):<code>
-('Execution time of test without OpenCL: ', 7.595532178878784, 's')
 ===============================================================
-('Platform name:', 'NVIDIA CUDA')
+Platform name: NVIDIA CUDA
-('Platform profile:', 'FULL_PROFILE')
+Platform profile: FULL_PROFILE
-('Platform vendor:', 'NVIDIA Corporation')
+Platform vendor: NVIDIA Corporation
-('Platform version:', 'OpenCL 1.1 CUDA 6.0.1')
+Platform version: OpenCL 1.2 CUDA 7.5.20
 ---------------------------------------------------------------
-('Device name:', 'GeForce GT 650M')
+Device name: GeForce GTX 980 Ti
-('Device type:', 'GPU')
+Device type: GPU
-('Device memory: ', 511, 'MB')
+Device memory:  6143 MB
-('Device max clock speed:', 405, 'MHz')
+Device max clock speed: 1190 MHz
-('Device compute units:', 2)
+Device compute units: 22
-Execution time of test: 0.0011792 s
+Device max work group size: 1024
+Device max work item sizes: [1024, 1024, 64]
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 32
+Execution time of test: 0.000522592 s
 Results OK
 ===============================================================
-('Platform name:', 'AMD Accelerated Parallel Processing')
+Platform name: NVIDIA CUDA
-('Platform profile:', 'FULL_PROFILE')
+Platform profile: FULL_PROFILE
-('Platform vendor:', 'Advanced Micro Devices, Inc.')
+Platform vendor: NVIDIA Corporation
-('Platform version:', 'OpenCL 1.2 AMD-APP (938.2)')
+Platform version: OpenCL 1.2 CUDA 7.5.20
 ---------------------------------------------------------------
-('Device name:', 'Intel(R) Core(TM) i7-3615QM CPU @ 2.30GHz')
+Device name: Quadro 600
-('Device type:', 'CPU')
+Device type: GPU
-('Device memory: ', 7942, 'MB')
+Device memory:  1023 MB
-('Device max clock speed:', 1388, 'MHz')
+Device max clock speed: 1280 MHz
-('Device compute units:', 8)
+Device compute units: 2
-Execution time of test: 0.00280508 s
+Device max work group size: 1024
+Device max work item sizes: [1024, 1024, 64]
+Data points: 8388608
+Workers: 256
+Preferred work group size multiple: 32
+Execution time of test: 0.00468445 s
 Results OK
 </code>
-Vous aurez noter que l'installation préalable d'un SDK AMD a permis de disposer d'un support OpenCL sur le processeur.
-====== Installation de PyFFT ======
-[[http://packages.python.org/pyfft/|PyFFT]] permet l'exploitation de PyCUDA et PyOpenCL pour effectuer tranquillement ses FFT directement à partir de ses scripts Python.
-Pour son installation, suivre la page : http://www.cbp.ens-lyon.fr/emmanuel.quemener/dokuwiki/doku.php?id=pyfft4squeeze

developpement/activites/integration/cuda4jessie.1447767117.txt.gz · Dernière modification : 2025/05/14 07:38 (modification externe)

Rechercher

Translations

Navigation

Piste :

toolbox