[NEON] Added vpx_highbd_sad* functions

    Total gain for 12-bit encoding:
    * ~7.8% for best profile
    * ~10% for rt profile

Change-Id: I89eda5c4372a5b628c9df84cdeb4c8486fc44789

diff --git a/test/sad_test.cc b/test/sad_test.cc
index 960bd49..4fb2af6 100644 (file)
--- a/test/sad_test.cc
+++ b/test/sad_test.cc
@@ -727,6 +727,45 @@ const SadMxNParam neon_tests[] = {
   SadMxNParam(8, 4, &vpx_sad8x4_neon),
   SadMxNParam(4, 8, &vpx_sad4x8_neon),
   SadMxNParam(4, 4, &vpx_sad4x4_neon),
+#if CONFIG_VP9_HIGHBITDEPTH
+  SadMxNParam(4, 4, &vpx_highbd_sad4x4_neon, 8),
+  SadMxNParam(4, 8, &vpx_highbd_sad4x8_neon, 8),
+  SadMxNParam(8, 4, &vpx_highbd_sad8x4_neon, 8),
+  SadMxNParam(8, 8, &vpx_highbd_sad8x8_neon, 8),
+  SadMxNParam(8, 16, &vpx_highbd_sad8x16_neon, 8),
+  SadMxNParam(16, 8, &vpx_highbd_sad16x8_neon, 8),
+  SadMxNParam(16, 16, &vpx_highbd_sad16x16_neon, 8),
+  SadMxNParam(16, 32, &vpx_highbd_sad16x32_neon, 8),
+  SadMxNParam(32, 32, &vpx_highbd_sad32x32_neon, 8),
+  SadMxNParam(32, 64, &vpx_highbd_sad32x64_neon, 8),
+  SadMxNParam(64, 32, &vpx_highbd_sad64x32_neon, 8),
+  SadMxNParam(64, 64, &vpx_highbd_sad64x64_neon, 8),
+  SadMxNParam(4, 4, &vpx_highbd_sad4x4_neon, 10),
+  SadMxNParam(4, 8, &vpx_highbd_sad4x8_neon, 10),
+  SadMxNParam(8, 4, &vpx_highbd_sad8x4_neon, 10),
+  SadMxNParam(8, 8, &vpx_highbd_sad8x8_neon, 10),
+  SadMxNParam(8, 16, &vpx_highbd_sad8x16_neon, 10),
+  SadMxNParam(16, 8, &vpx_highbd_sad16x8_neon, 10),
+  SadMxNParam(16, 16, &vpx_highbd_sad16x16_neon, 10),
+  SadMxNParam(16, 32, &vpx_highbd_sad16x32_neon, 10),
+  SadMxNParam(32, 32, &vpx_highbd_sad32x32_neon, 10),
+  SadMxNParam(32, 64, &vpx_highbd_sad32x64_neon, 10),
+  SadMxNParam(64, 32, &vpx_highbd_sad64x32_neon, 10),
+  SadMxNParam(64, 64, &vpx_highbd_sad64x64_neon, 10),
+  SadMxNParam(4, 4, &vpx_highbd_sad4x4_neon, 12),
+  SadMxNParam(4, 8, &vpx_highbd_sad4x8_neon, 12),
+  SadMxNParam(8, 4, &vpx_highbd_sad8x4_neon, 12),
+  SadMxNParam(8, 8, &vpx_highbd_sad8x8_neon, 12),
+  SadMxNParam(8, 16, &vpx_highbd_sad8x16_neon, 12),
+  SadMxNParam(16, 8, &vpx_highbd_sad16x8_neon, 12),
+  SadMxNParam(16, 16, &vpx_highbd_sad16x16_neon, 12),
+  SadMxNParam(16, 32, &vpx_highbd_sad16x32_neon, 12),
+  SadMxNParam(32, 32, &vpx_highbd_sad32x32_neon, 12),
+  SadMxNParam(32, 64, &vpx_highbd_sad32x64_neon, 12),
+  SadMxNParam(64, 32, &vpx_highbd_sad64x32_neon, 12),
+  SadMxNParam(64, 64, &vpx_highbd_sad64x64_neon, 12),
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
 };
 INSTANTIATE_TEST_SUITE_P(NEON, SADTest, ::testing::ValuesIn(neon_tests));
 
