Maksim Shabunin

Hmm... I'm observing test failures now (xuantie gcc 10.4.0, build with `-DCORE=C910V`, test with `-cpu c910v`): ``` [ RUN ] Imgproc_ColorRGB.accuracy /work/opencv/modules/ts/src/ts.cpp:618: Failure Failed failure reason: Bad accuracy test case...

Perhaps it has been partially fixed here: #26155

> * For stitching and other samples that are part of tutorial: move the code to `samples/cpp/tutorial_code//`. It's already widely used practice in OpenCV. I think it would be more...

We needs some way to limit memory usage, something similar to `OPENCV_IO_MAX_IMAGE_PIXELS` [environment variable](https://docs.opencv.org/4.x/d6/dea/tutorial_env_reference.html#autotoc_md990), but for animation. Otherwise it would be possible to exhaust memory by reading large animation, because...

I believe limiting just the number of frames is not enough (1) and is not convenient for end-users (2). 1. Single frame can be rather big - 16k*16k pixels and...

There is `hfloat` class for wrapping fp16 values (https://github.com/opencv/opencv/blob/4bf95ac2df822db8608c94d891f5edf8c549a03f/modules/core/include/opencv2/core/cvdef.h#L828). Perhaps we can try to add `_Float16` case there and use it instead.

I think we can try to cast `hfloat` to appropriate type (`_Float16`) internally like it is done in parts of NEON implementation: https://github.com/opencv/opencv/blob/a2d2ea6536df143ca118abc0fbe095e7dadbebd2/modules/core/include/opencv2/core/hal/intrin_neon.hpp#L2990-L2999 Though, I've found that NEON intrinsics are...

> Also, for extensions to RVV, it is acceptable to use RVV_, such as `RVV_ZVFH`? and what if non-RVV extensions are introduced later? Yes, let's use `RVV_ZVFH`. For non-RVV extensions...

Filed an issue: https://github.com/llvm/llvm-project/issues/97477

Filed new issue: https://github.com/llvm/llvm-project/issues/97849