Intel(R)_SDK_for_OpenCL_Applications_User_Guide by Sheng Soul

Intel® SDK for OpenCL* Applications 2013 User’s Guide

Contents Legal Information ............................................................................................................ 4 Getting Help and Support ................................................................................................. 5 Introduction ................................................................................................................... 5 Introducing Intel® SDK for OpenCL* Applications 2013 .............................................. 5 Related Documents................................................................................................. 5 What's New? .......................................................................................................... 6 Related Products .................................................................................................... 6 Overview of OpenCL* Support on Intel® Core™ Processors .................................................. 6 Components and Packages ...................................................................................... 6 The OpenCL* Platform on Intel® Architecture Processors ............................................ 8 Supported OpenCL* Features .......................................................................................... 10 OpenCL* 1.2 Full Profile ......................................................................................... 10 OpenCL* Installable Client Driver (ICD) ................................................................... 10 Shared Context ..................................................................................................... 10 OpenCL* Extensions and Optional Features .............................................................. 11 Utilizing CPU Device Resources ........................................................................................ 18 Seamless Vectorization Using the Implicit CPU Vectorization Module ............................ 18 Threading System ................................................................................................. 18 Developing with Intel® SDK for OpenCL* Applications 2013 ................................................ 18 Developing with OpenCL* Platform .......................................................................... 18 Configuring Microsoft Visual Studio* ........................................................................ 19 Working with the OpenCL* Installable Client Driver (ICD) ........................................... 20 Programming with the Intel® SDK for OpenCL* Applications 2013 ....................................... 21 Intel® SDK for OpenCL* Applications 2013 Development Tools ................................... 21 Building and Analyzing OpenCL* Kernel Performance ................................................. 22 Using the Intel® SDK for OpenCL* - Debugger ......................................................... 45 Tuning with the Intel® Graphics Performance Analyzers............................................. 47 Interoperability with the Intel® VTune™ Amplifier XE ................................................ 56 Appendix A - Supported Images Formats .......................................................................... 60 Read-Only Surface Formats .................................................................................... 60 Write-Only Surface Formats .................................................................................... 62 Read-Write Surface Formats ................................................................................... 64 Appendix B – OpenCL* Build and Linking Options ............................................................... 66 Preprocessor Options ............................................................................................. 66 Math Intrinsics Options .......................................................................................... 66 Optimization Options ............................................................................................. 66 Options for Warnings ............................................................................................. 67 Options Controlling the OpenCL* C Version ............................................................... 67

Optimization Notice Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. Notice revision #20110804

Legal Information INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL'S TERMS AND CONDITIONS OF SALE FOR SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. A "Mission Critical Application" is any application in which failure of the Intel Product could result, directly or indirectly, in personal injury or death. SHOULD YOU PURCHASE OR USE INTEL'S PRODUCTS FOR ANY SUCH MISSION CRITICAL APPLICATION, YOU SHALL INDEMNIFY AND HOLD INTEL AND ITS SUBSIDIARIES, SUBCONTRACTORS AND AFFILIATES, AND THE DIRECTORS, OFFICERS, AND EMPLOYEES OF EACH, HARMLESS AGAINST ALL CLAIMS COSTS, DAMAGES, AND EXPENSES AND REASONABLE ATTORNEYS' FEES ARISING OUT OF, DIRECTLY OR INDIRECTLY, ANY CLAIM OF PRODUCT LIABILITY, PERSONAL INJURY, OR DEATH ARISING IN ANY WAY OUT OF SUCH MISSION CRITICAL APPLICATION, WHETHER OR NOT INTEL OR ITS SUBCONTRACTOR WAS NEGLIGENT IN THE DESIGN, MANUFACTURE, OR WARNING OF THE INTEL PRODUCT OR ANY OF ITS PARTS. Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the absence or characteristics of any features or instructions marked "reserved" or "undefined". Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not finalize a design with this information. The products described in this document may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order. Copies of documents which have an order number and are referenced in this document, or other Intel literature, may be obtained by calling 1-800-548-4725, or go to: http://www.intel.com/design/literature.htm. Intel processor numbers are not a measure of performance. Processor numbers differentiate features within each processor family, not across different processor families. Go to: http://www.intel.com/products/processor_number/. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Intel, Intel logo, Intel Core, VTune, Xeon are trademarks of Intel Corporation in the U.S. and other countries. * Other names and brands may be claimed as the property of others. OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission from Khronos. Microsoft product screen shot(s) reprinted with permission from Microsoft Corporation. Copyright ÂŠ 2010-2013 Intel Corporation. All rights reserved.

