ref: 3b74066b10e5af032918fadc0add83370bec1449
parent: 1b275ab898c5ab891f8364041b66152d81b6ae4f
author: Linfeng Zhang <[email protected]>
date: Tue Oct 25 10:48:32 EDT 2016
Add high bitdepth intra prediction NEON optimization (mode dc) BUG=webm:1316 Change-Id: I984d6004ea2445e86f213fb6fa4d794a9955af8f
--- a/test/test_intra_pred_speed.cc
+++ b/test/test_intra_pred_speed.cc
@@ -475,6 +475,33 @@
vpx_highbd_tm_predictor_32x32_sse2)
#endif // HAVE_SSE2
+#if HAVE_NEON
+HIGHBD_INTRA_PRED_TEST(NEON, TestHighbdIntraPred4,
+ vpx_highbd_dc_predictor_4x4_neon,
+ vpx_highbd_dc_left_predictor_4x4_neon,
+ vpx_highbd_dc_top_predictor_4x4_neon,
+ vpx_highbd_dc_128_predictor_4x4_neon, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(NEON, TestHighbdIntraPred8,
+ vpx_highbd_dc_predictor_8x8_neon,
+ vpx_highbd_dc_left_predictor_8x8_neon,
+ vpx_highbd_dc_top_predictor_8x8_neon,
+ vpx_highbd_dc_128_predictor_8x8_neon, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(NEON, TestHighbdIntraPred16,
+ vpx_highbd_dc_predictor_16x16_neon,
+ vpx_highbd_dc_left_predictor_16x16_neon,
+ vpx_highbd_dc_top_predictor_16x16_neon,
+ vpx_highbd_dc_128_predictor_16x16_neon, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(NEON, TestHighbdIntraPred32,
+ vpx_highbd_dc_predictor_32x32_neon,
+ vpx_highbd_dc_left_predictor_32x32_neon,
+ vpx_highbd_dc_top_predictor_32x32_neon,
+ vpx_highbd_dc_128_predictor_32x32_neon, NULL, NULL, NULL,
+ NULL, NULL, NULL, NULL, NULL, NULL)
+#endif // HAVE_NEON
+
#endif // CONFIG_VP9_HIGHBITDEPTH
#include "test/test_libvpx.cc"
--- a/test/vp9_intrapred_test.cc
+++ b/test/vp9_intrapred_test.cc
@@ -412,5 +412,115 @@
&vpx_highbd_v_predictor_32x32_c, 32, 12)));
#endif // HAVE_SSE2
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+ NEON_TO_C_8, VP9HighbdIntraPredTest,
+ ::testing::Values(
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_neon,
+ &vpx_highbd_dc_128_predictor_4x4_c, 4, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_neon,
+ &vpx_highbd_dc_128_predictor_8x8_c, 8, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_neon,
+ &vpx_highbd_dc_128_predictor_16x16_c, 16, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_neon,
+ &vpx_highbd_dc_128_predictor_32x32_c, 32, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_neon,
+ &vpx_highbd_dc_left_predictor_4x4_c, 4, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_neon,
+ &vpx_highbd_dc_left_predictor_8x8_c, 8, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_neon,
+ &vpx_highbd_dc_left_predictor_16x16_c, 16, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_neon,
+ &vpx_highbd_dc_left_predictor_32x32_c, 32, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_neon,
+ &vpx_highbd_dc_predictor_4x4_c, 4, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_neon,
+ &vpx_highbd_dc_predictor_8x8_c, 8, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_16x16_neon,
+ &vpx_highbd_dc_predictor_16x16_c, 16, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_neon,
+ &vpx_highbd_dc_predictor_32x32_c, 32, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_neon,
+ &vpx_highbd_dc_top_predictor_4x4_c, 4, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_neon,
+ &vpx_highbd_dc_top_predictor_8x8_c, 8, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_neon,
+ &vpx_highbd_dc_top_predictor_16x16_c, 16, 8),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_neon,
+ &vpx_highbd_dc_top_predictor_32x32_c, 32, 8)));
+
+INSTANTIATE_TEST_CASE_P(
+ NEON_TO_C_10, VP9HighbdIntraPredTest,
