Math: implement BC2 flipping as well.

It's a weird format nobody uses, but since I have the full context in my
head now, it's easy to do.

src/Magnum/Math/ColorBatch.cpp

@@ -60,6 +60,36 @@ inline void yFlipBc1BlockInPlace(char* const data) {
     }
 }
 
+inline void yFlipBc2BlockInPlace(char* data) {
+    /* The 128-bit block is laid out as follows:
+
+       - 8 bytes for 4x4 4-bit alpha values, same order as BC1
+       - 2 bytes for first endpoint color
+       - 2 bytes for second endpoint color
+       - 4 bytes for 4x4 2-bit color indices, same order as BC1
+
+       Which means, swapping the alphas and then doing the same as with BC1
+       for the color indices in the second half.
+       https://learn.microsoft.com/cs-cz/windows/win32/direct3d10/d3d10-graphics-programming-guide-resources-block-compression#bc2 */
+    {
+        char tmp1 = data[0];
+        char tmp2 = data[1];
+        data[0] = data[6];
+        data[1] = data[7];
+        data[6] = tmp1;
+        data[7] = tmp2;
+    } {
+        char tmp1 = data[2];
+        char tmp2 = data[3];
+        data[2] = data[4];
+        data[3] = data[5];
+        data[4] = tmp1;
+        data[5] = tmp2;
+    }
+
+    yFlipBc1BlockInPlace(data + 8);
+}
+
 inline void yFlipBc4BlockInPlace(char* data) {
     /* The 64-bit block is laid out as follows:
 
@@ -173,6 +203,14 @@ void yFlipBc1InPlace(const Containers::StridedArrayView4D<char>& blocks) {
     );
 }
 
+void yFlipBc2InPlace(const Containers::StridedArrayView4D<char>& blocks) {
+    yFlipBlocksInPlace<16, yFlipBc2BlockInPlace>(blocks
+        #ifndef CORRADE_NO_ASSERT
+        , "Math::yFlipBc2InPlace():"
+        #endif
+    );
+}
+
 void yFlipBc3InPlace(const Containers::StridedArrayView4D<char>& blocks) {
     yFlipBlocksInPlace<16, yFlipBc3BlockInPlace>(blocks
         #ifndef CORRADE_NO_ASSERT

src/Magnum/Math/ColorBatch.h

@@ -57,7 +57,24 @@ expected to be contiguous with size of 8.
 */
 MAGNUM_EXPORT void yFlipBc1InPlace(const Corrade::Containers::StridedArrayView4D<char>& blocks);
 
-/** @todo BC2, if used at all for anything */
+/**
+@brief Y-flip BC2 texture blocks in-place
+@m_since_latest
+
+Performs a Y flip of given 3D image by flipping block order and modifying
+internal block representation to encode the same information, just upside down.
+No decoding or re-encoding of the block data is performed, thus the operation
+is lossless. However note that this operation flips full blocks --- if size of
+the actual image isn't whole blocks, the flipped image will be shifted compared
+to the original, possibly with garbage data appearing in the first few rows.
+
+First dimension is expected to be image slices, second block rows, third
+2D blocks, fourth the 128-bit 4x4 block data, i.e. the last dimension is
+expected to be contiguous with size of 16.
+@see @ref CompressedPixelFormat::Bc2RGBAUnorm,
+    @ref CompressedPixelFormat::Bc2RGBASrgb
+*/
+MAGNUM_EXPORT void yFlipBc2InPlace(const Corrade::Containers::StridedArrayView4D<char>& blocks);
 
 /**
 @brief Y-flip BC3 texture blocks in-place

src/Magnum/Math/Test/CMakeLists.txt

@@ -50,6 +50,7 @@ corrade_add_test(MathColorBatchTest ColorBatchTest.cpp
     LIBRARIES MagnumMathTestLib MagnumDebugTools
     FILES
         ColorBatchTestFiles/bc1.png
+        ColorBatchTestFiles/bc2.png
         ColorBatchTestFiles/bc3.png
         ColorBatchTestFiles/bc4.png
         ColorBatchTestFiles/bc5.png