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

Getting Help and Support You can get support with issues you face using the Intel® SDK for OpenCL* Applications support forum at intel.com/software/opencl. For information on SDK requirements, known issues and limitations, refer to the Intel® SDK for OpenCL* Applications 2013 - Release Notes at http://software.intel.com/en-us/articles/intel-sdk-foropencl-applications-2013-release-notes/.

Introduction Introducing Intel® SDK for OpenCL* Applications 2013 This User’s Guide contains general information about the Intel® SDK for OpenCL* Applications 2013, its features, tools, package contents, and basic usage guidelines. OpenCL (Open Computing Language) is an open standard for general-purpose parallel programming for diverse mix of multi-core CPUs, Graphics Processers, and other parallel processors.. OpenCL provides flexible execution model and uniform programming environment for software developers to write portable code for client systems running on both the CPU and graphics processors like Intel® HD Graphics. OpenCL is developed by multiple companies through the Khronos* OpenCL committee. Intel, a founding member of the OpenCL Working Group, has been a leading voice to ensure the OpenCL feature set supports OpenCL programmers on current and future Intel® Architecture.

About Intel® SDK for OpenCL* Applications 2013 Intel® SDK for OpenCL* Applications 2013 is a comprehensive software development environment for OpenCL applications on the 3rd and the 4th Generation Intel® Core™ processors, which support OpenCL 1.2 on Windows 7* and Windows 8* operating systems. For information on Intel® SDK for OpenCL* Applications 2013 limitations and known issues, please see the Intel® SDK for OpenCL* Applications 2013 Release Notes.

See Also Intel SDK for OpenCL Applications - Release Notes Intel HD Graphics Developers Web Site

Related Documents For information on SDK limitations, optimization guidelines, and samples, refer to the following links: 1. 2. 3. 4.

Intel® SDK for OpenCL* Applications - Release Notes Intel SDK for OpenCL Applications - Optimization Guide OpenCL 1.2 Specification at http://www.khronos.org/registry/cl/specs/opencl-1.2.pdf Intel SDK for OpenCL Applications - Samples Getting Started

Intel® SDK for OpenCL* Applications 2013 documentation, samples, and technical articles are available at intel.com/software/opencl/.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

What's New? The new Intel® SDK for OpenCL* Applications 2013 brings improvements and new features both in the OpenCL runtime and the development tools. The new Intel SDK for OpenCL Applications 2013 now works seamlessly with the latest edition of Intel® HD Graphics driver that includes certified OpenCL 1.2 support on the 3rd and the 4th Generation Intel® Core™ processors with Intel HD Graphics on Microsoft Windows* 7 and Windows 8* operating systems. OpenCL 1.2 support promises better performance and greater stability on both the CPU and the Intel HD Graphics. OpenCL 1.2 provides more flexibility in software design for OpenCL programmers with improved compilation, linking, and library support, and improved graphics and media surface sharing for full media and graphics acceleration on the Intel HD Graphics. This version of the SDK comprises developer tools, and provides a comprehensive environment for the build, debug, and tune stages of OpenCL application development. For performance analysis, this environment enables the leading Intel profiling applications, Intel® VTune™ Amplifier XE and Intel® Graphics Performance Analyzers (GPA). New tools available with the Intel SDK for OpenCL Applications 2013 ease build and optimization of OpenCL applications running on Intel HD Graphics. For more information on what is new, please see the SDK release notes.

See Also Intel VTune Amplifier XE website Intel Graphics Performance Analyzers website Using the Intel SDK for OpenCL Applications - Kernel Builder Working with the Intel SDK for OpenCL Applications - Offline Compiler - Command-Line Version

Related Products The following is the list of products related to Intel® SDK for OpenCL* Applications.

•

Intel® Graphics Performance Analyzers (Intel® GPA)

•

Intel® VTune™ Amplifier XE

•

Intel® Media SDK

•

Intel® Perceptual Computing SDK.

Overview of OpenCL* Support on Intel® Core™ Processors Components and Packages For full OpenCL* support on 3rd and 4th Generation Intel® Core™ Processors, the following components are available.

Intel® SDK for OpenCL* Applications 2013 Package with OpenCL* 1.2

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

The SDK package is a comprehensive software developer kit for OpenCL* applications on Intel® Architecture processors. This package includes Intel OpenCL headers, libraries, developer tools like kernel builder and kernel debugger, and integration with profiling tools. This SDK package also includes a CPU only runtime package with fully conformant support for OpenCL 1.2 features. The SDK installer also includes the standalone CPU only OpenCL 1.2 runtime.

