Skip to content
This repository has been archived by the owner on May 17, 2023. It is now read-only.

Latest commit

 

History

History
292 lines (201 loc) · 27.2 KB

mediasdk-tutorials-readme.md

File metadata and controls

292 lines (201 loc) · 27.2 KB

Intel® Media SDK Tutorials

LEGAL DISCLAIMER

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.

UNLESS OTHERWISE AGREED IN WRITING BY INTEL, THE INTEL PRODUCTS ARE NOT DESIGNED NOR INTENDED FOR ANY APPLICATION IN WHICH THE FAILURE OF THE INTEL PRODUCT COULD CREATE A SITUATION WHERE PERSONAL INJURY OR DEATH MAY OCCUR.

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 by visiting Intel's Web Site.

MPEG is an international standard for video compression/decompression promoted by ISO. Implementations of MPEG CODECs, or MPEG enabled platforms may require licenses from various entities, including Intel Corporation.

Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.

*Other names and brands may be claimed as the property of others.

Copyright © 2007-2020, Intel Corporation. All Rights reserved.

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

Table of contents

Introduction

Exploring Media SDK via a simplified set of samples

The Intel® Media Software Development Kit (Intel® Media SDK) gives developers access to specialized hardware acceleration capabilities for video decoding, encoding, and pre/postprocessing (VPP). Intel® Media SDK helps developers rapidly write software that accesses hardware acceleration for video codecs.

This quick-start tutorial illustrates how to use Intel® Media SDK by stepping from simple to complex. Concepts are added in layers. You are encouraged to use your favorite file difference tool to compare what was added in each step. While the tutorial cannot cover every possible usage, our goal is to provide starting points to shrink your TTM for a wide range of applications.

For simplicity and uniformity the tutorial focuses on the H.264 (AVC) video codec. Other codecs supported by Intel® Media SDK can be utilized in a similar way, often by a single configuration parameter. This unified approach can save significant development time and is a major advantage to working with Media SDK.

Detailed comments explain the behavior of the code. You are encouraged to read and experiment. For a deeper understanding of the SDK and details on specific parameters, please refer to SDK API Reference Manual. The set of samples packaged with the SDK showcases many more parameters and scenarios and are a valuable reference as well.

All tutorial samples are self-contained unless otherwise noted. To reduce redundancy and improve readability common code segments are located in the ‘common’ folder. These are tied to functions not directly related to Intel® Media SDK functionality, such as read/write of bit streams and raw frames. The ‘common’ folder also contains OS-specific device handling and surface allocation implementations.

The tutorial samples were developed and tested using recent versions of Intel® Media SDK for Windows and Linux. For questions, or to report issues with the tutorial samples presented in this article, please use MediaSDK GitHub.

Windows specific notes

  • Please note, some Windows tutorials also require installation of Microsoft Windows SDK*.
  • Since the introduction of Microsoft Window 8, Media SDK can now also be used with DirectX 11 devices and surfaces. Media SDK relies on features in DirectX 11.1, and therefore can't be used on Microsoft Windows 7. If target application must run on Microsoft Windows 7 then Media SDK DirectX9 capabilities should be used instead.

How to obtain input content

The tutorial assumes that the user has proper content to play with. Such content can be acquired from many sources on the web. One example is the Peach open movie project "Big Buck Bunny", which can be downloaded from here. The video elementary stream must be extracted before using. This can be done using ffmpeg with:

ffmpeg \
-i big_buck_bunny_1080p_h264.mov \
-an -vcodec copy -bsf h264_mp4toannexb \
-f h264 bbb1920x1080.264

You can decode this stream to raw YUV format using ffmpeg, the tutorial decode examples, or Media SDKs sample decode.

ffmpeg -i bbb1920x1080.264 bbb1920x1080.yuv
sample_decode h264 -i bbb1920x1080.264 -o bbb1920x1080.yuv

Storing all frames in raw format can take a lot of space. However, if disk capacity is a concern, the samples and tutorials will work on a subset of frames. Number of frames processed can be controlled via -vframes for ffmpeg or changing the decode sample/tutorial code.

