Subversion Repositories Games.Rick Dangerous

// -------------------------------------------------------------------------

// | xBRZ: "Scale by rules" - high quality image upscaling filter by Zenju |

// -------------------------------------------------------------------------

// using a modified approach of xBR:

// http://board.byuu.org/viewtopic.php?f=10&t=2248

//  - new rule set preserving small image features

//  - highly optimized for performance

//  - support alpha channel

//  - support multithreading

//  - support 64-bit architectures

//  - support processing image slices

//  - support scaling up to 6xBRZ

// -> map source (srcWidth * srcHeight) to target (scale * width x scale * height) image, optionally processing a half-open slice of rows [yFirst, yLast) only

// -> support for source/target pitch in bytes!

// -> if your emulator changes only a few image slices during each cycle (e.g. DOSBox) then there's no need to run xBRZ on the complete image:

//    Just make sure you enlarge the source image slice by 2 rows on top and 2 on bottom (this is the additional range the xBRZ algorithm is using during analysis)

//    CAVEAT: If there are multiple changed slices, make sure they do not overlap after adding these additional rows in order to avoid a memory race condition

//    in the target image data if you are using multiple threads for processing each enlarged slice!

// 

// THREAD-SAFETY: - parts of the same image may be scaled by multiple threads as long as the [yFirst, yLast) ranges do not overlap!

//                - there is a minor inefficiency for the first row of a slice, so avoid processing single rows only; suggestion: process at least 8-16 rows

#include <stddef.h> // for size_t

#include <stdint.h> // for uint32_t

#include <stdbool.h> // for bool

#include <memory.h> // for memset()

#include <limits.h>

#include <math.h>

// prototypes of exported functions

void xbrz_scale (size_t factor, const uint32_t *src, uint32_t *trg, int srcWidth, int srcHeight, bool has_alpha_channel);

void nearest_neighbor_scale (const uint32_t *src, int srcWidth, int srcHeight, uint32_t *trg, int trgWidth, int trgHeight);

// algorithm configuration

#define XBRZ_CFG_LUMINANCE_WEIGHT 1

#define XBRZ_CFG_EQUAL_COLOR_TOLERANCE 30

#define XBRZ_CFG_DOMINANT_DIRECTION_THRESHOLD 3.6

#define XBRZ_CFG_STEEP_DIRECTION_THRESHOLD 2.2

// blend types

#define BLEND_NONE     0

#define BLEND_NORMAL   1 // a normal indication to blend

#define BLEND_DOMINANT 2 // a strong indication to blend

// handy macros

#ifndef MIN

#define MIN(a,b) ((a) < (b) ? (a) : (b))

#endif // MIN

#ifndef MAX

#define MAX(a,b) ((a) > (b) ? (a) : (b))

#endif // MAX

#define GET_BYTE(val,byteno) ((uint8_t) (((val) >> ((byteno) << 3)) & 0xff))

#define GET_BLUE(val)  GET_BYTE (val, 0)

#define GET_GREEN(val) GET_BYTE (val, 1)

#define GET_RED(val)   GET_BYTE (val, 2)

#define GET_ALPHA(val) GET_BYTE (val, 3)

#define CALC_COLOR24(colFront,colBack,M,N) (uint8_t) ((((uint8_t) (colFront)) * ((unsigned int) (M)) + ((uint8_t) (colBack)) * (((unsigned int) (N)) - ((unsigned int) (M)))) / ((unsigned int) (N)))

#define CALC_COLOR32(colFront,colBack,weightFront,weightBack,weightSum) ((uint8_t) ((((uint8_t) (colFront)) * ((unsigned int) (weightFront)) + ((uint8_t) (colBack)) * ((unsigned int) (weightBack))) / ((unsigned int) (weightSum))))

#define BYTE_ADVANCE(buffer,offset) (((char *) buffer) + (offset))

// compress four blend types into a single byte

#define getTopL(b)    ((uint8_t) (0x3 & ((uint8_t) (b) >> 0)))

#define getTopR(b)    ((uint8_t) (0x3 & ((uint8_t) (b) >> 2)))

#define getBottomR(b) ((uint8_t) (0x3 & ((uint8_t) (b) >> 4)))

#define getBottomL(b) ((uint8_t) (0x3 & ((uint8_t) (b) >> 6)))

#define setTopL(b,blend_type)    *(b) |= (((uint8_t) (blend_type)) << 0) // buffer is assumed to be initialized before preprocessing!