Intel SDK for OpenCL - CPU Runtime Only Package with OpenCL 1.2 This standalone CPU only OpenCL 1.2 runtime can be installed on all generations of Intel Core Processors. It is also available for applications to ensure backward compatibility and coverage for Intel Core Processors without Intel® HD Graphics. This CPU runtime includes redistribution rights.

OpenCL 1.2 Support on CPU and Intel® HD Graphics The latest version of the Intel® HD Graphics driver includes the OpenCL* runtime and compiler for both the Intel® Processor (CPU) and the Intel HD Graphics (GPU). This driver is available with 3rd and 4th generation Intel Core Processors with the Intel HD Graphics. To develop OpenCL applications with the SDK, you need to download the latest edition of the driver. OpenCL applications developed with the Intel SDK for OpenCL Applications 2013 can run seamlessly on a wide range of the 3rd and the 4th Generation Intel Core Processors with Intel HD Graphics and preinstalled Intel HD Graphics driver.

Code Samples Free downloaded code samples and optimization best known methods for OpenCL* on Intel® Architecture are available online thought SDK page at intel.com/software/opencl/.

Intel® SDK for OpenCL* Applications Website The SDK web site at intel.com/software/opencl provides links to videos, technical articles, case studies, and support forums and is an integral component of this SDK.

The Intel SDK for OpenCL Applications Software Stack Use the Intel SDK for OpenCL Applications 2013 to develop applications and then seamlessly deploy and run them with the Intel HD Graphics driver.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

See Also Intel SDK for OpenCL Applications - Samples Getting Started Intel® Driver Update Utility website

The OpenCL* Platform on Intel® Architecture Processors Intel provides a single OpenCL* platform to seamlessly access the compute power across both Intel® Processors (CPUs) and Intel® HD Graphics in a standard manner. The Platform model for OpenCL on the 3rd and the 4th Generation Intel® Core™ processor family is defined in the following figure.

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

The OpenCL platform consists of two OpenCL devices: 1. Intel® Processor – a CPU device 2. Intel® HD Graphics – a GPU device.

Platform Info ID

Value

CL_PLATFORM_VENDOR Intel® Corporation CL_PLATFORM_NAME

Intel® OpenCL

NOTE The Intel platform on Intel® Xeon® processors or on previous generations of Intel Core processors includes only the CPU device. The following is the device type ID for Intel hardware with OpenCL 1.2 full profile: Device Type ID

Hardware Device

CL_DEVICE_TYPE_GPU Intel HD Graphics CL_DEVICE_TYPE_CPU Intel Core Processor

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

See Also Intel® GPA website

Using the Intel® Graphics Performance Analyzers (Intel® GPA) Platform Analyzer Introducing the Intel® Graphics Performance Analyzers Platform Analyzer The Intel® GPA Platform Analyzer visualizes the execution profile of the various tasks in your code over time. The Intel GPA Monitor collects real-time trace data during the application run and saves it to a trace capture file. The Intel GPA Platform Analyzer opens this file and provides a system-wide picture of: 1. how your code executes on software and hardware threads 2. how your code works with Intel® Media SDK and Microsoft* DXVA2 and how media-related workloads execute on various CPU and GPU cores in your system The tool infrastructure automatically aligns clocks across all cores in the entire system so that you can analyze CPU-based workloads together with GPU-based workloads within a unified time domain. The Intel SDK for OpenCL Applications 2013 provides built-in capturing support that takes advantage of the Intel GPA Platform Analyzer for execution of OpenCL kernels on CPU only. With the Intel GPA Platform Analyzer, you can trace the execution profiles of various OpenCL tasks on the CPU over a period of time and analyze it together with the graphics workloads (Microsoft DirectX* API) running on the GPU.

NOTE Intel SDK for OpenCL Applications 2013 is designed to work with Intel® GPA version 2012 R4 or higher. Intel SDK for OpenCL Applications 2013 provides built-in capturing support, with the following profiling features: 1. Context view - (available on Intel GPA 4.2 or higher) - enables you to examine the flow of the OpenCL* commands and their dependencies within the application context command queues. 2. API tracing - enables you to capture and measure the time of the application OpenCL API calls. 3. GPU Counters - provide GPU profiling by inspecting a set of GPU performance counter. This is applicable only for OpenCL tasks that run on Intel GPU device. 4. Device view (CPU only) - enables you to see the distribution of the application OpenCL commands (kernels and memory operations) across the system software threads. You can use the generated trace capture file later to identify critical bottlenecks, analyze information on the execution flow, and to improve your application performance.

