core: fix solveCubic numerical instability via coefficient normalization (fixes#27748) #28117
Summary
This PR fixes numerical instability in `cv::solveCubic` when the leading coefficient `a` is non-zero but extremely small relative to other coefficients (Issue #27748).
It introduces a **normalization step** that scales all coefficients by their maximum magnitude before solving. This ensures robust detection of when the equation should degenerate to a quadratic solver, without breaking valid cubic equations that happen to have small coefficients (e.g., scaled by 1e-9).
The Problem (Issue #27748)
The previous implementation checked `if (a == 0)` to decide whether to use the cubic or quadratic formula.
- When `a` is extremely small (e.g., 1e-17) but not exactly zero, and other coefficients are normal (e.g., 5.0), the standard cubic formula suffers from catastrophic cancellation and overflow, producing incorrect roots (e.g., 1e14).
The Fix
1. Normalization: The solver now finds `max_coeff = max(|a|, |b|, |c|, |d|)` and scales all coefficients by `1.0 / max_coeff`.
2. Relative Threshold: It then checks `if (abs(a) < epsilon)` on the *normalized* coefficients.
Why this is better than previous attempts
In a previous attempt (PR #28057), a simple absolute check `abs(a) < epsilon` was proposed. That approach was rejected because it failed for scaled equations.
Fixes#27748
Use MatShape instead of MatSize inside cv::Mat/cv::UMat #27757
**Merge together with https://github.com/opencv/opencv_contrib/pull/3996**
---
This PR continues cv::Mat/cv::UMat refactoring. See #26056, where `MatShape` was introduced. Now it's put inside cv::Mat/cv::UMat instead of a weird `MatSize`. MatSize is now an alias for MatShape:
**before:**
```
struct MatShape { ... };
struct MatSize { ... };
struct Mat {
...
int dims;
int rows;
int cols;
...
MatShape shape() const { ... /* constructs MatShape out of MatSize and returns it;
layout is always 'unknown', because we don't store it */ }
MatSize size; // size is not valid without the parent cv::Mat,
// because size.p may point to Mat::rows or to Mat::cols,
// depending on the dimensionality, and dims() returns Mat::dims.
MatStep step; // may allocate memory, depending on the dimensionality.
...
};
```
**after:**
```
struct MatShape { ... };
typedef MatShape MatSize; // they are now synonyms
struct Mat {
...
int dims;
int rows;
int cols;
...
MatShape shape() const { return size; } // just return the embedded shape (including the proper layout information)
MatSize size; // size is self-contained data structure that can be used without the parent cv::Mat.
// size.dims is now a copy of dims; size.p[*] contains copies of Mat::rows and Mat::cols when dims <= 2.
MatStep step; // does not allocate extra memory buffers.
...
};
```
There are several reasons to do that:
1. the main reason is to be able to store data layout (MatShape::layout) inside each cv::Mat/cv::UMat. This is necessary for the proper shape inference in DNN module. In particular, it's necessary for the next step of DNN inference optimization where we introduce block-layout-optimized convolution and other operations. Later on, we can use layout information to support non-interleaved images (e.g. RRR...GGG...BBB...) or even batches of such images in core/imgproc modules.
2. the other reason is to represent 3D/4D/5D etc. tensors as cv::Mat/cv::UMat instances more conveniently, without extra dynamic memory allocation. Before this patch we allocated some memory buffers dynamically to store shape & steps for more than 2D arrays. Now the whole cv::Mat/cv::UMat header can be stored completely on stack/in a container. Creating another copy of Mat/UMat header is now done more efficiently.
3. the third reason is to introduce the new coding pattern: `dst.create(src.size, <dst_type>);`. The pattern is suitable for most of element-wise (including cloning) and filtering operations. This pattern does not only look crisp and self-documenting, it will also automatically copy shape (including layout) from the source tensor into the destination matrix/tensor.
