The full documentation for MIVisionX is available at https://rocm.docs.amd.com/projects/MIVisionX/en/latest/doxygen/html/index.html
- OpenVX 1.3.2 full Vision Conformance Feature Set —
vxQueryImagenow supportsVX_IMAGE_IS_UNIFORMandVX_IMAGE_UNIFORM_VALUE;vxMinMaxLoccount scalars changed toVX_TYPE_SIZEper spec - HIP GPU architecture support for gfx115x (Radeon RX 9000 / gfx1150, gfx1151, gfx1152, gfx1153)
MIVISIONX_HIP_CU_COUNTenvironment variable to limit the number of compute units used by HIP kernels at runtime
- MinMaxLoc count-scalar type:
ovxKernel_MinMaxLocand allagoKernel_MinMaxLoc_*sub-kernels now declare and write count outputs asVX_TYPE_SIZEinstead ofVX_TYPE_UINT32, matching the OpenVX 1.3.2 spec and Khronos CTS - Updated
tests/openvx_api_tests/vxu_api/vxu_api.cppand all 17 MinMaxLoc GDF files intests/amd_openvx_gdfs/cpu/to useVX_TYPE_SIZE/scalar:SIZEfor count scalars
- README and sub-library documentation updated to explicitly state OpenVX 1.3.2 conformance throughout; spec links updated to the 1.3.2 URL
- Breaking change: MIVisionX has been streamlined to its core components. The toolkit now ships only the AMD OpenVX engine (
amd_openvx), the AMD RPP OpenVX extension (amd_openvx_extensions/amd_rpp), and theRunVXgraph executor, with continued support for theCPU,HIP, andOpenCLbackends. - MIVisionX is now built on top of the ROCm Core SDK and requires ROCm 7.13 or later. The Core SDK provides the HIP and OpenCL runtimes, the
amdclang++compiler, thehalflibrary, andRPP. - Windows builds now use CMake; the legacy Visual Studio
.sln/.vcxprojproject files have been removed.
- HIP Canny/Harris/NonMaxSupp kernels: out-of-bounds shared-memory reads when the top-left tile of the image falls on the image boundary. All six kernel families (
CannySobel_3x3/5x5/7x7L1/L2NORM,CannySuppThreshold_3x3,HarrisSobel_3x3/5x5/7x7,NonMaxSupp_3x3) computed a negative byte offset into the source image buffer for threads assigned to the first few rows or leftmost columns, resulting in undefined behavior. Addedgoffset >= 0guards before each halo load to skip the out-of-bounds access and leave the corresponding shared-memory region zeroed.
- OpenVX extensions:
amd_nn,amd_opencv,amd_media,amd_migraphx,amd_loomsl,amd_custom,amd_winml - Utilities:
runcl,loom_shell,mv_deploy,loom_io_media - The
MIVisionX-setup.pydependency installer; install the prerequisites (half,rpp-dev, and optionally OpenCV) directly with your package manager - The neural-net
model_compiler, thetoolkit, and the ML/Windowsapps(the OpenVX + OpenCVbubble_popandoptical_flowsample apps are retained) - Extension-specific test suites (
neural_network_tests,amd_opencv_tests,amd_media_tests,amd_migraphx_tests,zen_dnn_tests,opencv_benchmark) and the corresponding samples - Build options
NEURAL_NET,LOOM, andMIGRAPHX, and the legacy.travis.ymlCI - Dependencies on MIOpen, MIGraphX, rocBLAS, FFmpeg, and rocDecode
- Added the
PythonFunctionextension to VX_RPP - Added unified
rppt_*()function withRppBackendparameter for vx_rpp extensions to support runtime backend selection - CI - CPU (host backend) performance regression gate comparing PR vs main per-kernel and by geometric mean
- CI - AMD OpenVX vs OpenCV per-kernel CPU benchmark summary with a +/- 5% parity band
- Tests - Expanded CPU coverage with additional GDF graphs (Histogram, MinMaxLoc, ScaleImage) and a vision-coverage OpenVX API test
- Removed the batchPD extensions from VX_RPP
- Removed separate
rppt_*_host()andrppt_*_gpu()function calls for vx_rpp extensions
- OpenVX (host backend) - Optimized the S16-to-U8 range threshold kernel (
HafCpu_Threshold_U8_S16_Range) with an AVX2 (256-bit) code path
- Updated vx_rpp extension for Gaussian Filter
- Breaking change: This change requires updating to RPP version 3.1.0 with unified API support
- MIGraphX extension turned off by default
- AMD Clang++ - Location updated
${ROCM_PATH}/lib/llvm/bin - RPP required updated to RPP V2.2.1
- Memory leaks in OpenVX core, vx_nn, & vx_opencv
- Installation on CentOS/RedHat/SLES requires the manual installation of the
FFMPEG&OpenCVdev packages.