@@ -744,6 +783,47 @@ const SadMxNAvgParam avg_neon_tests[] = {
   SadMxNAvgParam(8, 4, &vpx_sad8x4_avg_neon),
   SadMxNAvgParam(4, 8, &vpx_sad4x8_avg_neon),
   SadMxNAvgParam(4, 4, &vpx_sad4x4_avg_neon),
+#if CONFIG_VP9_HIGHBITDEPTH
+  SadMxNAvgParam(4, 4, &vpx_highbd_sad4x4_avg_neon, 8),
+  SadMxNAvgParam(4, 8, &vpx_highbd_sad4x8_avg_neon, 8),
+  SadMxNAvgParam(8, 4, &vpx_highbd_sad8x4_avg_neon, 8),
+  SadMxNAvgParam(8, 8, &vpx_highbd_sad8x8_avg_neon, 8),
+  SadMxNAvgParam(8, 16, &vpx_highbd_sad8x16_avg_neon, 8),
+  SadMxNAvgParam(16, 8, &vpx_highbd_sad16x8_avg_neon, 8),
+  SadMxNAvgParam(16, 16, &vpx_highbd_sad16x16_avg_neon, 8),
+  SadMxNAvgParam(16, 32, &vpx_highbd_sad16x32_avg_neon, 8),
+  SadMxNAvgParam(32, 16, &vpx_highbd_sad32x16_avg_neon, 8),
+  SadMxNAvgParam(32, 32, &vpx_highbd_sad32x32_avg_neon, 8),
+  SadMxNAvgParam(32, 64, &vpx_highbd_sad32x64_avg_neon, 8),
+  SadMxNAvgParam(64, 32, &vpx_highbd_sad64x32_avg_neon, 8),
+  SadMxNAvgParam(64, 64, &vpx_highbd_sad64x64_avg_neon, 8),
+  SadMxNAvgParam(4, 4, &vpx_highbd_sad4x4_avg_neon, 10),
+  SadMxNAvgParam(4, 8, &vpx_highbd_sad4x8_avg_neon, 10),
+  SadMxNAvgParam(8, 4, &vpx_highbd_sad8x4_avg_neon, 10),
+  SadMxNAvgParam(8, 8, &vpx_highbd_sad8x8_avg_neon, 10),
+  SadMxNAvgParam(8, 16, &vpx_highbd_sad8x16_avg_neon, 10),
+  SadMxNAvgParam(16, 8, &vpx_highbd_sad16x8_avg_neon, 10),
+  SadMxNAvgParam(16, 16, &vpx_highbd_sad16x16_avg_neon, 10),
+  SadMxNAvgParam(16, 32, &vpx_highbd_sad16x32_avg_neon, 10),
+  SadMxNAvgParam(32, 16, &vpx_highbd_sad32x16_avg_neon, 10),
+  SadMxNAvgParam(32, 32, &vpx_highbd_sad32x32_avg_neon, 10),
+  SadMxNAvgParam(32, 64, &vpx_highbd_sad32x64_avg_neon, 10),
+  SadMxNAvgParam(64, 32, &vpx_highbd_sad64x32_avg_neon, 10),
+  SadMxNAvgParam(64, 64, &vpx_highbd_sad64x64_avg_neon, 10),
+  SadMxNAvgParam(4, 4, &vpx_highbd_sad4x4_avg_neon, 12),
+  SadMxNAvgParam(4, 8, &vpx_highbd_sad4x8_avg_neon, 12),
+  SadMxNAvgParam(8, 4, &vpx_highbd_sad8x4_avg_neon, 12),
+  SadMxNAvgParam(8, 8, &vpx_highbd_sad8x8_avg_neon, 12),
+  SadMxNAvgParam(8, 16, &vpx_highbd_sad8x16_avg_neon, 12),
+  SadMxNAvgParam(16, 8, &vpx_highbd_sad16x8_avg_neon, 12),
+  SadMxNAvgParam(16, 16, &vpx_highbd_sad16x16_avg_neon, 12),
+  SadMxNAvgParam(16, 32, &vpx_highbd_sad16x32_avg_neon, 12),
+  SadMxNAvgParam(32, 16, &vpx_highbd_sad32x16_avg_neon, 12),
+  SadMxNAvgParam(32, 32, &vpx_highbd_sad32x32_avg_neon, 12),
+  SadMxNAvgParam(32, 64, &vpx_highbd_sad32x64_avg_neon, 12),
+  SadMxNAvgParam(64, 32, &vpx_highbd_sad64x32_avg_neon, 12),
+  SadMxNAvgParam(64, 64, &vpx_highbd_sad64x64_avg_neon, 12),
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 };
 INSTANTIATE_TEST_SUITE_P(NEON, SADavgTest, ::testing::ValuesIn(avg_neon_tests));
 
