ref: 81a8138fc3a853df1b40784a237d5cee6f9fc5d9
parent: 5cd0aab81a007f7195ead1b5fdb7ae8545f885b4
author: Deb Mukherjee <debargha@google.com>
date: Mon Sep 15 08:59:19 EDT 2014
Adding high-bitdepth intra prediction functions Change-Id: I6f5cb101e2dc57c3d3f4d7e0ffb4ddbed027d111
--- a/test/test.mk
+++ b/test/test.mk
@@ -128,6 +128,7 @@
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += fdct8x8_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += variance_test.cc
LIBVPX_TEST_SRCS-$(CONFIG_VP9_ENCODER) += vp9_subtract_test.cc
+LIBVPX_TEST_SRCS-$(CONFIG_VP9) += vp9_intrapred_test.cc
ifeq ($(CONFIG_VP9_ENCODER),yes)
LIBVPX_TEST_SRCS-$(CONFIG_SPATIAL_SVC) += svc_test.cc
--- /dev/null
+++ b/test/vp9_intrapred_test.cc
@@ -1,0 +1,284 @@
+/*
+ * Copyright (c) 2014 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 <string>
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
+
+#include "./vpx_config.h"
+#include "./vp9_rtcd.h"
+#include "vp9/common/vp9_blockd.h"
+#include "vp9/common/vp9_pred_common.h"
+#include "vpx_mem/vpx_mem.h"
+#include "test/util.h"
+
+namespace {+
+using libvpx_test::ACMRandom;
+
+const int count_test_block = 100000;
+
+// Base class for VP9 intra prediction tests.
+class VP9IntraPredBase {+ public:
+ virtual ~VP9IntraPredBase() { libvpx_test::ClearSystemState(); }+
+ protected:
+ virtual void Predict(PREDICTION_MODE mode) = 0;
+
+ void CheckPrediction(int test_case_number, int *error_count) const {+ // For each pixel ensure that the calculated value is the same as reference.
+ for (int y = 0; y < block_size_; y++) {+ for (int x = 0; x < block_size_; x++) {+ *error_count += ref_dst_[x + y * stride_] != dst_[x + y * stride_];
+ if (*error_count == 1) {+ ASSERT_EQ(ref_dst_[x + y * stride_], dst_[x + y * stride_])
+ << " Failed on Test Case Number "<< test_case_number;
+ }
+ }
+ }
+ }
+
+ void RunTest(uint16_t* left_col, uint16_t* above_data,
+ uint16_t* dst, uint16_t* ref_dst) {+ ACMRandom rnd(ACMRandom::DeterministicSeed());
+ left_col_ = left_col;
+ dst_ = dst;
+ ref_dst_ = ref_dst;
+ above_row_ = above_data + 16;
+ int error_count = 0;
+ for (int i = 0; i < count_test_block; ++i) {+ // Fill edges with random data, try first with saturated values.
