stereo: performance optimization of sgm on Window-ARM64

[pratham-mcw]

Pull Request Readiness Checklist

• This PR adds an ARM64 NEON intrinsics-based optimization for the computeDisparityBinarySGBM function in stereo_binary_sgbm.cpp.
• The new implementation uses NEON vector instructions (e.g., vld1q_s16, vminq_s16, vqaddq_s16), allowing for efficient parallel computation. This is guarded under the CV_NEON macro and does not affect other platforms.
• This change is similar to existing SSE2 optimizations for x64 and brings the same performance benefits to ARM64.

Performance Improvements:

• The optimization significantly improves the performance of sgm on Windows ARM64 targets.
• The table below shows timing comparisons before and after the optimization:

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

[asmorkalov]

The modified code is not called in the existing perf tests. Please extend perf test coverage.

[pratham-mcw]

The modified code is not called in the existing perf tests. Please extend perf test coverage.

Hi @asmorkalov,
Thanks for the feedback! I've added debug statements in the added NEON code to verify the code execution.

#elif CV_NEON
    if ( useSIMD )
    {
        printf("NEON is supported\n");
        int16x8_t vP1 = vdupq_n_s16((short)P1);
        int16x8_t vDelta0 = vdupq_n_s16((short)delta0);
        int16x8_t vDelta1 = vdupq_n_s16((short)delta1);
        int16x8_t vDelta2 = vdupq_n_s16((short)delta2);
        int16x8_t vDelta3 = vdupq_n_s16((short)delta3);
        int16x8_t vMinL0 = vdupq_n_s16((short)MAX_COST);
        int16x8_t vCpd, vL0, vL1, vL2, vL3, vL0m1, vL0p1;
        int16x8_t vL1m1, vL1p1, vL2m1, vL2p1, vL3m1, vL3p1;
        for ( d = 0; d < D; d += 8 )
        {
            vCpd = vld1q_s16(Cp + d);
        // ... rest of the NEON implementation
    }
#endif

When running opencv_perf_stereo.exe, the debug statements are printed, confirming that the NEON-optimized path is being executed by the existing performance tests.
I've attached the log file containing the output of opencv_perf_stereo.exe
Steroe_Perf_Output.txt

[asmorkalov]

Perf results for my Jetson board:

Geometric mean (ms)

                             Name of Test                              4.x-1  patch-1  patch-1  
                                                                                          vs    
                                                                                        4.x-1   
                                                                                      (x-factor)
bm_perf::s_bm::(320x240, 8UC1, CV_8U)                                  7.073   6.840     1.03   
bm_perf::s_bm::(512x383, 8UC1, CV_8U)                                  18.183 17.796     1.02   
census_sparse_descriptor::descript_params::(127x61, 8UC1, CV_32S)      0.037   0.036     1.04   
census_sparse_descriptor::descript_params::(127x61, 8UC1, UNKNOWN)     0.037   0.035     1.03   
census_sparse_descriptor::descript_params::(640x480, 8UC1, CV_32S)     1.105   1.102     1.00   
census_sparse_descriptor::descript_params::(640x480, 8UC1, UNKNOWN)    1.107   1.104     1.00   
census_sparse_descriptor::descript_params::(1280x720, 8UC1, CV_32S)    3.294   3.305     1.00   
census_sparse_descriptor::descript_params::(1280x720, 8UC1, UNKNOWN)   3.293   3.297     1.00   
census_sparse_descriptor::descript_params::(1920x1080, 8UC1, CV_32S)   7.381   7.385     1.00   
census_sparse_descriptor::descript_params::(1920x1080, 8UC1, UNKNOWN)  7.383   7.384     1.00   
center_symetric_census::descript_params::(127x61, 8UC1, CV_32S)        0.043   0.043     1.00   
center_symetric_census::descript_params::(127x61, 8UC1, UNKNOWN)       0.044   0.043     1.02   
center_symetric_census::descript_params::(640x480, 8UC1, CV_32S)       1.330   1.319     1.01   
center_symetric_census::descript_params::(640x480, 8UC1, UNKNOWN)      1.319   1.321     1.00   
center_symetric_census::descript_params::(1280x720, 8UC1, CV_32S)      3.939   3.982     0.99   
center_symetric_census::descript_params::(1280x720, 8UC1, UNKNOWN)     3.942   3.945     1.00   
center_symetric_census::descript_params::(1920x1080, 8UC1, CV_32S)     8.857   8.869     1.00   
center_symetric_census::descript_params::(1920x1080, 8UC1, UNKNOWN)    8.851   8.859     1.00   
modified_census_transform::descript_params::(127x61, 8UC1, CV_32S)     0.118   0.119     1.00   
modified_census_transform::descript_params::(127x61, 8UC1, UNKNOWN)    0.118   0.119     0.99   
modified_census_transform::descript_params::(640x480, 8UC1, CV_32S)    4.780   4.787     1.00   
modified_census_transform::descript_params::(640x480, 8UC1, UNKNOWN)   4.770   4.818     0.99   
modified_census_transform::descript_params::(1280x720, 8UC1, CV_32S)   14.365 14.522     0.99   
modified_census_transform::descript_params::(1280x720, 8UC1, UNKNOWN)  14.355 14.496     0.99   
modified_census_transform::descript_params::(1920x1080, 8UC1, CV_32S)  32.361 32.632     0.99   
modified_census_transform::descript_params::(1920x1080, 8UC1, UNKNOWN) 32.331 32.602     0.99   
sgm_perf::s_bm::(320x240, 8UC1, CV_8U)                                 23.401 14.695     1.59   
sgm_perf::s_bm::(320x240, 8UC1, CV_16S)                                23.392 14.567     1.61   
sgm_perf::s_bm::(512x283, 8UC1, CV_8U)                                 44.305 27.594     1.61   
sgm_perf::s_bm::(512x283, 8UC1, CV_16S)                                44.492 28.086     1.58   
star_census_transform::descript_params::(127x61, 8UC1, CV_32S)         0.041   0.041     1.00   
star_census_transform::descript_params::(127x61, 8UC1, UNKNOWN)        0.041   0.041     1.01   
star_census_transform::descript_params::(640x480, 8UC1, CV_32S)        1.296   1.294     1.00   
star_census_transform::descript_params::(640x480, 8UC1, UNKNOWN)       1.296   1.294     1.00   
star_census_transform::descript_params::(1280x720, 8UC1, CV_32S)       3.893   3.892     1.00   
star_census_transform::descript_params::(1280x720, 8UC1, UNKNOWN)      3.885   3.883     1.00   
star_census_transform::descript_params::(1920x1080, 8UC1, CV_32S)      8.740   8.744     1.00   
star_census_transform::descript_params::(1920x1080, 8UC1, UNKNOWN)     8.739   8.739     1.00