+ ::testing::Values(
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_neon,
+ &vpx_highbd_dc_128_predictor_4x4_c, 4, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_neon,
+ &vpx_highbd_dc_128_predictor_8x8_c, 8, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_neon,
+ &vpx_highbd_dc_128_predictor_16x16_c, 16, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_neon,
+ &vpx_highbd_dc_128_predictor_32x32_c, 32, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_neon,
+ &vpx_highbd_dc_left_predictor_4x4_c, 4, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_neon,
+ &vpx_highbd_dc_left_predictor_8x8_c, 8, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_neon,
+ &vpx_highbd_dc_left_predictor_16x16_c, 16, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_neon,
+ &vpx_highbd_dc_left_predictor_32x32_c, 32, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_neon,
+ &vpx_highbd_dc_predictor_4x4_c, 4, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_neon,
+ &vpx_highbd_dc_predictor_8x8_c, 8, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_16x16_neon,
+ &vpx_highbd_dc_predictor_16x16_c, 16, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_neon,
+ &vpx_highbd_dc_predictor_32x32_c, 32, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_neon,
+ &vpx_highbd_dc_top_predictor_4x4_c, 4, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_neon,
+ &vpx_highbd_dc_top_predictor_8x8_c, 8, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_neon,
+ &vpx_highbd_dc_top_predictor_16x16_c, 16, 10),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_neon,
+ &vpx_highbd_dc_top_predictor_32x32_c, 32, 10)));
+
+INSTANTIATE_TEST_CASE_P(
+ NEON_TO_C_12, VP9HighbdIntraPredTest,
+ ::testing::Values(
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_4x4_neon,
+ &vpx_highbd_dc_128_predictor_4x4_c, 4, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_8x8_neon,
+ &vpx_highbd_dc_128_predictor_8x8_c, 8, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_16x16_neon,
+ &vpx_highbd_dc_128_predictor_16x16_c, 16, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_128_predictor_32x32_neon,
+ &vpx_highbd_dc_128_predictor_32x32_c, 32, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_4x4_neon,
+ &vpx_highbd_dc_left_predictor_4x4_c, 4, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_8x8_neon,
+ &vpx_highbd_dc_left_predictor_8x8_c, 8, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_16x16_neon,
+ &vpx_highbd_dc_left_predictor_16x16_c, 16, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_left_predictor_32x32_neon,
+ &vpx_highbd_dc_left_predictor_32x32_c, 32, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_4x4_neon,
+ &vpx_highbd_dc_predictor_4x4_c, 4, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_8x8_neon,
+ &vpx_highbd_dc_predictor_8x8_c, 8, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_16x16_neon,
+ &vpx_highbd_dc_predictor_16x16_c, 16, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_predictor_32x32_neon,
+ &vpx_highbd_dc_predictor_32x32_c, 32, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_4x4_neon,
+ &vpx_highbd_dc_top_predictor_4x4_c, 4, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_8x8_neon,
+ &vpx_highbd_dc_top_predictor_8x8_c, 8, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_16x16_neon,
+ &vpx_highbd_dc_top_predictor_16x16_c, 16, 12),
+ HighbdIntraPredParam(&vpx_highbd_dc_top_predictor_32x32_neon,
+ &vpx_highbd_dc_top_predictor_32x32_c, 32, 12)));
+#endif // HAVE_NEON
+
#endif // CONFIG_VP9_HIGHBITDEPTH
