scale_argb.cc

/*
 *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS. All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#include "libyuv/scale.h"

#include <assert.h>
#include <string.h>

#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"  // For CopyARGB
#include "libyuv/row.h"
#include "libyuv/scale_row.h"

#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif

static __inline int Abs(int v) {
  return v >= 0 ? v : -v;
}

// ScaleARGB ARGB, 1/2
// This is an optimized version for scaling down a ARGB to 1/2 of
// its original size.

static void ScaleARGBDown2(int /* src_width */, int /* src_height */,
                           int dst_width, int dst_height,
                           int src_stride, int dst_stride,
                           const uint8* src_argb, uint8* dst_argb,
                           int x, int dx, int y, int dy,
                           FilterMode filtering) {
  assert(dx == 65536 * 2);  // Test scale factor of 2.
  assert((dy & 0x1ffff) == 0);  // Test vertical scale is multiple of 2.
  // Advance to odd row, even column.
  if (filtering) {
    src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
  } else {
    src_argb += (y >> 16) * src_stride + ((x >> 16) - 1) * 4;
  }
  int row_stride = src_stride * (dy >> 16);
  void (*ScaleARGBRowDown2)(const uint8* src_argb, ptrdiff_t src_stride,
                            uint8* dst_argb, int dst_width) =
    filtering == kFilterNone ? ScaleARGBRowDown2_C :
        (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_C :
        ScaleARGBRowDown2Box_C);
#if defined(HAS_SCALEARGBROWDOWN2_SSE2)
  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4) &&
      IS_ALIGNED(src_argb, 16) && IS_ALIGNED(row_stride, 16) &&
      IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
    ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_SSE2 :
        (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2 :
        ScaleARGBRowDown2Box_SSE2);
  }
#elif defined(HAS_SCALEARGBROWDOWN2_NEON)
  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 8) &&
      IS_ALIGNED(src_argb, 4) && IS_ALIGNED(row_stride, 4)) {
    ScaleARGBRowDown2 = filtering ? ScaleARGBRowDown2Box_NEON :
        ScaleARGBRowDown2_NEON;
  }
#endif

  if (filtering == kFilterLinear) {
    src_stride = 0;
  }
  for (int y = 0; y < dst_height; ++y) {
    ScaleARGBRowDown2(src_argb, src_stride, dst_argb, dst_width);
    src_argb += row_stride;
    dst_argb += dst_stride;
  }
}

// ScaleARGB ARGB Even
// This is an optimized version for scaling down a ARGB to even
// multiple of its original size.
static void ScaleARGBDownEven(int src_width, int src_height,
                              int dst_width, int dst_height,
                              int src_stride, int dst_stride,
                              const uint8* src_argb, uint8* dst_argb,
                              int x, int dx, int y, int dy,
                              FilterMode filtering) {
  assert(IS_ALIGNED(src_width, 2));
  assert(IS_ALIGNED(src_height, 2));
  int col_step = dx >> 16;
  int row_stride = (dy >> 16) * src_stride;
  src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
  void (*ScaleARGBRowDownEven)(const uint8* src_argb, ptrdiff_t src_stride,
                               int src_step, uint8* dst_argb, int dst_width) =
      filtering ? ScaleARGBRowDownEvenBox_C : ScaleARGBRowDownEven_C;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
  if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 4) &&
      IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
    ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_SSE2 :
        ScaleARGBRowDownEven_SSE2;
  }
#elif defined(HAS_SCALEARGBROWDOWNEVEN_NEON)
  if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 4) &&
      IS_ALIGNED(src_argb, 4)) {
    ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_NEON :
        ScaleARGBRowDownEven_NEON;
  }
#endif

  if (filtering == kFilterLinear) {
    src_stride = 0;
  }
  for (int y = 0; y < dst_height; ++y) {
    ScaleARGBRowDownEven(src_argb, src_stride, col_step, dst_argb, dst_width);
    src_argb += row_stride;
    dst_argb += dst_stride;
  }
}

