[Gin890]

Make the ImageScaleBrightness, ImageFilter and BinaryMatrix kernel tests real tests. Each test now checks every SIMD and NEON implementation that the CPU supports against a scalar reference, on real images.

Before, these tests were debug code. They were not registered, so CI did not run them. They tested only the implementation that the dispatch picks, and had timing loops.

The new tests found three bugs. Each fix is its own commit:

• x86 scale_brightness(): SSE4.1, AVX2 and AVX-512 rounded each channel to the nearest integer, but Default and NEON truncate. Now all of them truncate. On x86, this changes some channels by 1 in ExactImageMatcher and ImageMatchDetector.
• PackedBinaryMatrix::submatrix(): if the width is a multiple of 64, or the height is a multiple of the tile height, it cleared the whole last column or row of tiles. The tile height depends on the CPU (4 to 64 rows), so the result was different on each machine. SparseBinaryMatrix::submatrix() got this fix in ecdc7fc. In production, this changes SubObjectTemplateMatcher::rmsd_with_background_replace().
• filter_rgb32_range() with several filters in one call: it passed a field that is not set yet as the replace flag, so the result was undefined. It now passes each filter's own flag. The PokemonLA MountDetector uses this function. This broke in 5bb2f45. A new test checks this function and to_blackwhite_rgb32_range() with several filters.

Other changes:

• Remove the two Euclidean filter tests from the ImageFilter tests. Kernels::RGB32Euclidean (RGB32 Euclidean filters: count the max distance as in range on x86 SIMD #1637) already tests them.
• The BinaryMatrix tile checks now also run on the Default tile. The x86 tiles do not compile in that file, so the matrix test and the Waterfill test check them.

The images are from the CommandLineTests repo. No width or height is a multiple of 8, so the tests also check the leftover pixels after the SIMD loops and the tiles at the edges. #1554 also changes these three test files, so it will need a merge fix.

0 test files added, 3 test files changed:

• SerialPrograms/Source/Kernels/ImageScaleBrightness/Kernels_ImageScaleBrightness_Tests.cpp
• SerialPrograms/Source/Kernels/ImageFilters/Kernels_ImageFilter_Tests.cpp
• SerialPrograms/Source/Kernels/BinaryMatrix/Kernels_BinaryMatrix_Tests.cpp

Test results: all tests pass on all six CI runners. These runs used the first 5 commits on the previous main (d315488), before the multi-filter fix and its test. Each implementation runs on 4 images.

Runners	Implementations tested
Ubuntu clang, Windows Clang (AVX-512)	Brightness and range filters: Default, SSE4.1 or SSE4.2, AVX2, AVX-512 Mask and binary matrix: 64x4 Default, 64x8 SSE4.2, 64x16 AVX2, 64x32 and 64x64 AVX-512 Tile: 64x4 Default
Ubuntu Default, Windows Default, Mac Intel (AVX2)	Brightness and range filters: Default, SSE4.1 or SSE4.2, AVX2 Mask and binary matrix: 64x4 Default, 64x8 SSE4.2, 64x16 AVX2 Tile: 64x4 Default
Mac (Apple Silicon)	Brightness and range filters: Default, NEON Mask and binary matrix: 64x4 Default, 64x8 NEON Tile: 64x4 Default, 64x8 NEON NEON PartialWordAccess

🤖 Generated with Claude Code