#define setTopR(b,blend_type)    *(b) |= (((uint8_t) (blend_type)) << 2)

#define setBottomR(b,blend_type) *(b) |= (((uint8_t) (blend_type)) << 4)

#define setBottomL(b,blend_type) *(b) |= (((uint8_t) (blend_type)) << 6)

typedef struct blendresult_s

{

   uint8_t

      blend_f, blend_g,

      blend_j, blend_k;

} blendresult_t;

typedef struct kernel_3x3_s

{

   uint32_t

      a, b, c,

      d, e, f,

      g, h, i;

} kernel_3x3_t;

typedef struct kernel_4x4_s //kernel for preprocessing step

{

   uint32_t

      a, b, c, d,

      e, f, g, h,

      i, j, k, l,

      m, n, o, p;

} kernel_4x4_t;

typedef struct colorformat_s

{

   int bpp;

   void (*alphagrad) (uint32_t *pixBack, uint32_t pixFront, unsigned int M, unsigned int N);

   double (*dist) (uint32_t pix1, uint32_t pix2);

} colorformat_t;

typedef struct outmatrix_s

{

   size_t size;

   uint32_t* ptr;

   int stride;

} outmatrix_t;

typedef uint32_t *(outmatrixreffunc_t) (outmatrix_t *mat, size_t I, size_t J);

typedef struct scaler_s

{

   int factor;

   void (*blend_line_shallow)           (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref);

   void (*blend_line_steep)             (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref);

   void (*blend_line_steep_and_shallow) (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref);

   void (*blend_line_diagonal)          (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref);

   void (*blend_corner)                 (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref);

} scaler_t;

/////////////////////////////////

// shallow line scaling functions

static void blend_line_shallow_2x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 2 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2 - 1, 1), col, 3, 4);

}

static void blend_line_shallow_3x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 3 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 3 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 3 - 1, 1), col, 3, 4);

   *outmatrix_ref (out, 3 - 1, 2) = col;

}

static void blend_line_shallow_4x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 4 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 4 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 4 - 1, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 4 - 2, 3), col, 3, 4);

   *outmatrix_ref (out, 4 - 1, 2) = col;

   *outmatrix_ref (out, 4 - 1, 3) = col;

}

static void blend_line_shallow_5x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 5 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 3, 4), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 1, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 2, 3), col, 3, 4);

   *outmatrix_ref (out, 5 - 1, 2) = col;

   *outmatrix_ref (out, 5 - 1, 3) = col;

   *outmatrix_ref (out, 5 - 1, 4) = col;

   *outmatrix_ref (out, 5 - 2, 4) = col;

}

static void blend_line_shallow_6x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 6 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 3, 4), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 1, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 2, 3), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 3, 5), col, 3, 4);

   *outmatrix_ref (out, 6 - 1, 2) = col;

   *outmatrix_ref (out, 6 - 1, 3) = col;

   *outmatrix_ref (out, 6 - 1, 4) = col;

   *outmatrix_ref (out, 6 - 1, 5) = col;

   *outmatrix_ref (out, 6 - 2, 4) = col;

   *outmatrix_ref (out, 6 - 2, 5) = col;

}

///////////////////////////////

// steep line scaling functions

static void blend_line_steep_2x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 2 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 2 - 1), col, 3, 4);

}

static void blend_line_steep_3x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 3 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 3 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 3 - 1), col, 3, 4);

   *outmatrix_ref (out, 2, 3 - 1) = col;

}

static void blend_line_steep_4x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 4 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 4 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 4 - 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 4 - 2), col, 3, 4);

   *outmatrix_ref (out, 2, 4 - 1) = col;

   *outmatrix_ref (out, 3, 4 - 1) = col;

}

static void blend_line_steep_5x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 5 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 5 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 4, 5 - 3), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 5 - 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 5 - 2), col, 3, 4);

   *outmatrix_ref (out, 2, 5 - 1) = col;

   *outmatrix_ref (out, 3, 5 - 1) = col;

   *outmatrix_ref (out, 4, 5 - 1) = col;

   *outmatrix_ref (out, 4, 5 - 2) = col;

}

static void blend_line_steep_6x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 6 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 6 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 4, 6 - 3), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 6 - 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 6 - 2), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 5, 6 - 3), col, 3, 4);

   *outmatrix_ref (out, 2, 6 - 1) = col;

   *outmatrix_ref (out, 3, 6 - 1) = col;

   *outmatrix_ref (out, 4, 6 - 1) = col;

   *outmatrix_ref (out, 5, 6 - 1) = col;