// Scale ARGB down with bilinear interpolation.
SAFEBUFFERS
static void ScaleARGBBilinearDown(int src_height,
                                  int dst_width, int dst_height,
                                  int src_stride, int dst_stride,
                                  const uint8* src_argb, uint8* dst_argb,
                                  int x, int dx, int y, int dy,
                                  FilterMode filtering) {
  assert(src_height > 0);
  assert(dst_width > 0);
  assert(dst_height > 0);
  int xlast = x + (dst_width - 1) * dx;
  int xl = (dx >= 0) ? x : xlast;
  int xr = (dx >= 0) ? xlast : x;
  xl = (xl >> 16) & ~3;  // Left edge aligned.
  xr = (xr >> 16) + 1;  // Right most pixel used.
  int clip_src_width = (((xr - xl) + 1 + 3) & ~3) * 4;  // Width aligned to 4.
  src_argb += xl * 4;
  x -= (xl << 16);
  assert(clip_src_width <= kMaxStride);
  // TODO(fbarchard): Remove clip_src_width alignment checks.
  SIMD_ALIGNED(uint8 row[kMaxStride + 16]);
  void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
      ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
      InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
  if (TestCpuFlag(kCpuHasSSE2) && clip_src_width >= 16) {
    InterpolateRow = InterpolateRow_Any_SSE2;
    if (IS_ALIGNED(clip_src_width, 16)) {
      InterpolateRow = InterpolateRow_Unaligned_SSE2;
      if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16)) {
        InterpolateRow = InterpolateRow_SSE2;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3) && clip_src_width >= 16) {
    InterpolateRow = InterpolateRow_Any_SSSE3;
    if (IS_ALIGNED(clip_src_width, 16)) {
      InterpolateRow = InterpolateRow_Unaligned_SSSE3;
      if (IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16)) {
        InterpolateRow = InterpolateRow_SSSE3;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_AVX2)
  if (TestCpuFlag(kCpuHasAVX2) && clip_src_width >= 32) {
    InterpolateRow = InterpolateRow_Any_AVX2;
    if (IS_ALIGNED(clip_src_width, 32)) {
      InterpolateRow = InterpolateRow_AVX2;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_NEON)
  if (TestCpuFlag(kCpuHasNEON) && clip_src_width >= 16) {
    InterpolateRow = InterpolateRow_Any_NEON;
    if (IS_ALIGNED(clip_src_width, 16)) {
      InterpolateRow = InterpolateRow_NEON;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
  if (TestCpuFlag(kCpuHasMIPS_DSPR2) && clip_src_width >= 4 &&
      IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4)) {
    InterpolateRow = InterpolateRow_Any_MIPS_DSPR2;
    if (IS_ALIGNED(clip_src_width, 4)) {
      InterpolateRow = InterpolateRow_MIPS_DSPR2;
    }
  }
#endif
  void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
      int dst_width, int x, int dx) = ScaleARGBFilterCols_C;
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3)) {
    ScaleARGBFilterCols = ScaleARGBFilterCols_SSSE3;
  }
#endif
  const int max_y = (src_height - 1) << 16;
  for (int j = 0; j < dst_height; ++j) {
    if (y > max_y) {
      y = max_y;
    }
    int yi = y >> 16;
    const uint8* src = src_argb + yi * src_stride;
    if (filtering == kFilterLinear) {
      ScaleARGBFilterCols(dst_argb, src, dst_width, x, dx);
    } else {
      int yf = (y >> 8) & 255;
      InterpolateRow(row, src, src_stride, clip_src_width, yf);
      ScaleARGBFilterCols(dst_argb, row, dst_width, x, dx);
    }
    dst_argb += dst_stride;
    y += dy;
  }
}