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

Enabling OpenCL* API Tracing The OpenCL* API Tracing feature enables you to view detailed timing information about your OpenCL* workload by inspecting the time the host API took to execute. It demonstrates which certain parts of the Host API took more time to execute. The OpenCL* API Tracing is disabled by default. To enable it, you need to control the following environment variables:

Environment Variable

Default Value

CL_GPA_CONFIG_ENABLE_API_TRACING True

Generating a Trace Capture File You can generate OpenCL* traces using the Profiler mechanism or runtime instrumentation (for CPU device only).

Using Profiler Mechanism This mechanism uses a special DLL to capture the OpenCL information, needed for Intel速 GPA trace generation. This method enables all profiling features. You can use it to generate traces for tasks running on Intel速 HD Graphics (or any other OpenCL device). To work in the profiling mode, you should make several changes to the system registry. These changes include changing the target runtime DLL (instead of the OpenCL driver) and creating several Registry variables used by the Profiler.

Using Runtime Instrumentation (CPU only). This method generates OpenCL tracing information by reporting the current state within the OpenCL driver. This method provides low level of information and is limited to tasks executed on CPU device only.

NOTE The CPU runtime instrumentation is activates automatically when you switch to profiling mode. You can manually enable or disable it through the following environment variable

Environment Variable Value CL_CONFIG_USE_GPA

True/False

Intel速 SDK for OpenCL* Application provides a configuration utility which will help developers to enable (and disable) Profiling mode and make all necessary changes in system registry.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

When running the profiling control, the current working mode is selected. To see the working mode, in the profiler configuration window click the requested mode, then click the Apply button.

NOTE The OpenCL Profiler Configuration Utility may cause an overhead penalty. Expect OpenCL programs to run slower. To restore performance, set back the original OpenCL settings after you finish using the Profiler to generate GPA traces. To set back the original OpenCL settings,run the configuration utility and choose Use Intel OpenCL Platform, then click Apply and Exit To generate a trace file follow these steps: 1. Enable Intel GPA tracing by launching the Profiling configuration utility and switching to profiling mode (in case using Profiling mode) or setting the required environment variable described in the Enabling OpenCL API Tracing section (in case of using runtime tracing mode). 2. Run the Intel GPA Monitor application.

3. Click the Manage Profiles... button and check the Capture Application Startup option under the Tracing tab.

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

4. Add the full path of the executable in the Command Line text box and click the Run button to run the application. 5. When the execution finishes, the trace file appears under the output folder of Intel GPA trace files, for example: C:\Users\<user name>\Documents\GPA_2012_R4 where <user name> is the name of your local user account. You can load the generated OpenCL trace data file into the Intel GPA applications. After loading the trace file, tasks and markers items appear in the application timeline view.

See Also Enabling OpenCL API Tracing Intel® GPA Getting Started Guide

OpenCL* Device View The OpenCL* Device View enables you to see the distribution of the application OpenCL* commands (kernels and memory operations) across the system software threads.

Tasks Two types of tasks are available on the timeline tracker:

•

The OpenCL Kernels. Each task on the timeline tracker represents a cluster of workgroups which belongs to the same kernel. All tasks which belong to the same kernel are marked is the same color; each kernel has a different color.

•

Memory operations. In addition to the OpenCL kernels, there are tasks on the timeline tracker which represent memory operations: Read, Write, Copy, Map and Un-map. All of these commands are red colored ( ).

Markers For each command in the queue, the trace enables you to see a marker on the time ruler, indicating the state of the command. The following markers are supported:

• •

A marker indicating the time command was entered to the commands queued. A marker indicating the time command submitted to the device.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

• •

A marker indicating the time command started running. A marker indicating the time command completed.

The default view shows the queued and the completed markers only.

Controlling the Markers Display You can change the markers display in the time ruler by adding and modifying the following environment variables: Environment Variable

Default Value

CL_GPA_CONFIG_SHOW_QUEUED_MARKER

True

CL_GPA_CONFIG_SHOW_SUBMITTED_MARKER False CL_GPA_CONFIG_SHOW_RUNNING_MARKER

False

CL_GPA_CONFIG_SHOW_COMPLETED_MARKER True

OpenCL* Context View The OpenCL* Context View enables you to examine the flow of the OpenCL commands and their dependencies within the application context command queues. You can view detailed timing information about your OpenCL workload by inspecting the time required to queue commands for the device, and the time the device took to execute items from the queue.