Tutorials Samples Index

Tutorials are devided into few sections:

Section Description
1 Media SDK "Hello World". Start and query a session
2-4 Decode, encode, and VPP as single component pipelines
5 Transcode: The most common compound workload
6 Advanced/compound Media SDK scenarios
Name Description
simple_1_session Sets up Intel® Media SDK session and perform queries to determine selected implementation and which API version is used
simple_2_decode Decodes AVC stream into YUV file using system memory surfaces, showcasing simple synchronous decode pipeline flow
simple_2_decode_vmem Adds use of video memory surfaces for improved decode performance
simple_2_decode_hevc10 Decodes HEVC 10bit stream into YUV file using system memory surfaces, showcasing simple synchronous decode pipeline flow
simple_3_encode Encodes YUV frames from file into AVC stream using surfaces in system memory, showcasing simple synchronous encode pipeline flow
simple_3_encode_vmem Adds use of video memory surfaces for improved encode performance
simple_3_encode_vmem_async Adds asynchronous operation to previous example, resulting in further improved performance
simple_3_encode_hevc10 Encodes YUV 10 bit frames from file into HEVC stream using surfaces in system memory, showcasing simple synchronous encode pipeline flow
simple_4_vpp_resize_denoise Showcases video frame processing (VPP) using system memory surfaces. Highlights frame resize and denoise filter processing
simple_4_vpp_resize_denoise_vmem Adds use of video memory surfaces for improved VPP performance
simple_5_transcode Transcodes (decode+encode) AVC stream to another AVC stream using system memory surfaces
simple_5_transcode_opaque Same as previous sample but uses the Intel® Media SDK opaque memory feature. The opaque memory type hides surface allocation specifics and allows the SDK to select the best type for the execution in HW or SW
simple_5_transcode_opaque_async Adds asynchronous operation to the transcode pipeline implementation, resulting in further improved performance
simple_5_transcode_vmem Same as "simple_5_transcode" sample but uses video memory surfaces instead. While opaque surfaces use video memory internally, application-level video memory allocation is required to integrate components not in Media SDK.
simple_5_transcode_opaque_async vppresize Same as "simple_5_transcode_opaque_async" sample but pipeline includes VPP resize.
simple_6_decode_vpp_postproc Similar to the simple_2_decode sample but adds VPP post-processing capabilities to showcase resize and ProcAmp
simple_6_encode_vmem_lowlatency Similar to the simple_3_encode_vmem sample with additional code to illustrate how to configure an encode pipeline for low latency and how to measure latency
simple_6_transcode_opaque_lowlatency Similar to the simple_5_transcode_opaque sample with additional code to illustrate how to configure a transcode pipeline for low latency and how to measure latency
simple_6_encode_vmem_vpp_preproc Similar to the simple 3_encode_vmem sample but adds VPP preprocessing capabilities to show frame color conversion from RGB32(4) to NV12
simple_7_encode Sets up Intel® Media SDK session and perform queries to determine device capabilities

Tutorial Section 1

Hello World

This tutorial sample showcases a very simple "hello world" type Intel® Media SDK use case.

The sample shows how to initialize an Intel® Media SDK session (MFXVideoSession) using the target selection option "MFX IMPL AUTO ANY", which is recommended as a default setting since it is appropriate for nearly all cases. Initialization Differences Between Windows and Linux There are not many differences between Media SDK for Windows and for Linux, but initialization showcases some of the main ones. Please note that the main function is the same for all operating systems but there is some OS specific code in the common directory.

Windows The Windows releases contain software and hardware implementations. MFX IMPL AUTO ANY implies the session will be initialized to use HW acceleration (regardless in which adapter the Intel HD Graphics device resides) if available on the processor. If HW acceleration is not available, the Intel® Media SDK defaults to SW implementation.

