Programming/OpenCL - Revision history

Pysdlb: /* Next Steps */

2022-11-16T11:08:39Z

‎Next Steps

Pysdlb: /* Navigation */

2022-11-16T11:08:25Z

‎Navigation

Pysdlb at 15:31, 8 November 2022

2022-11-08T15:31:21Z

Pysdlb: /* Introduction to openCL */

2022-11-08T15:30:37Z

‎Introduction to openCL

Pysdlb at 12:22, 2 November 2022

2022-11-02T12:22:58Z

Pysdlb at 09:14, 24 May 2021

2021-05-24T09:14:51Z

Pysdlb at 08:59, 24 August 2020

2020-08-24T08:59:28Z

Pysdlb at 10:45, 10 June 2019

2019-06-10T10:45:12Z

Pysdlb: Pysdlb moved page Programming/openCL to Programming/OpenCL without leaving a redirect

2019-02-07T15:03:22Z

Pysdlb moved page Programming/openCL to Programming/OpenCL without leaving a redirect

Pysdlb at 11:49, 7 February 2019

2019-02-07T11:49:29Z

@@ Line 360: / Line 360: @@
 * [[programming/C-Plusplus|C++ Programming]]
 * [[programming/Fortran|Fortran Programming]]
 * [https://handsonopencl.github.io/ OpenCL tutorial]
 * [https://www.nersc.gov/assets/pubs_presos/MattsonTutorialSC14.pdf OpenCL course]
 {{Librariespagenav}}

← Older revision		Revision as of 11:08, 16 November 2022
Line 364:		Line 364:
	* [https://www.nersc.gov/assets/pubs_presos/MattsonTutorialSC14.pdf OpenCL course]		* [https://www.nersc.gov/assets/pubs_presos/MattsonTutorialSC14.pdf OpenCL course]

−	~~==Navigation==~~	+	{{Librariespagenav}}
−
−	* [[Main_Page\|Home]]
−	* [[Applications\|Application support]]
−	* [[General\|General]]
−	~~* [[Programming\|Programming support]] *~~

@@ Line 55: / Line 55: @@
 #include "err_code.h"
-//pick up device type from compiler command line or from
+//pick up device type from the compiler command line or from
 //the default type
 #ifndef DEVICE

← Older revision		Revision as of 15:30, 8 November 2022
Line 1:		Line 1:
−	==Introduction to ~~openCL~~==	+	==Introduction to OpenCL==
−
−

	==OpenCL programming==		==OpenCL programming==

@@ Line 5: / Line 5: @@
 ==OpenCL programming==
-The first thing to realise when trying to port a code to a GPU is that they do not share the same memory as the CPU. In other words, a GPU does not have direct access to the host memory. The host memory is generally larger, but slower than the GPU memory. To use a GPU, data must therefore be transferred from the main program to the GPU through the PCI bus, which has a much lower bandwidth than either memory. This means that managing data transfer between the host and the GPU will be of paramount importance. Transferring the data and the code onto the device is called offloading. In summary, memory divided into '''host memory''' and '''device memory'''.
+The first thing to realise when trying to port a code to a GPU is that they do not share the same memory as the CPU. In other words, a GPU does not have direct access to the host memory. The host memory is generally larger, but slower than the GPU memory. To use a GPU, data must therefore be transferred from the main program to the GPU through the PCI bus, which has a much lower bandwidth than either memory. This means that managing data transfer between the host and the GPU will be of paramount importance. Transferring the data and the code onto the device is called offloading. In summary, memory is divided into '''host memory''' and '''device memory'''.
@@ Line 20: / Line 20: @@
 * CUDA is mature and efficient, it has many tools and libraries. (However, it is only usable by NVIDIA GPU architectures.)
-* OpenCL designed for various different processors including AMD/NVIDIA CPU/GPUs, DSPs, and FPGAs (many platforms heterogeneous).
+* OpenCL is designed for various different processors including AMD/NVIDIA CPU/GPUs, DSPs, and FPGAs (many platforms heterogeneous).
-However, it not as mature and as widely used as CUDA C and programming appears to be verbose compared to CUDA.
+However, it is not as mature and as widely used as CUDA C and programming appears to be verbose compared to CUDA.
-* OpenCL has a huge code overhead on the Host side with setting up and running OpenCL code. Even though our smallest Kernel size might be only 15 lines, there is an additional 1500 lines of code on the Host side in support. In our opinion this is one of the biggest drawbacks to the OpenCL platform, is becoming familiar with the considerable complexity of OpenCL API and execution models in general.
+* OpenCL has a huge code overhead on the Host side with setting up and running OpenCL code. Even though our smallest Kernel size might be only 15 lines, there is an additional 1500 lines of code on the Host side in support. In our opinion, this is one of the biggest drawbacks to the OpenCL platform, which is becoming familiar with the considerable complexity of OpenCL API and execution models in general.
@@ Line 357: / Line 357: @@
-==Further Information==
+==Next Steps==
 * [[programming/C|C Programming]]
 * [[programming/C-Plusplus|C++ Programming]]
 * [[programming/Fortran|Fortran Programming]]
 * [https://handsonopencl.github.io/ OpenCL tutorial]

