What's New

1

Updated the KAE documents.

• User-mode support for kernel 4.19 is enabled (kernel-mode encryption/decryption and compression/decompression interfaces are not supported).
• The hardware fault isolation status and isolation thresholds of host OS PF devices can be queried through VFs in KVM scenarios.

KAE Documentation

2

Updated the KQMalloc documents.

• The performance of single-threaded applications is greatly improved.
• The heap flag buffer is redesigned.
• The logic for frequently allocating large objects with a maximum size of 64 MB is optimized to adapt to MySQL-like workloads.
• The memory footprint in producer-consumer benchmarks is significantly reduced without compromising performance.
• Four environment variable configuration items are added.
• Memory allocation performance is significantly enhanced under the GREAT benchmark.

System Library KQMalloc Documentation

3

Updated the kpglibc documents.

The memcmp function is continuously optimized based on the Kunpeng 950 processor.

System Library kpglibc Documentation

4

Added the Snappy documents.

• Performance optimization for the open-source Snappy compression library targets the AArch64 architecture. It leverages techniques such as hash table optimization and adaptive skipping steps to enhance the performance of the compression algorithm. The interfaces remain fully consistent with open-source Snappy.
• Patch introduction, interface description, and patch enabling documents are provided.

Compression Library Snappy Documentation

5

Added the zstd documents.

• Performance optimization for the open-source Zstandard (zstd) compression library targets the Kunpeng platform. It focuses on common execution paths, including block compression, block decompression, streaming compression, and streaming decompression. The optimizations do not modify the user-facing APIs, ensuring that applications can still be developed and integrated using the public C APIs of open-source zstd.
• Patch introduction, interface description, and patch enabling documents are provided.

Compression Library zstd Documentation

6

Added the LZ4 documents.

• Performance optimization for the open-source LZ4 compression library targets the Kunpeng platform, including NEON vectorized memory copy, branch prediction, and data prefetch techniques.
• Patch introduction, interface description, and patch enabling documents are provided.

Compression Library LZ4 Documentation

7

Added the zlib documents.

• Performance optimization for the open-source zlib compression library targets the Linux AArch64 platform. It reduces instruction and memory access overhead in the hot paths of compression and decompression.
• Patch introduction, interface description, and patch enabling documents are provided.

Compression Library zlib Documentation

8

Changed the category of the Protobuf library and updated its documents.

• The Protobuf library is categorized under "KSERL (Serialization)".
• The update introduces support for Map and StringBlock parsing optimization, alongside the integration of the C++23 resize_and_overwrite feature.

Serialization Library Protobuf Documentation

9

Changed the category of the Sonic-Cpp library.

The Sonic-Cpp library is categorized under "KSERL (Serialization)".

Serialization Library Sonic-Cpp Documentation

10

Added the KSERL fbthrift documents.

• A patch repository is released based on the open-source fbthrift serialization framework. This release introduces batch integer array encoding for this framework.
• Patch introduction, interface description, and patch enabling documents are provided.

Serialization Library fbthrift Documentation

11

Added the KSERL simdjson documents.

• A patch repository is released based on the open-source simdjson library. It introduces a table lookup method to serialization and the AArch64 SVE/SVE2 instruction set to deserialization stage 1, accelerating character string scan and character classification.
• Patch introduction, interface description, and patch enabling documents are provided.

Serialization Library simdjson Documentation

12

Added the KSERL RapidJSON documents.

• A patch repository is released based on open-source RapidJSON. It includes fast-path optimizations for string parsing and serialization, improving the performance of parsing UTF-8 memory input and generating StringBuffer output.
• Patch introduction, interface description, and patch enabling documents are provided.

Serialization Library RapidJSON Documentation

1

Migrated the Kunpeng Accelerator Engine (KAE) to the GitCode open-source platform.

• Open-source software packages and documents of KAE 2.0 are provided on GitCode.
• The KAEZlib library supports the asynchronous mode. The KAE 2.0 decompression module supports Snappy hardware-based acceleration.

KAE Documentation

2

Migrated the Kunpeng Unified Acceleration Framework (KUAF) to the GitCode open-source platform.

Open-source software packages and documents are provided on GitCode.

KUAF Documentation

3

Added the Sonic-Cpp library to the KSL.