@@ -761,6 +841,44 @@ const SadMxNx4Param x4d_neon_tests[] = {
   SadMxNx4Param(8, 4, &vpx_sad8x4x4d_neon),
   SadMxNx4Param(4, 8, &vpx_sad4x8x4d_neon),
   SadMxNx4Param(4, 4, &vpx_sad4x4x4d_neon),
+#if CONFIG_VP9_HIGHBITDEPTH
+  SadMxNx4Param(4, 4, &vpx_highbd_sad4x4x4d_neon, 8),
+  SadMxNx4Param(4, 8, &vpx_highbd_sad4x8x4d_neon, 8),
+  SadMxNx4Param(8, 4, &vpx_highbd_sad8x4x4d_neon, 8),
+  SadMxNx4Param(8, 8, &vpx_highbd_sad8x8x4d_neon, 8),
+  SadMxNx4Param(8, 16, &vpx_highbd_sad8x16x4d_neon, 8),
+  SadMxNx4Param(16, 8, &vpx_highbd_sad16x8x4d_neon, 8),
+  SadMxNx4Param(16, 16, &vpx_highbd_sad16x16x4d_neon, 8),
+  SadMxNx4Param(16, 32, &vpx_highbd_sad16x32x4d_neon, 8),
+  SadMxNx4Param(32, 32, &vpx_highbd_sad32x32x4d_neon, 8),
+  SadMxNx4Param(32, 64, &vpx_highbd_sad32x64x4d_neon, 8),
+  SadMxNx4Param(64, 32, &vpx_highbd_sad64x32x4d_neon, 8),
+  SadMxNx4Param(64, 64, &vpx_highbd_sad64x64x4d_neon, 8),
+  SadMxNx4Param(4, 4, &vpx_highbd_sad4x4x4d_neon, 10),
+  SadMxNx4Param(4, 8, &vpx_highbd_sad4x8x4d_neon, 10),
+  SadMxNx4Param(8, 4, &vpx_highbd_sad8x4x4d_neon, 10),
+  SadMxNx4Param(8, 8, &vpx_highbd_sad8x8x4d_neon, 10),
+  SadMxNx4Param(8, 16, &vpx_highbd_sad8x16x4d_neon, 10),
+  SadMxNx4Param(16, 8, &vpx_highbd_sad16x8x4d_neon, 10),
+  SadMxNx4Param(16, 16, &vpx_highbd_sad16x16x4d_neon, 10),
+  SadMxNx4Param(16, 32, &vpx_highbd_sad16x32x4d_neon, 10),
+  SadMxNx4Param(32, 32, &vpx_highbd_sad32x32x4d_neon, 10),
+  SadMxNx4Param(32, 64, &vpx_highbd_sad32x64x4d_neon, 10),
+  SadMxNx4Param(64, 32, &vpx_highbd_sad64x32x4d_neon, 10),
+  SadMxNx4Param(64, 64, &vpx_highbd_sad64x64x4d_neon, 10),
+  SadMxNx4Param(4, 4, &vpx_highbd_sad4x4x4d_neon, 12),
+  SadMxNx4Param(4, 8, &vpx_highbd_sad4x8x4d_neon, 12),
+  SadMxNx4Param(8, 4, &vpx_highbd_sad8x4x4d_neon, 12),
+  SadMxNx4Param(8, 8, &vpx_highbd_sad8x8x4d_neon, 12),
+  SadMxNx4Param(8, 16, &vpx_highbd_sad8x16x4d_neon, 12),
+  SadMxNx4Param(16, 8, &vpx_highbd_sad16x8x4d_neon, 12),
+  SadMxNx4Param(16, 16, &vpx_highbd_sad16x16x4d_neon, 12),
+  SadMxNx4Param(16, 32, &vpx_highbd_sad16x32x4d_neon, 12),
+  SadMxNx4Param(32, 32, &vpx_highbd_sad32x32x4d_neon, 12),
+  SadMxNx4Param(32, 64, &vpx_highbd_sad32x64x4d_neon, 12),
+  SadMxNx4Param(64, 32, &vpx_highbd_sad64x32x4d_neon, 12),
+  SadMxNx4Param(64, 64, &vpx_highbd_sad64x64x4d_neon, 12),
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 };
 INSTANTIATE_TEST_SUITE_P(NEON, SADx4Test, ::testing::ValuesIn(x4d_neon_tests));
 #endif  // HAVE_NEON

