Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918

1 files changed, 227 insertions, 0 deletions

diff --git a/third_party/aom/aom_dsp/aom_convolve.c b/third_party/aom/aom_dsp/aom_convolve.c
index 4dac6aacc..c903ea52d 100644
--- a/third_party/aom/aom_dsp/aom_convolve.c
+++ b/third_party/aom/aom_dsp/aom_convolve.c
@@ -41,6 +41,29 @@ static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                   uint8_t *dst, ptrdiff_t dst_stride,
+                                   const InterpKernel *x_filters, int x0_qn,
+                                   int x_step_qn, int w, int h) {
+  int x, y;
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_qn = x0_qn;
+    for (x = 0; x < w; ++x) {
+      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];  // q8
+      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
+      assert(x_filter_idx < SUBPEL_SHIFTS);
+      const int16_t *const x_filter = x_filters[x_filter_idx];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
+      dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+      x_qn += x_step_qn;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
 static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const InterpKernel *x_filters, int x0_q4,
@@ -63,6 +86,30 @@ static void convolve_avg_horiz(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                       uint8_t *dst, ptrdiff_t dst_stride,
+                                       const InterpKernel *x_filters, int x0_qn,
+                                       int x_step_qn, int w, int h) {
+  int x, y;
+  src -= SUBPEL_TAPS / 2 - 1;
+  for (y = 0; y < h; ++y) {
+    int x_qn = x0_qn;
+    for (x = 0; x < w; ++x) {
+      const uint8_t *const src_x = &src[x_qn >> SCALE_SUBPEL_BITS];
+      const int x_filter_idx = (x_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS;
+      assert(x_filter_idx < SUBPEL_SHIFTS);
+      const int16_t *const x_filter = x_filters[x_filter_idx];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k) sum += src_x[k] * x_filter[k];
+      dst[x] = ROUND_POWER_OF_TWO(
+          dst[x] + clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)), 1);
+      x_qn += x_step_qn;
+    }
+    src += src_stride;
+    dst += dst_stride;
+  }
+}
+
 static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const InterpKernel *y_filters, int y0_q4,
@@ -86,6 +133,31 @@ static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                  uint8_t *dst, ptrdiff_t dst_stride,
+                                  const InterpKernel *y_filters, int y0_qn,
+                                  int y_step_qn, int w, int h) {
+  int x, y;
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+  for (x = 0; x < w; ++x) {
+    int y_qn = y0_qn;
+    for (y = 0; y < h; ++y) {
+      const unsigned char *src_y =
+          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter =
+          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
+      y_qn += y_step_qn;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
 static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
                               const InterpKernel *y_filters, int y0_q4,
@@ -112,6 +184,34 @@ static void convolve_avg_vert(const uint8_t *src, ptrdiff_t src_stride,
   }
 }
 
+static void convolve_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                      uint8_t *dst, ptrdiff_t dst_stride,
+                                      const InterpKernel *y_filters, int y0_qn,
+                                      int y_step_qn, int w, int h) {
+  int x, y;
+  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
+
+  for (x = 0; x < w; ++x) {
+    int y_qn = y0_qn;
+    for (y = 0; y < h; ++y) {
+      const unsigned char *src_y =
+          &src[(y_qn >> SCALE_SUBPEL_BITS) * src_stride];
+      const int16_t *const y_filter =
+          y_filters[(y_qn & SCALE_SUBPEL_MASK) >> SCALE_EXTRA_BITS];
+      int k, sum = 0;
+      for (k = 0; k < SUBPEL_TAPS; ++k)
+        sum += src_y[k * src_stride] * y_filter[k];
+      dst[y * dst_stride] = ROUND_POWER_OF_TWO(
+          dst[y * dst_stride] +
+              clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS)),
+          1);
+      y_qn += y_step_qn;
+    }
+    ++src;
+    ++dst;
+  }
+}
+
 static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      ptrdiff_t dst_stride, const InterpKernel *const x_filters,
                      int x0_q4, int x_step_q4,
@@ -146,6 +246,41 @@ static void convolve(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                 dst_stride, y_filters, y0_q4, y_step_q4, w, h);
 }
 
