Update aom to v1.0.0

Update aom to commit id d14c5bb4f336ef1842046089849dee4a301fbbf0.

1 files changed, 214 insertions, 834 deletions

diff --git a/third_party/aom/av1/common/warped_motion.c b/third_party/aom/av1/common/warped_motion.c
index 34374af69..ae6f07657 100644
--- a/third_party/aom/av1/common/warped_motion.c
+++ b/third_party/aom/av1/common/warped_motion.c
@@ -15,7 +15,8 @@
 #include <math.h>
 #include <assert.h>
 
-#include "./av1_rtcd.h"
+#include "config/av1_rtcd.h"
+
 #include "av1/common/warped_motion.h"
 #include "av1/common/scale.h"
 
@@ -91,78 +92,11 @@ static const int error_measure_lut[512] = {
 };
 /* clang-format on */
 
-static ProjectPointsFunc get_project_points_type(TransformationType type) {
-  switch (type) {
-    case VERTRAPEZOID: return project_points_vertrapezoid;
-    case HORTRAPEZOID: return project_points_hortrapezoid;
-    case HOMOGRAPHY: return project_points_homography;
-    case AFFINE: return project_points_affine;
-    case ROTZOOM: return project_points_rotzoom;
-    case TRANSLATION: return project_points_translation;
-    default: assert(0); return NULL;
-  }
-}
-
-void project_points_translation(const int32_t *mat, int *points, int *proj,
-                                const int n, const int stride_points,
-                                const int stride_proj, const int subsampling_x,
-                                const int subsampling_y) {
-  int i;
-  for (i = 0; i < n; ++i) {
-    const int x = *(points++), y = *(points++);
-    if (subsampling_x)
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((x * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[0]),
-          WARPEDDIFF_PREC_BITS + 1);
-    else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((x * (1 << WARPEDMODEL_PREC_BITS)) + mat[0]), WARPEDDIFF_PREC_BITS);
-    if (subsampling_y)
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((y * (1 << (WARPEDMODEL_PREC_BITS + 1))) + mat[1]),
-          WARPEDDIFF_PREC_BITS + 1);
-    else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          ((y * (1 << WARPEDMODEL_PREC_BITS))) + mat[1], WARPEDDIFF_PREC_BITS);
-    points += stride_points - 2;
-    proj += stride_proj - 2;
-  }
-}
-
-void project_points_rotzoom(const int32_t *mat, int *points, int *proj,
-                            const int n, const int stride_points,
-                            const int stride_proj, const int subsampling_x,
-                            const int subsampling_y) {
-  int i;
-  for (i = 0; i < n; ++i) {
-    const int x = *(points++), y = *(points++);
-    if (subsampling_x)
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          mat[2] * 2 * x + mat[3] * 2 * y + mat[0] +
-              (mat[2] + mat[3] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
-          WARPEDDIFF_PREC_BITS + 1);
-    else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(mat[2] * x + mat[3] * y + mat[0],
-                                            WARPEDDIFF_PREC_BITS);
-    if (subsampling_y)
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
-          -mat[3] * 2 * x + mat[2] * 2 * y + mat[1] +
-              (-mat[3] + mat[2] - (1 << WARPEDMODEL_PREC_BITS)) / 2,
-          WARPEDDIFF_PREC_BITS + 1);
-    else
-      *(proj++) = ROUND_POWER_OF_TWO_SIGNED(-mat[3] * x + mat[2] * y + mat[1],
-                                            WARPEDDIFF_PREC_BITS);
-    points += stride_points - 2;
-    proj += stride_proj - 2;
-  }
-}
-
 void project_points_affine(const int32_t *mat, int *points, int *proj,
                            const int n, const int stride_points,
                            const int stride_proj, const int subsampling_x,
                            const int subsampling_y) {
-  int i;
-  for (i = 0; i < n; ++i) {
+  for (int i = 0; i < n; ++i) {
     const int x = *(points++), y = *(points++);
     if (subsampling_x)
       *(proj++) = ROUND_POWER_OF_TWO_SIGNED(
@@ -185,301 +119,6 @@ void project_points_affine(const int32_t *mat, int *points, int *proj,
   }
 }
 
