Update aom to v1.0.0

Update aom to commit id d14c5bb4f336ef1842046089849dee4a301fbbf0.

1 files changed, 231 insertions, 78 deletions

diff --git a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
index 62acf3ed3..fdfadc886 100644
--- a/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
+++ b/third_party/aom/aom_dsp/x86/highbd_variance_sse2.c
@@ -12,13 +12,17 @@
 #include <assert.h>
 #include <emmintrin.h>  // SSE2
 
-#include "./aom_config.h"
-#include "./aom_dsp_rtcd.h"
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_dsp/x86/synonyms.h"
 
 #include "aom_ports/mem.h"
 
-#include "./av1_rtcd.h"
 #include "av1/common/filter.h"
+#include "av1/common/onyxc_int.h"
+#include "av1/common/reconinter.h"
 
 typedef uint32_t (*high_variance_fn_t)(const uint16_t *src, int src_stride,
                                        const uint16_t *ref, int ref_stride,
@@ -185,13 +189,11 @@ VAR_FN(16, 16, 16, 8);
 VAR_FN(16, 8, 8, 7);
 VAR_FN(8, 16, 8, 7);
 VAR_FN(8, 8, 8, 6);
-#if CONFIG_EXT_PARTITION_TYPES
 VAR_FN(16, 4, 16, 6);
 VAR_FN(8, 32, 8, 8);
-VAR_FN(32, 8, 16, 8);
+VAR_FN(32, 8, 8, 8);
 VAR_FN(16, 64, 16, 10);
 VAR_FN(64, 16, 16, 10);
-#endif
 
 #undef VAR_FN
 
@@ -398,7 +400,6 @@ DECLS(sse2);
     return (var >= 0) ? (uint32_t)var : 0;                                     \
   }
 
-#if CONFIG_EXT_PARTITION_TYPES
 #define FNS(opt)                        \
   FN(64, 64, 16, 6, 6, opt, (int64_t)); \
   FN(64, 32, 16, 6, 5, opt, (int64_t)); \
@@ -416,20 +417,6 @@ DECLS(sse2);
   FN(32, 8, 16, 5, 3, opt, (int64_t));  \
   FN(16, 64, 16, 4, 6, opt, (int64_t)); \
   FN(64, 16, 16, 6, 4, opt, (int64_t))
-#else
-#define FNS(opt)                        \
-  FN(64, 64, 16, 6, 6, opt, (int64_t)); \
-  FN(64, 32, 16, 6, 5, opt, (int64_t)); \
-  FN(32, 64, 16, 5, 6, opt, (int64_t)); \
-  FN(32, 32, 16, 5, 5, opt, (int64_t)); \
-  FN(32, 16, 16, 5, 4, opt, (int64_t)); \
-  FN(16, 32, 16, 4, 5, opt, (int64_t)); \
-  FN(16, 16, 16, 4, 4, opt, (int64_t)); \
-  FN(16, 8, 16, 4, 3, opt, (int64_t));  \
-  FN(8, 16, 8, 3, 4, opt, (int64_t));   \
-  FN(8, 8, 8, 3, 3, opt, (int64_t));    \
-  FN(8, 4, 8, 3, 2, opt, (int64_t))
-#endif
 
 FNS(sse2);
 
@@ -577,7 +564,6 @@ DECLS(sse2);
     return (var >= 0) ? (uint32_t)var : 0;                                     \
   }
 
-#if CONFIG_EXT_PARTITION_TYPES
 #define FNS(opt)                        \
   FN(64, 64, 16, 6, 6, opt, (int64_t)); \
   FN(64, 32, 16, 6, 5, opt, (int64_t)); \
@@ -595,30 +581,104 @@ DECLS(sse2);
   FN(32, 8, 16, 5, 3, opt, (int64_t));  \
   FN(16, 64, 16, 4, 6, opt, (int64_t)); \
   FN(64, 16, 16, 6, 4, opt, (int64_t));
-#else
-#define FNS(opt)                        \
-  FN(64, 64, 16, 6, 6, opt, (int64_t)); \
-  FN(64, 32, 16, 6, 5, opt, (int64_t)); \
-  FN(32, 64, 16, 5, 6, opt, (int64_t)); \
-  FN(32, 32, 16, 5, 5, opt, (int64_t)); \
-  FN(32, 16, 16, 5, 4, opt, (int64_t)); \
-  FN(16, 32, 16, 4, 5, opt, (int64_t)); \
-  FN(16, 16, 16, 4, 4, opt, (int64_t)); \
-  FN(16, 8, 16, 4, 3, opt, (int64_t));  \
-  FN(8, 16, 8, 3, 4, opt, (int64_t));   \
-  FN(8, 8, 8, 3, 3, opt, (int64_t));    \
-  FN(8, 4, 8, 3, 2, opt, (int64_t));
-#endif
 