diff --git a/vpx_dsp/arm/highbd_sad_neon.c b/vpx_dsp/arm/highbd_sad_neon.c
new file mode 100644 (file)
index 0000000..ecb52ce
--- /dev/null
+++ b/vpx_dsp/arm/highbd_sad_neon.c
@@ -0,0 +1,225 @@
+/*
+ *  Copyright (c) 2022 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+
+#include "vpx/vpx_integer.h"
+#include "vpx_dsp/arm/mem_neon.h"
+#include "vpx_dsp/arm/sum_neon.h"
+
+static VPX_FORCE_INLINE uint32_t highbd_sad4_neon(const uint8_t *src_ptr,
+                                                  int src_stride,
+                                                  const uint8_t *ref_ptr,
+                                                  int ref_stride, int width,
+                                                  int height) {
+  int i, j;
+  uint32x4_t sum_abs_diff = vdupq_n_u32(0);
+  const uint16_t *src16_ptr = CONVERT_TO_SHORTPTR(src_ptr);
+  const uint16_t *ref16_ptr = CONVERT_TO_SHORTPTR(ref_ptr);
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j += 4) {
+      const uint16x4_t src_u16 = vld1_u16(src16_ptr + j);
+      const uint16x4_t ref_u16 = vld1_u16(ref16_ptr + j);
+      sum_abs_diff = vabal_u16(sum_abs_diff, src_u16, ref_u16);
+    }
+    src16_ptr += src_stride;
+    ref16_ptr += ref_stride;
+  }
+
+  return horizontal_add_uint32x4(sum_abs_diff);
+}
+
+static VPX_FORCE_INLINE uint32_t highbd_sad8_neon(const uint8_t *src_ptr,
+                                                  int src_stride,
+                                                  const uint8_t *ref_ptr,
+                                                  int ref_stride, int width,
+                                                  int height) {
+  int i, j;
+  uint32x4_t sum_abs_diff = vdupq_n_u32(0);
+  const uint16_t *src16_ptr = CONVERT_TO_SHORTPTR(src_ptr);
+  const uint16_t *ref16_ptr = CONVERT_TO_SHORTPTR(ref_ptr);
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j += 8) {
+      const uint16x8_t src_u16 = vld1q_u16(src16_ptr + j);
+      const uint16x8_t ref_u16 = vld1q_u16(ref16_ptr + j);
+      sum_abs_diff =
+          vabal_u16(sum_abs_diff, vget_low_u16(src_u16), vget_low_u16(ref_u16));
+      sum_abs_diff = vabal_u16(sum_abs_diff, vget_high_u16(src_u16),
+                               vget_high_u16(ref_u16));
+    }
+    src16_ptr += src_stride;
+    ref16_ptr += ref_stride;
+  }
+
+  return horizontal_add_uint32x4(sum_abs_diff);
+}
+
+static VPX_FORCE_INLINE uint32_t highbd_sad4_avg_neon(
+    const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,
+    int ref_stride, const uint8_t *second_pred, int width, int height) {
+  int i, j;
+  uint32x4_t sum_abs_diff = vdupq_n_u32(0);
+  const uint16_t *src16_ptr = CONVERT_TO_SHORTPTR(src_ptr);