   *outmatrix_ref (out, 4, 6 - 2) = col;

   *outmatrix_ref (out, 5, 6 - 2) = col;

}

///////////////////////////////////////////

// steep and shallow line scaling functions

static void blend_line_steep_and_shallow_2x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 0, 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 1), col, 5, 6); // [!] fixes 7/8 used in xBR

}

static void blend_line_steep_and_shallow_3x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 2, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 0, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 2), col, 3, 4);

   *outmatrix_ref (out, 2, 2) = col;

}

static void blend_line_steep_and_shallow_4x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 3, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 3), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 0, 3), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 2), col, 1, 3); // [!] fixes 1/4 used in xBR

   *outmatrix_ref (out, 3, 3) = col;

   *outmatrix_ref (out, 3, 2) = col;

   *outmatrix_ref (out, 2, 3) = col;

}

static void blend_line_steep_and_shallow_5x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 5 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 5 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 5 - 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 5 - 1, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 3), col, 2, 3);

   *outmatrix_ref (out, 2, 5 - 1) = col;

   *outmatrix_ref (out, 3, 5 - 1) = col;

   *outmatrix_ref (out, 4, 5 - 1) = col;

   *outmatrix_ref (out, 5 - 1, 2) = col;

   *outmatrix_ref (out, 5 - 1, 3) = col;

}

static void blend_line_steep_and_shallow_6x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 0, 6 - 1), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 2, 6 - 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 1, 6 - 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 3, 6 - 2), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 1, 0), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 2, 2), col, 1, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 1, 1), col, 3, 4);

   color_format->alphagrad (outmatrix_ref (out, 6 - 2, 3), col, 3, 4);

   *outmatrix_ref (out, 2, 6 - 1) = col;

   *outmatrix_ref (out, 3, 6 - 1) = col;

   *outmatrix_ref (out, 4, 6 - 1) = col;

   *outmatrix_ref (out, 5, 6 - 1) = col;

   *outmatrix_ref (out, 4, 6 - 2) = col;

   *outmatrix_ref (out, 5, 6 - 2) = col;

   *outmatrix_ref (out, 6 - 1, 2) = col;

   *outmatrix_ref (out, 6 - 1, 3) = col;

}

//////////////////////////////////

// diagonal line scaling functions

static void blend_line_diagonal_2x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 1, 1), col, 1, 2);

}

static void blend_line_diagonal_3x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 1, 2), col, 1, 8); // conflict with other rotations for this odd scale

   color_format->alphagrad (outmatrix_ref (out, 2, 1), col, 1, 8);

   color_format->alphagrad (outmatrix_ref (out, 2, 2), col, 7, 8);

}

static void blend_line_diagonal_4x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 4 - 1, 4 / 2), col, 1, 2);

   color_format->alphagrad (outmatrix_ref (out, 4 - 2, 4 / 2 + 1), col, 1, 2);

   *outmatrix_ref (out, 4 - 1, 4 - 1) = col;

}

static void blend_line_diagonal_5x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 5 - 1, 5 / 2 + 0), col, 1, 8); // conflict with other rotations for this odd scale

   color_format->alphagrad (outmatrix_ref (out, 5 - 2, 5 / 2 + 1), col, 1, 8);

   color_format->alphagrad (outmatrix_ref (out, 5 - 3, 5 / 2 + 2), col, 1, 8);

   color_format->alphagrad (outmatrix_ref (out, 4, 3), col, 7, 8);

   color_format->alphagrad (outmatrix_ref (out, 3, 4), col, 7, 8);

   *outmatrix_ref (out, 4, 4) = col;

}

static void blend_line_diagonal_6x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   color_format->alphagrad (outmatrix_ref (out, 6 - 1, 6 / 2 + 0), col, 1, 2);

   color_format->alphagrad (outmatrix_ref (out, 6 - 2, 6 / 2 + 1), col, 1, 2);

   color_format->alphagrad (outmatrix_ref (out, 6 - 3, 6 / 2 + 2), col, 1, 2);

   *outmatrix_ref (out, 6 - 2, 6 - 1) = col;

   *outmatrix_ref (out, 6 - 1, 6 - 1) = col;

   *outmatrix_ref (out, 6 - 1, 6 - 2) = col;

}

///////////////////////////

// corner scaling functions

static void blend_corner_2x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   // model a round corner

   color_format->alphagrad (outmatrix_ref (out, 1, 1), col, 21, 100); // exact: 1 - pi/4 = 0.2146018366