4. in the future we might add `colorspace` member to MatShape that will allow to distinguish RGB from BGR or NV12. `dst.create(src.size, <dst_type>);` will then copy the colorspace information as well.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Improve solveCubic accuracy #27347
### Pull Request Readiness Checklist
Fix#27323
```
2e-13 * x^3 + x^2 - 2 * x + 1 = 0 -> x^3 + 5e12 * x^2 - 1e13 * x + 5e12 = 0
```
The problem that coefficients have quite big magnitudes and current calculations are subject to round-off error
```
Q = (a1 * a1 - 3 * a2) * (1./9)
R = (2 * a1 * a1 * a1 - 9 * a1 * a2 + 27 * a3) * (1./54)
Qcubed = Q * Q * Q = a1^6/729 - (a1^4 a2)/81 + (a1^2 a2^2)/27 - a2^3/27
R * R = R^2 = a1^6/729 - (a1^4 a2)/81 + (a1^2 a2^2)/36 + (a1^3 a3)/27 - (a1 a2 a3)/6 + a3^2/4
d = Qcubed - R * R
```
Let `a1`, `a2`, `a3` have quite big same magnitudes, then we see that `Qcubed` and `R * R` have same terms `a1^6/729` and `-(a1^4 a2)/81` (which will be reduced in `d`), but they level out the other terms (these terms have `6`th and `5`th degree and other terms - less or equal than `4`th degree).
So, if these terms will participate in the calculation, this will lead to a huge round-off error.
But if we expand the expression, then round-off error should be less
```
d = Qcubed - R * R = 1/108 (a1^2 a2^2 - 4 a2^3 - 4 a1^3 a3 + 18 a1 a2 a3 - 27 a3^2)
```
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Add tests for solveCubic #27331
### Pull Request Readiness Checklist
Related to #27323
I found only randomized tests with number of roots always equal to `1` or `3`, `x^3 = 0` and some simple test for Java and Swift.
Obviously, they don't cover all cases (implementation has strong branching and number of roots can be equal to `-1`, `0` and `2` additionally).
So, I think it will be useful to try explicitly cover more cases (and implementation branches correspondingly)
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
[HAL RVV] unify and impl polar_to_cart | add perf test #26999
### Summary
1. Implement through the existing `cv_hal_polarToCart32f` and `cv_hal_polarToCart64f` interfaces.
2. Add `polarToCart` performance tests
3. Make `cv::polarToCart` use CALL_HAL in the same way as `cv::cartToPolar`
4. To achieve the 3rd point, the original implementation was moved, and some modifications were made.
Tested through:
```sh
opencv_test_core --gtest_filter="*PolarToCart*:*Core_CartPolar_reverse*"
opencv_perf_core --gtest_filter="*PolarToCart*" --perf_min_samples=300 --perf_force_samples=300
```
### HAL performance test
***UPDATE***: Current implementation is no more depending on vlen.
**NOTE**: Due to the 4th point in the summary above, the `scalar` and `ui` test is based on the modified code of this PR. The impact of this patch on `scalar` and `ui` is evaluated in the next section, `Effect of Point 4`.
Vlen 256 (Muse Pi):
```
Name of Test scalar ui rvv ui rvv
vs vs
scalar scalar
(x-factor) (x-factor)
PolarToCart::PolarToCartFixture::(127x61, 32FC1) 0.315 0.110 0.034 2.85 9.34
PolarToCart::PolarToCartFixture::(127x61, 64FC1) 0.423 0.163 0.045 2.59 9.34
PolarToCart::PolarToCartFixture::(640x480, 32FC1) 13.695 4.325 1.278 3.17 10.71
PolarToCart::PolarToCartFixture::(640x480, 64FC1) 17.719 7.118 2.105 2.49 8.42
PolarToCart::PolarToCartFixture::(1280x720, 32FC1) 40.678 13.114 3.977 3.10 10.23
PolarToCart::PolarToCartFixture::(1280x720, 64FC1) 53.124 21.298 6.519 2.49 8.15
PolarToCart::PolarToCartFixture::(1920x1080, 32FC1) 95.158 29.465 8.894 3.23 10.70
PolarToCart::PolarToCartFixture::(1920x1080, 64FC1) 119.262 47.743 14.129 2.50 8.44
```
### Effect of Point 4
To make `cv::polarToCart` behave the same as `cv::cartToPolar`, the implementation detail of the former has been moved to the latter's location (from `mathfuncs.cpp` to `mathfuncs_core.simd.hpp`).