// Scale ARGB up with bilinear interpolation.
SAFEBUFFERS
static void ScaleARGBBilinearUp(int src_width, int src_height,
                                int dst_width, int dst_height,
                                int src_stride, int dst_stride,
                                const uint8* src_argb, uint8* dst_argb,
                                int x, int dx, int y, int dy,
                                FilterMode filtering) {
  assert(src_width > 0);
  assert(src_height > 0);
  assert(dst_width > 0);
  assert(dst_height > 0);
  assert(dst_width * 4 <= kMaxStride);
  void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
      ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
      InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
  if (TestCpuFlag(kCpuHasSSE2) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_SSE2;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_Unaligned_SSE2;
      if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
        InterpolateRow = InterpolateRow_SSE2;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_SSSE3;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_Unaligned_SSSE3;
      if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
        InterpolateRow = InterpolateRow_SSSE3;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_AVX2)
  if (TestCpuFlag(kCpuHasAVX2) && dst_width >= 8) {
    InterpolateRow = InterpolateRow_Any_AVX2;
    if (IS_ALIGNED(dst_width, 8)) {
      InterpolateRow = InterpolateRow_AVX2;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_NEON)
  if (TestCpuFlag(kCpuHasNEON) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_NEON;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_NEON;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
  if (TestCpuFlag(kCpuHasMIPS_DSPR2) && dst_width >= 1 &&
      IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
    InterpolateRow = InterpolateRow_MIPS_DSPR2;
  }
#endif
  void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
      int dst_width, int x, int dx) =
      filtering ? ScaleARGBFilterCols_C : ScaleARGBCols_C;
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
  if (filtering && TestCpuFlag(kCpuHasSSSE3)) {
    ScaleARGBFilterCols = ScaleARGBFilterCols_SSSE3;
  }
#endif
#if defined(HAS_SCALEARGBCOLS_SSE2)
  if (!filtering && TestCpuFlag(kCpuHasSSE2)) {
    ScaleARGBFilterCols = ScaleARGBCols_SSE2;
  }
#endif
  if (!filtering && src_width * 2 == dst_width && x < 0x8000) {
    ScaleARGBFilterCols = ScaleARGBColsUp2_C;
#if defined(HAS_SCALEARGBCOLSUP2_SSE2)
    if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) &&
        IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) &&
        IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
      ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2;
    }
#endif
  }

  const int max_y = (src_height - 1) << 16;
  if (y > max_y) {
    y = max_y;
  }
  int yi = y >> 16;
  const uint8* src = src_argb + yi * src_stride;
  SIMD_ALIGNED(uint8 row[2 * kMaxStride]);
  uint8* rowptr = row;
  int rowstride = kMaxStride;
  int lasty = yi;

  ScaleARGBFilterCols(rowptr, src, dst_width, x, dx);
  if (src_height > 1) {
    src += src_stride;
  }
  ScaleARGBFilterCols(rowptr + rowstride, src, dst_width, x, dx);
  src += src_stride;

  for (int j = 0; j < dst_height; ++j) {
    yi = y >> 16;
    if (yi != lasty) {
      if (y > max_y) {
        y = max_y;
        yi = y >> 16;
      }
      if (yi != lasty) {
        ScaleARGBFilterCols(rowptr, src, dst_width, x, dx);
        rowptr += rowstride;
        rowstride = -rowstride;
        lasty = yi;
        src += src_stride;
      }
    }
    if (filtering == kFilterLinear) {
      InterpolateRow(dst_argb, rowptr, 0, dst_width * 4, 0);
    } else {
      int yf = (y >> 8) & 255;
      InterpolateRow(dst_argb, rowptr, rowstride, dst_width * 4, yf);
    }
    dst_argb += dst_stride;
    y += dy;
  }
}

#ifdef YUVSCALEUP
// Scale YUV to ARGB up with bilinear interpolation.
SAFEBUFFERS
static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
                                     int dst_width, int dst_height,
                                     int src_stride_y,
                                     int src_stride_u,
                                     int src_stride_v,
                                     int dst_stride_argb,
                                     const uint8* src_y,
                                     const uint8* src_u,
                                     const uint8* src_v,
                                     uint8* dst_argb,
                                     int x, int dx, int y, int dy,
                                     FilterMode filtering) {
  assert(src_width > 0);
  assert(src_height > 0);
  assert(dst_width > 0);
  assert(dst_height > 0);
  assert(dst_width * 4 <= kMaxStride);