-void project_points_hortrapezoid(const int32_t *mat, int *points, int *proj,
-                                 const int n, const int stride_points,
-                                 const int stride_proj, const int subsampling_x,
-                                 const int subsampling_y) {
-  int i;
-  int64_t x, y, Z;
-  int64_t xp, yp;
-  for (i = 0; i < n; ++i) {
-    x = *(points++), y = *(points++);
-    x = (subsampling_x ? 4 * x + 1 : 2 * x);
-    y = (subsampling_y ? 4 * y + 1 : 2 * y);
-
-    Z = (mat[7] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
-    xp = (mat[2] * x + mat[3] * y + 2 * mat[0]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-    yp = (mat[5] * y + 2 * mat[1]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-
-    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
-    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;
-
-    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    *(proj++) = (int)xp;
-    *(proj++) = (int)yp;
-
-    points += stride_points - 2;
-    proj += stride_proj - 2;
-  }
-}
-
-void project_points_vertrapezoid(const int32_t *mat, int *points, int *proj,
-                                 const int n, const int stride_points,
-                                 const int stride_proj, const int subsampling_x,
-                                 const int subsampling_y) {
-  int i;
-  int64_t x, y, Z;
-  int64_t xp, yp;
-  for (i = 0; i < n; ++i) {
-    x = *(points++), y = *(points++);
-    x = (subsampling_x ? 4 * x + 1 : 2 * x);
-    y = (subsampling_y ? 4 * y + 1 : 2 * y);
-
-    Z = (mat[6] * x + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
-    xp = (mat[2] * x + 2 * mat[0]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-    yp = (mat[4] * x + mat[5] * y + 2 * mat[1]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-
-    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
-    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;
-
-    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    *(proj++) = (int)xp;
-    *(proj++) = (int)yp;
-
-    points += stride_points - 2;
-    proj += stride_proj - 2;
-  }
-}
-
-void project_points_homography(const int32_t *mat, int *points, int *proj,
-                               const int n, const int stride_points,
-                               const int stride_proj, const int subsampling_x,
-                               const int subsampling_y) {
-  int i;
-  int64_t x, y, Z;
-  int64_t xp, yp;
-  for (i = 0; i < n; ++i) {
-    x = *(points++), y = *(points++);
-    x = (subsampling_x ? 4 * x + 1 : 2 * x);
-    y = (subsampling_y ? 4 * y + 1 : 2 * y);
-
-    Z = (mat[6] * x + mat[7] * y + (1 << (WARPEDMODEL_ROW3HOMO_PREC_BITS + 1)));
-    xp = (mat[2] * x + mat[3] * y + 2 * mat[0]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-    yp = (mat[4] * x + mat[5] * y + 2 * mat[1]) *
-         (1 << (WARPEDPIXEL_PREC_BITS + WARPEDMODEL_ROW3HOMO_PREC_BITS -
-                WARPEDMODEL_PREC_BITS));
-
-    xp = xp > 0 ? (xp + Z / 2) / Z : (xp - Z / 2) / Z;
-    yp = yp > 0 ? (yp + Z / 2) / Z : (yp - Z / 2) / Z;
-
-    if (subsampling_x) xp = (xp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    if (subsampling_y) yp = (yp - (1 << (WARPEDPIXEL_PREC_BITS - 1))) / 2;
-    *(proj++) = (int)xp;
-    *(proj++) = (int)yp;
-
-    points += stride_points - 2;
-    proj += stride_proj - 2;
-  }
-}
-
-static const int16_t
-    filter_ntap[WARPEDPIXEL_PREC_SHIFTS][WARPEDPIXEL_FILTER_TAPS] = {
-#if WARPEDPIXEL_PREC_BITS == 6
-      { 0, 0, 128, 0, 0, 0 },      { 0, -1, 128, 2, -1, 0 },
-      { 1, -3, 127, 4, -1, 0 },    { 1, -4, 126, 6, -2, 1 },
-      { 1, -5, 126, 8, -3, 1 },    { 1, -6, 125, 11, -4, 1 },
-      { 1, -7, 124, 13, -4, 1 },   { 2, -8, 123, 15, -5, 1 },
-      { 2, -9, 122, 18, -6, 1 },   { 2, -10, 121, 20, -6, 1 },
-      { 2, -11, 120, 22, -7, 2 },  { 2, -12, 119, 25, -8, 2 },
-      { 3, -13, 117, 27, -8, 2 },  { 3, -13, 116, 29, -9, 2 },
-      { 3, -14, 114, 32, -10, 3 }, { 3, -15, 113, 35, -10, 2 },
-      { 3, -15, 111, 37, -11, 3 }, { 3, -16, 109, 40, -11, 3 },
-      { 3, -16, 108, 42, -12, 3 }, { 4, -17, 106, 45, -13, 3 },
-      { 4, -17, 104, 47, -13, 3 }, { 4, -17, 102, 50, -14, 3 },
-      { 4, -17, 100, 52, -14, 3 }, { 4, -18, 98, 55, -15, 4 },
-      { 4, -18, 96, 58, -15, 3 },  { 4, -18, 94, 60, -16, 4 },
-      { 4, -18, 91, 63, -16, 4 },  { 4, -18, 89, 65, -16, 4 },
-      { 4, -18, 87, 68, -17, 4 },  { 4, -18, 85, 70, -17, 4 },
-      { 4, -18, 82, 73, -17, 4 },  { 4, -18, 80, 75, -17, 4 },
-      { 4, -18, 78, 78, -18, 4 },  { 4, -17, 75, 80, -18, 4 },
-      { 4, -17, 73, 82, -18, 4 },  { 4, -17, 70, 85, -18, 4 },
-      { 4, -17, 68, 87, -18, 4 },  { 4, -16, 65, 89, -18, 4 },
-      { 4, -16, 63, 91, -18, 4 },  { 4, -16, 60, 94, -18, 4 },
-      { 3, -15, 58, 96, -18, 4 },  { 4, -15, 55, 98, -18, 4 },
-      { 3, -14, 52, 100, -17, 4 }, { 3, -14, 50, 102, -17, 4 },
-      { 3, -13, 47, 104, -17, 4 }, { 3, -13, 45, 106, -17, 4 },
-      { 3, -12, 42, 108, -16, 3 }, { 3, -11, 40, 109, -16, 3 },
-      { 3, -11, 37, 111, -15, 3 }, { 2, -10, 35, 113, -15, 3 },
-      { 3, -10, 32, 114, -14, 3 }, { 2, -9, 29, 116, -13, 3 },
-      { 2, -8, 27, 117, -13, 3 },  { 2, -8, 25, 119, -12, 2 },
-      { 2, -7, 22, 120, -11, 2 },  { 1, -6, 20, 121, -10, 2 },
-      { 1, -6, 18, 122, -9, 2 },   { 1, -5, 15, 123, -8, 2 },
-      { 1, -4, 13, 124, -7, 1 },   { 1, -4, 11, 125, -6, 1 },
-      { 1, -3, 8, 126, -5, 1 },    { 1, -2, 6, 126, -4, 1 },
-      { 0, -1, 4, 127, -3, 1 },    { 0, -1, 2, 128, -1, 0 },
-#elif WARPEDPIXEL_PREC_BITS == 5
-      { 0, 0, 128, 0, 0, 0 },      { 1, -3, 127, 4, -1, 0 },
-      { 1, -5, 126, 8, -3, 1 },    { 1, -7, 124, 13, -4, 1 },
-      { 2, -9, 122, 18, -6, 1 },   { 2, -11, 120, 22, -7, 2 },
-      { 3, -13, 117, 27, -8, 2 },  { 3, -14, 114, 32, -10, 3 },
-      { 3, -15, 111, 37, -11, 3 }, { 3, -16, 108, 42, -12, 3 },
-      { 4, -17, 104, 47, -13, 3 }, { 4, -17, 100, 52, -14, 3 },
-      { 4, -18, 96, 58, -15, 3 },  { 4, -18, 91, 63, -16, 4 },
-      { 4, -18, 87, 68, -17, 4 },  { 4, -18, 82, 73, -17, 4 },
-      { 4, -18, 78, 78, -18, 4 },  { 4, -17, 73, 82, -18, 4 },
-      { 4, -17, 68, 87, -18, 4 },  { 4, -16, 63, 91, -18, 4 },
-      { 3, -15, 58, 96, -18, 4 },  { 3, -14, 52, 100, -17, 4 },
-      { 3, -13, 47, 104, -17, 4 }, { 3, -12, 42, 108, -16, 3 },
-      { 3, -11, 37, 111, -15, 3 }, { 3, -10, 32, 114, -14, 3 },
-      { 2, -8, 27, 117, -13, 3 },  { 2, -7, 22, 120, -11, 2 },
-      { 1, -6, 18, 122, -9, 2 },   { 1, -4, 13, 124, -7, 1 },
-      { 1, -3, 8, 126, -5, 1 },    { 0, -1, 4, 127, -3, 1 },
-#endif  // WARPEDPIXEL_PREC_BITS == 6
-    };
-
-static int32_t do_ntap_filter(const int32_t *const p, int x) {
-  int i;
-  int32_t sum = 0;
-  for (i = 0; i < WARPEDPIXEL_FILTER_TAPS; ++i) {
-    sum += p[i - WARPEDPIXEL_FILTER_TAPS / 2 + 1] * filter_ntap[x][i];
-  }
-  return sum;
-}
-
-static int32_t do_cubic_filter(const int32_t *const p, int x) {
-  if (x == 0) {
-    return p[0] * (1 << WARPEDPIXEL_FILTER_BITS);
-  } else if (x == (1 << WARPEDPIXEL_PREC_BITS)) {
-    return p[1] * (1 << WARPEDPIXEL_FILTER_BITS);
-  } else {
-    const int64_t v1 = (int64_t)x * x * x * (3 * (p[0] - p[1]) + p[2] - p[-1]);
-    const int64_t v2 =
-        (int64_t)x * x * (2 * p[-1] - 5 * p[0] + 4 * p[1] - p[2]);
-    const int64_t v3 = x * (p[1] - p[-1]);
-    const int64_t v4 = 2 * p[0];
-    return (int32_t)ROUND_POWER_OF_TWO_SIGNED(
-        (v4 * (1 << (3 * WARPEDPIXEL_PREC_BITS))) +
-            (v3 * (1 << (2 * WARPEDPIXEL_PREC_BITS))) +
-            (v2 * (1 << WARPEDPIXEL_PREC_BITS)) + v1,
-        3 * WARPEDPIXEL_PREC_BITS + 1 - WARPEDPIXEL_FILTER_BITS);
-  }
-}
-
-static INLINE void get_subcolumn(int taps, const uint8_t *const ref,
-                                 int32_t *col, int stride, int x, int y_start) {
-  int i;
-  for (i = 0; i < taps; ++i) {
-    col[i] = ref[(i + y_start) * stride + x];
-  }
-}
-
-static uint8_t bi_ntap_filter(const uint8_t *const ref, int x, int y,
-                              int stride) {
-  int32_t val, arr[WARPEDPIXEL_FILTER_TAPS];
-  int k;
-  const int i = (int)x >> WARPEDPIXEL_PREC_BITS;
-  const int j = (int)y >> WARPEDPIXEL_PREC_BITS;
-  for (k = 0; k < WARPEDPIXEL_FILTER_TAPS; ++k) {
-    int32_t arr_temp[WARPEDPIXEL_FILTER_TAPS];
-    get_subcolumn(WARPEDPIXEL_FILTER_TAPS, ref, arr_temp, stride,
-                  i + k + 1 - WARPEDPIXEL_FILTER_TAPS / 2,
-                  j + 1 - WARPEDPIXEL_FILTER_TAPS / 2);
-    arr[k] = do_ntap_filter(arr_temp + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
-                            y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
-  }
-  val = do_ntap_filter(arr + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
-                       x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
-  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
-  return (uint8_t)clip_pixel(val);
-}
-
-static uint8_t bi_cubic_filter(const uint8_t *const ref, int x, int y,
-                               int stride) {
-  int32_t val, arr[4];
-  int k;
-  const int i = (int)x >> WARPEDPIXEL_PREC_BITS;
-  const int j = (int)y >> WARPEDPIXEL_PREC_BITS;
-  for (k = 0; k < 4; ++k) {
-    int32_t arr_temp[4];
-    get_subcolumn(4, ref, arr_temp, stride, i + k - 1, j - 1);
-    arr[k] =
-        do_cubic_filter(arr_temp + 1, y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
-  }
-  val = do_cubic_filter(arr + 1, x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
-  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
-  return (uint8_t)clip_pixel(val);
-}
-
-static uint8_t bi_linear_filter(const uint8_t *const ref, int x, int y,
-                                int stride) {
-  const int ix = x >> WARPEDPIXEL_PREC_BITS;
-  const int iy = y >> WARPEDPIXEL_PREC_BITS;
-  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
-  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
-  int32_t val;
-  val = ROUND_POWER_OF_TWO_SIGNED(
-      ref[iy * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sy) *
-              (WARPEDPIXEL_PREC_SHIFTS - sx) +
-          ref[iy * stride + ix + 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) * sx +
-          ref[(iy + 1) * stride + ix] * sy * (WARPEDPIXEL_PREC_SHIFTS - sx) +
-          ref[(iy + 1) * stride + ix + 1] * sy * sx,
-      WARPEDPIXEL_PREC_BITS * 2);
-  return (uint8_t)clip_pixel(val);
-}
-
-static uint8_t warp_interpolate(const uint8_t *const ref, int x, int y,
-                                int width, int height, int stride) {
-  const int ix = x >> WARPEDPIXEL_PREC_BITS;
-  const int iy = y >> WARPEDPIXEL_PREC_BITS;
-  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
-  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
-  int32_t v;
-
-  if (ix < 0 && iy < 0)
-    return ref[0];
-  else if (ix < 0 && iy >= height - 1)
-    return ref[(height - 1) * stride];
-  else if (ix >= width - 1 && iy < 0)
-    return ref[width - 1];
-  else if (ix >= width - 1 && iy >= height - 1)
-    return ref[(height - 1) * stride + (width - 1)];
-  else if (ix < 0) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[iy * stride] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
-            ref[(iy + 1) * stride] * sy,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel(v);
-  } else if (iy < 0) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) + ref[ix + 1] * sx,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel(v);
-  } else if (ix >= width - 1) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[iy * stride + width - 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
-            ref[(iy + 1) * stride + width - 1] * sy,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel(v);
-  } else if (iy >= height - 1) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[(height - 1) * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) +
-            ref[(height - 1) * stride + ix + 1] * sx,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel(v);
-  } else if (ix >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
-             iy >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
-             ix < width - WARPEDPIXEL_FILTER_TAPS / 2 &&
-             iy < height - WARPEDPIXEL_FILTER_TAPS / 2) {
-    return bi_ntap_filter(ref, x, y, stride);
-  } else if (ix >= 1 && iy >= 1 && ix < width - 2 && iy < height - 2) {
-    return bi_cubic_filter(ref, x, y, stride);
-  } else {
-    return bi_linear_filter(ref, x, y, stride);
-  }
-}
-
 // For warping, we really use a 6-tap filter, but we do blocks of 8 pixels
 // at a time. The zoom/rotation/shear in the model are applied to the
 // "fractional" position of each pixel, which therefore varies within
@@ -683,15 +322,14 @@ static const uint16_t div_lut[DIV_LUT_NUM + 1] = {
   8240,  8224,  8208,  8192,
 };
 