It enables you to know which commands took more time to execute. Each track in the context view represents an OpenCL commands queue. These tracks can be easily identified by the turquoise ( ) color of their left headers. Two types of tracks exist:

• •

in-order queue (representing in-order OpenCL commands queue) out-of-order queue (representing out-of-order OpenCL commands queue).

The tasks within these tracks represent OpenCL commands (OpenCL kernels or memory operations). All commands have different colors. Each task is logically divided into two parts:

• •

The wait time of an OpenCL command in the queue, before scheduling it for execution on the device. The actual execution time of an OpenCL command.

The two parts share the same color, although with different brightness levels. The colors of the commands in the context view trackers are identical to the colors of the corresponding commands in the device view tracks.

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

GPU Counters When using Profiler, you can inspect GPU counters values for OpenCL* tasks running on Intel速 HD Graphics (GPU) device. You can access GPU counters by clicking any OpenCL task in the Context view then clicking the Metadata tab, which appears in the bottom-right window.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

GPU Counters Description Counter Name

Counter Explanation

Compute Shader Active Time

Total time the compute shader spent active on all cores.

Compute Shader Stall Time – Core Stall

Total time in clocks the compute shader spent stalled on all cores – and the entire core was stalled as well.

Number of CS threads loaded

Number of CS threads loaded at any given time in the EUs.

Using the Intel® Graphics Performance Analyzers System Analyzer Intel® Graphics Performance Analyzers 2013 (Intel® GPA) introduced support of various metrics collections in the Intel GPA System Analyzer. You can use the following types of OpenCL*-related metrics: 1. CPU-specific metrics, such as core utilization 2. Execution unit (EU) metrics, of which GPU EU Active, Idle, or Stalled are most important

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

3. Memory metrics, such as GPU Memory Read and Write operations. 4. Power metrics for CPU, Intel® HD Graphics, and both devices simultaneously. Refer to the Intel GPA Online Help for details. You can also use the Intel GPA System Analyzer to profile non-DirectX* applications, if they are instrumented by Intel® ITT API. In this case the Metric Tree Control panel shows only CPU metrics.

NOTE “GPU busy” metric does not include general computations like OpenCL*, so it actually means that the GPU is currently processing 3D rendering contexts. The following screenshot shows the Intel GPA System Analyzer in action:

Metric Tree Control Panel on the left side displays the tree of available metrics. Metric Chart Area shows up to 16 charts that visualize the metrics performance. To add a metric to the chart, drag and drop it from the list of available metrics. In the screenshot above, you can see the following metrics beyond the general “Frame Time” chart:

• •

Family of execution unit metrics for this specific application. The percentage of GPU execution units stalled is quite high, which might indicate high bandwidth pressure (below) or a stall in other fixed function units, such as mathbox. GPU Memory Reads/Writes. The traffic of read\write operations is intensive. Use the Intel® SDK for OpenCL* Applications Optimization Guide for hints on memory performance.

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

Refer to the Intel GPA Online Help for information on how to inspect metrics and find performance bottlenecks.

See Also Intel SDK for OpenCL Applications - Optimization Guide Intel GPA website

Interoperability with the Intel® VTune™ Amplifier XE Introducing the Intel® VTune™ Amplifier XE Intel® SDK for OpenCL* Applications 2013 enables you to see the assembly code of your OpenCL kernels Just-in-time (JIT) code and to analyze its performance through sampling profiling (call-graph profiling is not supported in version 2012) using the graphical interface of the Intel® VTune™ Amplifier XE Performance Profiler tool. You must acquire the Intel VTune Amplifier XE 2011 separately. For more information, see the VTune Amplifier XE 2011 page at http://software.intel.com/en-us/articles/intel-vtune-amplifier-xe/.

NOTE The profiling support of Intel SDK for OpenCL Applications 2013 is designed to work with the Intel VTune Amplifier XE 2011 and Intel VTune Performance Analyzer 9.1 or higher. Using other versions may cause undefined results.

