Add Neon implementation of aom_highbd_comp_avg_pred

Add Neon implementation of aom_highbd_comp_avg_pred function.

The implementation is mostly a backport of this libvpx change[1].

[1]https://chromium-review.googlesource.com/c/webm/libvpx/+/4239538

Change-Id: Ie60ea9edd7a7742b8948d59e86b17f344d7c0eb5

aom_dsp/aom_dsp.cmake

@@ -293,6 +293,7 @@ if(CONFIG_AV1_ENCODER)
 
     list(APPEND AOM_DSP_ENCODER_INTRIN_NEON
                 "${AOM_ROOT}/aom_dsp/arm/highbd_avg_neon.c"
+                "${AOM_ROOT}/aom_dsp/arm/highbd_avg_pred_neon.c"
                 "${AOM_ROOT}/aom_dsp/arm/highbd_hadamard_neon.c"
                 "${AOM_ROOT}/aom_dsp/arm/highbd_quantize_neon.c"
                 "${AOM_ROOT}/aom_dsp/arm/highbd_sad_neon.c"

aom_dsp/aom_dsp_rtcd_defs.pl

@@ -1725,6 +1725,7 @@ ()
 
   if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
     add_proto qw/void aom_highbd_comp_avg_pred/, "uint8_t *comp_pred8, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride";
+    specialize qw/aom_highbd_comp_avg_pred neon/;
 
     add_proto qw/void aom_highbd_dist_wtd_comp_avg_pred/, "uint8_t *comp_pred8, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param";
     specialize qw/aom_highbd_dist_wtd_comp_avg_pred sse2/;

aom_dsp/arm/highbd_avg_pred_neon.c

@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) 2023 The WebM project authors. All Rights Reserved.
+ * Copyright (c) 2023, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <arm_neon.h>
+#include <assert.h>
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+void aom_highbd_comp_avg_pred_neon(uint8_t *comp_pred8, const uint8_t *pred8,
+                                   int width, int height, const uint8_t *ref8,
+                                   int ref_stride) {
+  const uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
+  const uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+  uint16_t *comp_pred = CONVERT_TO_SHORTPTR(comp_pred8);
+
+  int i = height;
+  if (width > 8) {
+    do {
+      int j = 0;
+      do {
+        const uint16x8_t p = vld1q_u16(pred + j);
+        const uint16x8_t r = vld1q_u16(ref + j);
+
+        uint16x8_t avg = vrhaddq_u16(p, r);
+        vst1q_u16(comp_pred + j, avg);
+
+        j += 8;
+      } while (j < width);
+
+      comp_pred += width;
+      pred += width;
+      ref += ref_stride;
+    } while (--i != 0);
+  } else if (width == 8) {
+    do {
+      const uint16x8_t p = vld1q_u16(pred);
+      const uint16x8_t r = vld1q_u16(ref);
+
+      uint16x8_t avg = vrhaddq_u16(p, r);
+      vst1q_u16(comp_pred, avg);
+
+      comp_pred += width;
+      pred += width;
+      ref += ref_stride;
+    } while (--i != 0);
+  } else {
+    assert(width == 4);
+    do {
+      const uint16x4_t p = vld1_u16(pred);
+      const uint16x4_t r = vld1_u16(ref);
+
+      uint16x4_t avg = vrhadd_u16(p, r);
+      vst1_u16(comp_pred, avg);
+
+      comp_pred += width;
+      pred += width;
+      ref += ref_stride;
+    } while (--i != 0);
+  }
+}

test/comp_mask_pred_test.cc

@@ -828,5 +828,13 @@ TEST_P(AV1HighbdCompAvgPredTest, DISABLED_Speed) {
   RunSpeedTest(GET_PARAM(0), GET_PARAM(1));
 }
 
+#if HAVE_NEON
+INSTANTIATE_TEST_SUITE_P(
+    NEON, AV1HighbdCompAvgPredTest,
+    ::testing::Combine(::testing::Values(&aom_highbd_comp_avg_pred_neon),
+                       ::testing::ValuesIn(kValidBlockSize),
+                       ::testing::Range(8, 13, 2)));
+#endif
+
 #endif  // CONFIG_AV1_HIGHBITDEPTH
 }  // namespace