-#if CONFIG_WARPED_MOTION
 // Decomposes a divisor D such that 1/D = y/2^shift, where y is returned
 // at precision of DIV_LUT_PREC_BITS along with the shift.
 static int16_t resolve_divisor_64(uint64_t D, int16_t *shift) {
-  int64_t e, f;
+  int64_t f;
   *shift = (int16_t)((D >> 32) ? get_msb((unsigned int)(D >> 32)) + 32
                                : get_msb((unsigned int)D));
   // e is obtained from D after resetting the most significant 1 bit.
-  e = D - ((uint64_t)1 << *shift);
+  const int64_t e = D - ((uint64_t)1 << *shift);
   // Get the most significant DIV_LUT_BITS (8) bits of e into f
   if (*shift > DIV_LUT_BITS)
     f = ROUND_POWER_OF_TWO_64(e, *shift - DIV_LUT_BITS);
@@ -702,13 +340,12 @@ static int16_t resolve_divisor_64(uint64_t D, int16_t *shift) {
   // Use f as lookup into the precomputed table of multipliers
   return div_lut[f];
 }
-#endif  // CONFIG_WARPED_MOTION
 
 static int16_t resolve_divisor_32(uint32_t D, int16_t *shift) {
-  int32_t e, f;
+  int32_t f;
   *shift = get_msb(D);
   // e is obtained from D after resetting the most significant 1 bit.
-  e = D - ((uint32_t)1 << *shift);
+  const int32_t e = D - ((uint32_t)1 << *shift);
   // Get the most significant DIV_LUT_BITS (8) bits of e into f
   if (*shift > DIV_LUT_BITS)
     f = ROUND_POWER_OF_TWO(e, *shift - DIV_LUT_BITS);
@@ -743,16 +380,13 @@ int get_shear_params(WarpedMotionParams *wm) {
   wm->beta = clamp(mat[3], INT16_MIN, INT16_MAX);
   int16_t shift;
   int16_t y = resolve_divisor_32(abs(mat[2]), &shift) * (mat[2] < 0 ? -1 : 1);
-  int64_t v;
-  v = ((int64_t)mat[4] * (1 << WARPEDMODEL_PREC_BITS)) * y;
+  int64_t v = ((int64_t)mat[4] * (1 << WARPEDMODEL_PREC_BITS)) * y;
   wm->gamma =
       clamp((int)ROUND_POWER_OF_TWO_SIGNED_64(v, shift), INT16_MIN, INT16_MAX);
   v = ((int64_t)mat[3] * mat[4]) * y;
   wm->delta = clamp(mat[5] - (int)ROUND_POWER_OF_TWO_SIGNED_64(v, shift) -
                         (1 << WARPEDMODEL_PREC_BITS),
                     INT16_MIN, INT16_MAX);
-  if (!is_affine_shear_allowed(wm->alpha, wm->beta, wm->gamma, wm->delta))
-    return 0;
 
   wm->alpha = ROUND_POWER_OF_TWO_SIGNED(wm->alpha, WARP_PARAM_REDUCE_BITS) *
               (1 << WARP_PARAM_REDUCE_BITS);
@@ -762,171 +396,24 @@ int get_shear_params(WarpedMotionParams *wm) {
               (1 << WARP_PARAM_REDUCE_BITS);
   wm->delta = ROUND_POWER_OF_TWO_SIGNED(wm->delta, WARP_PARAM_REDUCE_BITS) *
               (1 << WARP_PARAM_REDUCE_BITS);
-  return 1;
-}
-
-#if CONFIG_HIGHBITDEPTH
-static INLINE void highbd_get_subcolumn(int taps, const uint16_t *const ref,
-                                        int32_t *col, int stride, int x,
-                                        int y_start) {
-  int i;
-  for (i = 0; i < taps; ++i) {
-    col[i] = ref[(i + y_start) * stride + x];
-  }
-}
-
-static uint16_t highbd_bi_ntap_filter(const uint16_t *const ref, int x, int y,
-                                      int stride, int bd) {
-  int32_t val, arr[WARPEDPIXEL_FILTER_TAPS];
-  int k;
-  const int i = (int)x >> WARPEDPIXEL_PREC_BITS;
-  const int j = (int)y >> WARPEDPIXEL_PREC_BITS;
-  for (k = 0; k < WARPEDPIXEL_FILTER_TAPS; ++k) {
-    int32_t arr_temp[WARPEDPIXEL_FILTER_TAPS];
-    highbd_get_subcolumn(WARPEDPIXEL_FILTER_TAPS, ref, arr_temp, stride,
-                         i + k + 1 - WARPEDPIXEL_FILTER_TAPS / 2,
-                         j + 1 - WARPEDPIXEL_FILTER_TAPS / 2);
-    arr[k] = do_ntap_filter(arr_temp + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
-                            y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
-  }
-  val = do_ntap_filter(arr + WARPEDPIXEL_FILTER_TAPS / 2 - 1,
-                       x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
-  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
-  return (uint16_t)clip_pixel_highbd(val, bd);
-}
-
-static uint16_t highbd_bi_cubic_filter(const uint16_t *const ref, int x, int y,
-                                       int stride, int bd) {
-  int32_t val, arr[4];
-  int k;
-  const int i = (int)x >> WARPEDPIXEL_PREC_BITS;
-  const int j = (int)y >> WARPEDPIXEL_PREC_BITS;
-  for (k = 0; k < 4; ++k) {
-    int32_t arr_temp[4];
-    highbd_get_subcolumn(4, ref, arr_temp, stride, i + k - 1, j - 1);
-    arr[k] =
-        do_cubic_filter(arr_temp + 1, y - (j * (1 << WARPEDPIXEL_PREC_BITS)));
-  }
-  val = do_cubic_filter(arr + 1, x - (i * (1 << WARPEDPIXEL_PREC_BITS)));
-  val = ROUND_POWER_OF_TWO_SIGNED(val, WARPEDPIXEL_FILTER_BITS * 2);
-  return (uint16_t)clip_pixel_highbd(val, bd);
-}
 
-static uint16_t highbd_bi_linear_filter(const uint16_t *const ref, int x, int y,
-                                        int stride, int bd) {
-  const int ix = x >> WARPEDPIXEL_PREC_BITS;
-  const int iy = y >> WARPEDPIXEL_PREC_BITS;
-  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
-  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
-  int32_t val;
-  val = ROUND_POWER_OF_TWO_SIGNED(
-      ref[iy * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sy) *
-              (WARPEDPIXEL_PREC_SHIFTS - sx) +
-          ref[iy * stride + ix + 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) * sx +
-          ref[(iy + 1) * stride + ix] * sy * (WARPEDPIXEL_PREC_SHIFTS - sx) +
-          ref[(iy + 1) * stride + ix + 1] * sy * sx,
-      WARPEDPIXEL_PREC_BITS * 2);
-  return (uint16_t)clip_pixel_highbd(val, bd);
-}
+  if (!is_affine_shear_allowed(wm->alpha, wm->beta, wm->gamma, wm->delta))
+    return 0;
 