} // namespace
--- /dev/null
+++ b/vpx_dsp/arm/highbd_intrapred_neon.c
@@ -1,0 +1,285 @@
+/*
+ * Copyright (c) 2016 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"
+
+//------------------------------------------------------------------------------
+// DC 4x4
+
+static INLINE uint16x4_t dc_sum_4(const uint16_t *ref) {+ const uint16x4_t ref_u16 = vld1_u16(ref);
+ const uint16x4_t p0 = vpadd_u16(ref_u16, ref_u16);
+ return vpadd_u16(p0, p0);
+}
+
+static INLINE void dc_store_4x4(uint16_t *dst, ptrdiff_t stride,
+ const uint16x4_t dc) {+ const uint16x4_t dc_dup = vdup_lane_u16(dc, 0);
+ int i;
+ for (i = 0; i < 4; ++i, dst += stride) {+ vst1_u16(dst, dc_dup);
+ }
+}
+
+void vpx_highbd_dc_predictor_4x4_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t a = vld1_u16(above);
+ const uint16x4_t l = vld1_u16(left);
+ uint16x4_t sum;
+ uint16x4_t dc;
+ (void)bd;
+ sum = vadd_u16(a, l);
+ sum = vpadd_u16(sum, sum);
+ sum = vpadd_u16(sum, sum);
+ dc = vrshr_n_u16(sum, 3);
+ dc_store_4x4(dst, stride, dc);
+}
+
+void vpx_highbd_dc_left_predictor_4x4_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_4(left);
+ const uint16x4_t dc = vrshr_n_u16(sum, 2);
+ (void)above;
+ (void)bd;
+ dc_store_4x4(dst, stride, dc);
+}
+
+void vpx_highbd_dc_top_predictor_4x4_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_4(above);
+ const uint16x4_t dc = vrshr_n_u16(sum, 2);
+ (void)left;
+ (void)bd;
+ dc_store_4x4(dst, stride, dc);
+}
+
+void vpx_highbd_dc_128_predictor_4x4_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t dc = vdup_n_u16(1 << (bd - 1));
+ (void)above;
+ (void)left;
+ dc_store_4x4(dst, stride, dc);
+}
+
+//------------------------------------------------------------------------------
+// DC 8x8
+
+static INLINE uint16x4_t dc_sum_8(const uint16_t *ref) {+ const uint16x8_t ref_u16 = vld1q_u16(ref);
+ uint16x4_t sum = vadd_u16(vget_low_u16(ref_u16), vget_high_u16(ref_u16));
+ sum = vpadd_u16(sum, sum);
+ return vpadd_u16(sum, sum);
+}
+
+static INLINE void dc_store_8x8(uint16_t *dst, ptrdiff_t stride,
+ const uint16x4_t dc) {+ const uint16x8_t dc_dup = vdupq_lane_u16(dc, 0);
+ int i;
+ for (i = 0; i < 8; ++i, dst += stride) {+ vst1q_u16(dst, dc_dup);
+ }
+}
+
+void vpx_highbd_dc_predictor_8x8_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x8_t above_u16 = vld1q_u16(above);
+ const uint16x8_t left_u16 = vld1q_u16(left);
+ const uint16x8_t p0 = vaddq_u16(above_u16, left_u16);
+ uint16x4_t sum = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
+ uint16x4_t dc;
+ (void)bd;
+ sum = vpadd_u16(sum, sum);
+ sum = vpadd_u16(sum, sum);
+ dc = vrshr_n_u16(sum, 4);
+ dc_store_8x8(dst, stride, dc);
+}
+
+void vpx_highbd_dc_left_predictor_8x8_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_8(left);
+ const uint16x4_t dc = vrshr_n_u16(sum, 3);
+ (void)above;
+ (void)bd;
+ dc_store_8x8(dst, stride, dc);
+}
+
+void vpx_highbd_dc_top_predictor_8x8_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_8(above);
+ const uint16x4_t dc = vrshr_n_u16(sum, 3);
+ (void)left;
+ (void)bd;
+ dc_store_8x8(dst, stride, dc);
+}
+
+void vpx_highbd_dc_128_predictor_8x8_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t dc = vdup_n_u16(1 << (bd - 1));
+ (void)above;
+ (void)left;
+ dc_store_8x8(dst, stride, dc);
+}
+
+//------------------------------------------------------------------------------
+// DC 16x16
+
+static INLINE uint16x4_t dc_sum_16(const uint16_t *ref) {+ const uint16x8x2_t ref_u16 = vld2q_u16(ref);
+ const uint16x8_t p0 = vaddq_u16(ref_u16.val[0], ref_u16.val[1]);