In the "initialize" function (called from main), associating a display handle with the session is not necessary except when using video memory surfaces.

Linux The Linux releases do NOT include a software implementation. MFX IMPL AUTO ANY will attempt to start the session with hardware acceleration. If the hardware implementation cannot be found initialization is not successful.

In the "initialize" function (called from main), associating a display handle with the session is ALWAYS necessary for Linux. Please refer to Working with VA API Applications section for more details.

Session queries (all OS) After initialization, the session is queried to determine the actual target (via QueryIMPL) that was selected. For Windows this could be HW or SW, though HW will be chosen if your processor and driver support accelerated media processing. For Linux the implementation can only be HW. The highest supported API version is returned via QueryVersion.

simple_7_codec

This tutorial sample showcases Intel® Media SDK features via call Query functions: MFXVideoDECODE_Query, MFXVideoENCODE_Query, MFXVideoVPP_Query. Since this tutorial check all supported coded, it will configure a set of the video parameters for each codec by the predefined filling function. Each filling function will try the maximum resolutions, user might change the resolution based on his platform

Tutorial Section 2: Decode

simple_2_decode

This Intel® Media SDK tutorial sample illustrates the most simplistic way of implementing HW decode using system memory surfaces.

The basic goal of this example is to illustrate why asynchronous operation using video memory surfaces is necessary. While it is simpler to use system memory synchronously, as in this example, this introduces unnecessary bottlenecks:

Surfaces must be copied from video memory to system memory. While this must happen in any case for decode which writes frames to disk, buffering is not as efficient in this scenario. For a single decode (or possibly even several), stalls in the processing pipeline can't be filled easily. Since the GPU is not in constant use it may fall out of turbo mode. Based on the above analysis we should be able to improve the performance of the workload by using video memory surfaces instead of system memory surfaces. The next tutorial sample will explore such scenario.

simple_2_decode_vmem

This Intel® Media SDK tutorial sample operates in the same way as the previous "simple_2_decode" sample except that it uses video memory surfaces instead of system memory surfaces.

Video memory surfaces allow greater efficiency by avoiding explicit copies. Further improvement may be achieved by making the decode pipeline asynchronous. We'll explore this approach further when we discuss encoding workloads in the following tutorial sections. Improved GPU utilization can also be achieved by executing several decode workloads concurrently.

Additional details for Windows developers: Since the introduction of Microsoft Windows* 8, Intel® Media SDK can be used with DirectX11 devices and surfaces. Note that Intel® Media SDK relies on the features part of DirectX 11.1, and can therefore not be used on Microsoft Windows 7. If your target application must run on Microsoft Windows 7, use the DirectX 9 path via Intel® Media SDK.

Tutorial samples illustrating use of D3D surfaces (such as in this sample) have two Microsoft Visual Studio solution/project (sln/prj) files - for DirectX9 and DirectX11 usages.

simple_2_decode_hevc10

This Intel® Media SDK tutorial sample operates in the same way as the "simple_2_decode" sample except that it decode H.265 10 bit stream.

Tutorial Section 3: Encode

simple_3_encode

This Intel® Media SDK tutorial sample illustrates the most simplistic way of implementing HW encode using system memory surfaces.

This is intended as a simple starting point, but implicit copies and synchronous implementation limit performance.

simple_3_encode_vmem

This Intel® Media SDK tutorial sample operates in the same way as the "sample_3_encode" workload except that it is using video memory surfaces instead of system memory surfaces.

For more details on this topic, please refer to "simple_2_decode" sample description.

By moving from system to video memory implicit copies are eliminated, thus improving GPU load and overall performance. CPU utilization should also decrease slightly.

To improve performance and achieve greater GPU utilization we must move away from the synchronous encoding approach towards asynchronous workload behavior. The next tutorial section will explore an asynchronous encode pipeline.

simple_3_encode_vmem_async

This Intel® Media SDK tutorial sample keeps multiple encode tasks in flight simultaneously, and SyncOperation() is not called until absolutely necessary (when all surface input buffers have been exhausted).