See Also Setting a New Profiling Project Intel® VTune™ Amplifier XE 2011 at Visual Computing Source portal

Setting a New Profiling Project The instructions below refer to the Intel® VTune™ Amplifier XE 2011. Take into consideration that user interface of the Intel VTune Performance Analyzer may differ slightly. To run profiling on the Intel VTune Amplifier XE follow these steps: 1. Create a new sampling project by selecting File > New > Project... 2. In the Create a Project dialog, enter your new project name and click Create Project. 3. In the Project Properties window, select the application to run, including the working directory and command arguments. 4. Click Modify... next to the user-defined environment variables text box and add the following line to the User-defined Environment Variables table: CL_CONFIG_USE_VTUNE=True 5. Click OK in all open windows to save the new settings. 6. Click the New Analysis button to run and analyze your application. 7. In the Analysis Type window, select the type you need and click Start to run the analysis.

Viewing the OpenCL* Kernel’s Assembly Code To view the OpenCL* Kernel Assembly Code, do the following:

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

1. Wait until the sampling activity finishes. 2. click the Hotspots Bottom-up button at the navigation toolbar. 3. Select the /Function option in the Data Grouping selection box and look for your application OpenCL kernels in the Functions data grid:

If you are running several applications with identical OpenCL kernel names simultaneously, select the /Process /Function /Thread option in the Data Grouping selection box and look for the application process in the Processes data grid. The OpenCL kernel that belongs to the application is located under the application name in the application tree view, as shown in the example below:

4. Double-click the selected OpenCL kernel to see the assembly source code and its relevant sampling information:

Source Level Profiling of Your OpenCL* Kernel Through the source level profiling mode, you can perform hotspots analysis against the original OpenCL* kernel code. Hotspots analysis enables you to analyze kernel performance and bottlenecks of

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

an OpenCL* kernel through sampling profiling using the GUI of the Intel® VTune™ Amplifier XE Performance Profiler tool. In the build options string parameter in the clBuildProgram function: 1. Add the –profiling flag. 2. Specify the full path of the file -s <full path to the OpenCL* source file> If the path includes spaces, enclose the entire path with double quotes (“<path_to_opencl_source_file>\”). For example: err = clBuildProgram( g_program, 0, NULL, “-profiling -s \“<path_to_opencl_source_file>\””, NULL, NULL);

See Also Interoperability with the Intel® VTune™ Amplifier XE on Linux* OS

OpenCL* Tracing Support with the Intel® VTune™ Amplifier XE Introducing OpenCL* Tracing The Intel® Vtune™ Amplifier XE provides an experimental feature, which supports visual display of the execution profile of the various tasks in your code over a period of time. The tool collects real-time trace data during the application run and provides a thread-oriented view of you application. Besides OpenCL* commands, it also collects system-level threading issues, and Intel® Threading Building Blocks (Intel® TBB) events, to give you a detailed view of the running system. To download Intel Vtune Amplifier XE, go to: http://software.intel.com/en-us/articles/vcsource-tools-vtune-amplifier-xe/.

Capturing OpenCL* Trace You need to do the following to configure the Intel® Vtune™ Amplifier XE work: 1. Set the environment variables: o AMPLXE_EXPERIMENTAL=all o CL_CONFIG_USE_GPA=true o CL_CONFIG_USE_ITT=true

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

2. In the Intel Vtune Amplifier XE, invoke the New Analysis window and set Analyze user tasks check box.

3. Click Start. 4. Once the collection is done, switch to Bottom-Up view to see the tasks collected in the timeline:

5. For task-focused analysis, switch to Task Analysis viewpoint:

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

6. Use the Task Time and CPU Time panes to view the task data:

Appendix A - Supported Images Formats Read-Only Surface Formats cl_channel_order cl_channel_type