• The open-source Sonic-Cpp patch and documents are provided on GitCode.
• The performance of the serialization and deserialization functions in Sonic-Cpp is optimized using the vectorized instruction set based on Kunpeng processors. This library is applicable to Kunpeng 950 and the function interfaces are consistent with those of open-source Sonic-Cpp.

System Library Sonic-Cpp Documentation

4

Added the protobuf library to the KSL.

• The open-source protobuf patch and documents are provided on GitCode.
• The performance of some protobuf functions is optimized on Kunpeng processors.

System Library Protobuf Documentation

5

Migrated the KQMalloc documents to the GitCode open-source platform.

Binary software packages and documents are provided on GitCode.

System Library KQMalloc Documentation

6

Migrated the kpglibc documents to the GitCode open-source platform.

• Open-source software packages and documents are provided on GitCode.
• The memcpy, memmove, memcmp, and memset functions are optimized for the Kunpeng 950 processor, which brings acceleration benefits in most scenarios.

System Library kpglibc Documentation

7

Migrated the AVX2KI documents to the GitCode open-source platform.

Binary software packages and documents are provided on GitCode.

System Library AVX2KI Documentation

8

Migrated the Hyperscan documents.

Hyperscan documents are migrated to the Intelligent Data Analytics site.

BoostIDA Hyperscan Documentation

9

Adjusted the categorization of the Kunpeng Vector Signal Processing (KVSIP) library.

The category of KVSIP is changed to KSL (system library).

System Library KVSIP Documentation

10

Migrated the KAIL documents to the GitCode open-source platform.

• Binary software packages and documents are provided on GitCode.
• KDNN adds a MatMul implementation for s8/u8 data types based on MMLA instructions. It supports post-ops (post-processing operations) and the FusedMatMul operator.

KAIL Documentation

11

Migrated the HMPP documents to the GitCode open-source platform.

Binary software packages and documents are provided on GitCode.

Media Library HMPP Documentation

12

Added the Kunpeng Video Codec Library (KVCL) to the KMML media library.

• Binary software packages and documents are provided on GitCode.
• KVCL contains top common operators for video encoding and optimizes operators for Kunpeng hardware.

Media Library KVCL Documentation

13

Migrated media library documents.

The x265 encoding library, HW265Enc, KPCV, and KIL documents are migrated to the BoostMedia documentation.

BoostMedia Before You Start

1

Updated the Kunpeng Accelerator Engine User Guide.

KAE 2.0 supports Kunpeng 950 processors.

Asynchronous interfaces of the KAELz4 library support the polling mode.

KAE Documentation

2

Updated the KUAF Developer Guide.

Added support for asymmetric algorithms RSA and SM2 in the KUAF.

KUAF Documentation

3

Added the KACC_crypto library.

The KACC_crypto library leverages CPU SIMD and pipelining to improve single-core computing efficiency, optimize the SHA1_MB algorithm, and enhance HMAC performance. It can connect to VPP applications.

KACC_crypto Library Developer Guide

4

Updated the BiSheng JDK Acceleration Library User Guide.

Added support for the JProfilecache and DynamicMaxHeap features in BiSheng JDK 8.

BiSheng JDK Acceleration Library User Guide

5

Added the Kunpeng Acceleration JSON (KACC_JSON) library.

KACC_JSON is a high-performance JSON function library provided by Huawei based on the Kunpeng platform. It uses SVE256 instructions to enhance JSON serialization and deserialization. By leveraging Kunpeng hardware advantages, KACC_JSON delivers higher performance than Gson, an open-source Java library.

Kunpeng Acceleration JSON Library Developer Guide

6

Updated the Kunpeng System Library Developer Guide.

Added KHSEL enhancements, including the optimization of the Hyperscan multi-pattern matching algorithm and the backend validation of long strings for the Rose interpreter.

Kunpeng System Library Developer Guide

7

Updated the Hyperscan Compilation Guide.

Added the false positive blocking feature for the multi-pattern matching algorithm and optimized backend validation of long rules for the Rose interpreter.

Hyperscan Compilation Guide

8

Updated the HMPP User Guide.

• Added four signal library function interfaces.
• Added 47 image library function interfaces.

HMPP User Guide

9

Updated the HW265Enc User Guide.

Applied affinity optimization to the HW265 encoding library and verified the adaptation and performance of this library based on new Kunpeng 920 processor models.

HW265Enc User Guide

10

Updated the Kunpeng Computer Vision Library User Guide.

• Added security verification.
• Removed the KP_LITHO_IMG operator API. If you need to use the API, refer to a historical version.
• Updated the software package name.

