High Performance Computing with OpenCL

Introduction

This project explores the use of OpenCL in high performance computing for heterogenous platforms (i.e., CPU/GPU). We present updated C/C++ code adhering to OpenCL 2.0 standard for the exercises in the “OpenCL in Action” book by Matthew Scarpino. Read the book. It is a good comprehensive book, suitable for beginners, which teaches OpenCL beginning from the basics.

The following tools will be used in this project:

• C/C++
• OpenCL
• CMake
• Intel CPU - we can test our OpenCL code on a CPU without the need to have a GPU

Code

Find the source code in the repository.

Learning Outcome

At the end of this project, we should be able to:

• Use OpenCL in C/C++ for high performance computing.
• Build C/C++ code with CMake.

Project Structure

The project structure is as follows:

opencl                                 # Repository root
├── apps                               # Source code
|   ├── Ch1
|   |   ├── matvec.c                   # C source file
|   |   └── matvec.cl                  # Cl source file
|   ├── Ch2
|   |   ├── context_count.c            # C source file
|   |   ├── device_ext_test.c
|   |   ...
|   .
|   .
|   .
|   ├── Ch8
|   |   ├── buffer_test.cl             # Cl source file
|   |   ├── buffer_test.cpp            # C++ source file
|   |   ...
|   └── CMakeLists.txt                 # CMake build file
├── assets                             # Reference materials 
|   ├── libpng-1.4.0-manual.pdf      
|   ├── OpenCL device model.jpg      
|   ├── OpenCL operators.jpg
|   ├── OpenCL scalar data types.jpg
|   └── OpenCL vector data types.jpg
├── bin                                # Executables
|   ├── ...                              
|   ...
├── libs                               # Local libraries
│   └── util                     
|       ├── CMakeLists.txt             # Library-level CMake build file
|       ├── util.c                     
|       ├── util.cpp                   
|       ├── util.h                     
|       └── util.hpp                   
├── .dockerignore
├── .gitignore
├── CMakeLists.txt                     # Top-level CMake build file
└── README.md                                 

Steps to get started with OpenCL

1. Install OpenCL SDK
   • For Intel processors: website
2. Install and run clinfo to show complete information of OpenCL platforms and devices.

   $ sudo apt install clinfo
   $ clinfo
   

Exercise list

For easy reference, the solution code is organised by chapter and named according to the exercise.

Build the C/C++ code using:

$ cd /path/to/repository/root/opencl
$ cmake -E make_directory build
$ cmake -E chdir ./build cmake -DCMAKE_BUILD_TYPE=Release ..
$ cmake --build ./build

Run the desired code as follows. Assuming the desired code is callback_8_6, then run:

$ cd /path/to/repository/root/opencl
$ ./bin/callback_8_6

The complete code listing is given in the table below.

Chapter	Exercise	Name
1 Introducing OpenCL	1.1	matvec
2 Host programming: fundamental data structures	2.1	platform_ext_test
	2.2	device_ext_test
	2.3	context_count
3 Host programming: data transfer and partitioning	3.2	buffer_test
	3.3	map_copy
4 Kernel programming: data types and device memory	4.1	hello_kernel
	4.2	double_test
	4.3	float_config
	4.4	vector_widths
	4.5	vector_bytes
5 Kernel programming: operators and functions	5.1	op_test
	5.2	id_check
	5.3	mod_round
	5.4	polar_rect
	5.5	mad_test
	5.6	shuffle_test
	5.7	select_test
7 Events, profiling, and synchronization	7.2	callback
	7.3	user_event
	7.6	profile_read
	7.7	profile_items
	7.8	atomic
	7.9	mutex
8 Development with C++	8.1	full_context_8_1
	8.2	create_kernels_8_2
	8.4	buffer_test_8_4
	8.5	map_copy_8_5
	8.6	callback_8_6
	8.7	user_event_8_7
	8.8	profile_8_8