add upstream 11.0

2 files changed, 154 insertions, 150 deletions

diff --git a/xBRZ/src/xbrz.cpp b/xBRZ/src/xbrz.cpp
index 4e9fe6dc..50660b84 100644
--- a/xBRZ/src/xbrz.cpp
+++ b/xBRZ/src/xbrz.cpp
@@ -467,7 +467,10 @@ public:
 
     void readPonm(Kernel_4x4& ker, int x) const //(x, y) is at kernel position E
     {
-        [[likely]] if (const int x_p2 = x + 2; 0 <= x_p2 && x_p2 < srcWidth_)
+#if __has_cpp_attribute(likely)
+        [[likely]]
+#endif
+        if (const int x_p2 = x + 2; 0 <= x_p2 && x_p2 < srcWidth_)
         {
             ker.p = s_m1 ? s_m1[x_p2] : 0;
             ker.o = s_0  ? s_0 [x_p2] : 0;
@@ -520,6 +523,16 @@ private:
 };
 
 
+inline
+void fillBlock(uint32_t* trg, int trgWidth, uint32_t col, int blockSize)
+{
+    for (int y = 0; y < blockSize; ++y, trg += trgWidth)
+        //    std::fill(trg, trg + blockSize, col);
+        for (int x = 0; x < blockSize; ++x)
+            trg[x] = col;
+}
+
+
 template <class Scaler, class ColorDistance, class OobReader> //scaler policy: see "Scaler2x" reference implementation
 void scaleImage(const uint32_t* src, uint32_t* trg, int srcWidth, int srcHeight, const xbrz::ScalerCfg& cfg, int yFirst, int yLast)
 {
@@ -670,7 +683,10 @@ void scaleImage(const uint32_t* src, uint32_t* trg, int srcWidth, int srcHeight,
                 addTopR(blend_xy1, res.blend_h); //set 2nd known corner for (x, y + 1)
                 preProcBuf[x] = blend_xy1; //store on current buffer position for use on next row
 
-                [[likely]] if (x + 1 < srcWidth)
+#if __has_cpp_attribute(likely)
+                [[likely]]
+#endif
+                if (x + 1 < srcWidth)
                 {
                     //blend_xy1 -> blend_x1y1
                     clearAddTopL(blend_xy1, res.blend_i); //set 1st known corner for (x + 1, y + 1) and buffer for use on next column
@@ -680,7 +696,8 @@ void scaleImage(const uint32_t* src, uint32_t* trg, int srcWidth, int srcHeight,
             }
 
             //fill block of size scale * scale with the given color
-            fillBlock(out, trgWidth * sizeof(uint32_t), ker4.e, Scaler::scale, Scaler::scale);
+            fillBlock(out, trgWidth, ker4.e, Scaler::scale);
+
             //place *after* preprocessing step, to not overwrite the results while processing the last pixel!
 