+  const uint16_t *ref16_ptr = CONVERT_TO_SHORTPTR(ref_ptr);
+  const uint16_t *pred_ptr = CONVERT_TO_SHORTPTR(second_pred);
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j += 4) {
+      const uint16x4_t a_u16 = vld1_u16(src16_ptr + j);
+      const uint16x4_t b_u16 = vld1_u16(ref16_ptr + j);
+      const uint16x4_t c_u16 = vld1_u16(pred_ptr + j);
+      const uint16x4_t avg = vrhadd_u16(b_u16, c_u16);
+      sum_abs_diff = vabal_u16(sum_abs_diff, a_u16, avg);
+    }
+    src16_ptr += src_stride;
+    ref16_ptr += ref_stride;
+    pred_ptr += width;
+  }
+
+  return horizontal_add_uint32x4(sum_abs_diff);
+}
+
+static VPX_FORCE_INLINE uint32_t highbd_sad8_avg_neon(
+    const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,
+    int ref_stride, const uint8_t *second_pred, int width, int height) {
+  int i, j;
+  uint32x4_t sum_abs_diff = vdupq_n_u32(0);
+  const uint16_t *src16_ptr = CONVERT_TO_SHORTPTR(src_ptr);
+  const uint16_t *ref16_ptr = CONVERT_TO_SHORTPTR(ref_ptr);
+  const uint16_t *pred_ptr = CONVERT_TO_SHORTPTR(second_pred);
+  for (i = 0; i < height; i++) {
+    for (j = 0; j < width; j += 8) {
+      const uint16x8_t a_u16 = vld1q_u16(src16_ptr + j);
+      const uint16x8_t b_u16 = vld1q_u16(ref16_ptr + j);
+      const uint16x8_t c_u16 = vld1q_u16(pred_ptr + j);
+      const uint16x8_t avg = vrhaddq_u16(b_u16, c_u16);
+      sum_abs_diff =
+          vabal_u16(sum_abs_diff, vget_low_u16(a_u16), vget_low_u16(avg));
+      sum_abs_diff =
+          vabal_u16(sum_abs_diff, vget_high_u16(a_u16), vget_high_u16(avg));
+    }
+    src16_ptr += src_stride;
+    ref16_ptr += ref_stride;
+    pred_ptr += width;
+  }
+
+  return horizontal_add_uint32x4(sum_abs_diff);
+}
+
+#define highbd_sad4MxN(m, n)                                                 \
+  unsigned int vpx_highbd_sad##m##x##n##_neon(                               \
+      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,        \
+      int ref_stride) {                                                      \
+    return highbd_sad4_neon(src_ptr, src_stride, ref_ptr, ref_stride, m, n); \
+  }
+
+#define highbd_sadMxN(m, n)                                                  \
+  unsigned int vpx_highbd_sad##m##x##n##_neon(                               \
+      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,        \
+      int ref_stride) {                                                      \
+    return highbd_sad8_neon(src_ptr, src_stride, ref_ptr, ref_stride, m, n); \
+  }
+
+#define highbd_sad4MxN_avg(m, n)                                          \
+  unsigned int vpx_highbd_sad##m##x##n##_avg_neon(                        \
+      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,     \