+ for (int x = -1; x <= block_size_*2; x++) {+ if (i == 0) {+ above_row_[x] = mask_;
+ } else {+ above_row_[x] = rnd.Rand16() & mask_;
+ }
+ }
+ for (int y = 0; y < block_size_; y++) {+ if (i == 0) {+ left_col_[y] = mask_;
+ } else {+ left_col_[y] = rnd.Rand16() & mask_;
+ }
+ }
+ Predict(DC_PRED);
+ CheckPrediction(i, &error_count);
+ }
+ ASSERT_EQ(0, error_count);
+ }
+
+ int block_size_;
+ uint16_t *above_row_;
+ uint16_t *left_col_;
+ uint16_t *dst_;
+ uint16_t *ref_dst_;
+ ptrdiff_t stride_;
+ int mask_;
+};
+
+typedef void (*intra_pred_fn_t)(
+ uint16_t *dst, ptrdiff_t stride, const uint16_t *above,
+ const uint16_t *left, int bps);
+typedef std::tr1::tuple<intra_pred_fn_t,
+ intra_pred_fn_t, int, int> intra_pred_params_t;
+class VP9IntraPredTest
+ : public VP9IntraPredBase,
+ public ::testing::TestWithParam<intra_pred_params_t> {+
+ virtual void SetUp() {+ pred_fn_ = GET_PARAM(0);
+ ref_fn_ = GET_PARAM(1);
+ block_size_ = GET_PARAM(2);
+ bit_depth_ = GET_PARAM(3);
+ stride_ = block_size_ * 3;
+ mask_ = (1 << bit_depth_) - 1;
+ }
+
+ virtual void Predict(PREDICTION_MODE mode) {+ const uint16_t *const_above_row = above_row_;
+ const uint16_t *const_left_col = left_col_;
+ ref_fn_(ref_dst_, stride_, const_above_row, const_left_col, bit_depth_);
+ ASM_REGISTER_STATE_CHECK(pred_fn_(dst_, stride_, const_above_row,
+ const_left_col, bit_depth_));
+ }
+ intra_pred_fn_t pred_fn_;
+ intra_pred_fn_t ref_fn_;
+ int bit_depth_;
+};
+
+TEST_P(VP9IntraPredTest, IntraPredTests) {+ // max block size is 32
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, left_col, 2*32);
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, above_data, 2*32+32);
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, dst, 3 * 32 * 32);
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, ref_dst, 3 * 32 * 32);
+ RunTest(left_col, above_data, dst, ref_dst);
+}
+
+using std::tr1::make_tuple;
+
+#if HAVE_SSE2
+#if CONFIG_VP9_HIGHBITDEPTH
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+ &vp9_high_dc_predictor_32x32_c, 32, 8),
+ make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+ &vp9_high_tm_predictor_16x16_c, 16, 8),
+ make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+ &vp9_high_tm_predictor_32x32_c, 32, 8),
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 8),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 8),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 8),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 8),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 8),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_8, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 8),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 8),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 8),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 8),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 8),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 8),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 8)));
+#endif
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+ &vp9_high_dc_predictor_32x32_c, 32, 10),
+ make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+ &vp9_high_tm_predictor_16x16_c, 16, 10),
+ make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+ &vp9_high_tm_predictor_32x32_c, 32, 10),
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 10),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 10),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 10),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 10),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 10),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_10, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 10),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 10),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 10),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 10),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 10),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 10),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 10)));