-static uint16_t highbd_warp_interpolate(const uint16_t *const ref, int x, int y,
-                                        int width, int height, int stride,
-                                        int bd) {
-  const int ix = x >> WARPEDPIXEL_PREC_BITS;
-  const int iy = y >> WARPEDPIXEL_PREC_BITS;
-  const int sx = x - (ix * (1 << WARPEDPIXEL_PREC_BITS));
-  const int sy = y - (iy * (1 << WARPEDPIXEL_PREC_BITS));
-  int32_t v;
-
-  if (ix < 0 && iy < 0)
-    return ref[0];
-  else if (ix < 0 && iy > height - 1)
-    return ref[(height - 1) * stride];
-  else if (ix > width - 1 && iy < 0)
-    return ref[width - 1];
-  else if (ix > width - 1 && iy > height - 1)
-    return ref[(height - 1) * stride + (width - 1)];
-  else if (ix < 0) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[iy * stride] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
-            ref[(iy + 1) * stride] * sy,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel_highbd(v, bd);
-  } else if (iy < 0) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) + ref[ix + 1] * sx,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel_highbd(v, bd);
-  } else if (ix > width - 1) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[iy * stride + width - 1] * (WARPEDPIXEL_PREC_SHIFTS - sy) +
-            ref[(iy + 1) * stride + width - 1] * sy,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel_highbd(v, bd);
-  } else if (iy > height - 1) {
-    v = ROUND_POWER_OF_TWO_SIGNED(
-        ref[(height - 1) * stride + ix] * (WARPEDPIXEL_PREC_SHIFTS - sx) +
-            ref[(height - 1) * stride + ix + 1] * sx,
-        WARPEDPIXEL_PREC_BITS);
-    return clip_pixel_highbd(v, bd);
-  } else if (ix >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
-             iy >= WARPEDPIXEL_FILTER_TAPS / 2 - 1 &&
-             ix < width - WARPEDPIXEL_FILTER_TAPS / 2 &&
-             iy < height - WARPEDPIXEL_FILTER_TAPS / 2) {
-    return highbd_bi_ntap_filter(ref, x, y, stride, bd);
-  } else if (ix >= 1 && iy >= 1 && ix < width - 2 && iy < height - 2) {
-    return highbd_bi_cubic_filter(ref, x, y, stride, bd);
-  } else {
-    return highbd_bi_linear_filter(ref, x, y, stride, bd);
-  }
+  return 1;
 }
 
 static INLINE int highbd_error_measure(int err, int bd) {
   const int b = bd - 8;
   const int bmask = (1 << b) - 1;
   const int v = (1 << b);
-  int e1, e2;
   err = abs(err);
-  e1 = err >> b;
-  e2 = err & bmask;
+  const int e1 = err >> b;
+  const int e2 = err & bmask;
   return error_measure_lut[255 + e1] * (v - e2) +
          error_measure_lut[256 + e1] * e2;
 }
 
-static void highbd_warp_plane_old(const WarpedMotionParams *const wm,
-                                  const uint8_t *const ref8, int width,
-                                  int height, int stride,
-                                  const uint8_t *const pred8, int p_col,
-                                  int p_row, int p_width, int p_height,
-                                  int p_stride, int subsampling_x,
-                                  int subsampling_y, int x_scale, int y_scale,
-                                  int bd, ConvolveParams *conv_params) {
-  int i, j;
-  ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
-  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
-  const uint16_t *const ref = CONVERT_TO_SHORTPTR(ref8);
-  if (projectpoints == NULL) return;
-  for (i = p_row; i < p_row + p_height; ++i) {
-    for (j = p_col; j < p_col + p_width; ++j) {
-      int in[2], out[2];
-      in[0] = j;
-      in[1] = i;
-      projectpoints(wm->wmmat, in, out, 1, 2, 2, subsampling_x, subsampling_y);
-      out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, SCALE_SUBPEL_BITS);
-      out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, SCALE_SUBPEL_BITS);
-      if (conv_params->do_average)
-        pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
-            pred[(j - p_col) + (i - p_row) * p_stride] +
-                highbd_warp_interpolate(ref, out[0], out[1], width, height,
-                                        stride, bd),
-            1);
-      else
-        pred[(j - p_col) + (i - p_row) * p_stride] = highbd_warp_interpolate(
-            ref, out[0], out[1], width, height, stride, bd);
-    }
-  }
-}
-
 /* Note: For an explanation of the warp algorithm, and some notes on bit widths
     for hardware implementations, see the comments above av1_warp_affine_c
 */
@@ -938,37 +425,23 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
                               ConvolveParams *conv_params, int16_t alpha,
                               int16_t beta, int16_t gamma, int16_t delta) {
   int32_t tmp[15 * 8];
-  int i, j, k, l, m;
-#if CONFIG_CONVOLVE_ROUND
-  const int use_conv_params = conv_params->round == CONVOLVE_OPT_NO_ROUND;
   const int reduce_bits_horiz =
-      use_conv_params ? conv_params->round_0 : HORSHEAR_REDUCE_PREC_BITS;
-  const int max_bits_horiz =
-      use_conv_params
-          ? bd + FILTER_BITS + 1 - conv_params->round_0
-          : bd + WARPEDPIXEL_FILTER_BITS + 1 - HORSHEAR_REDUCE_PREC_BITS;
-  const int offset_bits_horiz =
-      use_conv_params ? bd + FILTER_BITS - 1 : bd + WARPEDPIXEL_FILTER_BITS - 1;
-  const int offset_bits_vert =
-      use_conv_params
-          ? bd + 2 * FILTER_BITS - conv_params->round_0
-          : bd + 2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS;
-  if (use_conv_params) {
-    conv_params->do_post_rounding = 1;
-  }
-  assert(FILTER_BITS == WARPEDPIXEL_FILTER_BITS);
-#else
-  const int reduce_bits_horiz = HORSHEAR_REDUCE_PREC_BITS;
-  const int max_bits_horiz =
-      bd + WARPEDPIXEL_FILTER_BITS + 1 - HORSHEAR_REDUCE_PREC_BITS;
-  const int offset_bits_horiz = bd + WARPEDPIXEL_FILTER_BITS - 1;
-  const int offset_bits_vert =
-      bd + 2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS;
-#endif
+      conv_params->round_0 +
+      AOMMAX(bd + FILTER_BITS - conv_params->round_0 - 14, 0);
+  const int reduce_bits_vert = conv_params->is_compound
+                                   ? conv_params->round_1
+                                   : 2 * FILTER_BITS - reduce_bits_horiz;
+  const int max_bits_horiz = bd + FILTER_BITS + 1 - reduce_bits_horiz;
+  const int offset_bits_horiz = bd + FILTER_BITS - 1;
+  const int offset_bits_vert = bd + 2 * FILTER_BITS - reduce_bits_horiz;
+  const int round_bits =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
   (void)max_bits_horiz;
+  assert(IMPLIES(conv_params->is_compound, conv_params->dst != NULL));
 
-  for (i = p_row; i < p_row + p_height; i += 8) {
-    for (j = p_col; j < p_col + p_width; j += 8) {
+  for (int i = p_row; i < p_row + p_height; i += 8) {
+    for (int j = p_col; j < p_col + p_width; j += 8) {
       // Calculate the center of this 8x8 block,
       // project to luma coordinates (if in a subsampled chroma plane),
       // apply the affine transformation,
@@ -980,9 +453,9 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
       const int32_t x4 = dst_x >> subsampling_x;
       const int32_t y4 = dst_y >> subsampling_y;
 
-      int32_t ix4 = x4 >> WARPEDMODEL_PREC_BITS;
+      const int32_t ix4 = x4 >> WARPEDMODEL_PREC_BITS;
       int32_t sx4 = x4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
-      int32_t iy4 = y4 >> WARPEDMODEL_PREC_BITS;
+      const int32_t iy4 = y4 >> WARPEDMODEL_PREC_BITS;
       int32_t sy4 = y4 & ((1 << WARPEDMODEL_PREC_BITS) - 1);
 
       sx4 += alpha * (-4) + beta * (-4);
@@ -992,15 +465,11 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
       sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
 
       // Horizontal filter
-      for (k = -7; k < 8; ++k) {
-        int iy = iy4 + k;
-        if (iy < 0)
-          iy = 0;
-        else if (iy > height - 1)
-          iy = height - 1;
+      for (int k = -7; k < 8; ++k) {
+        const int iy = clamp(iy4 + k, 0, height - 1);
 
         int sx = sx4 + beta * (k + 4);
-        for (l = -4; l < 4; ++l) {
+        for (int l = -4; l < 4; ++l) {
           int ix = ix4 + l - 3;
           const int offs = ROUND_POWER_OF_TWO(sx, WARPEDDIFF_PREC_BITS) +
                            WARPEDPIXEL_PREC_SHIFTS;
@@ -1008,12 +477,8 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
           const int16_t *coeffs = warped_filter[offs];
 
           int32_t sum = 1 << offset_bits_horiz;
-          for (m = 0; m < 8; ++m) {
-            int sample_x = ix + m;
-            if (sample_x < 0)
-              sample_x = 0;
-            else if (sample_x > width - 1)
-              sample_x = width - 1;
+          for (int m = 0; m < 8; ++m) {
+            const int sample_x = clamp(ix + m, 0, width - 1);
             sum += ref[iy * stride + sample_x] * coeffs[m];
           }
           sum = ROUND_POWER_OF_TWO(sum, reduce_bits_horiz);
@@ -1024,46 +489,50 @@ void av1_highbd_warp_affine_c(const int32_t *mat, const uint16_t *ref,
       }
 