+      int ref_stride, const uint8_t *second_pred) {                       \
+    return highbd_sad4_avg_neon(src_ptr, src_stride, ref_ptr, ref_stride, \
+                                second_pred, m, n);                       \
+  }
+
+#define highbd_sadMxN_avg(m, n)                                           \
+  unsigned int vpx_highbd_sad##m##x##n##_avg_neon(                        \
+      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr,     \
+      int ref_stride, const uint8_t *second_pred) {                       \
+    return highbd_sad8_avg_neon(src_ptr, src_stride, ref_ptr, ref_stride, \
+                                second_pred, m, n);                       \
+  }
+
+#define highbd_sadMxNx4D(m, n)                                                 \
+  void vpx_highbd_sad##m##x##n##x4d_neon(                                      \
+      const uint8_t *src_ptr, int src_stride,                                  \
+      const uint8_t *const ref_array[4], int ref_stride,                       \
+      uint32_t sad_array[4]) {                                                 \
+    int i;                                                                     \
+    for (i = 0; i < 4; ++i) {                                                  \
+      sad_array[i] = vpx_highbd_sad##m##x##n##_neon(src_ptr, src_stride,       \
+                                                    ref_array[i], ref_stride); \
+    }                                                                          \
+  }
+
+/* clang-format off */
+// 4x4
+highbd_sad4MxN(4, 4)
+highbd_sad4MxN_avg(4, 4)
+highbd_sadMxNx4D(4, 4)
+
+// 4x8
+highbd_sad4MxN(4, 8)
+highbd_sad4MxN_avg(4, 8)
+highbd_sadMxNx4D(4, 8)
+
+// 8x4
+highbd_sadMxN(8, 4)
+highbd_sadMxN_avg(8, 4)
+highbd_sadMxNx4D(8, 4)
+
+// 8x8
+highbd_sadMxN(8, 8)
+highbd_sadMxN_avg(8, 8)
+highbd_sadMxNx4D(8, 8)
+
+// 8x16
+highbd_sadMxN(8, 16)
+highbd_sadMxN_avg(8, 16)
+highbd_sadMxNx4D(8, 16)
+
+// 16x8
+highbd_sadMxN(16, 8)
+highbd_sadMxN_avg(16, 8)
+highbd_sadMxNx4D(16, 8)
+
+// 16x16
+highbd_sadMxN(16, 16)
+highbd_sadMxN_avg(16, 16)
+highbd_sadMxNx4D(16, 16)
+
+// 16x32
+highbd_sadMxN(16, 32)
+highbd_sadMxN_avg(16, 32)
+highbd_sadMxNx4D(16, 32)
+
+// 32x16
+highbd_sadMxN(32, 16)
+highbd_sadMxN_avg(32, 16)
+highbd_sadMxNx4D(32, 16)
+
+// 32x32
+highbd_sadMxN(32, 32)
+highbd_sadMxN_avg(32, 32)
+highbd_sadMxNx4D(32, 32)
+
+// 32x64
+highbd_sadMxN(32, 64)
+highbd_sadMxN_avg(32, 64)
+highbd_sadMxNx4D(32, 64)
+
+// 64x32
+highbd_sadMxN(64, 32)
+highbd_sadMxN_avg(64, 32)
+highbd_sadMxNx4D(64, 32)
+
+// 64x64
+highbd_sadMxN(64, 64)
+highbd_sadMxN_avg(64, 64)
+highbd_sadMxNx4D(64, 64)
+    /* clang-format on */