GPU Device CPU Device

CL_RGBA

CL_UNORM_INT8

Yes

CL_RGBA

CL_UNORM_INT16

Yes

CL_RGBA

CL_SIGNED_INT8

Yes

CL_RGBA

CL_SIGNED_INT16

Yes

CL_RGBA

CL_SIGNED_INT32

Yes

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

CL_RGBA

CL_UNSIGNED_INT8

Yes

CL_RGBA

CL_UNSIGNED_INT16

Yes

CL_RGBA

CL_UNSIGNED_INT32

Yes

CL_RGBA

CL_HALF_FLOAT

Yes

CL_RGBA

CL_FLOAT

Yes

CL_R

CL_FLOAT

Yes

CL_R

CL_UNORM_INT8

Yes

CL_R

CL_UNORM_INT16

Yes

CL_R

CL_SIGNED_INT8

Yes

CL_R

CL_SIGNED_INT16

Yes

CL_R

CL_SIGNED_INT32

Yes

CL_R

CL_UNSIGNED_INT8

Yes

CL_R

CL_UNSIGNED_INT16

Yes

CL_R

CL_UNSIGNED_INT32

Yes

CL_R

CL_HALF_FLOAT

Yes

CL_INTENSITY

CL_UNORM_INT8

Yes

CL_INTENSITY

CL_UNORM_INT16

Yes

CL_INTENSITY

CL_HALF_FLOAT

Yes

CL_INTENSITY

CL_FLOAT

Yes

CL_LUMINANCE

CL_UNORM_INT8

Yes

CL_LUMINANCE

CL_UNORM_INT16

Yes

CL_LUMINANCE

CL_HALF_FLOAT

Yes

CL_LUMINANCE

CL_FLOAT

Yes

CL_A

CL_UNORM_INT8

Yes

CL_A

CL_UNORM_INT16

Yes

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

CL_A

CL_HALF_FLOAT

Yes

CL_A

CL_FLOAT

Yes

CL_RG

CL_UNORM_INT8

Yes

CL_RG

CL_UNORM_INT16

Yes

CL_RG

CL_SIGNED_INT8

Yes

CL_RG

CL_SIGNED_INT16

Yes

CL_RG

CL_SIGNED_INT32

Yes

CL_RG

CL_UNSIGNED_INT8

Yes

CL_RG

CL_UNSIGNED_INT16

Yes

CL_RG

CL_UNSIGNED_INT32

Yes

CL_RG

CL_HALF_FLOAT

Yes

CL_RG

CL_FLOAT

Yes

Write-Only Surface Formats cl_channel_order cl_channel_type

GPU Device CPU Device

CL_RGBA

CL_UNORM_INT8

Yes

CL_RGBA

CL_UNORM_INT16

Yes

CL_RGBA

CL_SIGNED_INT8

Yes

CL_RGBA

CL_SIGNED_INT16

Yes

CL_RGBA

CL_SIGNED_INT32

Yes

CL_RGBA

CL_UNSIGNED_INT8

Yes

CL_RGBA

CL_UNSIGNED_INT16

Yes

CL_RGBA

CL_UNSIGNED_INT32

Yes

CL_RGBA

CL_HALF_FLOAT

Yes

CL_RGBA

CL_FLOAT

Yes

CL_R

CL_FLOAT

Yes

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

CL_R

CL_UNORM_INT8

Yes

CL_R

CL_UNORM_INT16

Yes

CL_R

CL_SIGNED_INT8

Yes

CL_R

CL_SIGNED_INT16

Yes

CL_R

CL_SIGNED_INT32

Yes

CL_R

CL_UNSIGNED_INT8

Yes

CL_R

CL_UNSIGNED_INT16

Yes

CL_R

CL_UNSIGNED_INT32

Yes

CL_R

CL_HALF_FLOAT

Yes

CL_INTENSITY

CL_UNORM_INT8

CL_INTENSITY

CL_UNORM_INT16

CL_INTENSITY

CL_HALF_FLOAT

CL_INTENSITY

CL_FLOAT

Yes

CL_LUMINANCE

CL_UNORM_INT8

CL_LUMINANCE

CL_UNORM_INT16

CL_LUMINANCE

CL_HALF_FLOAT

CL_LUMINANCE

CL_FLOAT

Yes

CL_A

CL_UNORM_INT8

Yes

CL_A

CL_UNORM_INT16

CL_A

CL_HALF_FLOAT

CL_A

CL_FLOAT

CL_RG

CL_UNORM_INT8

Yes

CL_RG

CL_UNORM_INT16

Yes

CL_RG

CL_SIGNED_INT8

Yes

CL_RG

CL_SIGNED_INT16

Yes

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

CL_RG

CL_SIGNED_INT32

Yes

CL_RG

CL_UNSIGNED_INT8

Yes

CL_RG

CL_UNSIGNED_INT16

Yes

CL_RG

CL_UNSIGNED_INT32

Yes

CL_RG

CL_HALF_FLOAT

Yes

CL_RG

CL_FLOAT

Yes

Read-Write Surface Formats cl_channel_order cl_channel_type

GPU Device CPU Device

CL_RGBA

CL_UNORM_INT8

Yes

CL_RGBA

CL_UNORM_INT16

Yes

CL_RGBA

CL_SIGNED_INT8

Yes

CL_RGBA

CL_SIGNED_INT16

Yes

CL_RGBA

CL_SIGNED_INT32

Yes

CL_RGBA

CL_UNSIGNED_INT8

Yes

CL_RGBA

CL_UNSIGNED_INT16

Yes

CL_RGBA

CL_UNSIGNED_INT32

Yes

CL_RGBA

CL_HALF_FLOAT

Yes

CL_RGBA

CL_FLOAT

Yes

CL_R

CL_FLOAT

Yes

CL_R

CL_UNORM_INT8

Yes

CL_R

CL_UNORM_INT16

Yes

CL_R

CL_SIGNED_INT8

Yes

CL_R

CL_SIGNED_INT16

Yes

CL_R

CL_SIGNED_INT32

Yes

CL_R

CL_UNSIGNED_INT8

Yes

Intel速 SDK for OpenCL* Applications - User's guide for Windows* OS

CL_R

CL_UNSIGNED_INT16