             //blend all four corners of current pixel
@@ -1145,6 +1162,7 @@ void xbrz::scale(size_t factor, const uint32_t* src, uint32_t* trg, int srcWidth
     static_assert(SCALE_FACTOR_MAX == 6);
     switch (colFmt)
     {
+        //*INDENT-OFF*
         case ColorFormat::RGB:
             switch (factor)
             {
@@ -1177,6 +1195,7 @@ void xbrz::scale(size_t factor, const uint32_t* src, uint32_t* trg, int srcWidth
                 case 6: return scaleImage<Scaler6x<ColorGradientARGB>, ColorDistanceUnbufferedARGB, OobReaderTransparent>(src, trg, srcWidth, srcHeight, cfg, yFirst, yLast);
             }
             break;
+        //*INDENT-ON*
     }
     assert(false);
 }
@@ -1186,8 +1205,10 @@ bool xbrz::equalColorTest(uint32_t col1, uint32_t col2, ColorFormat colFmt, doub
 {
     switch (colFmt)
     {
-        case ColorFormat::RGB:  return ColorDistanceRGB::dist(col1, col2, luminanceWeight) < equalColorTolerance;
-        case ColorFormat::ARGB: return ColorDistanceARGB::dist(col1, col2, luminanceWeight) < equalColorTolerance;
+        case ColorFormat::RGB:
+            return ColorDistanceRGB::dist(col1, col2, luminanceWeight) < equalColorTolerance;
+        case ColorFormat::ARGB:
+            return ColorDistanceARGB::dist(col1, col2, luminanceWeight) < equalColorTolerance;
         case ColorFormat::ARGB_UNBUFFERED:
             return ColorDistanceUnbufferedARGB::dist(col1, col2, luminanceWeight) < equalColorTolerance;
     }
@@ -1199,18 +1220,32 @@ bool xbrz::equalColorTest(uint32_t col1, uint32_t col2, ColorFormat colFmt, doub
 void xbrz::bilinearScale(const uint32_t* src, int srcWidth, int srcHeight,
                          /**/  uint32_t* trg, int trgWidth, int trgHeight)
 {
-    bilinearScale(src, srcWidth, srcHeight, srcWidth * sizeof(uint32_t),
-                  trg, trgWidth, trgHeight, trgWidth * sizeof(uint32_t),
-    0, trgHeight, [](uint32_t pix) { return pix; });
+    const auto imgReader = [src, srcWidth](int x, int y, BytePixel& pix)
+    {
+        static_assert(sizeof(pix) == sizeof(uint32_t));
+        std::memcpy(pix, src + y * srcWidth + x, sizeof(pix));
+    };
+
+    const auto imgWriter = [trg](const xbrz::BytePixel& pix) mutable { std::memcpy(trg++, pix, sizeof(pix)); };
+
+    bilinearScale(imgReader, srcWidth, srcHeight,
+                  imgWriter, trgWidth, trgHeight, 0, trgHeight);
 }
 
 
 void xbrz::nearestNeighborScale(const uint32_t* src, int srcWidth, int srcHeight,
                                 /**/  uint32_t* trg, int trgWidth, int trgHeight)
 {
-    nearestNeighborScale(src, srcWidth, srcHeight, srcWidth * sizeof(uint32_t),
-                         trg, trgWidth, trgHeight, trgWidth * sizeof(uint32_t),
-    0, trgHeight, [](uint32_t pix) { return pix; });
+    const auto imgReader = [src, srcWidth](int x, int y, BytePixel& pix)
+    {
+        static_assert(sizeof(pix) == sizeof(uint32_t));
+        std::memcpy(pix, src + y * srcWidth + x, sizeof(pix));
+    };
+
+    const auto imgWriter = [trg](const xbrz::BytePixel& pix) mutable { std::memcpy(trg++, pix, sizeof(pix)); };
+
+    nearestNeighborScale(imgReader, srcWidth, srcHeight,
+                         imgWriter, trgWidth, trgHeight, 0, trgHeight);
 }
 
 
diff --git a/xBRZ/src/xbrz_tools.h b/xBRZ/src/xbrz_tools.h
index 9d8e8247..404849ca 100644
--- a/xBRZ/src/xbrz_tools.h
+++ b/xBRZ/src/xbrz_tools.h
@@ -33,8 +33,8 @@ inline unsigned char getRed  (uint32_t pix) { return getByte<2>(pix); }
 inline unsigned char getGreen(uint32_t pix) { return getByte<1>(pix); }
 inline unsigned char getBlue (uint32_t pix) { return getByte<0>(pix); }
 
-inline uint32_t makePixel(unsigned char a, unsigned char r, unsigned char g, unsigned char b) { return (a << 24) | (r << 16) | (g << 8) | b; }
-inline uint32_t makePixel(                 unsigned char r, unsigned char g, unsigned char b) { return             (r << 16) | (g << 8) | b; }
+inline uint32_t makePixel(uint32_t a, uint32_t r, uint32_t g, uint32_t b) { return (a << 24) | (r << 16) | (g << 8) | b; }
+inline uint32_t makePixel(            uint32_t r, uint32_t g, uint32_t b) { return             (r << 16) | (g << 8) | b; }
 