+ uint16x4_t sum = vadd_u16(vget_low_u16(p0), vget_high_u16(p0));
+ sum = vpadd_u16(sum, sum);
+ return vpadd_u16(sum, sum);
+}
+
+static INLINE void dc_store_16x16(uint16_t *dst, ptrdiff_t stride,
+ const uint16x4_t dc) {+ uint16x8x2_t dc_dup;
+ int i;
+ dc_dup.val[0] = dc_dup.val[1] = vdupq_lane_u16(dc, 0);
+ for (i = 0; i < 16; ++i, dst += stride) {+ vst2q_u16(dst, dc_dup);
+ }
+}
+
+void vpx_highbd_dc_predictor_16x16_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x8x2_t a = vld2q_u16(above);
+ const uint16x8x2_t l = vld2q_u16(left);
+ const uint16x8_t pa = vaddq_u16(a.val[0], a.val[1]);
+ const uint16x8_t pl = vaddq_u16(l.val[0], l.val[1]);
+ const uint16x8_t pal0 = vaddq_u16(pa, pl);
+ uint16x4_t pal1 = vadd_u16(vget_low_u16(pal0), vget_high_u16(pal0));
+ uint32x2_t sum;
+ uint16x4_t dc;
+ (void)bd;
+ pal1 = vpadd_u16(pal1, pal1);
+ sum = vpaddl_u16(pal1);
+ dc = vreinterpret_u16_u32(vrshr_n_u32(sum, 5));
+ dc_store_16x16(dst, stride, dc);
+}
+
+void vpx_highbd_dc_left_predictor_16x16_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_16(left);
+ const uint16x4_t dc = vrshr_n_u16(sum, 4);
+ (void)above;
+ (void)bd;
+ dc_store_16x16(dst, stride, dc);
+}
+
+void vpx_highbd_dc_top_predictor_16x16_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t sum = dc_sum_16(above);
+ const uint16x4_t dc = vrshr_n_u16(sum, 4);
+ (void)left;
+ (void)bd;
+ dc_store_16x16(dst, stride, dc);
+}
+
+void vpx_highbd_dc_128_predictor_16x16_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t dc = vdup_n_u16(1 << (bd - 1));
+ (void)above;
+ (void)left;
+ dc_store_16x16(dst, stride, dc);
+}
+
+//------------------------------------------------------------------------------
+// DC 32x32
+
+static INLINE uint32x2_t dc_sum_32(const uint16_t *ref) {+ const uint16x8x4_t r = vld4q_u16(ref);
+ const uint16x8_t p0 = vaddq_u16(r.val[0], r.val[1]);
+ const uint16x8_t p1 = vaddq_u16(r.val[2], r.val[3]);
+ const uint16x8_t p2 = vaddq_u16(p0, p1);
+ uint16x4_t sum = vadd_u16(vget_low_u16(p2), vget_high_u16(p2));
+ sum = vpadd_u16(sum, sum);
+ return vpaddl_u16(sum);
+}
+
+static INLINE void dc_store_32x32(uint16_t *dst, ptrdiff_t stride,
+ const uint16x4_t dc) {+ uint16x8x2_t dc_dup;
+ int i;
+ dc_dup.val[0] = dc_dup.val[1] = vdupq_lane_u16(dc, 0);
+
+ for (i = 0; i < 32; ++i) {+ vst2q_u16(dst, dc_dup);
+ dst += 16;
+ vst2q_u16(dst, dc_dup);
+ dst += stride - 16;
+ }
+}
+
+void vpx_highbd_dc_predictor_32x32_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x8x4_t a = vld4q_u16(above);
+ const uint16x8x4_t l = vld4q_u16(left);
+ const uint16x8_t pa0 = vaddq_u16(a.val[0], a.val[1]);
+ const uint16x8_t pa1 = vaddq_u16(a.val[2], a.val[3]);
+ const uint16x8_t pl0 = vaddq_u16(l.val[0], l.val[1]);
+ const uint16x8_t pl1 = vaddq_u16(l.val[2], l.val[3]);
+ const uint16x8_t pa = vaddq_u16(pa0, pa1);
+ const uint16x8_t pl = vaddq_u16(pl0, pl1);
+ const uint16x8_t pal0 = vaddq_u16(pa, pl);
+ const uint16x4_t pal1 = vadd_u16(vget_low_u16(pal0), vget_high_u16(pal0));
+ uint32x2_t sum = vpaddl_u16(pal1);
+ uint16x4_t dc;
+ (void)bd;
+ sum = vpadd_u32(sum, sum);
+ dc = vreinterpret_u16_u32(vrshr_n_u32(sum, 6));
+ dc_store_32x32(dst, stride, dc);
+}
+
+void vpx_highbd_dc_left_predictor_32x32_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint32x2_t sum = dc_sum_32(left);
+ const uint16x4_t dc = vreinterpret_u16_u32(vrshr_n_u32(sum, 5));
+ (void)above;
+ (void)bd;
+ dc_store_32x32(dst, stride, dc);
+}
+