#### Reason for Changes:
This function works as follows:
$y = \text{mag} \times \sin(\text{angle})$ and $x = \text{mag} \times \cos(\text{angle})$. The original implementation first calculates the values of $\sin$ and $\cos$, storing the results in the output buffers $x$ and $y$, and then multiplies the result by $\text{mag}$.
However, when the function is used as an in-place operation (one of the output buffers is also an input buffer), the original implementation allocates an extra buffer to store the $\sin$ and $\cos$ values in case the $\text{mag}$ value gets overwritten. This extra buffer allocation prevents `cv::polarToCart` from functioning in the same way as `cv::cartToPolar`.
Therefore, the multiplication is now performed immediately without storing intermediate values. Since the original implementation also had AVX2 optimizations, I have applied the same optimizations to the AVX2 version of this implementation.
***UPDATE***: UI use v_sincos from #25892 now. The original implementation has AVX2 optimizations but is slower much than current UI so it's removed, and AVX2 perf test is below. Scalar implementation isn't changed because it's faster than using UI's method.
#### Test Result
`scalar` and `ui` test is done on Muse PI, and AVX2 test is done on Intel(R) Xeon(R) Gold 6140 CPU @ 2.30GHz.
`scalar` test:
```
Name of Test orig pr pr
vs
orig
(x-factor)
PolarToCart::PolarToCartFixture::(127x61, 32FC1) 0.333 0.294 1.13
PolarToCart::PolarToCartFixture::(127x61, 64FC1) 0.385 0.403 0.96
PolarToCart::PolarToCartFixture::(640x480, 32FC1) 14.749 12.343 1.19
PolarToCart::PolarToCartFixture::(640x480, 64FC1) 19.419 16.743 1.16
PolarToCart::PolarToCartFixture::(1280x720, 32FC1) 44.155 37.822 1.17
PolarToCart::PolarToCartFixture::(1280x720, 64FC1) 62.108 50.358 1.23
PolarToCart::PolarToCartFixture::(1920x1080, 32FC1) 99.011 85.769 1.15
PolarToCart::PolarToCartFixture::(1920x1080, 64FC1) 127.740 112.874 1.13
```
`ui` test:
```
Name of Test orig pr pr
vs
orig
(x-factor)
PolarToCart::PolarToCartFixture::(127x61, 32FC1) 0.306 0.110 2.77
PolarToCart::PolarToCartFixture::(127x61, 64FC1) 0.455 0.163 2.79
PolarToCart::PolarToCartFixture::(640x480, 32FC1) 13.381 4.325 3.09
PolarToCart::PolarToCartFixture::(640x480, 64FC1) 21.851 7.118 3.07
PolarToCart::PolarToCartFixture::(1280x720, 32FC1) 39.975 13.114 3.05
PolarToCart::PolarToCartFixture::(1280x720, 64FC1) 67.006 21.298 3.15
PolarToCart::PolarToCartFixture::(1920x1080, 32FC1) 90.362 29.465 3.07
PolarToCart::PolarToCartFixture::(1920x1080, 64FC1) 129.637 47.743 2.72
```
AVX2 test:
```
Name of Test orig pr pr
vs
orig
(x-factor)
PolarToCart::PolarToCartFixture::(127x61, 32FC1) 0.019 0.009 2.11
PolarToCart::PolarToCartFixture::(127x61, 64FC1) 0.022 0.013 1.74
PolarToCart::PolarToCartFixture::(640x480, 32FC1) 0.788 0.355 2.22
PolarToCart::PolarToCartFixture::(640x480, 64FC1) 1.102 0.618 1.78
PolarToCart::PolarToCartFixture::(1280x720, 32FC1) 2.383 1.042 2.29
PolarToCart::PolarToCartFixture::(1280x720, 64FC1) 3.758 2.316 1.62
PolarToCart::PolarToCartFixture::(1920x1080, 32FC1) 5.577 2.559 2.18
PolarToCart::PolarToCartFixture::(1920x1080, 64FC1) 9.710 6.424 1.51
```
A slight performance loss occurs because the check for whether $mag$ is nullptr is performed with every calculation, instead of being done once per batch. This is to reuse current `SinCos_32f` function.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [ ] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
Fixed the simd bugs of iPow8u and iPow16u #26061
Add the following cases in opencv_perf_core:
* OCL_PowFixture_iPow.iPow/0, where GetParam() = (640x480, 8UC1)
* OCL_PowFixture_iPow.iPow/2, where GetParam() = (640x480, 16UC1)
iPow8u and iPow16u failed to call to simd accelerating while executing.