}

static void blend_corner_3x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   // model a round corner

   color_format->alphagrad (outmatrix_ref (out, 2, 2), col, 45, 100); // exact: 0.4545939598

   //color_format->alphagrad (outmatrix_ref (out, 2, 1), col,  7, 256); // 0.02826017254 -> negligible + avoid conflicts with other rotations for this odd scale

   //color_format->alphagrad (outmatrix_ref (out, 1, 2), col,  7, 256); // 0.02826017254

}

static void blend_corner_4x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   // model a round corner

   color_format->alphagrad (outmatrix_ref (out, 3, 3), col, 68, 100); // exact: 0.6848532563

   color_format->alphagrad (outmatrix_ref (out, 3, 2), col,  9, 100); // 0.08677704501

   color_format->alphagrad (outmatrix_ref (out, 2, 3), col,  9, 100); // 0.08677704501

}

static void blend_corner_5x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   // model a round corner

   color_format->alphagrad (outmatrix_ref (out, 4, 4), col, 86, 100); // exact: 0.8631434088

   color_format->alphagrad (outmatrix_ref (out, 4, 3), col, 23, 100); // 0.2306749731

   color_format->alphagrad (outmatrix_ref (out, 3, 4), col, 23, 100); // 0.2306749731

   //color_format->alphagrad (outmatrix_ref (out, 4, 2), col,  1,  64); // 0.01676812367 -> negligible + avoid conflicts with other rotations for this odd scale

   //color_format->alphagrad (outmatrix_ref (out, 2, 4), col,  1,  64); // 0.01676812367

}

static void blend_corner_6x (uint32_t col, outmatrix_t *out, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref)

{

   // model a round corner

   color_format->alphagrad (outmatrix_ref (out, 5, 5), col, 97, 100); // exact: 0.9711013910

   color_format->alphagrad (outmatrix_ref (out, 4, 5), col, 42, 100); // 0.4236372243

   color_format->alphagrad (outmatrix_ref (out, 5, 4), col, 42, 100); // 0.4236372243

   color_format->alphagrad (outmatrix_ref (out, 5, 3), col,  6, 100); // 0.05652034508

   color_format->alphagrad (outmatrix_ref (out, 3, 5), col,  6, 100); // 0.05652034508

}

/////////////////////////////////////

// scaler objects for various factors

static const scaler_t scalers[] =

{

   { 2, blend_line_shallow_2x, blend_line_steep_2x, blend_line_steep_and_shallow_2x, blend_line_diagonal_2x, blend_corner_2x },

   { 3, blend_line_shallow_3x, blend_line_steep_3x, blend_line_steep_and_shallow_3x, blend_line_diagonal_3x, blend_corner_3x },

   { 4, blend_line_shallow_4x, blend_line_steep_4x, blend_line_steep_and_shallow_4x, blend_line_diagonal_4x, blend_corner_4x },

   { 5, blend_line_shallow_5x, blend_line_steep_5x, blend_line_steep_and_shallow_5x, blend_line_diagonal_5x, blend_corner_5x },

   { 6, blend_line_shallow_6x, blend_line_steep_6x, blend_line_steep_and_shallow_6x, blend_line_diagonal_6x, blend_corner_6x },

};

/////////////////////////////////////////////////////

// alpha gradient functions for various color formats

static void alphagrad24 (uint32_t *pixBack, uint32_t pixFront, unsigned int M, unsigned int N)

{

   // blend front color with opacity M / N over opaque background: http://en.wikipedia.org/wiki/Alpha_compositing#Alpha_blending

   *pixBack = ((CALC_COLOR24 (GET_RED   (pixFront), GET_RED   (*pixBack), M, N) << 16)

             | (CALC_COLOR24 (GET_GREEN (pixFront), GET_GREEN (*pixBack), M, N) <<  8)

             | (CALC_COLOR24 (GET_BLUE  (pixFront), GET_BLUE  (*pixBack), M, N) <<  0));

}

static void alphagrad32 (uint32_t *pixBack, uint32_t pixFront, unsigned int M, unsigned int N)

{

   // find intermediate color between two colors with alpha channels (=> NO alpha blending!!!)

   const unsigned int weightFront = GET_ALPHA (pixFront) * M;

   const unsigned int weightBack = GET_ALPHA (*pixBack) * (N - M);

   const unsigned int weightSum = weightFront + weightBack;

   *pixBack = (weightSum == 0 ? 0 :

               (((uint8_t) (weightSum / N))                                                                     << 24)

               | (CALC_COLOR32 (GET_RED   (pixFront), GET_RED   (*pixBack), weightFront, weightBack, weightSum) << 16)

               | (CALC_COLOR32 (GET_GREEN (pixFront), GET_GREEN (*pixBack), weightFront, weightBack, weightSum) <<  8)

               | (CALC_COLOR32 (GET_BLUE  (pixFront), GET_BLUE  (*pixBack), weightFront, weightBack, weightSum) <<  0));

}

/////////////////////////////////////////////////////

// color distance functions for various color formats

static double dist24 (uint32_t pix1, uint32_t pix2)

{

   // 30% perf boost compared to plain distYCbCr()!

