Openvino e Neural Compute Stick 2

Per poter utilizzare Neural Compute Stick 2 di Intel si deve installare Openvino Toolkit
La procedura di installazione e' lineare su Ubuntu e non porta problemi

Ogni volta che si vorra' utilizzare l'SDK si deve ricordare di impostare le variabili di ambiente con

source /home/luca/openvino/bin/setupvars.sh

 

Per utilizzare il Neural Compute Stick e' necessario installare i driver con la procedura

sudo usermod -a -G users "$(whoami)"

 

sudo cp /home/luca/openvino/inference_engine/external/97-myriad-usbboot.rules /etc/udev/rules.d/

 

sudo udevadm control --reload-rules

 

sudo udevadm trigger

 

sudo ldconfig

 

 

Openvino funziona con processori Intel di sesta generazione o superiori, con Neural Compute Stick 2 ma ammette anche l'accelerazione grafica delle GPU Intel tramite OpenCL andando nella directory openvino/install_dependencies/ e digitando

./install_NEO_OCL_driver.sh

Per iniziare a provare Openvino si possono usare gli esempi (sia in C che in Python) riportati al seguente indirizzo. 

Un'altra sorgente di informazioni si trovano a NCAPPZOO

Una volta effettuato il make il file eseguibile si trovera' nella cartella inference_engine_demos_build/intel64/Release (nello stesso momenti, oltre alla compilazione degli eseguibili, vengono scaricati da Internet anche i modelli della rete neurale)

Provando l'esempio Age e Gender di NCAPPZOO si deve andare nella directory ~/intel/ncappzoo/networks/age_gender_net, si installano le dipendenze delle librerie Python e si processa l'immagine di esempio (i parametri del programma hanno dei valori di default ma possono cambiati da linea di comando)

age_gender_net: Starting application...
   - Plugin:       Myriad
   - IR File:      ./age-gender-recognition-retail-0013.xml
   - Input Shape:  [1, 3, 62, 62]
   - Age Output Shape: [1, 1, 1, 1]
   - Gender Output Shape: [1, 2, 1, 1]

 Gender prediction is 98.4% Male
 Age prediction is 59 years old.

in alcuni casi ho dovuto modificare a mano i file Python per problemi con la codifica UTF-8 aggiungendo la stringa # -*- coding: utf-8 -*-

sd

Installare sviluppo OpenCL su Ubuntu

questa procedura e' stata provata su un Lenovo T460 con una Intel HD 520

I pacchetti da installare sono i seguenti

$ sudo apt install ocl-icd-libopencl1
$ sudo apt install opencl-headers
$ sudo apt install clinfo
$ sudo apt install ocl-icd-opencl-dev
$ sudo apt install beignet

Per verificare l'installazione si puo' usare 

clinfo

 --------------------------------------

Number of platforms                               1
  Platform Name                                   Intel Gen OCL Driver
  Platform Vendor                                 Intel
  Platform Version                                OpenCL 2.0 beignet 1.3
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_3d_image_writes cl_khr_image2d_from_buffer cl_khr_depth_images cl_khr_spir cl_khr_icd cl_intel_accelerator cl_intel_subgroups cl_intel_subgroups_short cl_khr_gl_sharing
  Platform Extensions function suffix             Intel

  Platform Name                                   Intel Gen OCL Driver
Number of devices                                 1
  Device Name                                     Intel(R) HD Graphics Skylake ULT GT2
  Device Vendor                                   Intel
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.0 beignet 1.3
  Driver Version                                  1.3
  Device OpenCL C Version                         OpenCL C 2.0 beignet 1.3
  Device Type                                     GPU
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               24
  Max clock frequency                             1000MHz
  Device Partition                                (core)
    Max number of sub-devices                     1
    Supported partition types                     None, None, None
  Max work item dimensions                        3
  Max work item sizes                             512x512x512
  Max work group size                             512
  Preferred work group size multiple              16
  Preferred / native vector sizes                 
    char                                                16 / 8       
    short                                                8 / 8       
    int                                                  4 / 4       
    long                                                 2 / 2       
    half                                                 0 / 8        (cl_khr_fp16)
    float                                                4 / 4       
    double                                               0 / 2        (n/a)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
  Single-precision Floating-point support         (core)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Double-precision Floating-point support         (n/a)
  Address bits                                    32, Little-Endian
  Global memory size                              3869245440 (3.604GiB)
  Error Correction support                        No
  Max memory allocation                           2901409792 (2.702GiB)
  Unified memory for Host and Device              Yes
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   No
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           0 bytes
    Global                                        0 bytes
    Local                                         0 bytes
  Max size for global variable                    65536 (64KiB)
  Preferred total size of global vars             65536 (64KiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        8192 (8KiB)
  Global Memory cache line size                   64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            65536 pixels
    Max 1D or 2D image array size                 2048 images
    Base address alignment for 2D image buffers   4096 bytes
    Pitch alignment for 2D image buffers          1 pixels
    Max 2D image size                             8192x8192 pixels
    Max 3D image size                             8192x8192x2048 pixels
    Max number of read image args                 128
    Max number of write image args                8
    Max number of read/write image args           8
  Max number of pipe args                         16
  Max active pipe reservations                    1
  Max pipe packet size                            1024
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Max number of constant args                     8
  Max constant buffer size                        134217728 (128MiB)
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        No
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                16384 (16KiB)
    Max size                                      262144 (256KiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Profiling timer resolution                      80ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            Yes
    SPIR versions                                 1.2
  printf() buffer size                            1048576 (1024KiB)
  Built-in kernels                                __cl_copy_region_align4;__cl_copy_region_align16;__cl_cpy_region_unalign_same_offset;__cl_copy_region_unalign_dst_offset;__cl_copy_region_unalign_src_offset;__cl_copy_buffer_rect;__cl_copy_image_1d_to_1d;__cl_copy_image_2d_to_2d;__cl_copy_image_3d_to_2d;__cl_copy_image_2d_to_3d;__cl_copy_image_3d_to_3d;__cl_copy_image_2d_to_buffer;__cl_copy_image_3d_to_buffer;__cl_copy_buffer_to_image_2d;__cl_copy_buffer_to_image_3d;__cl_fill_region_unalign;__cl_fill_region_align2;__cl_fill_region_align4;__cl_fill_region_align8_2;__cl_fill_region_align8_4;__cl_fill_region_align8_8;__cl_fill_region_align8_16;__cl_fill_region_align128;__cl_fill_image_1d;__cl_fill_image_1d_array;__cl_fill_image_2d;__cl_fill_image_2d_array;__cl_fill_image_3d;
  Device Extensions                               cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_3d_image_writes cl_khr_image2d_from_buffer cl_khr_depth_images cl_khr_spir cl_khr_icd cl_intel_accelerator cl_intel_subgroups cl_intel_subgroups_short cl_khr_gl_sharing cl_khr_fp16