 FNS(sse2);
 
 #undef FNS
 #undef FN
 
-void aom_highbd_upsampled_pred_sse2(uint16_t *comp_pred, int width, int height,
+void aom_highbd_upsampled_pred_sse2(MACROBLOCKD *xd,
+                                    const struct AV1Common *const cm,
+                                    int mi_row, int mi_col, const MV *const mv,
+                                    uint16_t *comp_pred, int width, int height,
                                     int subpel_x_q3, int subpel_y_q3,
                                     const uint8_t *ref8, int ref_stride,
                                     int bd) {
+  // expect xd == NULL only in tests
+  if (xd != NULL) {
+    const MB_MODE_INFO *mi = xd->mi[0];
+    const int ref_num = 0;
+    const int is_intrabc = is_intrabc_block(mi);
+    const struct scale_factors *const sf =
+        is_intrabc ? &cm->sf_identity : &xd->block_refs[ref_num]->sf;
+    const int is_scaled = av1_is_scaled(sf);
+
+    if (is_scaled) {
+      // Note: This is mostly a copy from the >=8X8 case in
+      // build_inter_predictors() function, with some small tweaks.
+      uint8_t *comp_pred8 = CONVERT_TO_BYTEPTR(comp_pred);
+
+      // Some assumptions.
+      const int plane = 0;
+
+      // Get pre-requisites.
+      const struct macroblockd_plane *const pd = &xd->plane[plane];
+      const int ssx = pd->subsampling_x;
+      const int ssy = pd->subsampling_y;
+      assert(ssx == 0 && ssy == 0);
+      const struct buf_2d *const dst_buf = &pd->dst;
+      const struct buf_2d *const pre_buf =
+          is_intrabc ? dst_buf : &pd->pre[ref_num];
+      const int mi_x = mi_col * MI_SIZE;
+      const int mi_y = mi_row * MI_SIZE;
+
+      // Calculate subpel_x/y and x/y_step.
+      const int row_start = 0;  // Because ss_y is 0.
+      const int col_start = 0;  // Because ss_x is 0.
+      const int pre_x = (mi_x + MI_SIZE * col_start) >> ssx;
+      const int pre_y = (mi_y + MI_SIZE * row_start) >> ssy;
+      int orig_pos_y = pre_y << SUBPEL_BITS;
+      orig_pos_y += mv->row * (1 << (1 - ssy));
+      int orig_pos_x = pre_x << SUBPEL_BITS;
+      orig_pos_x += mv->col * (1 << (1 - ssx));
+      int pos_y = sf->scale_value_y(orig_pos_y, sf);
+      int pos_x = sf->scale_value_x(orig_pos_x, sf);
+      pos_x += SCALE_EXTRA_OFF;
+      pos_y += SCALE_EXTRA_OFF;
+
+      const int top = -AOM_LEFT_TOP_MARGIN_SCALED(ssy);
+      const int left = -AOM_LEFT_TOP_MARGIN_SCALED(ssx);
+      const int bottom = (pre_buf->height + AOM_INTERP_EXTEND)
+                         << SCALE_SUBPEL_BITS;
+      const int right = (pre_buf->width + AOM_INTERP_EXTEND)
+                        << SCALE_SUBPEL_BITS;
+      pos_y = clamp(pos_y, top, bottom);
+      pos_x = clamp(pos_x, left, right);
+
+      const uint8_t *const pre =
+          pre_buf->buf0 + (pos_y >> SCALE_SUBPEL_BITS) * pre_buf->stride +
+          (pos_x >> SCALE_SUBPEL_BITS);
+      const SubpelParams subpel_params = { sf->x_step_q4, sf->y_step_q4,
+                                           pos_x & SCALE_SUBPEL_MASK,
+                                           pos_y & SCALE_SUBPEL_MASK };
+
+      // Get warp types.
+      const WarpedMotionParams *const wm =
+          &xd->global_motion[mi->ref_frame[ref_num]];
+      const int is_global = is_global_mv_block(mi, wm->wmtype);
+      WarpTypesAllowed warp_types;
+      warp_types.global_warp_allowed = is_global;
+      warp_types.local_warp_allowed = mi->motion_mode == WARPED_CAUSAL;
+
+      // Get convolve parameters.
+      ConvolveParams conv_params = get_conv_params(ref_num, 0, plane, xd->bd);
+      const InterpFilters filters =
+          av1_broadcast_interp_filter(EIGHTTAP_REGULAR);
+
+      // Get the inter predictor.
+      const int build_for_obmc = 0;
+      av1_make_inter_predictor(pre, pre_buf->stride, comp_pred8, width,
+                               &subpel_params, sf, width, height, &conv_params,
+                               filters, &warp_types, mi_x >> pd->subsampling_x,