       // Vertical filter
-      for (k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) {
+      for (int k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) {
         int sy = sy4 + delta * (k + 4);
-        for (l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) {
+        for (int l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) {
           const int offs = ROUND_POWER_OF_TWO(sy, WARPEDDIFF_PREC_BITS) +
                            WARPEDPIXEL_PREC_SHIFTS;
           assert(offs >= 0 && offs <= WARPEDPIXEL_PREC_SHIFTS * 3);
           const int16_t *coeffs = warped_filter[offs];
 
           int32_t sum = 1 << offset_bits_vert;
-          for (m = 0; m < 8; ++m) {
+          for (int m = 0; m < 8; ++m) {
             sum += tmp[(k + m + 4) * 8 + (l + 4)] * coeffs[m];
           }
-#if CONFIG_CONVOLVE_ROUND
-          if (use_conv_params) {
+
+          if (conv_params->is_compound) {
             CONV_BUF_TYPE *p =
                 &conv_params
                      ->dst[(i - p_row + k + 4) * conv_params->dst_stride +
                            (j - p_col + l + 4)];
-            sum = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
-                  (1 << (offset_bits_horiz + FILTER_BITS -
-                         conv_params->round_0 - conv_params->round_1)) -
-                  (1 << (offset_bits_vert - conv_params->round_1));
-            if (conv_params->do_average)
-              *p += sum;
-            else
+            sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert);
+            if (conv_params->do_average) {
+              uint16_t *dst16 =
+                  &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
+              int32_t tmp32 = *p;
+              if (conv_params->use_jnt_comp_avg) {
+                tmp32 = tmp32 * conv_params->fwd_offset +
+                        sum * conv_params->bck_offset;
+                tmp32 = tmp32 >> DIST_PRECISION_BITS;
+              } else {
+                tmp32 += sum;
+                tmp32 = tmp32 >> 1;
+              }
+              tmp32 = tmp32 - (1 << (offset_bits - conv_params->round_1)) -
+                      (1 << (offset_bits - conv_params->round_1 - 1));
+              *dst16 =
+                  clip_pixel_highbd(ROUND_POWER_OF_TWO(tmp32, round_bits), bd);
+            } else {
               *p = sum;
+            }
           } else {
-#else
-          {
-#endif
             uint16_t *p =
                 &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
-            sum = ROUND_POWER_OF_TWO(sum, VERSHEAR_REDUCE_PREC_BITS);
+            sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert);
             assert(0 <= sum && sum < (1 << (bd + 2)));
-            uint16_t px =
-                clip_pixel_highbd(sum - (1 << (bd - 1)) - (1 << bd), bd);
-            if (conv_params->do_average)
-              *p = ROUND_POWER_OF_TWO(*p + px, 1);
-            else
-              *p = px;
+            *p = clip_pixel_highbd(sum - (1 << (bd - 1)) - (1 << bd), bd);
           }
           sy += gamma;
         }
@@ -1076,32 +545,25 @@ static void highbd_warp_plane(WarpedMotionParams *wm, const uint8_t *const ref8,
                               int width, int height, int stride,
                               const uint8_t *const pred8, int p_col, int p_row,
                               int p_width, int p_height, int p_stride,
-                              int subsampling_x, int subsampling_y, int x_scale,
-                              int y_scale, int bd,
+                              int subsampling_x, int subsampling_y, int bd,
                               ConvolveParams *conv_params) {
+  assert(wm->wmtype <= AFFINE);
   if (wm->wmtype == ROTZOOM) {
     wm->wmmat[5] = wm->wmmat[2];
     wm->wmmat[4] = -wm->wmmat[3];
   }
-  if ((wm->wmtype == ROTZOOM || wm->wmtype == AFFINE) &&
-      x_scale == SCALE_SUBPEL_SHIFTS && y_scale == SCALE_SUBPEL_SHIFTS) {
-    const int32_t *const mat = wm->wmmat;
-    const int16_t alpha = wm->alpha;
-    const int16_t beta = wm->beta;
-    const int16_t gamma = wm->gamma;
-    const int16_t delta = wm->delta;
-
-    const uint16_t *const ref = CONVERT_TO_SHORTPTR(ref8);
-    uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
-    av1_highbd_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row,
-                           p_width, p_height, p_stride, subsampling_x,
-                           subsampling_y, bd, conv_params, alpha, beta, gamma,
-                           delta);
-  } else {
-    highbd_warp_plane_old(wm, ref8, width, height, stride, pred8, p_col, p_row,
-                          p_width, p_height, p_stride, subsampling_x,
-                          subsampling_y, x_scale, y_scale, bd, conv_params);
-  }
+  const int32_t *const mat = wm->wmmat;
+  const int16_t alpha = wm->alpha;
+  const int16_t beta = wm->beta;
+  const int16_t gamma = wm->gamma;
+  const int16_t delta = wm->delta;
+
+  const uint16_t *const ref = CONVERT_TO_SHORTPTR(ref8);
+  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  av1_highbd_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row,
+                         p_width, p_height, p_stride, subsampling_x,
+                         subsampling_y, bd, conv_params, alpha, beta, gamma,
+                         delta);
 }
 
 static int64_t highbd_frame_error(const uint16_t *const ref, int stride,
@@ -1120,25 +582,25 @@ static int64_t highbd_frame_error(const uint16_t *const ref, int stride,
 static int64_t highbd_warp_error(
     WarpedMotionParams *wm, const uint8_t *const ref8, int width, int height,
     int stride, const uint8_t *const dst8, int p_col, int p_row, int p_width,
-    int p_height, int p_stride, int subsampling_x, int subsampling_y,
-    int x_scale, int y_scale, int bd, int64_t best_error) {
+    int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd,
+    int64_t best_error) {
   int64_t gm_sumerr = 0;
-  int warp_w, warp_h;
-  int error_bsize_w = AOMMIN(p_width, WARP_ERROR_BLOCK);
-  int error_bsize_h = AOMMIN(p_height, WARP_ERROR_BLOCK);
+  const int error_bsize_w = AOMMIN(p_width, WARP_ERROR_BLOCK);
+  const int error_bsize_h = AOMMIN(p_height, WARP_ERROR_BLOCK);
   uint16_t tmp[WARP_ERROR_BLOCK * WARP_ERROR_BLOCK];
 
-  ConvolveParams conv_params = get_conv_params(0, 0, 0);
+  ConvolveParams conv_params = get_conv_params(0, 0, 0, bd);
+  conv_params.use_jnt_comp_avg = 0;
   for (int i = p_row; i < p_row + p_height; i += WARP_ERROR_BLOCK) {
     for (int j = p_col; j < p_col + p_width; j += WARP_ERROR_BLOCK) {
       // avoid warping extra 8x8 blocks in the padded region of the frame
       // when p_width and p_height are not multiples of WARP_ERROR_BLOCK
-      warp_w = AOMMIN(error_bsize_w, p_col + p_width - j);
-      warp_h = AOMMIN(error_bsize_h, p_row + p_height - i);
+      const int warp_w = AOMMIN(error_bsize_w, p_col + p_width - j);
+      const int warp_h = AOMMIN(error_bsize_h, p_row + p_height - i);
       highbd_warp_plane(wm, ref8, width, height, stride,
                         CONVERT_TO_BYTEPTR(tmp), j, i, warp_w, warp_h,
-                        WARP_ERROR_BLOCK, subsampling_x, subsampling_y, x_scale,
-                        y_scale, bd, &conv_params);
+                        WARP_ERROR_BLOCK, subsampling_x, subsampling_y, bd,
+                        &conv_params);
 
       gm_sumerr += highbd_frame_error(
           tmp, WARP_ERROR_BLOCK, CONVERT_TO_SHORTPTR(dst8) + j + i * p_stride,
@@ -1148,41 +610,11 @@ static int64_t highbd_warp_error(
   }
   return gm_sumerr;
 }
-#endif  // CONFIG_HIGHBITDEPTH
 
 static INLINE int error_measure(int err) {
   return error_measure_lut[255 + err];
 }
 
-static void warp_plane_old(const WarpedMotionParams *const wm,
-                           const uint8_t *const ref, int width, int height,
-                           int stride, uint8_t *pred, int p_col, int p_row,
-                           int p_width, int p_height, int p_stride,
-                           int subsampling_x, int subsampling_y, int x_scale,
-                           int y_scale, ConvolveParams *conv_params) {
-  int i, j;
-  ProjectPointsFunc projectpoints = get_project_points_type(wm->wmtype);
-  if (projectpoints == NULL) return;
-  for (i = p_row; i < p_row + p_height; ++i) {
-    for (j = p_col; j < p_col + p_width; ++j) {
-      int in[2], out[2];
-      in[0] = j;
-      in[1] = i;
-      projectpoints(wm->wmmat, in, out, 1, 2, 2, subsampling_x, subsampling_y);
-      out[0] = ROUND_POWER_OF_TWO_SIGNED(out[0] * x_scale, SCALE_SUBPEL_BITS);
-      out[1] = ROUND_POWER_OF_TWO_SIGNED(out[1] * y_scale, SCALE_SUBPEL_BITS);
-      if (conv_params->do_average)
-        pred[(j - p_col) + (i - p_row) * p_stride] = ROUND_POWER_OF_TWO(
-            pred[(j - p_col) + (i - p_row) * p_stride] +
-                warp_interpolate(ref, out[0], out[1], width, height, stride),
-            1);
-      else
-        pred[(j - p_col) + (i - p_row) * p_stride] =
-            warp_interpolate(ref, out[0], out[1], width, height, stride);
-    }
-  }
-}
-
 /* The warp filter for ROTZOOM and AFFINE models works as follows:
    * Split the input into 8x8 blocks
    * For each block, project the point (4, 4) within the block, to get the
@@ -1237,10 +669,10 @@ static void warp_plane_old(const WarpedMotionParams *const wm,
     This allows the derivation of the appropriate bit widths and offsets for
     the various intermediate values: If
 