+#endif
+
+#if ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_32x32_sse2,
+ &vp9_high_dc_predictor_32x32_c, 32, 12),
+ make_tuple(&vp9_high_tm_predictor_16x16_sse2,
+ &vp9_high_tm_predictor_16x16_c, 16, 12),
+ make_tuple(&vp9_high_tm_predictor_32x32_sse2,
+ &vp9_high_tm_predictor_32x32_c, 32, 12),
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 12),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 12),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 12),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 12),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 12),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#else
+INSTANTIATE_TEST_CASE_P(SSE2_TO_C_12, VP9IntraPredTest,
+ ::testing::Values(
+ make_tuple(&vp9_high_dc_predictor_4x4_sse,
+ &vp9_high_dc_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_dc_predictor_8x8_sse2,
+ &vp9_high_dc_predictor_8x8_c, 8, 12),
+ make_tuple(&vp9_high_dc_predictor_16x16_sse2,
+ &vp9_high_dc_predictor_16x16_c, 16, 12),
+ make_tuple(&vp9_high_v_predictor_4x4_sse,
+ &vp9_high_v_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_v_predictor_8x8_sse2,
+ &vp9_high_v_predictor_8x8_c, 8, 12),
+ make_tuple(&vp9_high_v_predictor_16x16_sse2,
+ &vp9_high_v_predictor_16x16_c, 16, 12),
+ make_tuple(&vp9_high_v_predictor_32x32_sse2,
+ &vp9_high_v_predictor_32x32_c, 32, 12),
+ make_tuple(&vp9_high_tm_predictor_4x4_sse,
+ &vp9_high_tm_predictor_4x4_c, 4, 12),
+ make_tuple(&vp9_high_tm_predictor_8x8_sse2,
+ &vp9_high_tm_predictor_8x8_c, 8, 12)));
+#endif
+#endif // CONFIG_VP9_HIGHBITDEPTH
+#endif // HAVE_SSE2
+} // namespace
--- a/vp9/common/vp9_common.h
+++ b/vp9/common/vp9_common.h
@@ -65,6 +65,18 @@
}
#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE uint16_t clip_pixel_high(int val, int bd) {+ switch (bd) {+ case 8:
+ default:
+ return (uint16_t)clamp(val, 0, 255);
+ case 10:
+ return (uint16_t)clamp(val, 0, 1023);
+ case 12:
+ return (uint16_t)clamp(val, 0, 4095);
+ }
+}
+
#define CONVERT_TO_SHORTPTR(x) ((uint16_t*)(((uintptr_t)x) << 1))
#define CONVERT_TO_BYTEPTR(x) ((uint8_t*)(((uintptr_t)x) >> 1 ))
#endif // CONFIG_VP9_HIGHBITDEPTH
--- a/vp9/common/vp9_reconintra.c
+++ b/vp9/common/vp9_reconintra.c
@@ -40,12 +40,292 @@
type##_predictor(dst, stride, size, above, left); \
}
+#if CONFIG_VP9_HIGHBITDEPTH
+#define intra_pred_high_sized(type, size) \
+ void vp9_high_##type##_predictor_##size##x##size##_c( \
+ uint16_t *dst, ptrdiff_t stride, const uint16_t *above, \
+ const uint16_t *left, int bd) { \+ high_##type##_predictor(dst, stride, size, above, left, bd); \
+ }
+
#define intra_pred_allsizes(type) \
intra_pred_sized(type, 4) \
intra_pred_sized(type, 8) \
intra_pred_sized(type, 16) \
+ intra_pred_sized(type, 32) \
+ intra_pred_high_sized(type, 4) \
+ intra_pred_high_sized(type, 8) \
+ intra_pred_high_sized(type, 16) \
+ intra_pred_high_sized(type, 32)
+
+#else
+
+#define intra_pred_allsizes(type) \
+ intra_pred_sized(type, 4) \
+ intra_pred_sized(type, 8) \
+ intra_pred_sized(type, 16) \
intra_pred_sized(type, 32)
+#endif // CONFIG_VP9_HIGHBITDEPTH
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE void high_d207_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) above;
+ (void) bd;
+ int r, c;
+
+ // First column.
+ for (r = 0; r < bs - 1; ++r) {+ dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1], 1);
+ }
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // Second column.
+ for (r = 0; r < bs - 2; ++r) {+ dst[r * stride] = ROUND_POWER_OF_TWO(left[r] + left[r + 1] * 2 +
+ left[r + 2], 2);
+ }
+ dst[(bs - 2) * stride] = ROUND_POWER_OF_TWO(left[bs - 2] +
+ left[bs - 1] * 3, 2);
+ dst[(bs - 1) * stride] = left[bs - 1];
+ dst++;
+
+ // Rest of last row.