src/Magnum/Math/Test/ColorBatchTest.cpp

@@ -192,6 +192,31 @@ const struct {
         '\xdd', '\xff', '\xa8', '\x6b', '\x09', '\x25', '\x95', '\x55',
         '\x79', '\xd6', '\xa7', '\x39', '\x35', '\xd5', '\x55', '\x5c'
     }}},
+    {"BC2", CompressedPixelFormat::Bc2RGBAUnorm, {1, 4}, {InPlaceInit, {
+        /* ./extract-interesting-blocks.py testcard_bc2.dds bc2.png --offset 88
+           (image taken from the bcdec repository test files) */
+        /* [7, 52], 2.427 */
+        '\xf8', '\xff', '\x91', '\xff', '\x20', '\xfa', '\x00', '\xc4',
+        '\xff', '\xff', '\x0f', '\xb3', '\x55', '\x55', '\x54', '\x50',
+        /* [6, 51], 2.427 */
+        '\x90', '\xff', '\x20', '\xfb', '\x00', '\xd4', '\x00', '\x60',
+        '\xff', '\xff', '\x0f', '\xb3', '\x54', '\x54', '\x50', '\x40',
+        /* [53, 8], 2.412 */
+        '\x6d', '\x00', '\xdf', '\x06', '\xff', '\x7e', '\xff', '\xff',
+        '\xff', '\xff', '\xf6', '\x63', '\x05', '\x15', '\x55', '\x55',
+        /* [10, 8], 2.412 */
+        '\x00', '\xd6', '\x60', '\xfd', '\xe7', '\xff', '\xff', '\xff',
+        '\xff', '\xff', '\xf6', '\x63', '\x50', '\x54', '\x55', '\x55',
+    }}, yFlipBc2InPlace, "bc2.png", {InPlaceInit, {
+        '\xff', '\xff', '\xe7', '\xff', '\x60', '\xfd', '\x00', '\xd6',
+        '\xff', '\xff', '\xf6', '\x63', '\x55', '\x55', '\x54', '\x50',
+        '\xff', '\xff', '\xff', '\x7e', '\xdf', '\x06', '\x6d', '\x00',
+        '\xff', '\xff', '\xf6', '\x63', '\x55', '\x55', '\x15', '\x05',
+        '\x00', '\x60', '\x00', '\xd4', '\x20', '\xfb', '\x90', '\xff',
+        '\xff', '\xff', '\x0f', '\xb3', '\x40', '\x50', '\x54', '\x54',
+        '\x00', '\xc4', '\x20', '\xfa', '\x91', '\xff', '\xf8', '\xff',
+        '\xff', '\xff', '\x0f', '\xb3', '\x50', '\x54', '\x55', '\x55'
+    }}},
     {"BC3", CompressedPixelFormat::Bc3RGBAUnorm, {1, 4}, {InPlaceInit, {
         /* ./extract-interesting-blocks.py dice_bc3.dds bc3.png --offset 148
            (image taken from the bcdec repository test files) */

src/Magnum/PixelFormat.h

@@ -1078,7 +1078,8 @@ enum class CompressedPixelFormat: UnsignedInt {
      * @ref Vk::PixelFormat::CompressedBc2RGBAUnorm;
      * @m_class{m-doc-external} [DXGI_FORMAT_BC2_UNORM](https://docs.microsoft.com/en-us/windows/win32/api/dxgiformat/ne-dxgiformat-dxgi_format)
      * or @m_class{m-doc-external} [MTLPixelFormatBC2_RGBA](https://developer.apple.com/documentation/metal/mtlpixelformat/mtlpixelformatbc2_rgba?language=objc).
-     * @see @relativeref{Trade,BcDecImageConverter}
+     * @see @ref Math::yFlipBc2InPlace(),
+     *      @relativeref{Trade,BcDecImageConverter}
      * @m_keywords{DXGI_FORMAT_BC2_UNORM MTLPixelFormatBC2_RGBA}
      */
     Bc2RGBAUnorm,
@@ -1094,7 +1095,8 @@ enum class CompressedPixelFormat: UnsignedInt {
      * @ref Vk::PixelFormat::CompressedBc2RGBASrgb;
      * @m_class{m-doc-external} [DXGI_FORMAT_BC2_UNORM_SRGB](https://docs.microsoft.com/en-us/windows/win32/api/dxgiformat/ne-dxgiformat-dxgi_format)
      * or @m_class{m-doc-external} [MTLPixelFormatBC2_RGBA_sRGB](https://developer.apple.com/documentation/metal/mtlpixelformat/mtlpixelformatbc2_rgba_srgb?language=objc).
-     * @see @relativeref{Trade,BcDecImageConverter}
+     * @see @ref Math::yFlipBc2InPlace(),
+     *      @relativeref{Trade,BcDecImageConverter}
      * @m_keywords{DXGI_FORMAT_BC2_UNORM_SRGB MTLPixelFormatBC2_RGBA_sRGB}
      * @m_since{2019,10}
      */