-    F := WARPEDPIXEL_FILTER_BITS = 7 (or else the above ranges need adjusting)
+    F := FILTER_BITS = 7 (or else the above ranges need adjusting)
          So a *single* filter stage maps a k-bit input to a (k + F + 1)-bit
          intermediate value.
-    H := HORSHEAR_REDUCE_PREC_BITS
+    H := ROUND0_BITS
     V := VERSHEAR_REDUCE_PREC_BITS
     (and note that we must have H + V = 2*F for the output to have the same
      scale as the input)
@@ -1275,38 +707,23 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
                        ConvolveParams *conv_params, int16_t alpha, int16_t beta,
                        int16_t gamma, int16_t delta) {
   int32_t tmp[15 * 8];
-  int i, j, k, l, m;
   const int bd = 8;
-#if CONFIG_CONVOLVE_ROUND
-  const int use_conv_params = conv_params->round == CONVOLVE_OPT_NO_ROUND;
-  const int reduce_bits_horiz =
-      use_conv_params ? conv_params->round_0 : HORSHEAR_REDUCE_PREC_BITS;
-  const int max_bits_horiz =
-      use_conv_params
-          ? bd + FILTER_BITS + 1 - conv_params->round_0
-          : bd + WARPEDPIXEL_FILTER_BITS + 1 - HORSHEAR_REDUCE_PREC_BITS;
-  const int offset_bits_horiz =
-      use_conv_params ? bd + FILTER_BITS - 1 : bd + WARPEDPIXEL_FILTER_BITS - 1;
-  const int offset_bits_vert =
-      use_conv_params
-          ? bd + 2 * FILTER_BITS - conv_params->round_0
-          : bd + 2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS;
-  if (use_conv_params) {
-    conv_params->do_post_rounding = 1;
-  }
-  assert(FILTER_BITS == WARPEDPIXEL_FILTER_BITS);
-#else
-  const int reduce_bits_horiz = HORSHEAR_REDUCE_PREC_BITS;
-  const int max_bits_horiz =
-      bd + WARPEDPIXEL_FILTER_BITS + 1 - HORSHEAR_REDUCE_PREC_BITS;
-  const int offset_bits_horiz = bd + WARPEDPIXEL_FILTER_BITS - 1;
-  const int offset_bits_vert =
-      bd + 2 * WARPEDPIXEL_FILTER_BITS - HORSHEAR_REDUCE_PREC_BITS;
-#endif
+  const int reduce_bits_horiz = conv_params->round_0;
+  const int reduce_bits_vert = conv_params->is_compound
+                                   ? conv_params->round_1
+                                   : 2 * FILTER_BITS - reduce_bits_horiz;
+  const int max_bits_horiz = bd + FILTER_BITS + 1 - reduce_bits_horiz;
+  const int offset_bits_horiz = bd + FILTER_BITS - 1;
+  const int offset_bits_vert = bd + 2 * FILTER_BITS - reduce_bits_horiz;
+  const int round_bits =
+      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
+  const int offset_bits = bd + 2 * FILTER_BITS - conv_params->round_0;
   (void)max_bits_horiz;
+  assert(IMPLIES(conv_params->is_compound, conv_params->dst != NULL));
+  assert(IMPLIES(conv_params->do_average, conv_params->is_compound));
 
-  for (i = p_row; i < p_row + p_height; i += 8) {
-    for (j = p_col; j < p_col + p_width; j += 8) {
+  for (int i = p_row; i < p_row + p_height; i += 8) {
+    for (int j = p_col; j < p_col + p_width; j += 8) {
       // Calculate the center of this 8x8 block,
       // project to luma coordinates (if in a subsampled chroma plane),
       // apply the affine transformation,
@@ -1330,17 +747,13 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
       sy4 &= ~((1 << WARP_PARAM_REDUCE_BITS) - 1);
 
       // Horizontal filter
-      for (k = -7; k < 8; ++k) {
+      for (int k = -7; k < 8; ++k) {
         // Clamp to top/bottom edge of the frame
-        int iy = iy4 + k;
-        if (iy < 0)
-          iy = 0;
-        else if (iy > height - 1)
-          iy = height - 1;
+        const int iy = clamp(iy4 + k, 0, height - 1);
 
         int sx = sx4 + beta * (k + 4);
 
-        for (l = -4; l < 4; ++l) {
+        for (int l = -4; l < 4; ++l) {
           int ix = ix4 + l - 3;
           // At this point, sx = sx4 + alpha * l + beta * k
           const int offs = ROUND_POWER_OF_TWO(sx, WARPEDDIFF_PREC_BITS) +
@@ -1349,13 +762,9 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
           const int16_t *coeffs = warped_filter[offs];
 
           int32_t sum = 1 << offset_bits_horiz;
-          for (m = 0; m < 8; ++m) {
+          for (int m = 0; m < 8; ++m) {
             // Clamp to left/right edge of the frame
-            int sample_x = ix + m;
-            if (sample_x < 0)
-              sample_x = 0;
-            else if (sample_x > width - 1)
-              sample_x = width - 1;
+            const int sample_x = clamp(ix + m, 0, width - 1);
 
             sum += ref[iy * stride + sample_x] * coeffs[m];
           }
@@ -1367,9 +776,9 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
       }
 
       // Vertical filter
-      for (k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) {
+      for (int k = -4; k < AOMMIN(4, p_row + p_height - i - 4); ++k) {
         int sy = sy4 + delta * (k + 4);
-        for (l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) {
+        for (int l = -4; l < AOMMIN(4, p_col + p_width - j - 4); ++l) {
           // At this point, sy = sy4 + gamma * l + delta * k
           const int offs = ROUND_POWER_OF_TWO(sy, WARPEDDIFF_PREC_BITS) +
                            WARPEDPIXEL_PREC_SHIFTS;
@@ -1377,36 +786,40 @@ void av1_warp_affine_c(const int32_t *mat, const uint8_t *ref, int width,
           const int16_t *coeffs = warped_filter[offs];
 
           int32_t sum = 1 << offset_bits_vert;
-          for (m = 0; m < 8; ++m) {
+          for (int m = 0; m < 8; ++m) {
             sum += tmp[(k + m + 4) * 8 + (l + 4)] * coeffs[m];
           }
-#if CONFIG_CONVOLVE_ROUND
-          if (use_conv_params) {
+
+          if (conv_params->is_compound) {
             CONV_BUF_TYPE *p =
                 &conv_params
                      ->dst[(i - p_row + k + 4) * conv_params->dst_stride +
                            (j - p_col + l + 4)];
-            sum = ROUND_POWER_OF_TWO(sum, conv_params->round_1) -
-                  (1 << (offset_bits_horiz + FILTER_BITS -
-                         conv_params->round_0 - conv_params->round_1)) -
-                  (1 << (offset_bits_vert - conv_params->round_1));
-            if (conv_params->do_average)
-              *p += sum;
-            else
+            sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert);
+            if (conv_params->do_average) {
+              uint8_t *dst8 =
+                  &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
+              int32_t tmp32 = *p;
+              if (conv_params->use_jnt_comp_avg) {
+                tmp32 = tmp32 * conv_params->fwd_offset +
+                        sum * conv_params->bck_offset;
+                tmp32 = tmp32 >> DIST_PRECISION_BITS;
+              } else {
+                tmp32 += sum;
+                tmp32 = tmp32 >> 1;
+              }
+              tmp32 = tmp32 - (1 << (offset_bits - conv_params->round_1)) -
+                      (1 << (offset_bits - conv_params->round_1 - 1));
+              *dst8 = clip_pixel(ROUND_POWER_OF_TWO(tmp32, round_bits));
+            } else {
               *p = sum;
+            }
           } else {
-#else
-          {
-#endif
             uint8_t *p =
                 &pred[(i - p_row + k + 4) * p_stride + (j - p_col + l + 4)];
-            sum = ROUND_POWER_OF_TWO(sum, VERSHEAR_REDUCE_PREC_BITS);
+            sum = ROUND_POWER_OF_TWO(sum, reduce_bits_vert);
             assert(0 <= sum && sum < (1 << (bd + 2)));
-            uint8_t px = clip_pixel(sum - (1 << (bd - 1)) - (1 << bd));
-            if (conv_params->do_average)
-              *p = ROUND_POWER_OF_TWO(*p + px, 1);
-            else
-              *p = px;
+            *p = clip_pixel(sum - (1 << (bd - 1)) - (1 << bd));
           }
           sy += gamma;
         }
@@ -1419,27 +832,20 @@ static void warp_plane(WarpedMotionParams *wm, const uint8_t *const ref,
                        int width, int height, int stride, uint8_t *pred,
                        int p_col, int p_row, int p_width, int p_height,
                        int p_stride, int subsampling_x, int subsampling_y,
-                       int x_scale, int y_scale, ConvolveParams *conv_params) {
+                       ConvolveParams *conv_params) {
+  assert(wm->wmtype <= AFFINE);
   if (wm->wmtype == ROTZOOM) {
     wm->wmmat[5] = wm->wmmat[2];
     wm->wmmat[4] = -wm->wmmat[3];
   }
-  if ((wm->wmtype == ROTZOOM || wm->wmtype == AFFINE) &&
-      x_scale == SCALE_SUBPEL_SHIFTS && y_scale == SCALE_SUBPEL_SHIFTS) {
-    const int32_t *const mat = wm->wmmat;
-    const int16_t alpha = wm->alpha;
-    const int16_t beta = wm->beta;
-    const int16_t gamma = wm->gamma;
-    const int16_t delta = wm->delta;
-
-    av1_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row,
-                    p_width, p_height, p_stride, subsampling_x, subsampling_y,
-                    conv_params, alpha, beta, gamma, delta);
-  } else {
-    warp_plane_old(wm, ref, width, height, stride, pred, p_col, p_row, p_width,
-                   p_height, p_stride, subsampling_x, subsampling_y, x_scale,
-                   y_scale, conv_params);
-  }
+  const int32_t *const mat = wm->wmmat;
+  const int16_t alpha = wm->alpha;
+  const int16_t beta = wm->beta;
+  const int16_t gamma = wm->gamma;
+  const int16_t delta = wm->delta;
+  av1_warp_affine(mat, ref, width, height, stride, pred, p_col, p_row, p_width,
+                  p_height, p_stride, subsampling_x, subsampling_y, conv_params,
+                  alpha, beta, gamma, delta);
 }
 