+ for (c = 0; c < bs - 2; ++c)
+ dst[(bs - 1) * stride + c] = left[bs - 1];
+
+ for (r = bs - 2; r >= 0; --r) {+ for (c = 0; c < bs - 2; ++c)
+ dst[r * stride + c] = dst[(r + 1) * stride + c - 2];
+ }
+}
+
+static INLINE void high_d63_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) left;
+ (void) bd;
+ int r, c;
+ for (r = 0; r < bs; ++r) {+ for (c = 0; c < bs; ++c) {+ dst[c] = r & 1 ? ROUND_POWER_OF_TWO(above[r/2 + c] +
+ above[r/2 + c + 1] * 2 +
+ above[r/2 + c + 2], 2)
+ : ROUND_POWER_OF_TWO(above[r/2 + c] +
+ above[r/2 + c + 1], 1);
+ }
+ dst += stride;
+ }
+}
+
+static INLINE void high_d45_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) left;
+ (void) bd;
+ int r, c;
+ for (r = 0; r < bs; ++r) {+ for (c = 0; c < bs; ++c) {+ dst[c] = r + c + 2 < bs * 2 ? ROUND_POWER_OF_TWO(above[r + c] +
+ above[r + c + 1] * 2 +
+ above[r + c + 2], 2)
+ : above[bs * 2 - 1];
+ }
+ dst += stride;
+ }
+}
+
+static INLINE void high_d117_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) bd;
+ int r, c;
+
+ // first row
+ for (c = 0; c < bs; c++)
+ dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c], 1);
+ dst += stride;
+
+ // second row
+ dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ for (c = 1; c < bs; c++)
+ dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+ dst += stride;
+
+ // the rest of first col
+ dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ for (r = 3; r < bs; ++r)
+ dst[(r - 2) * stride] = ROUND_POWER_OF_TWO(left[r - 3] + left[r - 2] * 2 +
+ left[r - 1], 2);
+
+ // the rest of the block
+ for (r = 2; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-2 * stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void high_d135_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) bd;
+ int r, c;
+ dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ for (c = 1; c < bs; c++)
+ dst[c] = ROUND_POWER_OF_TWO(above[c - 2] + above[c - 1] * 2 + above[c], 2);
+
+ dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ for (r = 2; r < bs; ++r)
+ dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+ left[r], 2);
+
+ dst += stride;
+ for (r = 1; r < bs; ++r) {+ for (c = 1; c < bs; c++)
+ dst[c] = dst[-stride + c - 1];
+ dst += stride;
+ }
+}
+
+static INLINE void high_d153_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) bd;
+ int r, c;
+ dst[0] = ROUND_POWER_OF_TWO(above[-1] + left[0], 1);
+ for (r = 1; r < bs; r++)
+ dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 1] + left[r], 1);
+ dst++;
+
+ dst[0] = ROUND_POWER_OF_TWO(left[0] + above[-1] * 2 + above[0], 2);
+ dst[stride] = ROUND_POWER_OF_TWO(above[-1] + left[0] * 2 + left[1], 2);
+ for (r = 2; r < bs; r++)
+ dst[r * stride] = ROUND_POWER_OF_TWO(left[r - 2] + left[r - 1] * 2 +
+ left[r], 2);
+ dst++;
+
+ for (c = 0; c < bs - 2; c++)
+ dst[c] = ROUND_POWER_OF_TWO(above[c - 1] + above[c] * 2 + above[c + 1], 2);
+ dst += stride;
+
+ for (r = 1; r < bs; ++r) {+ for (c = 0; c < bs - 2; c++)
+ dst[c] = dst[-stride + c - 2];
+ dst += stride;
+ }
+}
+
+static INLINE void high_v_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) left;
+ (void) bd;
+ int r;
+
+ for (r = 0; r < bs; r++) {+ vpx_memcpy(dst, above, bs * sizeof(uint16_t));
+ dst += stride;
+ }
+}
+
+static INLINE void high_h_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above, const uint16_t *left,
+ int bd) {+ (void) above;
+ (void) bd;
+ int r;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, left[r], bs);
+ dst += stride;
+ }
+}
+
+static INLINE void high_tm_predictor(uint16_t *dst, ptrdiff_t stride, int bs,
+ const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) bd;
+ int r, c;
+ int ytop_left = above[-1];
+
+ for (r = 0; r < bs; r++) {+ for (c = 0; c < bs; c++)
+ dst[c] = clip_pixel_high(left[r] + above[c] - ytop_left, bd);