 inline uint32_t rgb555to888(uint16_t pix) { return ((pix & 0x7C00) << 9) | ((pix & 0x03E0) << 6) | ((pix & 0x001F) << 3); }
 inline uint32_t rgb565to888(uint16_t pix) { return ((pix & 0xF800) << 8) | ((pix & 0x07E0) << 5) | ((pix & 0x001F) << 3); }
@@ -43,49 +43,30 @@ inline uint16_t rgb888to555(uint32_t pix) { return static_cast<uint16_t>(((pix &
 inline uint16_t rgb888to565(uint32_t pix) { return static_cast<uint16_t>(((pix & 0xF80000) >> 8) | ((pix & 0x00FC00) >> 5) | ((pix & 0x0000F8) >> 3)); }
 
 
-template <class Pix> inline
-Pix* byteAdvance(Pix* ptr, int bytes)
-{
-    using PixNonConst = typename std::remove_cv<Pix>::type;
-    using PixByte     = typename std::conditional<std::is_same<Pix, PixNonConst>::value, char, const char>::type;
-
-    static_assert(std::is_integral<PixNonConst>::value, "Pix* is expected to be cast-able to char*");
+using BytePixel = unsigned char[4]; //unspecified byte order
+static_assert(std::alignment_of_v<BytePixel> == 1); // :)
 
-    return reinterpret_cast<Pix*>(reinterpret_cast<PixByte*>(ptr) + bytes);
-}
 
-
-//fill block  with the given color
-template <class Pix> inline
-void fillBlock(Pix* trg, int pitch /*[bytes]*/, Pix col, int blockWidth, int blockHeight)
+template <class PixReader, class PixWriter> inline
+void unscaledCopy(PixReader srcReader /* (int x, int y, BytePixel& pix) */,
+                  PixWriter trgWriter /* (const BytePixel& pix)         */, int width, int height)
 {
-    //for (int y = 0; y < blockHeight; ++y, trg = byteAdvance(trg, pitch))
-    //    std::fill(trg, trg + blockWidth, col);
-
-    for (int y = 0; y < blockHeight; ++y, trg = byteAdvance(trg, pitch))
-        for (int x = 0; x < blockWidth; ++x)
-            trg[x] = col;
+    for (int y = 0; y < height; ++y)
+        for (int x = 0; x < width; ++x)
+        {
+            BytePixel pix; //uninitialized
+            srcReader(x, y, pix);
+            trgWriter(pix);
+        }
 }
 
 
 //nearest-neighbor (going over target image - slow for upscaling, since source is read multiple times missing out on cache! Fast for similar image sizes!)
-template <class PixSrc, class PixTrg, class PixConverter>
-void nearestNeighborScale(const PixSrc* src, int srcWidth, int srcHeight, int srcPitch /*[bytes]*/,
-                          /**/  PixTrg* trg, int trgWidth, int trgHeight, int trgPitch /*[bytes]*/,
-                          int yFirst, int yLast, PixConverter pixCvrt /*convert PixSrc to PixTrg*/)
+template <class PixReader, class PixWriter>
+void nearestNeighborScale(PixReader srcReader /* (int x, int y, BytePixel& pix) */, int srcWidth, int srcHeight,
+                          PixWriter trgWriter /* (const BytePixel& pix)         */, int trgWidth, int trgHeight,
+                          int yFirst, int yLast)
 {
-    static_assert(std::is_integral<PixSrc>::value, "PixSrc* is expected to be cast-able to char*");
-    static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
-
-    static_assert(std::is_same<decltype(pixCvrt(PixSrc())), PixTrg>::value, "PixConverter returning wrong pixel format");
-
-    if (srcPitch < srcWidth * static_cast<int>(sizeof(PixSrc))  ||
-        trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
-    {
-        assert(false);
-        return;
-    }
-
     yFirst = std::max(yFirst, 0);
     yLast  = std::min(yLast, trgHeight);
     if (yFirst >= yLast || srcHeight <= 0 || srcWidth <= 0) return;
@@ -93,92 +74,28 @@ void nearestNeighborScale(const PixSrc* src, int srcWidth, int srcHeight, int sr
     for (int y = yFirst; y < yLast; ++y)
     {
         const int ySrc = srcHeight * y / trgHeight;
-        const PixSrc* const srcLine = byteAdvance(src, ySrc * srcPitch);
-        /**/  PixTrg* const trgLine = byteAdvance(trg, y    * trgPitch);
 