 static int64_t frame_error(const uint8_t *const ref, int stride,
@@ -1459,14 +865,15 @@ static int64_t warp_error(WarpedMotionParams *wm, const uint8_t *const ref,
                           int width, int height, int stride,
                           const uint8_t *const dst, int p_col, int p_row,
                           int p_width, int p_height, int p_stride,
-                          int subsampling_x, int subsampling_y, int x_scale,
-                          int y_scale, int64_t best_error) {
+                          int subsampling_x, int subsampling_y,
+                          int64_t best_error) {
   int64_t gm_sumerr = 0;
   int warp_w, warp_h;
   int error_bsize_w = AOMMIN(p_width, WARP_ERROR_BLOCK);
   int error_bsize_h = AOMMIN(p_height, WARP_ERROR_BLOCK);
   uint8_t tmp[WARP_ERROR_BLOCK * WARP_ERROR_BLOCK];
-  ConvolveParams conv_params = get_conv_params(0, 0, 0);
+  ConvolveParams conv_params = get_conv_params(0, 0, 0, 8);
+  conv_params.use_jnt_comp_avg = 0;
 
   for (int i = p_row; i < p_row + p_height; i += WARP_ERROR_BLOCK) {
     for (int j = p_col; j < p_col + p_width; j += WARP_ERROR_BLOCK) {
@@ -1475,8 +882,7 @@ static int64_t warp_error(WarpedMotionParams *wm, const uint8_t *const ref,
       warp_w = AOMMIN(error_bsize_w, p_col + p_width - j);
       warp_h = AOMMIN(error_bsize_h, p_row + p_height - i);
       warp_plane(wm, ref, width, height, stride, tmp, j, i, warp_w, warp_h,
-                 WARP_ERROR_BLOCK, subsampling_x, subsampling_y, x_scale,
-                 y_scale, &conv_params);
+                 WARP_ERROR_BLOCK, subsampling_x, subsampling_y, &conv_params);
 
       gm_sumerr += frame_error(tmp, WARP_ERROR_BLOCK, dst + j + i * p_stride,
                                warp_w, warp_h, p_stride);
@@ -1486,70 +892,49 @@ static int64_t warp_error(WarpedMotionParams *wm, const uint8_t *const ref,
   return gm_sumerr;
 }
 
-int64_t av1_frame_error(
-#if CONFIG_HIGHBITDEPTH
-    int use_hbd, int bd,
-#endif  // CONFIG_HIGHBITDEPTH
-    const uint8_t *ref, int stride, uint8_t *dst, int p_width, int p_height,
-    int p_stride) {
-#if CONFIG_HIGHBITDEPTH
+int64_t av1_frame_error(int use_hbd, int bd, const uint8_t *ref, int stride,
+                        uint8_t *dst, int p_width, int p_height, int p_stride) {
   if (use_hbd) {
     return highbd_frame_error(CONVERT_TO_SHORTPTR(ref), stride,
                               CONVERT_TO_SHORTPTR(dst), p_width, p_height,
                               p_stride, bd);
   }
-#endif  // CONFIG_HIGHBITDEPTH
   return frame_error(ref, stride, dst, p_width, p_height, p_stride);
 }
 
-int64_t av1_warp_error(WarpedMotionParams *wm,
-#if CONFIG_HIGHBITDEPTH
-                       int use_hbd, int bd,
-#endif  // CONFIG_HIGHBITDEPTH
+int64_t av1_warp_error(WarpedMotionParams *wm, int use_hbd, int bd,
                        const uint8_t *ref, int width, int height, int stride,
                        uint8_t *dst, int p_col, int p_row, int p_width,
                        int p_height, int p_stride, int subsampling_x,
-                       int subsampling_y, int x_scale, int y_scale,
-                       int64_t best_error) {
+                       int subsampling_y, int64_t best_error) {
   if (wm->wmtype <= AFFINE)
     if (!get_shear_params(wm)) return 1;
-#if CONFIG_HIGHBITDEPTH
   if (use_hbd)
     return highbd_warp_error(wm, ref, width, height, stride, dst, p_col, p_row,
                              p_width, p_height, p_stride, subsampling_x,
-                             subsampling_y, x_scale, y_scale, bd, best_error);
-#endif  // CONFIG_HIGHBITDEPTH
+                             subsampling_y, bd, best_error);
   return warp_error(wm, ref, width, height, stride, dst, p_col, p_row, p_width,
-                    p_height, p_stride, subsampling_x, subsampling_y, x_scale,
-                    y_scale, best_error);
+                    p_height, p_stride, subsampling_x, subsampling_y,
+                    best_error);
 }
 
-void av1_warp_plane(WarpedMotionParams *wm,
-#if CONFIG_HIGHBITDEPTH
-                    int use_hbd, int bd,
-#endif  // CONFIG_HIGHBITDEPTH
+void av1_warp_plane(WarpedMotionParams *wm, int use_hbd, int bd,
                     const uint8_t *ref, int width, int height, int stride,
                     uint8_t *pred, int p_col, int p_row, int p_width,
                     int p_height, int p_stride, int subsampling_x,
-                    int subsampling_y, int x_scale, int y_scale,
-                    ConvolveParams *conv_params) {
-#if CONFIG_HIGHBITDEPTH
+                    int subsampling_y, ConvolveParams *conv_params) {
   if (use_hbd)
     highbd_warp_plane(wm, ref, width, height, stride, pred, p_col, p_row,
                       p_width, p_height, p_stride, subsampling_x, subsampling_y,
-                      x_scale, y_scale, bd, conv_params);
+                      bd, conv_params);
   else
-#endif  // CONFIG_HIGHBITDEPTH
     warp_plane(wm, ref, width, height, stride, pred, p_col, p_row, p_width,
-               p_height, p_stride, subsampling_x, subsampling_y, x_scale,
-               y_scale, conv_params);
+               p_height, p_stride, subsampling_x, subsampling_y, conv_params);
 }
 
-#if CONFIG_WARPED_MOTION
-#define LEAST_SQUARES_ORDER 2
-
 #define LS_MV_MAX 256  // max mv in 1/8-pel
-#define LS_STEP 2
+// Use LS_STEP = 8 so that 2 less bits needed for A, Bx, By.
+#define LS_STEP 8
 
 // Assuming LS_MV_MAX is < MAX_SB_SIZE * 8,
 // the precision needed is:
@@ -1570,13 +955,17 @@ void av1_warp_plane(WarpedMotionParams *wm,
 #define LS_MAT_MIN (-(1 << (LS_MAT_BITS - 1)))
 #define LS_MAT_MAX ((1 << (LS_MAT_BITS - 1)) - 1)
 
-#define LS_SUM(a) ((a)*4 + LS_STEP * 2)
-#define LS_SQUARE(a) \
-  (((a) * (a)*4 + (a)*4 * LS_STEP + LS_STEP * LS_STEP * 2) >> 2)
-#define LS_PRODUCT1(a, b) \
-  (((a) * (b)*4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP) >> 2)
-#define LS_PRODUCT2(a, b) \
-  (((a) * (b)*4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP * 2) >> 2)
+// By setting LS_STEP = 8, the least 2 bits of every elements in A, Bx, By are
+// 0. So, we can reduce LS_MAT_RANGE_BITS(2) bits here.
+#define LS_SQUARE(a)                                          \
+  (((a) * (a)*4 + (a)*4 * LS_STEP + LS_STEP * LS_STEP * 2) >> \
+   (2 + LS_MAT_DOWN_BITS))
+#define LS_PRODUCT1(a, b)                                           \
+  (((a) * (b)*4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP) >> \
+   (2 + LS_MAT_DOWN_BITS))
+#define LS_PRODUCT2(a, b)                                               \
+  (((a) * (b)*4 + ((a) + (b)) * 2 * LS_STEP + LS_STEP * LS_STEP * 2) >> \
+   (2 + LS_MAT_DOWN_BITS))
 