+void vpx_highbd_dc_top_predictor_32x32_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint32x2_t sum = dc_sum_32(above);
+ const uint16x4_t dc = vreinterpret_u16_u32(vrshr_n_u32(sum, 5));
+ (void)left;
+ (void)bd;
+ dc_store_32x32(dst, stride, dc);
+}
+
+void vpx_highbd_dc_128_predictor_32x32_neon(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ const uint16x4_t dc = vdup_n_u16(1 << (bd - 1));
+ (void)above;
+ (void)left;
+ dc_store_32x32(dst, stride, dc);
+}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -48,6 +48,7 @@
ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
DSP_SRCS-$(HAVE_SSE) += x86/highbd_intrapred_sse2.asm
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_intrapred_sse2.asm
+DSP_SRCS-$(HAVE_NEON) += arm/highbd_intrapred_neon.c
endif # CONFIG_VP9_HIGHBITDEPTH
ifneq ($(filter yes,$(CONFIG_POSTPROC) $(CONFIG_VP9_POSTPROC)),)
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -233,13 +233,16 @@
specialize qw/vpx_highbd_tm_predictor_4x4 sse2/;
add_proto qw/void vpx_highbd_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_4x4 sse2/;
+ specialize qw/vpx_highbd_dc_predictor_4x4 neon sse2/;
add_proto qw/void vpx_highbd_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_top_predictor_4x4 neon/;
add_proto qw/void vpx_highbd_dc_left_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_left_predictor_4x4 neon/;
add_proto qw/void vpx_highbd_dc_128_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_128_predictor_4x4 neon/;
add_proto qw/void vpx_highbd_d207_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
@@ -268,13 +271,16 @@
specialize qw/vpx_highbd_tm_predictor_8x8 sse2/;
add_proto qw/void vpx_highbd_dc_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_8x8 sse2/;;
+ specialize qw/vpx_highbd_dc_predictor_8x8 neon sse2/;
add_proto qw/void vpx_highbd_dc_top_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_top_predictor_8x8 neon/;
add_proto qw/void vpx_highbd_dc_left_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_left_predictor_8x8 neon/;
add_proto qw/void vpx_highbd_dc_128_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_128_predictor_8x8 neon/;
add_proto qw/void vpx_highbd_d207_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
@@ -303,13 +309,16 @@
specialize qw/vpx_highbd_tm_predictor_16x16 sse2/;
add_proto qw/void vpx_highbd_dc_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_16x16 sse2/;
+ specialize qw/vpx_highbd_dc_predictor_16x16 neon sse2/;
add_proto qw/void vpx_highbd_dc_top_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_top_predictor_16x16 neon/;
add_proto qw/void vpx_highbd_dc_left_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_left_predictor_16x16 neon/;
add_proto qw/void vpx_highbd_dc_128_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_128_predictor_16x16 neon/;
add_proto qw/void vpx_highbd_d207_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
@@ -338,13 +347,16 @@
specialize qw/vpx_highbd_tm_predictor_32x32 sse2/;
add_proto qw/void vpx_highbd_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
- specialize qw/vpx_highbd_dc_predictor_32x32 sse2/;
+ specialize qw/vpx_highbd_dc_predictor_32x32 neon sse2/;
add_proto qw/void vpx_highbd_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_top_predictor_32x32 neon/;
add_proto qw/void vpx_highbd_dc_left_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_left_predictor_32x32 neon/;
add_proto qw/void vpx_highbd_dc_128_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bd";
+ specialize qw/vpx_highbd_dc_128_predictor_32x32 neon/;
} # CONFIG_VP9_HIGHBITDEPTH
#