  void (*I422ToARGBRow)(const uint8* y_buf,
                        const uint8* u_buf,
                        const uint8* v_buf,
                        uint8* rgb_buf,
                        int width) = I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3) && src_width >= 8) {
    I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
    if (IS_ALIGNED(src_width, 8)) {
      I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3;
      if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
        I422ToARGBRow = I422ToARGBRow_SSSE3;
      }
    }
  }
#endif
#if defined(HAS_I422TOARGBROW_AVX2)
  if (TestCpuFlag(kCpuHasAVX2) && src_width >= 16) {
    I422ToARGBRow = I422ToARGBRow_Any_AVX2;
    if (IS_ALIGNED(src_width, 16)) {
      I422ToARGBRow = I422ToARGBRow_AVX2;
    }
  }
#endif
#if defined(HAS_I422TOARGBROW_NEON)
  if (TestCpuFlag(kCpuHasNEON) && src_width >= 8) {
    I422ToARGBRow = I422ToARGBRow_Any_NEON;
    if (IS_ALIGNED(src_width, 8)) {
      I422ToARGBRow = I422ToARGBRow_NEON;
    }
  }
#endif
#if defined(HAS_I422TOARGBROW_MIPS_DSPR2)
  if (TestCpuFlag(kCpuHasMIPS_DSPR2) && IS_ALIGNED(src_width, 4) &&
      IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) &&
      IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) &&
      IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) &&
      IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
    I422ToARGBRow = I422ToARGBRow_MIPS_DSPR2;
  }
#endif

  void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
      ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
      InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSE2)
  if (TestCpuFlag(kCpuHasSSE2) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_SSE2;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_Unaligned_SSE2;
      if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
        InterpolateRow = InterpolateRow_SSE2;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_SSSE3;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_Unaligned_SSSE3;
      if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
        InterpolateRow = InterpolateRow_SSSE3;
      }
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_AVX2)
  if (TestCpuFlag(kCpuHasAVX2) && dst_width >= 8) {
    InterpolateRow = InterpolateRow_Any_AVX2;
    if (IS_ALIGNED(dst_width, 8)) {
      InterpolateRow = InterpolateRow_AVX2;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROW_NEON)
  if (TestCpuFlag(kCpuHasNEON) && dst_width >= 4) {
    InterpolateRow = InterpolateRow_Any_NEON;
    if (IS_ALIGNED(dst_width, 4)) {
      InterpolateRow = InterpolateRow_NEON;
    }
  }
#endif
#if defined(HAS_INTERPOLATEROWS_MIPS_DSPR2)
  if (TestCpuFlag(kCpuHasMIPS_DSPR2) && dst_width >= 1 &&
      IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
    InterpolateRow = InterpolateRow_MIPS_DSPR2;
  }
#endif
  void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
      int dst_width, int x, int dx) = ScaleARGBFilterCols_C;
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
  if (TestCpuFlag(kCpuHasSSSE3)) {
    ScaleARGBFilterCols = ScaleARGBFilterCols_SSSE3;
  }
#endif
  const int max_y = (src_height - 1) << 16;
  if (y > max_y) {
    y = max_y;
  }
  const int kYShift = 1;  // Shift Y by 1 to convert Y plane to UV coordinate.
  int yi = y >> 16;
  int uv_yi = yi >> kYShift;
  const uint8* src_row_y = src_y + yi * src_stride_y;
  const uint8* src_row_u = src_u + uv_yi * src_stride_u;
  const uint8* src_row_v = src_v + uv_yi * src_stride_v;
  SIMD_ALIGNED(uint8 row[2 * kMaxStride]);
  SIMD_ALIGNED(uint8 argb_row[kMaxStride * 4]);
  uint8* rowptr = row;
  int rowstride = kMaxStride;
  int lasty = yi;

  ScaleARGBFilterCols(rowptr, src_row_y, dst_width, x, dx);
  if (src_height > 1) {
    src_row_y += src_stride_y;
    if (yi & 1) {
      src_row_u += src_stride_u;
      src_row_v += src_stride_v;
    }
  }
  ScaleARGBFilterCols(rowptr + rowstride, src_row_y, dst_width, x, dx);
  if (src_height > 2) {
    src_row_y += src_stride_y;
    if (!(yi & 1)) {
      src_row_u += src_stride_u;
      src_row_v += src_stride_v;
    }
  }