 #define USE_LIMITED_PREC_MULT 0
 
@@ -1655,22 +1044,24 @@ static int32_t get_mult_shift_diag(int64_t Px, int16_t iDet, int shift) {
 }
 #endif  // USE_LIMITED_PREC_MULT
 
-static int find_affine_int(int np, int *pts1, int *pts2, BLOCK_SIZE bsize,
-                           int mvy, int mvx, WarpedMotionParams *wm, int mi_row,
-                           int mi_col) {
+static int find_affine_int(int np, const int *pts1, const int *pts2,
+                           BLOCK_SIZE bsize, int mvy, int mvx,
+                           WarpedMotionParams *wm, int mi_row, int mi_col) {
   int32_t A[2][2] = { { 0, 0 }, { 0, 0 } };
   int32_t Bx[2] = { 0, 0 };
   int32_t By[2] = { 0, 0 };
-  int i, n = 0;
+  int i;
 
   const int bw = block_size_wide[bsize];
   const int bh = block_size_high[bsize];
-  const int isuy = (mi_row * MI_SIZE + AOMMAX(bh, MI_SIZE) / 2 - 1);
-  const int isux = (mi_col * MI_SIZE + AOMMAX(bw, MI_SIZE) / 2 - 1);
-  const int suy = isuy * 8;
-  const int sux = isux * 8;
+  const int rsuy = (AOMMAX(bh, MI_SIZE) / 2 - 1);
+  const int rsux = (AOMMAX(bw, MI_SIZE) / 2 - 1);
+  const int suy = rsuy * 8;
+  const int sux = rsux * 8;
   const int duy = suy + mvy;
   const int dux = sux + mvx;
+  const int isuy = (mi_row * MI_SIZE + rsuy);
+  const int isux = (mi_col * MI_SIZE + rsux);
 
   // Assume the center pixel of the block has exactly the same motion vector
   // as transmitted for the block. First shift the origin of the source
@@ -1694,13 +1085,15 @@ static int find_affine_int(int np, int *pts1, int *pts2, BLOCK_SIZE bsize,
   //
   // The loop below computes: A = P'P, Bx = P'q, By = P'r
   // We need to just compute inv(A).Bx and inv(A).By for the solutions.
-  int sx, sy, dx, dy;
   // Contribution from neighbor block
-  for (i = 0; i < np && n < LEAST_SQUARES_SAMPLES_MAX; i++) {
-    dx = pts2[i * 2] - dux;
-    dy = pts2[i * 2 + 1] - duy;
-    sx = pts1[i * 2] - sux;
-    sy = pts1[i * 2 + 1] - suy;
+  for (i = 0; i < np; i++) {
+    const int dx = pts2[i * 2] - dux;
+    const int dy = pts2[i * 2 + 1] - duy;
+    const int sx = pts1[i * 2] - sux;
+    const int sy = pts1[i * 2 + 1] - suy;
+    // (TODO)yunqing: This comparison wouldn't be necessary if the sample
+    // selection is done in find_samples(). Also, global offset can be removed
+    // while collecting samples.
     if (abs(sx - dx) < LS_MV_MAX && abs(sy - dy) < LS_MV_MAX) {
       A[0][0] += LS_SQUARE(sx);
       A[0][1] += LS_PRODUCT1(sx, sy);
@@ -1709,41 +1102,20 @@ static int find_affine_int(int np, int *pts1, int *pts2, BLOCK_SIZE bsize,
       Bx[1] += LS_PRODUCT1(sy, dx);
       By[0] += LS_PRODUCT1(sx, dy);
       By[1] += LS_PRODUCT2(sy, dy);
-      n++;
     }
   }
-  int downshift;
-  if (n >= 4)
-    downshift = LS_MAT_DOWN_BITS;
-  else if (n >= 2)
-    downshift = LS_MAT_DOWN_BITS - 1;
-  else
-    downshift = LS_MAT_DOWN_BITS - 2;
-
-  // Reduce precision by downshift bits
-  A[0][0] = clamp(ROUND_POWER_OF_TWO_SIGNED(A[0][0], downshift), LS_MAT_MIN,
-                  LS_MAT_MAX);
-  A[0][1] = clamp(ROUND_POWER_OF_TWO_SIGNED(A[0][1], downshift), LS_MAT_MIN,
-                  LS_MAT_MAX);
-  A[1][1] = clamp(ROUND_POWER_OF_TWO_SIGNED(A[1][1], downshift), LS_MAT_MIN,
-                  LS_MAT_MAX);
-  Bx[0] = clamp(ROUND_POWER_OF_TWO_SIGNED(Bx[0], downshift), LS_MAT_MIN,
-                LS_MAT_MAX);
-  Bx[1] = clamp(ROUND_POWER_OF_TWO_SIGNED(Bx[1], downshift), LS_MAT_MIN,
-                LS_MAT_MAX);
-  By[0] = clamp(ROUND_POWER_OF_TWO_SIGNED(By[0], downshift), LS_MAT_MIN,
-                LS_MAT_MAX);
-  By[1] = clamp(ROUND_POWER_OF_TWO_SIGNED(By[1], downshift), LS_MAT_MIN,
-                LS_MAT_MAX);
-
-  int64_t Px[2], Py[2], Det;
-  int16_t iDet, shift;
 
-  // These divided by the Det, are the least squares solutions
-  Px[0] = (int64_t)A[1][1] * Bx[0] - (int64_t)A[0][1] * Bx[1];
-  Px[1] = -(int64_t)A[0][1] * Bx[0] + (int64_t)A[0][0] * Bx[1];
-  Py[0] = (int64_t)A[1][1] * By[0] - (int64_t)A[0][1] * By[1];
-  Py[1] = -(int64_t)A[0][1] * By[0] + (int64_t)A[0][0] * By[1];
+  // Just for debugging, and can be removed later.
+  assert(A[0][0] >= LS_MAT_MIN && A[0][0] <= LS_MAT_MAX);
+  assert(A[0][1] >= LS_MAT_MIN && A[0][1] <= LS_MAT_MAX);
+  assert(A[1][1] >= LS_MAT_MIN && A[1][1] <= LS_MAT_MAX);
+  assert(Bx[0] >= LS_MAT_MIN && Bx[0] <= LS_MAT_MAX);
+  assert(Bx[1] >= LS_MAT_MIN && Bx[1] <= LS_MAT_MAX);
+  assert(By[0] >= LS_MAT_MIN && By[0] <= LS_MAT_MAX);
+  assert(By[1] >= LS_MAT_MIN && By[1] <= LS_MAT_MAX);
+
+  int64_t Det;
+  int16_t iDet, shift;
 
   // Compute Determinant of A
   Det = (int64_t)A[0][0] * A[1][1] - (int64_t)A[0][1] * A[0][1];
@@ -1755,6 +1127,14 @@ static int find_affine_int(int np, int *pts1, int *pts2, BLOCK_SIZE bsize,
     shift = 0;
   }
 
+  int64_t Px[2], Py[2];
+
+  // These divided by the Det, are the least squares solutions
+  Px[0] = (int64_t)A[1][1] * Bx[0] - (int64_t)A[0][1] * Bx[1];
+  Px[1] = -(int64_t)A[0][1] * Bx[0] + (int64_t)A[0][0] * Bx[1];
+  Py[0] = (int64_t)A[1][1] * By[0] - (int64_t)A[0][1] * By[1];
+  Py[1] = -(int64_t)A[0][1] * By[0] + (int64_t)A[0][0] * By[1];
+
   wm->wmmat[2] = get_mult_shift_diag(Px[0], iDet, shift);
   wm->wmmat[3] = get_mult_shift_ndiag(Px[1], iDet, shift);
   wm->wmmat[4] = get_mult_shift_ndiag(Py[0], iDet, shift);
@@ -1783,13 +1163,13 @@ int find_projection(int np, int *pts1, int *pts2, BLOCK_SIZE bsize, int mvy,
                     int mvx, WarpedMotionParams *wm_params, int mi_row,
                     int mi_col) {
   assert(wm_params->wmtype == AFFINE);
-  const int result = find_affine_int(np, pts1, pts2, bsize, mvy, mvx, wm_params,
-                                     mi_row, mi_col);
-  if (result == 0) {
-    // check compatibility with the fast warp filter
-    if (!get_shear_params(wm_params)) return 1;
-  }
 
-  return result;
+  if (find_affine_int(np, pts1, pts2, bsize, mvy, mvx, wm_params, mi_row,
+                      mi_col))
+    return 1;
+
+  // check compatibility with the fast warp filter
+  if (!get_shear_params(wm_params)) return 1;
+
+  return 0;
 }
-#endif  // CONFIG_WARPED_MOTION