Fix the bug by changing the input type of iPow_SIMD function.
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [ ] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
Check range for type-dependant function tables #25598
Address https://github.com/opencv/opencv/issues/24703
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
Added in-place support for cartToPolar and polarToCart #24893
- a fused hal::cartToPolar[32|64]f() is used instead of sequential hal::magnitude[32|64]f/hal::fastAtan[32|64]f
- ipp_polarToCart is skipped for in-place processing (it seems not to support it correctly)
relates to #24891
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [X] I agree to contribute to the project under Apache 2 License.
- [X] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [X] The PR is proposed to the proper branch
- [X] There is a reference to the original bug report and related work
- [X] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
C-API cleanup: apps, imgproc_c and some constants #25075
Merge with https://github.com/opencv/opencv_contrib/pull/3642
* Removed obsolete apps - traincascade and createsamples (please use older OpenCV versions if you need them). These apps relied heavily on C-API
* removed all mentions of imgproc C-API headers (imgproc_c.h, types_c.h) - they were empty, included core C-API headers
* replaced usage of several C constants with C++ ones (error codes, norm modes, RNG modes, PCA modes, ...) - most part of this PR (split into two parts - all modules and calib+3d - for easier backporting)
* removed imgproc C-API headers (as separate commit, so that other changes could be backported to 4.x)
Most of these changes can be backported to 4.x.
* started adding support for new types (16f, 16bf, 32u, 64u, 64s) to arithmetic functions
* fixed several tests; refactored and extended sum(), extended inRange().
* extended countNonZero(), mean(), meanStdDev(), minMaxIdx(), norm() and sum() to support new types (F16, BF16, U32, U64, S64)
* put missing CV_DEPTH_MAX to some function dispatcher tables
* extended findnonzero, hasnonzero with the new types support
* extended mixChannels() to support new types
* minor fix
* fixed a few compile errors on Linux and a few failures in core tests
* fixed a few more warnings and test failures
* trying to fix the remaining warnings and test failures. The test `MulTestGPU.MathOpTest` was disabled - not clear whether to set tolerance - it's not bit-exact operation, as possibly assumed by the test, due to the use of scale and possibly limited accuracy of the intermediate floating-point calculations.
* found that in the current snapshot G-API produces incorrect results in Mul, Div and AddWeighted (at least when using OpenCL on Windows x64 or MacOS x64). Disabled the respective tests.