         for (int x = 0; x < trgWidth; ++x)
         {
             const int xSrc = srcWidth * x / trgWidth;
-            trgLine[x] = pixCvrt(srcLine[xSrc]);
-        }
-    }
-}
-
-
-//nearest-neighbor (going over source image - fast for upscaling, since source is read only once
-template <class PixSrc, class PixTrg, class PixConverter>
-void nearestNeighborScaleOverSource(const PixSrc* src, int srcWidth, int srcHeight, int srcPitch /*[bytes]*/,
-                                    /**/  PixTrg* trg, int trgWidth, int trgHeight, int trgPitch /*[bytes]*/,
-                                    int yFirst, int yLast, PixConverter pixCvrt /*convert PixSrc to PixTrg*/)
-{
-    static_assert(std::is_integral<PixSrc>::value, "PixSrc* is expected to be cast-able to char*");
-    static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
-
-    static_assert(std::is_same<decltype(pixCvrt(PixSrc())), PixTrg>::value, "PixConverter returning wrong pixel format");
-
-    if (srcPitch < srcWidth * static_cast<int>(sizeof(PixSrc))  ||
-        trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
-    {
-        assert(false);
-        return;
-    }
-
-    yFirst = std::max(yFirst, 0);
-    yLast  = std::min(yLast, srcHeight);
-    if (yFirst >= yLast || trgWidth <= 0 || trgHeight <= 0) return;
-
-    for (int y = yFirst; y < yLast; ++y)
-    {
-        //mathematically: ySrc = floor(srcHeight * yTrg / trgHeight)
-        // => search for integers in: [ySrc, ySrc + 1) * trgHeight / srcHeight
-
-        //keep within for loop to support MT input slices!
-        const int yTrgFirst = ( y      * trgHeight + srcHeight - 1) / srcHeight; //=ceil(y * trgHeight / srcHeight)
-        const int yTrgLast  = ((y + 1) * trgHeight + srcHeight - 1) / srcHeight; //=ceil(((y + 1) * trgHeight) / srcHeight)
-        const int blockHeight = yTrgLast - yTrgFirst;
-
-        if (blockHeight > 0)
-        {
-            const PixSrc* srcLine = byteAdvance(src, y         * srcPitch);
-            /**/  PixTrg* trgLine = byteAdvance(trg, yTrgFirst * trgPitch);
-            int xTrgFirst = 0;
-
-            for (int x = 0; x < srcWidth; ++x)
-            {
-                const int xTrgLast = ((x + 1) * trgWidth + srcWidth - 1) / srcWidth;
-                const int blockWidth = xTrgLast - xTrgFirst;
-                if (blockWidth > 0)
-                {
-                    xTrgFirst = xTrgLast;
 
-                    const auto trgPix = pixCvrt(srcLine[x]);
-                    fillBlock(trgLine, trgPitch, trgPix, blockWidth, blockHeight);
-                    trgLine += blockWidth;
-                }
-            }
+            BytePixel pix; //uninitialized
+            srcReader(xSrc, ySrc, pix);
+            trgWriter(pix);
         }
     }
 }
 