diff --git a/vpx_dsp/vpx_dsp.mk b/vpx_dsp/vpx_dsp.mk
index 4f17425..1a03aed 100644 (file)
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -392,6 +392,7 @@ DSP_SRCS-$(HAVE_LSX)    += loongarch/subtract_lsx.c
 ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 DSP_SRCS-$(HAVE_SSE2) += x86/highbd_sad4d_sse2.asm
 DSP_SRCS-$(HAVE_SSE2) += x86/highbd_sad_sse2.asm
+DSP_SRCS-$(HAVE_NEON) += arm/highbd_sad_neon.c
 endif  # CONFIG_VP9_HIGHBITDEPTH
 
 endif  # CONFIG_ENCODERS

diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl
index cab74f9..72442ed 100644 (file)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -940,41 +940,43 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   # Single block SAD
   #
   add_proto qw/unsigned int vpx_highbd_sad64x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad64x64 sse2/;
+  specialize qw/vpx_highbd_sad64x64 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad64x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad64x32 sse2/;
+  specialize qw/vpx_highbd_sad64x32 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad32x64 sse2/;
+  specialize qw/vpx_highbd_sad32x64 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad32x32 sse2/;
+  specialize qw/vpx_highbd_sad32x32 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad32x16 sse2/;
+  specialize qw/vpx_highbd_sad32x16 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad16x32 sse2/;
+  specialize qw/vpx_highbd_sad16x32 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad16x16 sse2/;
+  specialize qw/vpx_highbd_sad16x16 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad16x8 sse2/;
+  specialize qw/vpx_highbd_sad16x8 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad8x16 sse2/;
+  specialize qw/vpx_highbd_sad8x16 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad8x8 sse2/;
+  specialize qw/vpx_highbd_sad8x8 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
-  specialize qw/vpx_highbd_sad8x4 sse2/;
+  specialize qw/vpx_highbd_sad8x4 sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vpx_highbd_sad4x8 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
+  specialize qw/vpx_highbd_sad4x4 neon/;
 
   #
   # Avg
@@ -988,83 +990,85 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   add_proto qw/void vpx_highbd_minmax_8x8/, "const uint8_t *s8, int p, const uint8_t *d8, int dp, int *min, int *max";
 
   add_proto qw/unsigned int vpx_highbd_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad64x64_avg sse2/;
+  specialize qw/vpx_highbd_sad64x64_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad64x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad64x32_avg sse2/;
+  specialize qw/vpx_highbd_sad64x32_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad32x64_avg sse2/;
+  specialize qw/vpx_highbd_sad32x64_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad32x32_avg sse2/;
+  specialize qw/vpx_highbd_sad32x32_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad32x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad32x16_avg sse2/;
+  specialize qw/vpx_highbd_sad32x16_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad16x32_avg sse2/;
+  specialize qw/vpx_highbd_sad16x32_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad16x16_avg sse2/;
+  specialize qw/vpx_highbd_sad16x16_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad16x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad16x8_avg sse2/;
+  specialize qw/vpx_highbd_sad16x8_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad8x16_avg sse2/;
+  specialize qw/vpx_highbd_sad8x16_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad8x8_avg sse2/;
+  specialize qw/vpx_highbd_sad8x8_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad8x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
-  specialize qw/vpx_highbd_sad8x4_avg sse2/;
+  specialize qw/vpx_highbd_sad8x4_avg sse2 neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad4x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vpx_highbd_sad4x8_avg neon/;
 
   add_proto qw/unsigned int vpx_highbd_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
+  specialize qw/vpx_highbd_sad4x4_avg neon/;
 
   #
   # Multi-block SAD, comparing a reference to N independent blocks
   #
   add_proto qw/void vpx_highbd_sad64x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad64x64x4d sse2/;
+  specialize qw/vpx_highbd_sad64x64x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad64x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad64x32x4d sse2/;
+  specialize qw/vpx_highbd_sad64x32x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad32x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad32x64x4d sse2/;
+  specialize qw/vpx_highbd_sad32x64x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad32x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad32x32x4d sse2/;
+  specialize qw/vpx_highbd_sad32x32x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad32x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad32x16x4d sse2/;
+  specialize qw/vpx_highbd_sad32x16x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad16x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad16x32x4d sse2/;
+  specialize qw/vpx_highbd_sad16x32x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad16x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad16x16x4d sse2/;
+  specialize qw/vpx_highbd_sad16x16x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad16x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad16x8x4d sse2/;
+  specialize qw/vpx_highbd_sad16x8x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad8x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad8x16x4d sse2/;
+  specialize qw/vpx_highbd_sad8x16x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad8x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad8x8x4d sse2/;
+  specialize qw/vpx_highbd_sad8x8x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad8x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad8x4x4d sse2/;
+  specialize qw/vpx_highbd_sad8x4x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad4x8x4d sse2/;
+  specialize qw/vpx_highbd_sad4x8x4d sse2 neon/;
 
   add_proto qw/void vpx_highbd_sad4x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_array[4], int ref_stride, uint32_t sad_array[4]";
-  specialize qw/vpx_highbd_sad4x4x4d sse2/;
+  specialize qw/vpx_highbd_sad4x4x4d sse2 neon/;
 
   #
   # Structured Similarity (SSIM)