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  Intel Gen OCL Driver
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   Success [Intel]
  clCreateContext(NULL, ...) [default]            Success [Intel]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 Intel Gen OCL Driver
    Device Name                                   Intel(R) HD Graphics Skylake ULT GT2
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 Intel Gen OCL Driver
    Device Name                                   Intel(R) HD Graphics Skylake ULT GT2
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 Intel Gen OCL Driver
    Device Name                                   Intel(R) HD Graphics Skylake ULT GT2

ICD loader properties
  ICD loader Name                                 OpenCL ICD Loader
  ICD loader Vendor                               OCL Icd free software
  ICD loader Version                              2.2.11
  ICD loader Profile                              OpenCL 2.1

-------------------------------------------

Ho scoperto che la Intel HD 520  non ha il supporto per la doppia precisione

Per compilare su Linux make base ha una forma del tipo

hello: hello.cpp
        g++ -std=c++0x -o hello hello.cpp -lOpenCL
 

OpenCL e Mandelbrot

Il calcolo dell'insieme di Mandelbrot e' forse uno dei casi piu' semplici in cui applicare il calcolo parallelo perche' l'elaborazione di  ogni cella e' indipendente da quelle vicine.

Ai giorni d'oggi per effettuare il calcolo parallelo, invece di usare un cluster di computer, si possono usare i diversi core del processore oppure usare la scheda grafica con la sua GPU (mediante le librerie CUDA di NVidia o le piu' generiche OpenCL che possono girare su AMD, NVidia e Intel)

La strategia utilizzata in CUDA e OpenCL e' piu' o meno la stessa ovvero si scrive il codice di un kernel, ovvero della elaborazione che verra' eseguita su ogni unita' di elaborazione, e poi il codice che gestisce la coda dei processi e la distribuzione alle varie unita' di elaborazione

Un esempio di kernel per Mandelbrot e' il seguente
-------------------------------------------------------------------
#pragma OPENCL EXTENSION cl_khr_fp64 : enable

double trans_x(int x, int N)
{
    return 3.0 * ((double)x / N) - 2.0;
}
double trans_y(int y, int N)
{
    return 3.0 * ((double)y / N) - 1.5;
}
double mag2(double r, double i)
{
    return r * r + i * i;
}
__kernel void mandel(__global double* out, int N, int depth, double escape2)
{
    size_t idx = get_global_id(0);

    double z0_r = trans_x(idx % N, N);
    double z0_i = trans_y(idx / N, N);

    double z_r = 0;
    double z_i = 0;
    int k = 0;
    for(; k <= depth && mag2(z_r, z_i) < escape2 ; ++k)
    {
        double t_r = z_r; double t_i = z_i;
        z_r = t_r * t_r - t_i * t_i + z0_r;
        z_i = 2 * t_r * t_i + z0_i;
    }
    out[idx] = log(k + 1.0 - log(log(max(mag2(z_r, z_i), escape2)) / 2.0) / log(2.0));
}

-------------------------------------------------------------------


Per mettere alla prova quanto vantaggio c'e' nell'usare OpenCL per il calcolo di Mandelbrot ho usato un portatile Surface Pro 2 con scheda video Intel HD 4400 ed il programma reperibile a questo link.

Come test ho provato il  Miniset level 2 (1000 iterazioni per punto) con i seguenti risultati

- C float, single thread :1.71 s
- C float, 4 thread : 0.61 s
- C float, 8 thread : 0.59 s
- C fixed-point 128 bit  2 thread : 22.5 s
- OpenCl CPU float  : 0.36 s
- OpenCL CPU  double : 0.4 s
- OpenCL GPU float : 0.096 s