 
-template <class PixTrg, class PixConverter>
-void bilinearScale(const uint32_t* src, int srcWidth, int srcHeight, int srcPitch,
-                   /**/    PixTrg* trg, int trgWidth, int trgHeight, int trgPitch,
-                   int yFirst, int yLast, PixConverter pixCvrt /*convert uint32_t to PixTrg*/)
+template <class PixReader, class PixWriter>
+void bilinearScale(PixReader srcReader /* (int x, int y, BytePixel& pix) */, int srcWidth, int srcHeight,
+                   PixWriter trgWriter /* (const BytePixel& pix)         */, int trgWidth, int trgHeight,
+                   int yFirst, int yLast)
 {
-    static_assert(std::is_integral<PixTrg>::value,                            "PixTrg* is expected to be cast-able to char*");
-    static_assert(std::is_same<decltype(pixCvrt(uint32_t())), PixTrg>::value, "PixConverter returning wrong pixel format");
-
-    if (srcPitch < srcWidth * static_cast<int>(sizeof(uint32_t)) ||
-        trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
-    {
-        assert(false);
-        return;
-    }
-
     yFirst = std::max(yFirst, 0);
     yLast  = std::min(yLast, trgHeight);
-    if (yFirst >= yLast || srcHeight <= 0 || srcWidth <= 0) return;
+    if (yFirst >= yLast || srcHeight <= 0 || srcWidth <= 0)
+        return;
 
     const double scaleX = static_cast<double>(trgWidth ) / srcWidth;
     const double scaleY = static_cast<double>(trgHeight) / srcHeight;
@@ -198,7 +115,8 @@ void bilinearScale(const uint32_t* src, int srcWidth, int srcHeight, int srcPitc
     {
         const int x1 = srcWidth * x / trgWidth;
         int x2 = x1 + 1;
-        if (x2 == srcWidth) --x2;
+        if (x2 == srcWidth)
+            --x2;
 
         const double xx1 = x / scaleX - x1;
         const double x2x = 1 - xx1;
@@ -210,58 +128,109 @@ void bilinearScale(const uint32_t* src, int srcWidth, int srcHeight, int srcPitc
     {
         const int y1 = srcHeight * y / trgHeight;
         int y2 = y1 + 1;
-        if (y2 == srcHeight) --y2;
+        if (y2 == srcHeight)
+            --y2;
 
         const double yy1 = y / scaleY - y1;
         const double y2y = 1 - yy1;
 
-        const uint32_t* const srcLine     = byteAdvance(src, y1 * srcPitch);
-        const uint32_t* const srcLineNext = byteAdvance(src, y2 * srcPitch);
-        PixTrg*         const trgLine     = byteAdvance(trg, y  * trgPitch);
-
         for (int x = 0; x < trgWidth; ++x)
         {
             //perf: do NOT "simplify" the variable layout without measurement!
-            const int     x1 = buf[x].x1;
-            const int     x2 = buf[x].x2;
-            const double xx1 = buf[x].xx1;
-            const double x2x = buf[x].x2x;
+            const CoeffsX& bufX = buf[x];
+            const int     x1 = bufX.x1;
+            const int     x2 = bufX.x2;
+            const double xx1 = bufX.xx1;
+            const double x2x = bufX.x2x;
 
             const double x2xy2y = x2x * y2y;
             const double xx1y2y = xx1 * y2y;
             const double x2xyy1 = x2x * yy1;
             const double xx1yy1 = xx1 * yy1;
 
-            auto interpolate = [=](int offset)
+            BytePixel pix11; //
+            BytePixel pix21; //uninitialized
+            BytePixel pix12; //
+            BytePixel pix22; //
+
+            srcReader(x1, y1, pix11); //
+            srcReader(x2, y1, pix21); //perf: srcReader has to (re-)calculate row using y
+            srcReader(x1, y2, pix12); //      => ~7% additional runtime
+            srcReader(x2, y2, pix22); //
+
+            const auto interpolate = [&](int offset)
             {
                 /* https://en.wikipedia.org/wiki/Bilinear_interpolation
                      (c11(x2 - x) + c21(x - x1)) * (y2 - y ) +
-                     (c12(x2 - x) + c22(x - x1)) * (y  - y1)                          */
-                const auto c11 = (srcLine    [x1] >> (8 * offset)) & 0xff;
-                const auto c21 = (srcLine    [x2] >> (8 * offset)) & 0xff;
-                const auto c12 = (srcLineNext[x1] >> (8 * offset)) & 0xff;
-                const auto c22 = (srcLineNext[x2] >> (8 * offset)) & 0xff;
-
-                return c11 * x2xy2y + c21 * xx1y2y +
-                       c12 * x2xyy1 + c22 * xx1yy1;
+                     (c12(x2 - x) + c22(x - x1)) * (y  - y1)      */
+                return static_cast<unsigned char>(pix11[offset] * x2xy2y + pix21[offset] * xx1y2y +
+                                                  pix12[offset] * x2xyy1 + pix22[offset] * xx1yy1 + 0.5);
             };
+            trgWriter(BytePixel{interpolate(0),
+                                interpolate(1),
+                                interpolate(2),
+                                interpolate(3)});
+        }
+    }
+}
 