finiteMask() and doubles for patchNaNs() #23098
Related to #22826
Connected PR in extra: [#1037@extra](https://github.com/opencv/opencv_extra/pull/1037)
### TODOs:
- [ ] Vectorize `finiteMask()` for 64FC3 and 64FC4
### Changes
This PR:
* adds a new function `finiteMask()`
* extends `patchNaNs()` by CV_64F support
* moves `patchNaNs()` and `finiteMask()` to a separate file
**NOTE:** now the function is called `finiteMask()` as discussed with the OpenCV core team
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
Backport to 4.x: patchNaNs() SIMD acceleration #24480
backport from #23098
connected PR in extra: [#1118@extra](https://github.com/opencv/opencv_extra/pull/1118)
### This PR contains:
* new SIMD code for `patchNaNs()`
* CPU perf test
<details>
<summary>Performance comparison</summary>
Geometric mean (ms)
|Name of Test|noopt|sse2|avx2|sse2 vs noopt (x-factor)|avx2 vs noopt (x-factor)|
|---|:-:|:-:|:-:|:-:|:-:|
|PatchNaNs::OCL_PatchNaNsFixture::(640x480, 32FC1)|0.019|0.017|0.018|1.11|1.07|
|PatchNaNs::OCL_PatchNaNsFixture::(640x480, 32FC4)|0.037|0.037|0.033|1.00|1.10|
|PatchNaNs::OCL_PatchNaNsFixture::(1280x720, 32FC1)|0.032|0.032|0.033|0.99|0.98|
|PatchNaNs::OCL_PatchNaNsFixture::(1280x720, 32FC4)|0.072|0.072|0.070|1.00|1.03|
|PatchNaNs::OCL_PatchNaNsFixture::(1920x1080, 32FC1)|0.051|0.051|0.050|1.00|1.01|
|PatchNaNs::OCL_PatchNaNsFixture::(1920x1080, 32FC4)|0.137|0.138|0.128|0.99|1.06|
|PatchNaNs::OCL_PatchNaNsFixture::(3840x2160, 32FC1)|0.137|0.128|0.129|1.07|1.06|
|PatchNaNs::OCL_PatchNaNsFixture::(3840x2160, 32FC4)|0.450|0.450|0.448|1.00|1.01|
|PatchNaNs::PatchNaNsFixture::(640x480, 32FC1)|0.149|0.029|0.020|5.13|7.44|
|PatchNaNs::PatchNaNsFixture::(640x480, 32FC2)|0.304|0.058|0.040|5.25|7.65|
|PatchNaNs::PatchNaNsFixture::(640x480, 32FC3)|0.448|0.086|0.059|5.22|7.55|
|PatchNaNs::PatchNaNsFixture::(640x480, 32FC4)|0.601|0.133|0.083|4.51|7.23|
|PatchNaNs::PatchNaNsFixture::(1280x720, 32FC1)|0.451|0.093|0.060|4.83|7.52|
|PatchNaNs::PatchNaNsFixture::(1280x720, 32FC2)|0.892|0.184|0.126|4.85|7.06|
|PatchNaNs::PatchNaNsFixture::(1280x720, 32FC3)|1.345|0.311|0.230|4.32|5.84|
|PatchNaNs::PatchNaNsFixture::(1280x720, 32FC4)|1.831|0.546|0.436|3.35|4.20|
|PatchNaNs::PatchNaNsFixture::(1920x1080, 32FC1)|1.017|0.250|0.160|4.06|6.35|
|PatchNaNs::PatchNaNsFixture::(1920x1080, 32FC2)|2.077|0.646|0.605|3.21|3.43|
|PatchNaNs::PatchNaNsFixture::(1920x1080, 32FC3)|3.134|1.053|0.961|2.97|3.26|
|PatchNaNs::PatchNaNsFixture::(1920x1080, 32FC4)|4.222|1.436|1.288|2.94|3.28|
|PatchNaNs::PatchNaNsFixture::(3840x2160, 32FC1)|4.225|1.401|1.277|3.01|3.31|
|PatchNaNs::PatchNaNsFixture::(3840x2160, 32FC2)|8.310|2.953|2.635|2.81|3.15|
|PatchNaNs::PatchNaNsFixture::(3840x2160, 32FC3)|12.396|4.455|4.252|2.78|2.92|
|PatchNaNs::PatchNaNsFixture::(3840x2160, 32FC4)|17.174|5.831|5.824|2.95|2.95|
</details>
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [x] I agree to contribute to the project under Apache 2 License.
- [x] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [x] The PR is proposed to the proper branch
- [x] There is a reference to the original bug report and related work
- [x] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [x] The feature is well documented and sample code can be built with the project CMake
* attempt to add 0d/1d mat support to OpenCV
* revised the patch; now 1D mat is treated as 1xN 2D mat rather than Nx1.