+                               mi_y >> pd->subsampling_y, plane, ref_num, mi,
+                               build_for_obmc, xd, cm->allow_warped_motion);
+      return;
+    }
+  }
+
+  const InterpFilterParams filter =
+      av1_get_interp_filter_params_with_block_size(EIGHTTAP_REGULAR, 8);
+
   if (!subpel_x_q3 && !subpel_y_q3) {
     uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
     if (width >= 8) {
@@ -648,54 +708,48 @@ void aom_highbd_upsampled_pred_sse2(uint16_t *comp_pred, int width, int height,
         ref += 2 * ref_stride;
       }
     }
+  } else if (!subpel_y_q3) {
+    const int16_t *const kernel =
+        av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+    aom_highbd_convolve8_horiz(ref8, ref_stride, CONVERT_TO_BYTEPTR(comp_pred),
+                               width, kernel, 16, NULL, -1, width, height, bd);
+  } else if (!subpel_x_q3) {
+    const int16_t *const kernel =
+        av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+    aom_highbd_convolve8_vert(ref8, ref_stride, CONVERT_TO_BYTEPTR(comp_pred),
+                              width, NULL, -1, kernel, 16, width, height, bd);
   } else {
-    InterpFilterParams filter;
-    filter = av1_get_interp_filter_params(EIGHTTAP_REGULAR);
-    if (!subpel_y_q3) {
-      const int16_t *kernel;
-      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
-      aom_highbd_convolve8_horiz(ref8, ref_stride,
-                                 CONVERT_TO_BYTEPTR(comp_pred), width, kernel,
-                                 16, NULL, -1, width, height, bd);
-    } else if (!subpel_x_q3) {
-      const int16_t *kernel;
-      kernel = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
-      aom_highbd_convolve8_vert(ref8, ref_stride, CONVERT_TO_BYTEPTR(comp_pred),
-                                width, NULL, -1, kernel, 16, width, height, bd);
-    } else {
-      DECLARE_ALIGNED(16, uint16_t,
-                      temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
-      const int16_t *kernel_x;
-      const int16_t *kernel_y;
-      int intermediate_height;
-      kernel_x = av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
-      kernel_y = av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
-      intermediate_height =
-          (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
-      assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
-      aom_highbd_convolve8_horiz(ref8 - ref_stride * ((filter.taps >> 1) - 1),
-                                 ref_stride, CONVERT_TO_BYTEPTR(temp),
-                                 MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
-                                 intermediate_height, bd);
-      aom_highbd_convolve8_vert(
-          CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1)),
-          MAX_SB_SIZE, CONVERT_TO_BYTEPTR(comp_pred), width, NULL, -1, kernel_y,
-          16, width, height, bd);
-    }
+    DECLARE_ALIGNED(16, uint16_t,
+                    temp[((MAX_SB_SIZE + 16) + 16) * MAX_SB_SIZE]);
+    const int16_t *const kernel_x =
+        av1_get_interp_filter_subpel_kernel(filter, subpel_x_q3 << 1);
+    const int16_t *const kernel_y =
+        av1_get_interp_filter_subpel_kernel(filter, subpel_y_q3 << 1);
+    const int intermediate_height =
+        (((height - 1) * 8 + subpel_y_q3) >> 3) + filter.taps;
+    assert(intermediate_height <= (MAX_SB_SIZE * 2 + 16) + 16);
+    aom_highbd_convolve8_horiz(ref8 - ref_stride * ((filter.taps >> 1) - 1),
+                               ref_stride, CONVERT_TO_BYTEPTR(temp),
+                               MAX_SB_SIZE, kernel_x, 16, NULL, -1, width,
+                               intermediate_height, bd);
+    aom_highbd_convolve8_vert(
+        CONVERT_TO_BYTEPTR(temp + MAX_SB_SIZE * ((filter.taps >> 1) - 1)),
+        MAX_SB_SIZE, CONVERT_TO_BYTEPTR(comp_pred), width, NULL, -1, kernel_y,
+        16, width, height, bd);
   }
 }
 