   // consumes 64 MB memory; using double is only 2% faster, but takes 128 MB

   static float diffToDist[256 * 256 * 256] = { 0 };

   static bool is_initialized = false;

   if (!is_initialized)

   {

      for (uint32_t i = 0; i < 256 * 256 * 256; ++i) //startup time: 114 ms on Intel Core i5 (four cores)

      {

         const int r_diff = GET_RED (i) * 2 - 0xFF;

         const int g_diff = GET_GREEN (i) * 2 - 0xFF;

         const int b_diff = GET_BLUE (i) * 2 - 0xFF;

         const double k_b = 0.0593; //ITU-R BT.2020 conversion

         const double k_r = 0.2627; //

         const double k_g = 1 - k_b - k_r;

         const double scale_b = 0.5 / (1 - k_b);

         const double scale_r = 0.5 / (1 - k_r);

         const double y = k_r * r_diff + k_g * g_diff + k_b * b_diff; //[!], analog YCbCr!

         const double c_b = scale_b * (b_diff - y);

         const double c_r = scale_r * (r_diff - y);

         diffToDist[i] = (float) (sqrt ((y * y) + (c_b * c_b) + (c_r * c_r)));

      }

      is_initialized = true;

   }

   const int r_diff = (int) GET_RED (pix1) - (int) GET_RED (pix2);

   const int g_diff = (int) GET_GREEN (pix1) - (int) GET_GREEN (pix2);

   const int b_diff = (int) GET_BLUE (pix1) - (int) GET_BLUE (pix2);

   return (diffToDist[  (((r_diff + 0xFF) / 2) << 16) // slightly reduce precision (division by 2) to squeeze value into single byte

                      | (((g_diff + 0xFF) / 2) <<  8)

                      | (((b_diff + 0xFF) / 2) <<  0)]);

}

static double dist32 (uint32_t pix1, uint32_t pix2)

{

   // Requirements for a color distance handling alpha channel: with a1, a2 in [0, 1]

   //    1. if a1 = a2, distance should be: a1 * distYCbCr()

   //    2. if a1 = 0,  distance should be: a2 * distYCbCr(black, white) = a2 * 255

   //    3. if a1 = 1,  ??? maybe: 255 * (1 - a2) + a2 * distYCbCr()

   // return MIN (a1, a2) * distYCbCrBuffered(pix1, pix2) + 255 * abs(a1 - a2);

   // => following code is 15% faster:

   const double d = dist24 (pix1, pix2);

   const double a1 = GET_ALPHA (pix1) / 255.0;

   const double a2 = GET_ALPHA (pix2) / 255.0;

   return (a1 < a2 ? a1 * d + 255 * (a2 - a1) : a2 * d + 255 * (a1 - a2));

}

///////////////////////////////////////

// color format objects for various bpp

static colorformat_t color_format_24 = { 24, alphagrad24, dist24 };

static colorformat_t color_format_32 = { 32, alphagrad32, dist32 };

//////////////////////////////////////////////////////////

// output matrix reference functions for various rotations

static uint32_t *outmatrixref_0   (outmatrix_t *mat, size_t I, size_t J) { return (mat->ptr + I * mat->stride + J); }

static uint32_t *outmatrixref_90  (outmatrix_t *mat, size_t I, size_t J) { return (mat->ptr + (mat->size - 1 - J) * mat->stride + I); }

static uint32_t *outmatrixref_180 (outmatrix_t *mat, size_t I, size_t J) { return (mat->ptr + (mat->size - 1 - I) * mat->stride + (mat->size - 1 - J)); }

static uint32_t *outmatrixref_270 (outmatrix_t *mat, size_t I, size_t J) { return (mat->ptr + J * mat->stride + (mat->size - 1 - I)); }