Kunpeng Computer Vision Library User Guide

1

Updated the Kunpeng Accelerator Engine User Guide.

• Optimized the decompression performance of KAELz4 and KAEZstd.
• Added examples of calling KAEZlib asynchronous decompression interfaces and KAELz4 asynchronous compression interfaces.

• KAE Documentation
• Best Practices

2

Updated the Kunpeng System Library Developer Guide.

Optimized the SVE implementations of memcpy and memset in the kpglibc library.

Kunpeng System Library Developer Guide

3

Added the x265 encoding library.

x265 is an open-source free software and function library for encoding HEVC/H.265 videos. It takes advantage of SVE, a vectorized instruction set architecture, to improve the video encoding performance on Arm devices.

x265 Encoding Library Developer Guide

4

Added the Kunpeng Image Library.

The Kunpeng Image Library optimizes the image encoding performance of open-source image libraries libwebp and libavif, as well as the sws_scale function in FFmpeg libswscale, through vector, algorithm, and process optimization.

Kunpeng Image Library Developer Guide

1

Updated the Kunpeng Accelerator Engine User Guide.

• Combined the Kunpeng Accelerator Engine Developer Guide (KAEzip) and the Kunpeng Accelerator Engine Developer Guide (KAE Encryption and Decryption) into the Kunpeng Accelerator Engine User Guide, and combined the Kunpeng Accelerator Engine FAQs (KAEzip) and the Kunpeng Accelerator Engine FAQs (KAE Encryption and Decryption) into the Kunpeng Accelerator Engine FAQs.
• Added the Kunpeng Accelerator Engine Release Notes and Kunpeng Accelerator Engine Best Practices.
• Adapted KAE 2.0 to TencentOS 5.4.
• Adapted the KAE 2.0 encryption and decryption module to BoringSSL RSA hardware-based acceleration.
• Added support for the asynchronous mode to the KAELz4 compression library.

• Release Notes
• Installation Guide
• Best Practices
• FAQs

2

Added the Kunpeng Unified Acceleration Framework (KUAF).

KUAF enhances the collaboration between the Kunpeng Accelerator Engine hardware and software libraries for faster compression.

KUAF Documentation

3

Updated the Kunpeng System Library Developer Guide.

• Separated the time function of the kpglibc library to libkpglibc_time.so and added the memcpy optimization.
• Optimized Flink's replaceALL function in C in the Kunpeng Hyperscan Enhanced Library (KHSEL), doubling the performance compared to the replaceALL function implemented in Java.
• Optimized the Hyperscan short-byte rule matching algorithm in KHSEL based on the new Kunpeng 920 processor model, improving the performance by 20%.

Kunpeng System Library Developer Guide

4

Updated the Hyperscan Compilation Guide.

Updated the content related to KHSEL.

Hyperscan Compilation Guide

5

Updated the HMPP User Guide.

Implemented Enhanced Voice Services (EVS) encoding for the audio library based on the Kunpeng 920 processor.

HMPP User Guide

6

Updated the Kunpeng Computer Vision Library User Guide.

• Optimized six OpenCV image preprocessing functions (morphologyEx, copyMakeBorder, bitwise_and, bitwise_or, kcv_normalAndNCHW, and t) in the CV image operator library based on Kunpeng 920 series processors, improving the average performance by 20%.
• Optimized two image preprocessing functions (crop and write_video) in the PyTorch image operator library based on Kunpeng 920 series processors, improving the average performance by 20%.

Kunpeng Computer Vision Library User Guide

7

Added the HW265Enc encoder.

HW265Enc is an in-house video encoder following the H.265/High Efficiency Video Coding (HEVC) standard.

HW265Enc User Guide

1

Updated the Kunpeng Accelerator Engine Developer Guide.

• Adapted to Tongsuo 8.4.0 in the KAE 2.0 encryption and decryption module based on Kunpeng 920 series processors.
• Adapted to open-source Gzip in the KAE 2.0 compression module based on Kunpeng 920 series processors and added the procedures for installing and testing the tool.

KAE Documentation

2

Updated the BiSheng JDK Acceleration Library User Guide.

Added the JBolt feature in BiSheng JDK 11.

BiSheng JDK Acceleration Library User Guide

3

Updated the Kunpeng System Library Developer Guide.