  for (int j = 0; j < dst_height; ++j) {
    yi = y >> 16;
    if (yi != lasty) {
      if (y > max_y) {
        y = max_y;
        yi = y >> 16;
      }
      if (yi != lasty) {
        // TODO(fbarchard): Convert the clipped region of row.
        I422ToARGBRow(src_row_y, src_row_u, src_row_v, argb_row, src_width);
        ScaleARGBFilterCols(rowptr, argb_row, dst_width, x, dx);
        rowptr += rowstride;
        rowstride = -rowstride;
        lasty = yi;
        src_row_y += src_stride_y;
        if (yi & 1) {
          src_row_u += src_stride_u;
          src_row_v += src_stride_v;
        }
      }
    }
    if (filtering == kFilterLinear) {
      InterpolateRow(dst_argb, rowptr, 0, dst_width * 4, 0);
    } else {
      int yf = (y >> 8) & 255;
      InterpolateRow(dst_argb, rowptr, rowstride, dst_width * 4, yf);
    }
    dst_argb += dst_stride_argb;
    y += dy;
  }
}
#endif

// Scale ARGB to/from any dimensions, without interpolation.
// Fixed point math is used for performance: The upper 16 bits
// of x and dx is the integer part of the source position and
// the lower 16 bits are the fixed decimal part.

static void ScaleARGBSimple(int src_width, int src_height,
                            int dst_width, int dst_height,
                            int src_stride, int dst_stride,
                            const uint8* src_argb, uint8* dst_argb,
                            int x, int dx, int y, int dy) {
  void (*ScaleARGBCols)(uint8* dst_argb, const uint8* src_argb,
      int dst_width, int x, int dx) = ScaleARGBCols_C;
#if defined(HAS_SCALEARGBCOLS_SSE2)
  if (TestCpuFlag(kCpuHasSSE2)) {
    ScaleARGBCols = ScaleARGBCols_SSE2;
  }
#endif
  if (src_width * 2 == dst_width && x < 0x8000) {
    ScaleARGBCols = ScaleARGBColsUp2_C;
#if defined(HAS_SCALEARGBCOLSUP2_SSE2)
    if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8) &&
        IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride, 16) &&
        IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride, 16)) {
      ScaleARGBCols = ScaleARGBColsUp2_SSE2;
    }
#endif
  }

  for (int i = 0; i < dst_height; ++i) {
    ScaleARGBCols(dst_argb, src_argb + (y >> 16) * src_stride,
                  dst_width, x, dx);
    dst_argb += dst_stride;
    y += dy;
  }
}

// ScaleARGB a ARGB.
// This function in turn calls a scaling function
// suitable for handling the desired resolutions.
static void ScaleARGB(const uint8* src, int src_stride,
                      int src_width, int src_height,
                      uint8* dst, int dst_stride,
                      int dst_width, int dst_height,
                      int clip_x, int clip_y, int clip_width, int clip_height,
                      FilterMode filtering) {
  // Negative src_height means invert the image.
  if (src_height < 0) {
    src_height = -src_height;
    src = src + (src_height - 1) * src_stride;
    src_stride = -src_stride;
  }
  // Initial source x/y coordinate and step values as 16.16 fixed point.
  int dx = 0;
  int dy = 0;
  int x = 0;
  int y = 0;
  if (filtering) {
    // Scale step for bilinear sampling renders last pixel once for upsample.
    if (dst_width <= Abs(src_width)) {
      dx = FixedDiv(Abs(src_width), dst_width);
      x = (dx >> 1) - 32768;
    } else if (dst_width > 1) {
      dx = FixedDiv(Abs(src_width) - 1, dst_width - 1);
    }
    if (dst_height <= src_height) {
      dy = FixedDiv(src_height,  dst_height);
      y = (dy >> 1) - 32768;
    } else if (dst_height > 1) {
      dy = FixedDiv(src_height - 1, dst_height - 1);
    }
  } else {
    // Scale step for point sampling duplicates all pixels equally.
    dx = FixedDiv(Abs(src_width), dst_width);
    dy = FixedDiv(src_height, dst_height);
    x = dx >> 1;
    y = dy >> 1;
  }
  // Negative src_width means horizontally mirror.
  if (src_width < 0) {
    x += (dst_width - 1) * dx;
    dx = -dx;
    src_width = -src_width;
  }
  if (clip_x) {
    x += clip_x * dx;
    dst += clip_x * 4;
  }
  if (clip_y) {
    y += clip_y * dy;
    dst += clip_y * dst_stride;
  }

