Closed zhaodongmcw closed 4 years ago
Please review colorconvert, nonlinearfilter, phase and threshold kernels of NEON intrinsics to merge into branch OpenVX1.3.
Add 34 kernels of NEON intrinsics
Add the rest kernels of NEON intrinsics
Fix the bug that some kernels do not have image remainder computation
Fix the bug that Harris_Score calling C reference
We do not have access to the "api-docs" repo, you should add :
VX_KERNEL_WEIGHTED_AVERAGE = VX_KERNEL_BASE(VX_ID_KHRONOS, VX_LIBRARY_KHR_BASE) + 0x40, in "include/VX/vx_kernels.h"
VX_API_ENTRY vx_node VX_API_CALL vxWeightedAverageImageNode(vx_graph graph, vx_image img1, vx_scalar alpha, vx_image img2, vx_image output); in "include/VX/vx_nodes.h"
VX_API_ENTRY vx_status VX_API_CALL vxuWeightedAverageImage(vx_context context, vx_image img1, vx_scalar alpha, vx_image img2, vx_image output); in "include/VX/vxu.h"
As per Implement the OpenVX 1.3 Adopter's program, we implement vxQueryMetaFormatAttribute and delay object(tensor, object array), please check.
Implement vxGetUserStructNameByEnum, vxCreateTensorFromHandle, vxSwapTensorHandle, vxMapTensorPatch and vxUnmapTensorPatch, please check.
By the way, we try to use with ifdefs OPENVX_CONFORMANCE_VISION to filter kernel absdiff, please check if it is the result expected by khronos.
Add feature set flags to Concerto makefiles: OPENVX_CONFORMANCE_VISION OPENVX_CONFORMANCE_NEURAL_NETWORKS OPENVX_CONFORMANCE_VISION_NNEF_IMPORT OPENVX_USE_ENHANCED_VISION
Please check.
@zhaodongmcw the OpenVX 1.3 headers are available on GitHub API Docs project. The 1.3 headers are on the openvx_1.3 branch, which you can use to build the sample implementation. https://github.com/KhronosGroup/OpenVX-api-docs/tree/openvx_1.3
Closing this pull request as the commits are done in GitLab
Please review absdiff kernel of NEON intrinsics to merge into branch OpenVX1.3.