• Integrated the KQMalloc versions for single-threaded and multi-threaded applications.
• Supports 3GPP-NGAP standard encoding and decoding in KSL_ASN1.
• Added the kpglibc library. It optimizes seven common glibc functions (strcpy, strcmp, memchr, memcmp, gettimeofday, memcpy, and memset) based on Kunpeng series processors.

Kunpeng System Library Developer Guide

4

Updated the HMPP User Guide.

Implemented Enhanced Voice Services (EVS) decoding for the audio library based on the Kunpeng 920 processor.

HMPP User Guide

5

Updated the Kunpeng Computer Vision Library User Guide.

Optimized five OpenCV image preprocessing functions (convertTo, contourArea, threshold, countNonZero, and warpAffine) based on Kunpeng 920 series processors.

Kunpeng Computer Vision Library User Guide

1

Updated the Kunpeng Accelerator Engine Developer Guide.

• Added support for OpenSSL 3.0.x by KAE 2.0.
• Added support for openEuler 22.03 LTS SP3/SP4 by KAE 2.0.
• Added the description of installing, upgrading, and uninstalling KAE 2.0 using RPM packages.

KAE Documentation

2

Updated the BiSheng JDK Acceleration Library User Guide.

Added the JBolt feature to the BiSheng JDK acceleration library.

BiSheng JDK Acceleration Library User Guide

3

Updated the Kunpeng System Library Developer Guide.

• Added support for Distinguished Encoding Rules (DER) and XML Encoding Rules (XER), changed the name of the compilation tool of KSL_ASN1, and updated encoding and decoding interfaces in KSL_ASN1.
• Added the Kunpeng Hyperscan Enhanced Library (KHSEL) for enhancing open-source Hyperscan.

Kunpeng System Library Developer Guide

4

Released the Kunpeng System Library Open Source Software Notice.

Added the open source software notice of the Kunpeng System Library.

Kunpeng System Library Open Source Software Notice

5

Updated the HAF User Guide.

Upgraded the HAF software package to 1.4.0.

HAF User Guide

6

Updated the Hyperscan Compilation Guide.

Added the KHSEL package based on Hyperscan 5.4.2.

Hyperscan Compilation Guide

7

Updated the HMPP User Guide.

Added wavelet transform and IIR filtering functions to the signal library.

HMPP User Guide

8

Added the Kunpeng Computer Vision (KPCV) library.

The Kunpeng Computer Vision Library (KPCV) leverages Kunpeng vector instructions to optimize OpenCV and PyTorch image operators.

Kunpeng Computer Vision Library User Guide

1

Added the BiSheng JDK acceleration library.

Provided enhanced heap dump and JBooster features in the BiSheng JDK acceleration library that adapts to the Kunpeng processor.

BiSheng JDK Acceleration Library User Guide

1

Updated the HMPP User Guide.

• Added the description of compiling the HMPP library.
• Modified the interface for setting the number of FFT threads.

HMPP User Guide

2

Updated the Kunpeng System Library Developer Guide.

Added the KSL_ASN1 and KQMallocmt sub-libraries.

Kunpeng System Library Developer Guide

3

Updated the Hyperscan Compilation Guide.

Added the compilation guide for Hyperscan 5.4.2 on openEuler.

Hyperscan Compilation Guide

1

Updated the HMPP User Guide.

Added an interface for setting the number of FFT threads to basic functions.

HMPP User Guide

2

Updated the Kunpeng System Library Developer Guide.

Optimized AVX functions.

Kunpeng System Library Developer Guide

1

Updated the Kunpeng Accelerator Engine Developer Guide.

Released KAE 2.0 to adapt to kernel 5.10 and updated the developer guide accordingly.

KAE Documentation

1

Updated the HMPP User Guide.

Added the GSMHR interface for audio encoding.

HMPP User Guide

2

Updated the Kunpeng Math Library Developer Guide.

• Added the implementation functions of the KML_MATH, KML_VML, and KML_SOLVER libraries.
• Added the KML_EIGENSOLVER library.

Kunpeng Math Library Developer Guide

3

Updated the Kunpeng Math Library Migration Guide.

Added content about the migration of the LAPACK library.

Kunpeng Math Library Migration Guide

4

Updated the Kunpeng System Library Developer Guide.

Added the Hyper Thread Library (HTL).

Kunpeng System Library Developer Guide

5

LOM DPDK

Added the DPDK 21.11.5 support for openEuler 22.03 LTS SP2.

DPDK User Guide