libvpx: Pull from upstream

Current HEAD: 7105df53d7dc13d5e575bc8df714ec8d1da36b06

Includes security fixes and performance improvements. Also removed the VP10
related code from the upstream repository.

BUG=23452792

Change-Id: I97452dff5b1f0756e19d621111797363cc533d46
(cherry picked from commit da49e34c1fb5e99681f4ad99c21d9cfd83eddb96)

1 files changed, 334 insertions, 30 deletions

diff --git a/libvpx/vp9/common/vp9_postproc.c b/libvpx/vp9/common/vp9_postproc.c
index abda4e682..71ab86150 100644
--- a/libvpx/vp9/common/vp9_postproc.c
+++ b/libvpx/vp9/common/vp9_postproc.c
@@ -16,20 +16,21 @@
 #include "./vpx_scale_rtcd.h"
 #include "./vp9_rtcd.h"
 
+#include "vpx_ports/mem.h"
+#include "vpx_ports/system_state.h"
 #include "vpx_scale/vpx_scale.h"
 #include "vpx_scale/yv12config.h"
 
 #include "vp9/common/vp9_onyxc_int.h"
 #include "vp9/common/vp9_postproc.h"
-#include "vp9/common/vp9_systemdependent.h"
 #include "vp9/common/vp9_textblit.h"
 
 #if CONFIG_VP9_POSTPROC
-static const short kernel5[] = {
+static const int16_t kernel5[] = {
   1, 1, 4, 1, 1
 };
 
-const short vp9_rv[] = {
+const int16_t vp9_rv[] = {
   8, 5, 2, 2, 8, 12, 4, 9, 8, 3,
   0, 3, 9, 0, 0, 0, 8, 3, 14, 4,
   10, 1, 11, 14, 1, 14, 9, 6, 12, 11,
@@ -76,6 +77,9 @@ const short vp9_rv[] = {
   0, 9, 5, 5, 11, 10, 13, 9, 10, 13,
 };
 
+static const uint8_t q_diff_thresh = 20;
+static const uint8_t last_q_thresh = 170;
+
 void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
                                      uint8_t *dst_ptr,
                                      int src_pixels_per_line,
@@ -85,10 +89,7 @@ void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
                                      int flimit) {
   uint8_t const *p_src;
   uint8_t *p_dst;
-  int row;
-  int col;
-  int i;
-  int v;
+  int row, col, i, v, kernel;
   int pitch = src_pixels_per_line;
   uint8_t d[8];
   (void)dst_pixels_per_line;
@@ -99,8 +100,8 @@ void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
     p_dst = dst_ptr;
 
     for (col = 0; col < cols; col++) {
-      int kernel = 4;
-      int v = p_src[col];
+      kernel = 4;
+      v = p_src[col];
 
       for (i = -2; i <= 2; i++) {
         if (abs(v - p_src[col + i * pitch]) > flimit)
@@ -122,7 +123,7 @@ void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
       d[i] = p_src[i];
 
     for (col = 0; col < cols; col++) {
-      int kernel = 4;
+      kernel = 4;
       v = p_src[col];
 
       d[col & 7] = v;
@@ -152,6 +153,81 @@ void vp9_post_proc_down_and_across_c(const uint8_t *src_ptr,
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_post_proc_down_and_across_c(const uint16_t *src_ptr,
+                                            uint16_t *dst_ptr,
+                                            int src_pixels_per_line,
+                                            int dst_pixels_per_line,
+                                            int rows,
+                                            int cols,
+                                            int flimit) {
+  uint16_t const *p_src;
+  uint16_t *p_dst;
+  int row, col, i, v, kernel;
+  int pitch = src_pixels_per_line;
+  uint16_t d[8];
+
+  for (row = 0; row < rows; row++) {
+    // post_proc_down for one row.
+    p_src = src_ptr;
+    p_dst = dst_ptr;
+
+    for (col = 0; col < cols; col++) {
+      kernel = 4;
+      v = p_src[col];
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i * pitch]) > flimit)
+          goto down_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i * pitch];
+      }
+
+      v = (kernel >> 3);
+
+    down_skip_convolve:
+      p_dst[col] = v;
+    }
+
+    /* now post_proc_across */
+    p_src = dst_ptr;
+    p_dst = dst_ptr;
+
+    for (i = 0; i < 8; i++)
+      d[i] = p_src[i];
+
+    for (col = 0; col < cols; col++) {
+      kernel = 4;
+      v = p_src[col];
+
+      d[col & 7] = v;
+
+      for (i = -2; i <= 2; i++) {
+        if (abs(v - p_src[col + i]) > flimit)
+          goto across_skip_convolve;
+
+        kernel += kernel5[2 + i] * p_src[col + i];
+      }
+
+      d[col & 7] = (kernel >> 3);
+
+    across_skip_convolve:
+      if (col >= 2)
+        p_dst[col - 2] = d[(col - 2) & 7];
+    }
+
+    /* handle the last two pixels */
+    p_dst[col - 2] = d[(col - 2) & 7];
+    p_dst[col - 1] = d[(col - 1) & 7];
+
+
+    /* next row */
+    src_ptr += pitch;
+    dst_ptr += dst_pixels_per_line;
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static int q2mbl(int x) {
   if (x < 20) x = 20;
 