+static void convolve_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const InterpKernel *const x_filters, int x0_qn,
+                             int x_step_qn, const InterpKernel *const y_filters,
+                             int y0_qn, int y_step_qn, int w, int h) {
+  // TODO(afergs): Update comment here
+  // Note: Fixed size intermediate buffer, temp, places limits on parameters.
+  // 2d filtering proceeds in 2 steps:
+  //   (1) Interpolate horizontally into an intermediate buffer, temp.
+  //   (2) Interpolate temp vertically to derive the sub-pixel result.
+  // Deriving the maximum number of rows in the temp buffer (135):
+  // --Smallest scaling factor is x1/2 ==> y_step_qn = 32 (Normative).
+  // --Largest block size is 64x64 pixels.
+  // --64 rows in the downscaled frame span a distance of (64 - 1) * 32 in the
+  //   original frame (in 1/16th pixel units).
+  // --Must round-up because block may be located at sub-pixel position.
+  // --Require an additional SUBPEL_TAPS rows for the 8-tap filter tails.
+  // --((64 - 1) * 32 + 15) >> 4 + 8 = 135.
+  uint8_t temp[MAX_EXT_SIZE * MAX_SB_SIZE];
+  int intermediate_height =
+      (((h - 1) * y_step_qn + y0_qn) >> SCALE_SUBPEL_BITS) + SUBPEL_TAPS;
+
+  assert(w <= MAX_SB_SIZE);
+  assert(h <= MAX_SB_SIZE);
+
+  assert(y_step_qn <= SCALE_SUBPEL_BITS * 2);
+  assert(x_step_qn <= SCALE_SUBPEL_BITS * 2);
+
+  convolve_horiz_scale_c(src - src_stride * (SUBPEL_TAPS / 2 - 1), src_stride,
+                         temp, MAX_SB_SIZE, x_filters, x0_qn, x_step_qn, w,
+                         intermediate_height);
+  convolve_vert_scale_c(temp + MAX_SB_SIZE * (SUBPEL_TAPS / 2 - 1), MAX_SB_SIZE,
+                        dst, dst_stride, y_filters, y0_qn, y_step_qn, w, h);
+}
+
 static const InterpKernel *get_filter_base(const int16_t *filter) {
   // NOTE: This assumes that the filter table is 256-byte aligned.
   // TODO(agrange) Modify to make independent of table alignment.
@@ -171,6 +306,21 @@ void aom_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                  w, h);
 }
 
+void aom_convolve8_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                 uint8_t *dst, ptrdiff_t dst_stride,
+                                 const int16_t *filter_x, int subpel_x,
+                                 int x_step_qn, const int16_t *filter_y,
+                                 int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  (void)subpel_y;
+  (void)filter_y;
+  (void)y_step_qn;
+
+  convolve_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
+                         x_step_qn, w, h);
+}
+
 void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                                uint8_t *dst, ptrdiff_t dst_stride,
                                const int16_t *filter_x, int x_step_q4,
@@ -186,6 +336,22 @@ void aom_convolve8_avg_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
                      x_step_q4, w, h);
 }
 
+void aom_convolve8_avg_horiz_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x, int subpel_x,
+                                     int x_step_qn, const int16_t *filter_y,
+                                     int subpel_y, int y_step_qn, int w,
+                                     int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  (void)subpel_y;
+  (void)filter_y;
+  (void)y_step_qn;
+
+  convolve_avg_horiz_scale_c(src, src_stride, dst, dst_stride, filters_x,
+                             subpel_x, x_step_qn, w, h);
+}
+
 void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                           uint8_t *dst, ptrdiff_t dst_stride,
                           const int16_t *filter_x, int x_step_q4,
@@ -201,6 +367,21 @@ void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                 w, h);
 }
 