-            const double bi = interpolate(0);
-            const double gi = interpolate(1);
-            const double ri = interpolate(2);
-            const double ai = interpolate(3);
 
-            const auto b = static_cast<uint32_t>(bi + 0.5);
-            const auto g = static_cast<uint32_t>(gi + 0.5);
-            const auto r = static_cast<uint32_t>(ri + 0.5);
-            const auto a = static_cast<uint32_t>(ai + 0.5);
+#if 0
+//nearest-neighbor (going over source image - fast for upscaling, since source is read only once
+template <class PixSrc, class PixTrg, class PixConverter>
+void nearestNeighborScaleOverSource(const PixSrc* src, int srcWidth, int srcHeight, int srcPitch /*[bytes]*/,
+                                    /**/  PixTrg* trg, int trgWidth, int trgHeight, int trgPitch /*[bytes]*/,
+                                    int yFirst, int yLast, PixConverter pixCvrt /*convert PixSrc to PixTrg*/)
+{
+    static_assert(std::is_integral<PixSrc>::value, "PixSrc* is expected to be cast-able to char*");
+    static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
 
-            const uint32_t trgPix = (a << 24) | (r << 16) | (g << 8) | b;
+    static_assert(std::is_same<decltype(pixCvrt(PixSrc())), PixTrg>::value, "PixConverter returning wrong pixel format");
 
-            trgLine[x] = pixCvrt(trgPix);
+    if (srcPitch < srcWidth * static_cast<int>(sizeof(PixSrc))  ||
+        trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
+    {
+        assert(false);
+        return;
+    }
+
+    yFirst = std::max(yFirst, 0);
+    yLast  = std::min(yLast, srcHeight);
+    if (yFirst >= yLast || trgWidth <= 0 || trgHeight <= 0) return;
+
+    for (int y = yFirst; y < yLast; ++y)
+    {
+        //mathematically: ySrc = floor(srcHeight * yTrg / trgHeight)
+        // => search for integers in: [ySrc, ySrc + 1) * trgHeight / srcHeight
+
+        //keep within for loop to support MT input slices!
+        const int yTrgFirst = ( y      * trgHeight + srcHeight - 1) / srcHeight; //=ceil(y * trgHeight / srcHeight)
+        const int yTrgLast  = ((y + 1) * trgHeight + srcHeight - 1) / srcHeight; //=ceil(((y + 1) * trgHeight) / srcHeight)
+        const int blockHeight = yTrgLast - yTrgFirst;
+
+        if (blockHeight > 0)
+        {
+            const PixSrc* srcLine = byteAdvance(src, y         * srcPitch);
+            /**/  PixTrg* trgLine = byteAdvance(trg, yTrgFirst * trgPitch);
+            int xTrgFirst = 0;
+
+            for (int x = 0; x < srcWidth; ++x)
+            {
+                const int xTrgLast = ((x + 1) * trgWidth + srcWidth - 1) / srcWidth;
+                const int blockWidth = xTrgLast - xTrgFirst;
+                if (blockWidth > 0)
+                {
+                    xTrgFirst = xTrgLast;
+
+                    const auto trgPix = pixCvrt(srcLine[x]);
+                    fillBlock(trgLine, trgPitch, trgPix, blockWidth, blockHeight);
+                    trgLine += blockWidth;
+                }
+            }
         }
     }
 }
+#endif
 }
 
 #endif //XBRZ_TOOLS_H_825480175091875