* a step towards 'green' tests
* another little step towards 'green' tests
* calib test failures seem to be fixed now
* more fixes _core & _dnn
* another step towards green ci; even 0D mat's (a.k.a. scalars) are now partly supported!
* * fixed strange bug in aruco/charuco detector, not sure why it did not work
* also fixed a few remaining failures (hopefully) in dnn & core
* disabled failing GAPI tests - too complex to dig into this compiler pipeline
* hopefully fixed java tests
* trying to fix some more tests
* quick followup fix
* continue to fix test failures and warnings
* quick followup fix
* trying to fix some more tests
* partly fixed support for 0D/scalar UMat's
* use updated parseReduce() from upstream
* trying to fix the remaining test failures
* fixed [ch]aruco tests in Python
* still trying to fix tests
* revert "fix" in dnn's CUDA tensor
* trying to fix dnn+CUDA test failures
* fixed 1D umat creation
* hopefully fixed remaining cuda test failures
* removed training whitespaces
Rewrite Universal Intrinsic code by using new API: Core module. #23980
The goal of this PR is to match and modify all SIMD code blocks guarded by `CV_SIMD` macro in the `opencv/modules/core` folder and rewrite them by using the new Universal Intrinsic API.
The patch is almost auto-generated by using the [rewriter](https://github.com/hanliutong/rewriter), related PR #23885.
Most of the files have been rewritten, but I marked this PR as draft because, the `CV_SIMD` macro also exists in the following files, and the reasons why they are not rewrited are:
1. ~~code design for fixed-size SIMD (v_int16x8, v_float32x4, etc.), need to manually rewrite.~~ Rewrited
- ./modules/core/src/stat.simd.hpp
- ./modules/core/src/matrix_transform.cpp
- ./modules/core/src/matmul.simd.hpp
2. Vector types are wrapped in other class/struct, that are not supported by the compiler in variable-length backends. Can not be rewrited directly.
- ./modules/core/src/mathfuncs_core.simd.hpp
```cpp
struct v_atan_f32
{
explicit v_atan_f32(const float& scale)
{
...
}
v_float32 compute(const v_float32& y, const v_float32& x)
{
...
}
...
v_float32 val90; // sizeless type can not used in a class
v_float32 val180;
v_float32 val360;
v_float32 s;
};
```
3. The API interface does not support/does not match
- ./modules/core/src/norm.cpp
Use `v_popcount`, ~~waiting for #23966~~ Fixed
- ./modules/core/src/has_non_zero.simd.hpp
Use illegal Universal Intrinsic API: For float type, there is no logical operation `|`. Further discussion needed
```cpp
/** @brief Bitwise OR
Only for integer types. */
template<typename _Tp, int n> CV_INLINE v_reg<_Tp, n> operator|(const v_reg<_Tp, n>& a, const v_reg<_Tp, n>& b);
template<typename _Tp, int n> CV_INLINE v_reg<_Tp, n>& operator|=(v_reg<_Tp, n>& a, const v_reg<_Tp, n>& b);
```
```cpp
#if CV_SIMD
typedef v_float32 v_type;
const v_type v_zero = vx_setzero_f32();
constexpr const int unrollCount = 8;
int step = v_type::nlanes * unrollCount;
int len0 = len & -step;
const float* srcSimdEnd = src+len0;
int countSIMD = static_cast<int>((srcSimdEnd-src)/step);
while(!res && countSIMD--)
{
v_type v0 = vx_load(src);
src += v_type::nlanes;
v_type v1 = vx_load(src);
src += v_type::nlanes;
....
src += v_type::nlanes;
v0 |= v1; //Illegal ?
....
//res = v_check_any(((v0 | v4) != v_zero));//beware : (NaN != 0) returns "false" since != is mapped to _CMP_NEQ_OQ and not _CMP_NEQ_UQ
res = !v_check_all(((v0 | v4) == v_zero));
}
v_cleanup();
#endif
```
### Pull Request Readiness Checklist
See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request
- [ ] I agree to contribute to the project under Apache 2 License.