- VX_AMD_MEDIA -
rocDecodeandrocJPEGsupport for hardware decode
- VX_RPP - Update blur
- HIP - HIP_CHECK for hipLaunchKernelGGL for gated launch
- AMD Custom V1.1.0 - OpenMP updates
- HALF - Fix half.hpp path updates
- AMD Custom - dependency linking errors resolved
- VX_RPP - Fix memory leak
- Packaging - Remove Meta Package dependency for HIP
- Installation on CentOS/RedHat/SLES requires the manual installation of the
FFMPEG&OpenCVdev packages.
- VX_AMD_MEDIA - rocDecode support for hardware decode
- VX_RPP extension : Version 3.1.0 release
- Add support to enable/disable BatchPD code in VX_RPP extensions by checking the RPP_LEGACY_SUPPORT flag.
- Update the parameters and kernel API of Blur, Fog, Jitter, LensCorrection, Rain, Pixelate, Vignette and ResizeCrop wrt tensor kernels replacing the legacy BatchPD API calls in VX_RPP extensions.
- Installation on CentOS/RedHat/SLES requires the manual installation of the
FFMPEG&OpenCVdev packages.
- Optimized audio augmentations support for VX_RPP
- OpenCV is now installed with the package installer on Ubuntu
- AMD Clang is now the default CXX and C compiler
- The version of OpenMP included in the ROCm LLVM project is now used instead of libomp-dev/devel
- Installation on CentOS/RedHat/SLES requires manually installing the FFMPEG & OpenCV dev packages.
- Hardware decode requires the ROCm graphics usecase.
- Optimized audio augmentations support for VX_RPP
- Setup: rocdecode install disabled
- Package: rocdecode dependency removed
- Setup: only core dependency packages installed
- MIVisionX package installation on RedHat/SLES requires the manual installation of the
OpenCVandFFMPEGdevelopment package.
- Optimized audio augmentations support for VX_RPP
- Support has been added for the following:
- Advanced GPUs
- PreEmphasis Filter augmentation in openVX extensions
- Spectrogram augmentation in openVX extensions
- Downmix and ToDecibels augmentations in openVX extensions
- Resample augmentation and Operator overloading nodes in openVX extensions
- NonSilentRegion and Slice augmentations in openVX extensions
- Mel-Filter bank and Normalize augmentations in openVX extensions
- CentOS 7 and SLES 15 SP5 support has been added to the setup script.
- The
FindPackagemodules have been updated withFindMIVision. - New tests have been added for all modules.
- rocAL is no longer installed with MIVisionX and must be installed separately.
- MIVisionX compatibility fix: Resample and pre-emphasis filter.
- Broken image links have been fixed in the documentation.
- Package install requires manually installing OpenCV
- Installation CentOS/RedHat/SLES requires manually installing the
FFMPEG Devpackage - Hardware decode requires the ROCm
graphicsusecase.
- Optimized audio augmentations support for VX_RPP.
- CTest: Tests for install verification
- Hardware support updates
- Doxygen support for API documentation
- CMakeList Cleanup
- Readme
- rocAL: PyBind Link to prebuilt library
- PyBind11
- RapidJSON
- Setup Updates
- RPP - Use package install
- Dockerfiles: Updates & bugfix
- CuPy - No longer installed with setup.py
- rocAL bug fix and updates
- Windows
10/11 - Linux distribution
- Ubuntu -
20.04/22.04 - CentOS -
7/8 - RHEL -
8/9 - SLES -
15-SP4
- Ubuntu -
- ROCm: rocm-core -
5.7.0.50700-6 - miopen-hip -
2.20.0.50700-63 - MIGraphX -
2.7.0.50700-63 - Protobuf - V3.12.4
- OpenCV - 4.6.0
- RPP - [1.5.0]
- FFMPEG - n4.4.2
- Dependencies for all preceding packages
- MIVisionX setup script -
V2.6.1
- OpenCV 4.X support for some applications is missing
- MIVisionX package install requires manual prerequisites installation
- OpenVX FP16 Support
- rocAL: CPU, HIP, & OCL backends
- AMD RPP: CPU, HIP, amd OCL backends
- MIVisionX Setup Support for RHEL
- Extended OS Support
- Docker Support for Ubuntu
22.04 - Tests
- CMakeList cleanup
- MIGraphX extension updates
- rocAL: Documentation
- CMakeList updates and cleanup
- rocAL: Changing Python lib path
- Docker support: Ubuntu 18 support has been dropped
- RPP: Link to Version 1.0.0