@@ -162,10 +238,46 @@ static int q2mbl(int x) {
 void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
                                  int rows, int cols, int flimit) {
   int r, c, i;
-
   uint8_t *s = src;
   uint8_t d[16];
 
+  for (r = 0; r < rows; r++) {
+    int sumsq = 0;
+    int sum = 0;
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i] * s[i];
+      sum += s[i];
+      d[i + 8] = 0;
+    }
+
+    for (c = 0; c < cols + 8; c++) {
+      int x = s[c + 7] - s[c - 8];
+      int y = s[c + 7] + s[c - 8];
+
+      sum += x;
+      sumsq += x * y;
+
+      d[c & 15] = s[c];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[c & 15] = (8 + sum + s[c]) >> 4;
+      }
+
+      s[c - 8] = d[(c - 8) & 15];
+    }
+    s += pitch;
+  }
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_mbpost_proc_across_ip_c(uint16_t *src, int pitch,
+                                        int rows, int cols, int flimit) {
+  int r, c, i;
+
+  uint16_t *s = src;
+  uint16_t d[16];
+
 
   for (r = 0; r < rows; r++) {
     int sumsq = 0;
@@ -196,6 +308,7 @@ void vp9_mbpost_proc_across_ip_c(uint8_t *src, int pitch,
     s += pitch;
   }
 }
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 
 void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
                             int rows, int cols, int flimit) {
@@ -207,7 +320,7 @@ void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
     int sumsq = 0;
     int sum   = 0;
     uint8_t d[16];
-    const short *rv2 = rv3 + ((c * 17) & 127);
+    const int16_t *rv2 = rv3 + ((c * 17) & 127);
 
     for (i = -8; i <= 6; i++) {
       sumsq += s[i * pitch] * s[i * pitch];
@@ -229,6 +342,40 @@ void vp9_mbpost_proc_down_c(uint8_t *dst, int pitch,
   }
 }
 
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_mbpost_proc_down_c(uint16_t *dst, int pitch,
+                                   int rows, int cols, int flimit) {
+  int r, c, i;
+  const int16_t *rv3 = &vp9_rv[63 & rand()];  // NOLINT
+
+  for (c = 0; c < cols; c++) {
+    uint16_t *s = &dst[c];
+    int sumsq = 0;
+    int sum = 0;
+    uint16_t d[16];
+    const int16_t *rv2 = rv3 + ((c * 17) & 127);
+
+    for (i = -8; i <= 6; i++) {
+      sumsq += s[i * pitch] * s[i * pitch];
+      sum += s[i * pitch];
+    }
+
+    for (r = 0; r < rows + 8; r++) {
+      sumsq += s[7 * pitch] * s[ 7 * pitch] - s[-8 * pitch] * s[-8 * pitch];
+      sum += s[7 * pitch] - s[-8 * pitch];
+      d[r & 15] = s[0];
+
+      if (sumsq * 15 - sum * sum < flimit) {
+        d[r & 15] = (rv2[r & 127] + sum + s[0]) >> 4;
+      }
+
+      s[-8 * pitch] = d[(r - 8) & 15];
+      s += pitch;
+    }
+  }
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
                                        YV12_BUFFER_CONFIG   *post,
                                        int                   q,
@@ -239,6 +386,51 @@ static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
   (void) low_var_thresh;
   (void) flag;
 