  // Special case for integer step values.
  if (((dx | dy) & 0xffff) == 0) {
    if (!dx || !dy) {  // 1 pixel wide and/or tall.
      filtering = kFilterNone;
    } else {
      // Optimized even scale down. ie 2, 4, 6, 8, 10x.
      if (!(dx & 0x10000) && !(dy & 0x10000)) {
        if ((dx >> 16) == 2) {
          // Optimized 1/2 horizontal.
          ScaleARGBDown2(src_width, src_height,
                         clip_width, clip_height,
                         src_stride, dst_stride, src, dst,
                         x, dx, y, dy, filtering);
          return;
        }
        ScaleARGBDownEven(src_width, src_height,
                          clip_width, clip_height,
                          src_stride, dst_stride, src, dst,
                          x, dx, y, dy, filtering);
        return;
      }
      // Optimized odd scale down. ie 3, 5, 7, 9x.
      if ((dx & 0x10000) && (dy & 0x10000)) {
        filtering = kFilterNone;
        if (dx == 0x10000 && dy == 0x10000) {
          // Straight copy.
          ARGBCopy(src + (y >> 16) * src_stride + (x >> 16) * 4, src_stride,
                   dst, dst_stride, clip_width, clip_height);
          return;
        }
      }
    }
  }
  if (dx == 0x10000 && (x & 0xffff) == 0) {
    // Arbitrary scale vertically, but unscaled vertically.
    ScalePlaneVertical(src_height,
                       clip_width, clip_height,
                       src_stride, dst_stride, src, dst,
                       x, y, dy, 4, filtering);
    return;
  }
  if (filtering && dy < 65536 && dst_width * 4 <= kMaxStride) {
    ScaleARGBBilinearUp(src_width, src_height,
                        clip_width, clip_height,
                        src_stride, dst_stride, src, dst,
                        x, dx, y, dy, filtering);
    return;
  }
  if (filtering && src_width * 4 < kMaxStride) {
    ScaleARGBBilinearDown(src_height,
                          clip_width, clip_height,
                          src_stride, dst_stride, src, dst,
                          x, dx, y, dy, filtering);
    return;
  }
  ScaleARGBSimple(src_width, src_height, clip_width, clip_height,
                  src_stride, dst_stride, src, dst,
                  x, dx, y, dy);
}

LIBYUV_API
int ARGBScaleClip(const uint8* src_argb, int src_stride_argb,
                  int src_width, int src_height,
                  uint8* dst_argb, int dst_stride_argb,
                  int dst_width, int dst_height,
                  int clip_x, int clip_y, int clip_width, int clip_height,
                  enum FilterMode filtering) {
  if (!src_argb || src_width == 0 || src_height == 0 ||
      !dst_argb || dst_width <= 0 || dst_height <= 0 ||
      clip_x < 0 || clip_y < 0 ||
      (clip_x + clip_width) > dst_width ||
      (clip_y + clip_height) > dst_height) {
    return -1;
  }
  ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
            dst_argb, dst_stride_argb, dst_width, dst_height,
            clip_x, clip_y, clip_width, clip_height, filtering);
  return 0;
}

// Scale an ARGB image.
LIBYUV_API
int ARGBScale(const uint8* src_argb, int src_stride_argb,
              int src_width, int src_height,
              uint8* dst_argb, int dst_stride_argb,
              int dst_width, int dst_height,
              FilterMode filtering) {
  if (!src_argb || src_width == 0 || src_height == 0 ||
      !dst_argb || dst_width <= 0 || dst_height <= 0) {
    return -1;
  }
  ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
            dst_argb, dst_stride_argb, dst_width, dst_height,
            0, 0, dst_width, dst_height, filtering);
  return 0;
}

#ifdef __cplusplus
}  // extern "C"
}  // namespace libyuv
#endif