///////////////////////////

// core algorithm functions

#ifdef _MSC_VER

#define FORCE_INLINE __forceinline

#elif defined __GNUC__

#define FORCE_INLINE __attribute__((always_inline)) inline

#else

#define FORCE_INLINE inline

#endif

static FORCE_INLINE void preprocess_corners (blendresult_t *result, const kernel_4x4_t *ker, colorformat_t *color_format)

{

   // detect blend direction

   // result: F, G, J, K corners of "GradientType"

   // input kernel area naming convention:

   // -----------------

   // | A | B | C | D |

   // ----|---|---|---|

   // | E | F | G | H |   //evaluate the four corners between F, G, J, K

   // ----|---|---|---|   //input pixel is at position F

   // | I | J | K | L |

   // ----|---|---|---|

   // | M | N | O | P |

   // -----------------

   memset (result, 0, sizeof (blendresult_t));

   if (((ker->f == ker->g) && (ker->j == ker->k)) || ((ker->f == ker->j) && (ker->g == ker->k)))

      return;

   const int weight = 4;

   double jg = color_format->dist (ker->i, ker->f) + color_format->dist (ker->f, ker->c) + color_format->dist (ker->n, ker->k) + color_format->dist (ker->k, ker->h) + weight * color_format->dist (ker->j, ker->g);

   double fk = color_format->dist (ker->e, ker->j) + color_format->dist (ker->j, ker->o) + color_format->dist (ker->b, ker->g) + color_format->dist (ker->g, ker->l) + weight * color_format->dist (ker->f, ker->k);

   if (jg < fk) // test sample: 70% of values max(jg, fk) / min(jg, fk) are between 1.1 and 3.7 with median being 1.8

   {

      const bool dominantGradient = XBRZ_CFG_DOMINANT_DIRECTION_THRESHOLD * jg < fk;

      if (ker->f != ker->g && ker->f != ker->j)

         result->blend_f = dominantGradient ? BLEND_DOMINANT : BLEND_NORMAL;

      if (ker->k != ker->j && ker->k != ker->g)

         result->blend_k = dominantGradient ? BLEND_DOMINANT : BLEND_NORMAL;

   }

   else if (fk < jg)

   {

      const bool dominantGradient = XBRZ_CFG_DOMINANT_DIRECTION_THRESHOLD * fk < jg;

      if (ker->j != ker->f && ker->j != ker->k)

         result->blend_j = dominantGradient ? BLEND_DOMINANT : BLEND_NORMAL;

      if (ker->g != ker->f && ker->g != ker->k)

         result->blend_g = dominantGradient ? BLEND_DOMINANT : BLEND_NORMAL;

   }

   return;

}

static FORCE_INLINE void blend_pixel (const scaler_t *scaler, const kernel_3x3_t *ker, uint32_t *target, int trgWidth, uint8_t blendInfo, colorformat_t *color_format, outmatrixreffunc_t outmatrix_ref) //result of preprocessing all four corners of pixel "e"

{

   // input kernel area naming convention:

   // -------------

   // | A | B | C |

   // ----|---|---|

   // | D | E | F | // input pixel is at position E

   // ----|---|---|

   // | G | H | I |

   // -------------

   uint32_t

      a, b, c,

      d, e, f,

      g, h, i;

   uint8_t blend;

   if      (outmatrix_ref == outmatrixref_270) { a = ker->c; b = ker->f; c = ker->i; d = ker->b; e = ker->e; f = ker->h; g = ker->a; h = ker->d; i = ker->g; blend = ((blendInfo << 6) | (blendInfo >> 2)) & 0xff; }

   else if (outmatrix_ref == outmatrixref_180) { a = ker->i; b = ker->h; c = ker->g; d = ker->f; e = ker->e; f = ker->d; g = ker->c; h = ker->b; i = ker->a; blend = ((blendInfo << 4) | (blendInfo >> 4)) & 0xff; }

   else if (outmatrix_ref == outmatrixref_90)  { a = ker->g; b = ker->d; c = ker->a; d = ker->h; e = ker->e; f = ker->b; g = ker->i; h = ker->f; i = ker->c; blend = ((blendInfo << 2) | (blendInfo >> 6)) & 0xff; }

   else                                        { a = ker->a; b = ker->b; c = ker->c; d = ker->d; e = ker->e; f = ker->f; g = ker->g; h = ker->h; i = ker->i; blend = blendInfo; } // blendInfo here is equivalent to ((blendInfo << 0) | (blendInfo >> 8)) & 0xff

   if (getBottomR (blend) >= BLEND_NORMAL)

   {

      uint32_t px;

      bool doLineBlend;