+#if CONFIG_VP9_HIGHBITDEPTH
+  if (source->flags & YV12_FLAG_HIGHBITDEPTH) {
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->y_buffer),
+                                         CONVERT_TO_SHORTPTR(post->y_buffer),
+                                         source->y_stride, post->y_stride,
+                                         source->y_height, source->y_width,
+                                         ppl);
+
+    vp9_highbd_mbpost_proc_across_ip(CONVERT_TO_SHORTPTR(post->y_buffer),
+                                     post->y_stride, post->y_height,
+                                     post->y_width, q2mbl(q));
+
+    vp9_highbd_mbpost_proc_down(CONVERT_TO_SHORTPTR(post->y_buffer),
+                                post->y_stride, post->y_height,
+                                post->y_width, q2mbl(q));
+
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->u_buffer),
+                                         CONVERT_TO_SHORTPTR(post->u_buffer),
+                                         source->uv_stride, post->uv_stride,
+                                         source->uv_height, source->uv_width,
+                                         ppl);
+    vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(source->v_buffer),
+                                         CONVERT_TO_SHORTPTR(post->v_buffer),
+                                         source->uv_stride, post->uv_stride,
+                                         source->uv_height, source->uv_width,
+                                         ppl);
+  } else {
+    vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
+                                  source->y_stride, post->y_stride,
+                                  source->y_height, source->y_width, ppl);
+
+    vp9_mbpost_proc_across_ip(post->y_buffer, post->y_stride, post->y_height,
+                              post->y_width, q2mbl(q));
+
+    vp9_mbpost_proc_down(post->y_buffer, post->y_stride, post->y_height,
+                         post->y_width, q2mbl(q));
+
+    vp9_post_proc_down_and_across(source->u_buffer, post->u_buffer,
+                                  source->uv_stride, post->uv_stride,
+                                  source->uv_height, source->uv_width, ppl);
+    vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
+                                  source->uv_stride, post->uv_stride,
+                                  source->uv_height, source->uv_width, ppl);
+  }
+#else
   vp9_post_proc_down_and_across(source->y_buffer, post->y_buffer,
                                 source->y_stride, post->y_stride,
                                 source->y_height, source->y_width, ppl);
@@ -255,6 +447,7 @@ static void deblock_and_de_macro_block(YV12_BUFFER_CONFIG   *source,
   vp9_post_proc_down_and_across(source->v_buffer, post->v_buffer,
                                 source->uv_stride, post->uv_stride,
                                 source->uv_height, source->uv_width, ppl);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 }
 
 void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
@@ -271,10 +464,26 @@ void vp9_deblock(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
   uint8_t *const dsts[3] = {dst->y_buffer, dst->u_buffer, dst->v_buffer};
   const int dst_strides[3] = {dst->y_stride, dst->uv_stride, dst->uv_stride};
 
-  for (i = 0; i < MAX_MB_PLANE; ++i)
+  for (i = 0; i < MAX_MB_PLANE; ++i) {
+#if CONFIG_VP9_HIGHBITDEPTH
+    assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
+           (dst->flags & YV12_FLAG_HIGHBITDEPTH));
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      vp9_highbd_post_proc_down_and_across(CONVERT_TO_SHORTPTR(srcs[i]),
+                                           CONVERT_TO_SHORTPTR(dsts[i]),
+                                           src_strides[i], dst_strides[i],
+                                           src_heights[i], src_widths[i], ppl);
+    } else {
+      vp9_post_proc_down_and_across(srcs[i], dsts[i],
+                                    src_strides[i], dst_strides[i],
+                                    src_heights[i], src_widths[i], ppl);
+    }
+#else
     vp9_post_proc_down_and_across(srcs[i], dsts[i],
                                   src_strides[i], dst_strides[i],
                                   src_heights[i], src_widths[i], ppl);
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+  }
 }
 
 void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
@@ -293,15 +502,34 @@ void vp9_denoise(const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst,
 
   for (i = 0; i < MAX_MB_PLANE; ++i) {
     const int src_stride = src_strides[i];
-    const uint8_t *const src = srcs[i] + 2 * src_stride + 2;
     const int src_width = src_widths[i] - 4;
     const int src_height = src_heights[i] - 4;
-
     const int dst_stride = dst_strides[i];
-    uint8_t *const dst = dsts[i] + 2 * dst_stride + 2;
 