- [ ] To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
- [ ] The PR is proposed to the proper branch
- [ ] There is a reference to the original bug report and related work
- [ ] There is accuracy test, performance test and test data in opencv_extra repository, if applicable
Patch to opencv_extra has the same branch name.
- [ ] The feature is well documented and sample code can be built with the project CMake
- to reduce binaries size of FFmpeg Windows wrapper
- MinGW linker doesn't support -ffunction-sections (used for FFmpeg Windows wrapper)
- move code to improve locality with its used dependencies
- move UMat::dot() to matmul.dispatch.cpp (Mat::dot() is already there)
- move UMat::inv() to lapack.cpp
- move UMat::mul() to arithm.cpp
- move UMat:eye() to matrix_operations.cpp (near setIdentity() implementation)
- move normalize(): convert_scale.cpp => norm.cpp
- move convertAndUnrollScalar(): arithm.cpp => copy.cpp
- move scalarToRawData(): array.cpp => copy.cpp
- move transpose(): matrix_operations.cpp => matrix_transform.cpp
- move flip(), rotate(): copy.cpp => matrix_transform.cpp (rotate90 uses flip and transpose)
- add 'OPENCV_CORE_EXCLUDE_C_API' CMake variable to exclude compilation of C-API functions from the core module
- matrix_wrap.cpp: add compile-time checks for CUDA/OpenGL calls
- the steps above allow to reduce FFmpeg wrapper size for ~1.5Mb (initial size of OpenCV part is about 3Mb)
backport is done to improve merge experience (less conflicts)
backport of commit: 65eb946756
- to reduce binaries size of FFmpeg Windows wrapper
- MinGW linker doesn't support -ffunction-sections (used for FFmpeg Windows wrapper)
- move code to improve locality with its used dependencies
- move UMat::dot() to matmul.dispatch.cpp (Mat::dot() is already there)
- move UMat::inv() to lapack.cpp
- move UMat::mul() to arithm.cpp
- move UMat:eye() to matrix_operations.cpp (near setIdentity() implementation)
- move normalize(): convert_scale.cpp => norm.cpp
- move convertAndUnrollScalar(): arithm.cpp => copy.cpp
- move scalarToRawData(): array.cpp => copy.cpp
- move transpose(): matrix_operations.cpp => matrix_transform.cpp
- move flip(), rotate(): copy.cpp => matrix_transform.cpp (rotate90 uses flip and transpose)
- add 'OPENCV_CORE_EXCLUDE_C_API' CMake variable to exclude compilation of C-API functions from the core module
- matrix_wrap.cpp: add compile-time checks for CUDA/OpenGL calls
- the steps above allow to reduce FFmpeg wrapper size for ~1.5Mb (initial size of OpenCV part is about 3Mb)
* js: update build script
- support emscipten 1.38.12 (wasm is ON by default)
- verbose build messages
* js: use builtin Math functions
* js: disable tracing code completelly
Updated integrations for:
cv::split
cv::merge
cv::insertChannel
cv::extractChannel
cv::Mat::convertTo - now with scaled conversions support
cv::LUT - disabled due to performance issues
Mat::copyTo
Mat::setTo
cv::flip
cv::copyMakeBorder - currently disabled
cv::polarToCart
cv::pow - ipp pow function was removed due to performance issues
cv::hal::magnitude32f/64f - disabled for <= SSE42, poor performance
cv::countNonZero
cv::minMaxIdx
cv::norm
cv::canny - new integration. Disabled for threaded;
cv::cornerHarris
cv::boxFilter
cv::bilateralFilter
cv::integral
Alexander Smorkalov
Vincent Rabaud
satyam yadav
Vadim Pisarevsky
Maxim Smolskiy
Alexander Alekhin
GenshinImpactStarts
Alexander Smorkalov
penghuiho
Alexander Smorkalov
Pierre Chatelier
Maksim Shabunin
Rostislav Vasilikhin
HAN Liutong
OpenCV Developers
Suleyman TURKMEN
Nikita Shulga
Vitaly Tuzov
Hamdi Sahloul
Alexander Alekhin
Tomoaki Teshima
Pavel Vlasov
Maksim Shabunin