Yes

CL_R

CL_UNSIGNED_INT32

Yes

CL_R

CL_HALF_FLOAT

Yes

CL_INTENSITY

CL_UNORM_INT8

CL_INTENSITY

CL_UNORM_INT16

CL_INTENSITY

CL_HALF_FLOAT

CL_INTENSITY

CL_FLOAT

Yes

CL_LUMINANCE

CL_UNORM_INT8

CL_LUMINANCE

CL_UNORM_INT16

CL_LUMINANCE

CL_HALF_FLOAT

CL_LUMINANCE

CL_FLOAT

Yes

CL_A

CL_UNORM_INT8

Yes

CL_A

CL_UNORM_INT16

CL_A

CL_HALF_FLOAT

CL_A

CL_FLOAT

CL_RG

CL_UNORM_INT8

Yes

CL_RG

CL_UNORM_INT16

Yes

CL_RG

CL_SIGNED_INT8

Yes

CL_RG

CL_SIGNED_INT16

Yes

CL_RG

CL_SIGNED_INT32

Yes

CL_RG

CL_UNSIGNED_INT8

Yes

CL_RG

CL_UNSIGNED_INT16

Yes

CL_RG

CL_UNSIGNED_INT32

Yes

CL_RG

CL_HALF_FLOAT

Yes

CL_RG

CL_FLOAT

Yes

Intel® SDK for OpenCL* Applications – User’s Guide for Windows* OS

Appendix B – OpenCL* Build and Linking Options Preprocessor Options Option

Description

-D <name>

Predefines name as a macro, with definition 1

GPU Device CPU Device

The contents of definition are tokenized and processed as if they appeared during -D <name=definition> translation phase three in a #define directive. In particular, the definition will be truncated by embedded newline characters -I <dir>

Add the directory <dir> to the list of directories to be searched for header files

Yes

Math Intrinsics Options GPU Device

CPU Device

Option

Description

-cl-single-precisionconstant

Treats double precision floating-point constant as single precision constant

Yes

-cl-denorms-are-zero

This option controls how single precision and double precision denormalized numbers are handled. If specified as a build option, the single precision denormalized numbers may be flushed to zero; double precision denormalized numbers may also be flushed to zero if the optional extension for double precision is supported

Yes

GPU Device

CPU Device

Optimization Options Option

Description

-cl-opt-disable

This option disables all optimizations. Optimizations are enabled by default.

Yes

-cl-mad-enable

Enables a * b + c to be replaced by a mad. The mad computes a * b + c

Yes

Intel® SDK for OpenCL* Applications - User's guide for Windows* OS

with reduced accuracy.

-cl-no-signed-zeros

Enables optimizations for floating-point arithmetic that ignore the signedness of zero. IEEE 754 arithmetic specifies the distinct behavior of +0.0 and -0.0 values, which then prohibits simplification of expressions such as x+0.0 or 0.0*x (even with-clfinite-math only). This option implies that the sign of a zero result is not significant.

Yes

Enables optimizations for floating-point arithmetic that, -cl-unsafe-mathoptimizations

1. assume that arguments and results are valid 2. may violate IEEE 754 standard 3. may violate the OpenCL* numerical compliance requirements.

-cl-finite-math-only

Enables optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or >±∞.

Yes

-cl-fast-relaxed-math

Sets the optimization options -cl-finite-math-only and -cl-unsafe-math-optimizations. This enables optimizations for floating-point arithmetic that may violate the IEEE 754 standard and the OpenCL* numerical compliance requirements.

Yes

Options for Warnings Option Description

GPU Device CPU Device

Inhibit all warning messages

Yes

-Werror Make all warnings into errors

Yes

-w

Options Controlling the OpenCL* C Version Option

Description

-cl-std= Determine the OpenCL* C language version to use.

GPU Device CPU Device Yes

Yes