For more details on implementing with video memory please refer to "simple_2_decode" sample description.

This example achieves efficient operation with video memory surfaces and asynchronous implementation, minimizing gaps in the GPU pipeline. Decode and encode operations are optimally scheduled internally by Media SDK so that fixed function hardware is fully utilized with many operations occurring simultaneously. As an added benefit of fully utilizing the GPU, this will cause it to remain in turbo mode1 providing a further boost to performance.

Marginal throughput improvements may be achieved by executing several encode workloads concurrently.

simple_3_encode_hevc10

This Intel® Media SDK tutorial sample operates in the same way as the "simple_3_encode" sample except that it encode H.265 10 bit stream.

Tutorial Section 4: VPP

simple_4_vpp_resize_denoise

This Intel® Media SDK tutorial sample we illustrate how to utilize Intel® Media SDK to do frame processing on frame surfaces using the SDKs VPP component.

We start with the most simplistic Intel® Media SDK VPP workload that uses system memory frame surfaces. This has several disadvantages, resulting in low GPU utilization and low performance (Specifically, implicit memory copies between CPU and GPU and synchronous implementation.) Note: many VPP operations are implemented on general purpose execution units (EUs) and do not directly use fixed function hardware. To address the above performance issues lets explore a modified VPP workload that uses video memory surfaces instead.

simple_4_vpp_resize_denoise_vmem

This Intel® Media SDK tutorial sample operates in the exact same way as the previous tutorial sample, "simple_4 vpp_resize_denoise" except that it is using video memory surfaces.

For more details on using video memory please refer to "simple_2_decode" sample description.

As with the previous workload, the RunFrameVPPAsync() call leads to the VPP Submit operation. However, in this case the task is submitted to the GPU almost immediately since the input surface already resides on a video memory surface.

Like the encode workloads discussed earlier (such as simple 3_encode_vmem), Intel® Media SDK uses a polling mechanism, via VPP Query, to determine if the GPU has fully processed the frame.The GPU is queried every 1ms until the frame is ready. Since the current workload is synchronous the SyncOperation() call will wait until the next VPP Query which can cause large gaps in GPU execution.

We already demonstrated how to achieve greater performance by making the Intel® Media SDK work in an asynchronous fashion, as in the simple 3_encode_vmem async workload. The same approach can be used for VPP processing so we will not explore this case further. It suffices to say that GPU utilization and performance can be improved significantly by applying the task concurrency approach for VPP.

This concludes the analysis of Intel® Media SDK decode, encode and VPP workloads. The next tutorial sections explore the behavior of workloads combining several Intel® Media SDK components.

This tutorial sample is found in the tutorial samples package under the name "simple_4_vpp_resize_denoise_vmem". The code is extensively documented with inline comments detailing each step required to setup and execute the use case.

Tutorial Section 5: Transcode

simple_5_transcode

This Intel® Media SDK tutorial we will explore transcode workloads, starting with the most simplistic transcode sample using system memory frame surfaces.

In this simple implementation there are barriers to full GPU utilization, as with the other examples: System memory adds implicit copies when using hardware acceleration Synchronous implementation means less efficient internal scheduling of decode/encode stages. Not as many opportunities to keep hardware pipeline fully loaded. Since we are using system memory surfaces we must copy the decoded surface to system memory first, then before encode the surface will be copied to video memory again. Both copy operations have a large impact on CPU load and performance.

As noted when exploring the encode workloads, the Encode Query polling method also introduces a slight inefficiency in the pipeline after the GPU has completed the encoding task.

The performance is also indirectly degraded by the fact that the GPU remains in lower frequency states due to the relatively low GPU activity.