+ dst += stride;
+ }
+}
+
+static INLINE void high_dc_128_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) above;
+ (void) left;
+ int r;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, 128 << (bd - 8), bs);
+ dst += stride;
+ }
+}
+
+static INLINE void high_dc_left_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) above;
+ (void) bd;
+ int i, r, expected_dc, sum = 0;
+
+ for (i = 0; i < bs; i++)
+ sum += left[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void high_dc_top_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) left;
+ (void) bd;
+ int i, r, expected_dc, sum = 0;
+
+ for (i = 0; i < bs; i++)
+ sum += above[i];
+ expected_dc = (sum + (bs >> 1)) / bs;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+
+static INLINE void high_dc_predictor(uint16_t *dst, ptrdiff_t stride,
+ int bs, const uint16_t *above,
+ const uint16_t *left, int bd) {+ (void) bd;
+ int i, r, expected_dc, sum = 0;
+ const int count = 2 * bs;
+
+ for (i = 0; i < bs; i++) {+ sum += above[i];
+ sum += left[i];
+ }
+
+ expected_dc = (sum + (count >> 1)) / count;
+
+ for (r = 0; r < bs; r++) {+ vpx_memset16(dst, expected_dc, bs);
+ dst += stride;
+ }
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
static INLINE void d207_predictor(uint8_t *dst, ptrdiff_t stride, int bs,
const uint8_t *above, const uint8_t *left) {int r, c;
@@ -293,6 +573,14 @@
static intra_pred_fn pred[INTRA_MODES][TX_SIZES];
static intra_pred_fn dc_pred[2][2][TX_SIZES];
+#if CONFIG_VP9_HIGHBITDEPTH
+typedef void (*intra_high_pred_fn)(uint16_t *dst, ptrdiff_t stride,
+ const uint16_t *above, const uint16_t *left,
+ int bd);
+static intra_high_pred_fn pred_high[INTRA_MODES][4];
+static intra_high_pred_fn dc_pred_high[2][2][4];
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
void vp9_init_intra_predictors() {#define INIT_ALL_SIZES(p, type) \
p[TX_4X4] = vp9_##type##_predictor_4x4; \
@@ -315,9 +603,164 @@
INIT_ALL_SIZES(dc_pred[1][0], dc_left);
INIT_ALL_SIZES(dc_pred[1][1], dc);
-#undef INIT_ALL_SIZES
+#if CONFIG_VP9_HIGHBITDEPTH
+ INIT_ALL_SIZES(pred_high[V_PRED], high_v);
+ INIT_ALL_SIZES(pred_high[H_PRED], high_h);
+ INIT_ALL_SIZES(pred_high[D207_PRED], high_d207);
+ INIT_ALL_SIZES(pred_high[D45_PRED], high_d45);
+ INIT_ALL_SIZES(pred_high[D63_PRED], high_d63);
+ INIT_ALL_SIZES(pred_high[D117_PRED], high_d117);
+ INIT_ALL_SIZES(pred_high[D135_PRED], high_d135);
+ INIT_ALL_SIZES(pred_high[D153_PRED], high_d153);
+ INIT_ALL_SIZES(pred_high[TM_PRED], high_tm);
+
+ INIT_ALL_SIZES(dc_pred_high[0][0], high_dc_128);
+ INIT_ALL_SIZES(dc_pred_high[0][1], high_dc_top);
+ INIT_ALL_SIZES(dc_pred_high[1][0], high_dc_left);
+ INIT_ALL_SIZES(dc_pred_high[1][1], high_dc);
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
+#undef intra_pred_allsizes
}
+#if CONFIG_VP9_HIGHBITDEPTH
+static void build_intra_predictors_high(const MACROBLOCKD *xd,
+ const uint8_t *ref8,
+ int ref_stride,
+ uint8_t *dst8,
+ int dst_stride,
+ PREDICTION_MODE mode,
+ TX_SIZE tx_size,
+ int up_available,
+ int left_available,
+ int right_available,
+ int x, int y,
+ int plane, int bd) {+ int i;
+ uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
+ uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, left_col, 64);
+ DECLARE_ALIGNED_ARRAY(16, uint16_t, above_data, 128 + 16);
+ uint16_t *above_row = above_data + 16;
+ const uint16_t *const_above_row = above_row;
+ const int bs = 4 << tx_size;
+ int frame_width, frame_height;
+ int x0, y0;
+ const struct macroblockd_plane *const pd = &xd->plane[plane];
+ // int base=128;
+ int base = 128 << (bd - 8);
+ // 127 127 127 .. 127 127 127 127 127 127
+ // 129 A B .. Y Z
+ // 129 C D .. W X
+ // 129 E F .. U V
+ // 129 G H .. S T T T T T
+
+ // Get current frame pointer, width and height.