@@ Line 5: / Line 5: @@
 ==OpenCL programming==
-The first thing to realise when trying to port a code to a GPU is that they do not share the same memory as the CPU. In other words, a GPU does not have direct access to the host memory. The host memory is generally larger, but slower than the GPU memory. To use a GPU, data must therefore be transferred from the main program to the GPU through the PCI bus, which has a much lower bandwidth than either memories. This means that managing data transfer between the host and the GPU will be of paramount importance. Transferring the data and the code onto the device is called offloading. In summary, memory divided into '''host memory''' and '''device memory'''.
+The first thing to realise when trying to port a code to a GPU is that they do not share the same memory as the CPU. In other words, a GPU does not have direct access to the host memory. The host memory is generally larger, but slower than the GPU memory. To use a GPU, data must therefore be transferred from the main program to the GPU through the PCI bus, which has a much lower bandwidth than either memory. This means that managing data transfer between the host and the GPU will be of paramount importance. Transferring the data and the code onto the device is called offloading. In summary, memory divided into '''host memory''' and '''device memory'''.
-OpenCL (Open Computing Language) is a framework for writing programs that execute across heterogeneous platforms consisting of central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs) and other processors or hardware accelerators. OpenCL specifies programming languages (based on C99 and C++11) for programming these devices and application programming interfaces (APIs) to control the platform and execute programs on the compute devices. OpenCL provides a standard interface for parallel computing using task- and data-based parallelism.
+OpenCL (Open Computing Language) is a framework for writing programs that execute across heterogeneous platforms consisting of central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs), and other processors or hardware accelerators. OpenCL specifies programming languages (based on C99 and C++11) for programming these devices and application programming interfaces (APIs) to control the platform and execute programs on the computing devices. OpenCL provides a standard interface for parallel computing using task- and data-based parallelism.
 The power behind OpenCL is the ability to replace intensive loops with functions (a '''kernel''') executing at each point in a problem domain
@@ Line 17: / Line 17: @@
 ===OpenCL vs CUDA===
-From one point of view these two paradigms look quite similar, however there are some differences
+From one point of view, these two paradigms look quite similar, however, there are some differences
 * CUDA is mature and efficient, it has many tools and libraries. (However, it is only usable by NVIDIA GPU architectures.)
-* OpenCL designed for various different processors including AMD/NVIDIA CPU/GPUs, DSPs and FPGAs (many platforms heterogeneous).
+* OpenCL designed for various different processors including AMD/NVIDIA CPU/GPUs, DSPs, and FPGAs (many platforms heterogeneous).
 However, it not as mature and as widely used as CUDA C and programming appears to be verbose compared to CUDA.
 * OpenCL has a huge code overhead on the Host side with setting up and running OpenCL code. Even though our smallest Kernel size might be only 15 lines, there is an additional 1500 lines of code on the Host side in support. In our opinion this is one of the biggest drawbacks to the OpenCL platform, is becoming familiar with the considerable complexity of OpenCL API and execution models in general.

@@ Line 368: / Line 368: @@
 * [https://www.nersc.gov/assets/pubs_presos/MattsonTutorialSC14.pdf OpenCL course]
-{|
+==Navigation==
-|style="width:5%; border-width: 0" | [[File:icon_home.png]]
 * [[Main_Page|Home]]
 * [[Applications|Application support]]
 * [[General|General]]
-* [[Training|Training]]
+* [[Programming|Programming support]] *

← Older revision	Revision as of 15:03, 7 February 2019
(No difference)

← Older revision		Revision as of 11:49, 7 February 2019
Line 309:		Line 309:

	* In order to compile your OpenCL program you must tell the compiler to use the OpenCL library with the flag '''–l OpenCL'''.		* In order to compile your OpenCL program you must tell the compiler to use the OpenCL library with the flag '''–l OpenCL'''.
		+

	===Python OpenCL===		===Python OpenCL===