+void aom_convolve8_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                uint8_t *dst, ptrdiff_t dst_stride,
+                                const int16_t *filter_x, int subpel_x,
+                                int x_step_qn, const int16_t *filter_y,
+                                int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  (void)subpel_x;
+  (void)filter_x;
+  (void)x_step_qn;
+
+  convolve_vert_scale_c(src, src_stride, dst, dst_stride, filters_y, subpel_y,
+                        y_step_qn, w, h);
+}
+
 void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                               uint8_t *dst, ptrdiff_t dst_stride,
                               const int16_t *filter_x, int x_step_q4,
@@ -216,6 +397,21 @@ void aom_convolve8_avg_vert_c(const uint8_t *src, ptrdiff_t src_stride,
                     y_step_q4, w, h);
 }
 
+void aom_convolve8_avg_vert_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                                    uint8_t *dst, ptrdiff_t dst_stride,
+                                    const int16_t *filter_x, int subpel_x,
+                                    int x_step_qn, const int16_t *filter_y,
+                                    int subpel_y, int y_step_qn, int w, int h) {
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  (void)subpel_x;
+  (void)filter_x;
+  (void)x_step_qn;
+
+  convolve_avg_vert_scale_c(src, src_stride, dst, dst_stride, filters_y,
+                            subpel_y, y_step_qn, w, h);
+}
+
 void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      ptrdiff_t dst_stride, const int16_t *filter_x,
                      int x_step_q4, const int16_t *filter_y, int y_step_q4,
@@ -230,6 +426,19 @@ void aom_convolve8_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
            filters_y, y0_q4, y_step_q4, w, h);
 }
 
+void aom_convolve8_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                           uint8_t *dst, ptrdiff_t dst_stride,
+                           const int16_t *filter_x, int subpel_x, int x_step_qn,
+                           const int16_t *filter_y, int subpel_y, int y_step_qn,
+                           int w, int h) {
+  const InterpKernel *const filters_x = get_filter_base(filter_x);
+
+  const InterpKernel *const filters_y = get_filter_base(filter_y);
+
+  convolve_scale_c(src, src_stride, dst, dst_stride, filters_x, subpel_x,
+                   x_step_qn, filters_y, subpel_y, y_step_qn, w, h);
+}
+
 void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                          ptrdiff_t dst_stride, const int16_t *filter_x,
                          int x_step_q4, const int16_t *filter_y, int y_step_q4,
@@ -245,6 +454,22 @@ void aom_convolve8_avg_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                      h);
 }
 
+void aom_convolve8_avg_scale_c(const uint8_t *src, ptrdiff_t src_stride,
+                               uint8_t *dst, ptrdiff_t dst_stride,
+                               const int16_t *filter_x, int subpel_x,
+                               int x_step_qn, const int16_t *filter_y,
+                               int subpel_y, int y_step_qn, int w, int h) {
+  /* Fixed size intermediate buffer places limits on parameters. */
+  DECLARE_ALIGNED(16, uint8_t, temp[MAX_SB_SIZE * MAX_SB_SIZE]);
+  assert(w <= MAX_SB_SIZE);
+  assert(h <= MAX_SB_SIZE);
+
+  aom_convolve8_scale_c(src, src_stride, temp, MAX_SB_SIZE, filter_x, subpel_x,
+                        x_step_qn, filter_y, subpel_y, y_step_qn, w, h);
+  aom_convolve_avg_c(temp, MAX_SB_SIZE, dst, dst_stride, NULL, 0, NULL, 0, w,
+                     h);
+}
+
 void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                          ptrdiff_t dst_stride, const int16_t *filter_x,
                          int filter_x_stride, const int16_t *filter_y,
@@ -332,6 +557,7 @@ void aom_scaled_avg_2d_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
                       filter_y, y_step_q4, w, h);
 }
 
+// TODO(afergs): Make sure this works too
 #if CONFIG_LOOP_RESTORATION
 static void convolve_add_src_horiz(const uint8_t *src, ptrdiff_t src_stride,
                                    uint8_t *dst, ptrdiff_t dst_stride,
@@ -569,6 +795,7 @@ void aom_convolve8_add_src_hip_c(const uint8_t *src, ptrdiff_t src_stride,
 }
 #endif  // CONFIG_LOOP_RESTORATION
 
+// TODO(afergs): Make sure this works too
 #if CONFIG_HIGHBITDEPTH
 static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                   uint8_t *dst8, ptrdiff_t dst_stride,