In the following sections we will explore how to enhance Intel® Media SDK transcode pipelines for improved GPU utilization leading to better performance.

simple_5_transcode_opaque

This Intel® Media SDK tutorial sections covering encode, decode and VPP the performance issues caused by relying on system memory were removed by using video memory surfaces. "Opaque memory" surfaces offer a simplified path to video memory optimization when working with transcode pipelines implemented entirely with Media SDK. With opaque memory surfaces are managed entirely by Media SDK so it can automatically optimize for the type of session requested. Video memory will be used for hardware sessions, system memory for software sessions.

The use of opaque memory eliminates two surface copies per frame which leads to lower CPU utilization and higher GPU utilization. As with encode, synchronous implementation is insufficient to keep the hardware pipeline busy and introduces many gaps. These gaps can be filled with work from the same video sequence with asynchronous implementation. Further efficiency can be gained by working with multiple transcodes simultaneously. For best results, individual pipelines should be asynchronous with multiple sessions running simultaneously.

simple_5_transcode_opaque_async

This Intel® Media SDK tutorial transcode sample we introduce asynchronous pipeline behavior using the same approach as we did in the simple_3_encode_vmem_async sample.

Overall GPU utilization can be improved significantly by implementing asynchronously. Since the GPU is highly utilized, the overall performance is also improved by the fact that the GPU is consistently residing in a high frequency state (due to Intel® Turbo Boost Technology1).

simple_5_transcode_opaque_async_vppresize

This Intel® Media SDK tutorial sample is essentially the same as the "simple_5_transcode_opaque_async" sample except that VPP processing is used to resize the content.

simple_5_transcode_vmem

This Intel® Media SDK tutorial sample is essentially the same as the "simple_5_transcode" sample except for that it uses D3D surfaces instead of system memory surfaces.

Like the "simple_2_decode" tutorial sample this sample supports both Microsoft DirectX* 9 and DirectX* 11 for Windows and VAAPI for Linux. For more details on this topic please refer to "simple_2_decode" sample description.

The use of surfaces in video memory leads to improved performance. In essence the behavior of this workload is the same as for "simple_5_transcode_opaque" (when executed on a processor that supports HW acceleration). However, since the application manages video memory these surfaces are available to integrate with components that are not in the standard Intel® Media SDK decode -> process -> encode pipeline.

Tutorial Section 6: Advanced Media SDK

How to create low latency pipelines and how to benchmark latency

Low latency video codec pipelines are important for many workloads, one example is video conferencing where minimal latency is desired. Media SDK supports configuration of encoder and decoder for low latency. Using a low latency configuration results in lower pipeline throughput, but for this specific use case its not an issue. Note that Media SDK also supports a wide range of other features useful for developing video conferencing or dynamic video streaming solutions. More information on how to use Media SDK for low latency workloads and other typical video conferencing usages can be found in this white paper: http://software.intel.com/en-us/articles/video-conferencing-features-of-intel-media-software

simple_6_encode_vmem_lowlatency

This sample illustrates how to configure an encode pipeline for low latency and how to measure latency.

simple_6_transcode_opaque_lowlatency

This sample illustrates how to configure Intel® Media SDK transcode pipeline for low latency and how to measure latency.

The approach is very similar to the description of low latency workloads in tutorial section, simple 6 encode d3d lowlatency.

How to use VPP for pre- or post-processing purposes

Aside from using VPP in a transcode pipeline, VPP can also be used for pre or post processing of frames. Common usages are image resize, effects or enhancements before rendering to display. Such usage is illustrated in the first sample below. The second sample below showcases how to VPP for pre processing of frames, such as color space conversion (common for pipelines where frame input originates from camera).

simple_6_decode_vpp_postproc

This sample is similar to simple_2_decode but also adds VPP post processing capabilities, showcasing frame resize and noise reduction.

simple_6_encode_vmem_vpp_preproc

This Intel® Media SDK tutorial sample is similar to simple_3_encode vmem but adds VPP pre-processing capabilities, showcasing VPP color conversion from RGB32(4) to NV12.