-    vp9_post_proc_down_and_across(src, dst, src_stride, dst_stride,
+#if CONFIG_VP9_HIGHBITDEPTH
+    assert((src->flags & YV12_FLAG_HIGHBITDEPTH) ==
+           (dst->flags & YV12_FLAG_HIGHBITDEPTH));
+    if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
+      const uint16_t *const src_plane = CONVERT_TO_SHORTPTR(
+          srcs[i] + 2 * src_stride + 2);
+      uint16_t *const dst_plane = CONVERT_TO_SHORTPTR(
+          dsts[i] + 2 * dst_stride + 2);
+      vp9_highbd_post_proc_down_and_across(src_plane, dst_plane, src_stride,
+                                           dst_stride, src_height, src_width,
+                                           ppl);
+    } else {
+      const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
+      uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
+
+      vp9_post_proc_down_and_across(src_plane, dst_plane, src_stride,
+                                    dst_stride, src_height, src_width, ppl);
+    }
+#else
+    const uint8_t *const src_plane = srcs[i] + 2 * src_stride + 2;
+    uint8_t *const dst_plane = dsts[i] + 2 * dst_stride + 2;
+    vp9_post_proc_down_and_across(src_plane, dst_plane, src_stride, dst_stride,
                                   src_height, src_width, ppl);
+#endif
   }
 }
 
@@ -316,7 +544,7 @@ static void fillrd(struct postproc_state *state, int q, int a) {
   double sigma;
   int ai = a, qi = q, i;
 
-  vp9_clear_system_state();
+  vpx_clear_system_state();
 
   sigma = ai + .5 + .6 * (63 - qi) / 63.0;
 
@@ -324,16 +552,15 @@ static void fillrd(struct postproc_state *state, int q, int a) {
    * a gaussian distribution with sigma determined by q.
    */
   {
-    double i;
     int next, j;
 
     next = 0;
 
     for (i = -32; i < 32; i++) {
-      int a = (int)(0.5 + 256 * gaussian(sigma, 0, i));
+      int a_i = (int)(0.5 + 256 * gaussian(sigma, 0, i));
 
-      if (a) {
-        for (j = 0; j < a; j++) {
+      if (a_i) {
+        for (j = 0; j < a_i; j++) {
           char_dist[next + j] = (char) i;
         }
 
@@ -366,6 +593,9 @@ void vp9_plane_add_noise_c(uint8_t *start, char *noise,
                            unsigned int width, unsigned int height, int pitch) {
   unsigned int i, j;
 
+  // TODO(jbb): why does simd code use both but c doesn't,  normalize and
+  // fix..
+  (void) bothclamp;
   for (i = 0; i < height; i++) {
     uint8_t *pos = start + i * pitch;
     char  *ref = (char *)(noise + (rand() & 0xff));  // NOLINT
@@ -382,9 +612,20 @@ void vp9_plane_add_noise_c(uint8_t *start, char *noise,
   }
 }
 
+static void swap_mi_and_prev_mi(VP9_COMMON *cm) {
+  // Current mip will be the prev_mip for the next frame.
+  MODE_INFO *temp = cm->postproc_state.prev_mip;
+  cm->postproc_state.prev_mip = cm->mip;
+  cm->mip = temp;
+
+  // Update the upper left visible macroblock ptrs.
+  cm->mi = cm->mip + cm->mi_stride + 1;
+  cm->postproc_state.prev_mi = cm->postproc_state.prev_mip + cm->mi_stride + 1;
+}
+
 int vp9_post_proc_frame(struct VP9Common *cm,
                         YV12_BUFFER_CONFIG *dest, vp9_ppflags_t *ppflags) {
-  const int q = MIN(63, cm->lf.filter_level * 10 / 6);
+  const int q = MIN(105, cm->lf.filter_level * 2);
   const int flags = ppflags->post_proc_flag;
   YV12_BUFFER_CONFIG *const ppbuf = &cm->post_proc_buffer;
   struct postproc_state *const ppstate = &cm->postproc_state;
@@ -397,17 +638,76 @@ int vp9_post_proc_frame(struct VP9Common *cm,
     return 0;
   }
 
-  vp9_clear_system_state();
+  vpx_clear_system_state();
 