-void aom_highbd_comp_avg_upsampled_pred_sse2(uint16_t *comp_pred,
-                                             const uint8_t *pred8, int width,
-                                             int height, int subpel_x_q3,
-                                             int subpel_y_q3,
-                                             const uint8_t *ref8,
-                                             int ref_stride, int bd) {
+void aom_highbd_comp_avg_upsampled_pred_sse2(
+    MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+    const MV *const mv, uint16_t *comp_pred, const uint8_t *pred8, int width,
+    int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
+    int ref_stride, int bd) {
   uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
   int n;
   int i;
-  aom_highbd_upsampled_pred(comp_pred, width, height, subpel_x_q3, subpel_y_q3,
-                            ref8, ref_stride, bd);
+  aom_highbd_upsampled_pred(xd, cm, mi_row, mi_col, mv, comp_pred, width,
+                            height, subpel_x_q3, subpel_y_q3, ref8, ref_stride,
+                            bd);
   /*The total number of pixels must be a multiple of 8 (e.g., 4x4).*/
   assert(!(width * height & 7));
   n = width * height >> 3;
@@ -707,3 +761,102 @@ void aom_highbd_comp_avg_upsampled_pred_sse2(uint16_t *comp_pred,
     pred += 8;
   }
 }
+
+static INLINE void highbd_compute_jnt_comp_avg(__m128i *p0, __m128i *p1,
+                                               const __m128i *w0,
+                                               const __m128i *w1,
+                                               const __m128i *r,
+                                               void *const result) {
+  assert(DIST_PRECISION_BITS <= 4);
+  __m128i mult0 = _mm_mullo_epi16(*p0, *w0);
+  __m128i mult1 = _mm_mullo_epi16(*p1, *w1);
+  __m128i sum = _mm_adds_epu16(mult0, mult1);
+  __m128i round = _mm_adds_epu16(sum, *r);
+  __m128i shift = _mm_srli_epi16(round, DIST_PRECISION_BITS);
+
+  xx_storeu_128(result, shift);
+}
+
+void aom_highbd_jnt_comp_avg_pred_sse2(uint16_t *comp_pred,
+                                       const uint8_t *pred8, int width,
+                                       int height, const uint8_t *ref8,
+                                       int ref_stride,
+                                       const JNT_COMP_PARAMS *jcp_param) {
+  int i;
+  const uint16_t wt0 = (uint16_t)jcp_param->fwd_offset;
+  const uint16_t wt1 = (uint16_t)jcp_param->bck_offset;
+  const __m128i w0 = _mm_set_epi16(wt0, wt0, wt0, wt0, wt0, wt0, wt0, wt0);
+  const __m128i w1 = _mm_set_epi16(wt1, wt1, wt1, wt1, wt1, wt1, wt1, wt1);
+  const uint16_t round = ((1 << DIST_PRECISION_BITS) >> 1);
+  const __m128i r =
+      _mm_set_epi16(round, round, round, round, round, round, round, round);
+  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+
+  if (width >= 8) {
+    // Read 8 pixels one row at a time
+    assert(!(width & 7));
+    for (i = 0; i < height; ++i) {
+      int j;
+      for (j = 0; j < width; j += 8) {
+        __m128i p0 = xx_loadu_128(ref);
+        __m128i p1 = xx_loadu_128(pred);
+
+        highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
+
+        comp_pred += 8;
+        pred += 8;
+        ref += 8;
+      }
+      ref += ref_stride - width;
+    }
+  } else {
+    // Read 4 pixels two rows at a time
+    assert(!(width & 3));
+    for (i = 0; i < height; i += 2) {
+      __m128i p0_0 = xx_loadl_64(ref + 0 * ref_stride);
+      __m128i p0_1 = xx_loadl_64(ref + 1 * ref_stride);
+      __m128i p0 = _mm_unpacklo_epi64(p0_0, p0_1);
+      __m128i p1 = xx_loadu_128(pred);
+
+      highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
+
+      comp_pred += 8;
+      pred += 8;
+      ref += 2 * ref_stride;
+    }
+  }
+}
+
+void aom_highbd_jnt_comp_avg_upsampled_pred_sse2(
+    MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+    const MV *const mv, uint16_t *comp_pred, const uint8_t *pred8, int width,
+    int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
+    int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param) {
+  uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  int n;
+  int i;
+  aom_highbd_upsampled_pred(xd, cm, mi_row, mi_col, mv, comp_pred, width,
+                            height, subpel_x_q3, subpel_y_q3, ref8, ref_stride,
+                            bd);
+  assert(!(width * height & 7));
+  n = width * height >> 3;
+
+  const uint16_t wt0 = (uint16_t)jcp_param->fwd_offset;
+  const uint16_t wt1 = (uint16_t)jcp_param->bck_offset;
+  const __m128i w0 = _mm_set_epi16(wt0, wt0, wt0, wt0, wt0, wt0, wt0, wt0);
+  const __m128i w1 = _mm_set_epi16(wt1, wt1, wt1, wt1, wt1, wt1, wt1, wt1);
+  const uint16_t round = ((1 << DIST_PRECISION_BITS) >> 1);
+  const __m128i r =
+      _mm_set_epi16(round, round, round, round, round, round, round, round);
+
+  for (i = 0; i < n; i++) {
+    __m128i p0 = xx_loadu_128(comp_pred);
+    __m128i p1 = xx_loadu_128(pred);
+
+    highbd_compute_jnt_comp_avg(&p0, &p1, &w0, &w1, &r, comp_pred);
+
+    comp_pred += 8;
+    pred += 8;
+  }
+}