- rocAL: Support updates
- Setup updates
- rocAL bug fix and updates
- AMD RPP bug fixes
- CMakeLists: Issues
- RPATH: Link issues
- Windows
10/11 - Linux distribution
- Ubuntu -
20.04/22.04 - CentOS -
7/8 - RHEL -
8/9 - SLES -
15-SP3
- Ubuntu -
- ROCm: rocm-core -
5.4.3.50403-121 - miopen-hip -
2.19.0.50403-121 - miopen-opencl -
2.18.0.50300-63 - MIGraphX -
2.4.0.50403-121 - Protobuf - V3.12.4
- OpenCV - 4.6.0
- RPP - 1.0.0
- FFMPEG - n4.4.2
- Dependencies for all preceding packages
- MIVisionX setup script -
V2.4.2
- OpenCV 4.X support for some applications is missing
- Zen DNN samples
- OpenCV extension tests
- rocAL SPACK support
- AMD custom extension
- MIGraphX updates
- Model compiler scripts
- CMakeList: Find HIP updates
- rocAL issues
- MIGraphX issues
- Windows
10/11 - Linux distribution
- Ubuntu -
20.04/22.04 - CentOS -
7/8 - SLES -
15-SP2
- Ubuntu -
- ROCm: rocm-core -
5.3.0.50300-36 - miopen-hip -
2.18.0.50300-36 - miopen-opencl -
2.18.0.50300-36 - MIGraphX -
2.3.0.50300-36 - Protobuf - V3.12.4
- OpenCV - 4.5.5
- RPP - 0.97
- FFMPEG - n4.4.2
- Dependencies for all preceding packages
- MIVisionX setup script -
V2.3.7
- OpenCV 4.X support for some applications is missing
- Docker support for ROCm
5.1.3
- MIGraphX: Implementation and samples
- DockerFiles: Updates to install ROCm 5.1.1 Plus
- Minor bugs in setup and test scripts
- Windows
10/11 - Linux distribution
- Ubuntu -
18.04/20.04 - CentOS -
7/8 - SLES -
15-SP2
- Ubuntu -
- ROCm: rocm-core -
5.1.3.50103-66 - miopen-hip -
2.16.0.50101-48 - miopen-opencl -
2.16.0.50101-48 - MIGraphX -
2.1.0.50101-48 - Protobuf - V3.12.4
- OpenCV - 4.5.5
- RPP - 0.93
- FFMPEG - n4.4.2
- Dependencies for all preceding packages
- MIVisionX setup script -
V2.3.2
- OpenCV 4.X support for some applications is missing
- New Tests:
AMD_MEDIA
- Readme updates
- HIP buffer transfer: Eliminate
cupyusage
- Backend: Default backend has been set to
HIP - DockerFiles: Updates to install ROCm 5.0 Plus
- RPP: Upgraded to V0.93
- Minor bugs and warnings
AMD_MEDIA: Bug Fixes
- Windows
10/11 - Linux distribution
- Ubuntu -
18.04/20.04 - CentOS -
7/8 - SLES -
15-SP2
- Ubuntu -
- ROCm: rocm-core -
5.0.0.50000-49 - rocm-cmake - rocm-5.1.1
- MIOpenGEMM - 1.1.5
- MIOpen - 2.14.0
- Protobuf - V3.12.0
- OpenCV - 4.5.5
- RPP - 0.93
- FFMPEG - n4.0.4
- Dependencies for all preceding packages
- MIVisionX Setup Script -
V2.1.1
CMakeList.txtwarnings: With ROCm CMake version MIVisionX will have CMake warnings
- Support for CMake 3.22.X
- Support for OpenCV 4.X.X
- Support for mv_compile with the HIP GPU backend
- Support for tensor_compare node (
less,greater,less_than,greater_than, andequalONNX operators)
- Code cleanup
- Readme updates
- License updates
- Minor bugs and warnings
- Inference server application: OpenCL backend
vxCreateThresholdfix: applications and sample
- Windows 10
- Linux distribution
- Ubuntu -
18.04/20.04 - CentOS -
7/8 - SLES -
15-SP2
- Ubuntu -
- ROCm: rocm-dev -
4.5.2.40502-164 - rocm-cmake - rocm-4.2.0
- MIOpenGEMM - 1.1.5
- MIOpen - 2.14.0
- Protobuf - V3.12.0
- OpenCV - 3.4.0
- RPP - 0.92
- FFMPEG - n4.0.4
- Dependencies for all preceding packages
- MIVisionX Setup Script -
V2.0.0
- Package install requires OpenCV
v3.4.Xto executeAMD OpenCV extensions
- OpenVX 1.3: vision feature set
- Conformance test script
- HIP backend support for OpenVX and OpenVX extensions
- Improved rocAL performance
- Improved the performance of OpenVX OpenCL backend functions
- Docker build files
MIVisionX-setup.pyinstall on Linux- Fixed out-of-bounds read for OpenVX OpenCL kernels
- OpenVX: Optical flow segfault fix
- Windows 10
- Linux distribution
- Ubuntu -
18.04/20.04 - CentOS -
7/8 - SLES -
15-SP2
- Ubuntu -
- ROCm: rocm-dev -
4.5.2.40502-164 - rocm-cmake - rocm-4.2.0
- MIOpenGEMM - 1.1.5
- MIOpen - 2.14.0
- Protobuf - V3.12.0
- OpenCV - 3.4.0
- RPP - 0.92
- FFMPEG - n4.0.4
- Dependencies for all preceding packages
- MIVisionX Setup Script -
V2.0.0
- Package install requires OpenCV
v3.4.Xto runAMD OpenCV extensions