-#if CONFIG_VP9_POSTPROC || CONFIG_INTERNAL_STATS
-  if (vp9_realloc_frame_buffer(&cm->post_proc_buffer, cm->width, cm->height,
+  // Alloc memory for prev_mip in the first frame.
+  if (cm->current_video_frame == 1) {
+    cm->postproc_state.last_base_qindex = cm->base_qindex;
+    cm->postproc_state.last_frame_valid = 1;
+    ppstate->prev_mip = vpx_calloc(cm->mi_alloc_size, sizeof(*cm->mip));
+    if (!ppstate->prev_mip) {
+      return 1;
+    }
+    ppstate->prev_mi = ppstate->prev_mip + cm->mi_stride + 1;
+    memset(ppstate->prev_mip, 0,
+           cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->mip));
+  }
+
+  // Allocate post_proc_buffer_int if needed.
+  if ((flags & VP9D_MFQE) && !cm->post_proc_buffer_int.buffer_alloc) {
+    if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
+      const int width = ALIGN_POWER_OF_TWO(cm->width, 4);
+      const int height = ALIGN_POWER_OF_TWO(cm->height, 4);
+
+      if (vpx_alloc_frame_buffer(&cm->post_proc_buffer_int, width, height,
+                                 cm->subsampling_x, cm->subsampling_y,
+#if CONFIG_VP9_HIGHBITDEPTH
+                                 cm->use_highbitdepth,
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+                                 VP9_ENC_BORDER_IN_PIXELS,
+                                 cm->byte_alignment) < 0) {
+        vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
+                           "Failed to allocate MFQE framebuffer");
+      }
+
+      // Ensure that postproc is set to all 0s so that post proc
+      // doesn't pull random data in from edge.
+      memset(cm->post_proc_buffer_int.buffer_alloc, 128,
+             cm->post_proc_buffer.frame_size);
+    }
+  }
+
+  if (vpx_realloc_frame_buffer(&cm->post_proc_buffer, cm->width, cm->height,
                                cm->subsampling_x, cm->subsampling_y,
-                               VP9_DEC_BORDER_IN_PIXELS, NULL, NULL, NULL) < 0)
+#if CONFIG_VP9_HIGHBITDEPTH
+                               cm->use_highbitdepth,
+#endif
+                               VP9_DEC_BORDER_IN_PIXELS, cm->byte_alignment,
+                               NULL, NULL, NULL) < 0)
     vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
                        "Failed to allocate post-processing buffer");
-#endif
 
-  if (flags & VP9D_DEMACROBLOCK) {
+  if ((flags & VP9D_MFQE) && cm->current_video_frame >= 2 &&
+      cm->postproc_state.last_frame_valid && cm->bit_depth == 8 &&
+      cm->postproc_state.last_base_qindex <= last_q_thresh &&
+      cm->base_qindex - cm->postproc_state.last_base_qindex >= q_diff_thresh) {
+    vp9_mfqe(cm);
+    // TODO(jackychen): Consider whether enable deblocking by default
+    // if mfqe is enabled. Need to take both the quality and the speed
+    // into consideration.
+    if ((flags & VP9D_DEMACROBLOCK) || (flags & VP9D_DEBLOCK)) {
+      vp8_yv12_copy_frame(ppbuf, &cm->post_proc_buffer_int);
+    }
+    if ((flags & VP9D_DEMACROBLOCK) && cm->post_proc_buffer_int.buffer_alloc) {
+      deblock_and_de_macro_block(&cm->post_proc_buffer_int, ppbuf,
+                                 q + (ppflags->deblocking_level - 5) * 10,
+                                 1, 0);
+    } else if (flags & VP9D_DEBLOCK) {
+      vp9_deblock(&cm->post_proc_buffer_int, ppbuf, q);
+    } else {
+      vp8_yv12_copy_frame(&cm->post_proc_buffer_int, ppbuf);
+    }
+  } else if (flags & VP9D_DEMACROBLOCK) {
     deblock_and_de_macro_block(cm->frame_to_show, ppbuf,
                                q + (ppflags->deblocking_level - 5) * 10, 1, 0);
   } else if (flags & VP9D_DEBLOCK) {
@@ -416,6 +716,9 @@ int vp9_post_proc_frame(struct VP9Common *cm,
     vp8_yv12_copy_frame(cm->frame_to_show, ppbuf);
   }
 
+  cm->postproc_state.last_base_qindex = cm->base_qindex;
+  cm->postproc_state.last_frame_valid = 1;
+
   if (flags & VP9D_ADDNOISE) {
     const int noise_level = ppflags->noise_level;
     if (ppstate->last_q != q ||
@@ -436,6 +739,7 @@ int vp9_post_proc_frame(struct VP9Common *cm,
   dest->uv_width = dest->y_width >> cm->subsampling_x;
   dest->uv_height = dest->y_height >> cm->subsampling_y;
 
+  swap_mi_and_prev_mi(cm);
   return 0;
 }
-#endif
+#endif  // CONFIG_VP9_POSTPROC