+ if (plane == 0) {+ frame_width = xd->cur_buf->y_width;
+ frame_height = xd->cur_buf->y_height;
+ } else {+ frame_width = xd->cur_buf->uv_width;
+ frame_height = xd->cur_buf->uv_height;
+ }
+
+ // Get block position in current frame.
+ x0 = (-xd->mb_to_left_edge >> (3 + pd->subsampling_x)) + x;
+ y0 = (-xd->mb_to_top_edge >> (3 + pd->subsampling_y)) + y;
+
+ // left
+ if (left_available) {+ if (xd->mb_to_bottom_edge < 0) {+ /* slower path if the block needs border extension */
+ if (y0 + bs <= frame_height) {+ for (i = 0; i < bs; ++i)
+ left_col[i] = ref[i * ref_stride - 1];
+ } else {+ const int extend_bottom = frame_height - y0;
+ for (i = 0; i < extend_bottom; ++i)
+ left_col[i] = ref[i * ref_stride - 1];
+ for (; i < bs; ++i)
+ left_col[i] = ref[(extend_bottom - 1) * ref_stride - 1];
+ }
+ } else {+ /* faster path if the block does not need extension */
+ for (i = 0; i < bs; ++i)
+ left_col[i] = ref[i * ref_stride - 1];
+ }
+ } else {+ // TODO(Peter): this value should probably change for high bitdepth
+ vpx_memset16(left_col, base + 1, bs);
+ }
+
+ // TODO(hkuang) do not extend 2*bs pixels for all modes.
+ // above
+ if (up_available) {+ const uint16_t *above_ref = ref - ref_stride;
+ if (xd->mb_to_right_edge < 0) {+ /* slower path if the block needs border extension */
+ if (x0 + 2 * bs <= frame_width) {+ if (right_available && bs == 4) {+ vpx_memcpy(above_row, above_ref, 2 * bs * sizeof(uint16_t));
+ } else {+ vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+ vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+ }
+ } else if (x0 + bs <= frame_width) {+ const int r = frame_width - x0;
+ if (right_available && bs == 4) {+ vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+ vpx_memset16(above_row + r, above_row[r - 1],
+ x0 + 2 * bs - frame_width);
+ } else {+ vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+ vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+ }
+ } else if (x0 <= frame_width) {+ const int r = frame_width - x0;
+ if (right_available && bs == 4) {+ vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+ vpx_memset16(above_row + r, above_row[r - 1],
+ x0 + 2 * bs - frame_width);
+ } else {+ vpx_memcpy(above_row, above_ref, r * sizeof(uint16_t));
+ vpx_memset16(above_row + r, above_row[r - 1],
+ x0 + 2 * bs - frame_width);
+ }
+ }
+ // TODO(Peter) this value should probably change for high bitdepth
+ above_row[-1] = left_available ? above_ref[-1] : (base+1);
+ } else {+ /* faster path if the block does not need extension */
+ if (bs == 4 && right_available && left_available) {+ const_above_row = above_ref;
+ } else {+ vpx_memcpy(above_row, above_ref, bs * sizeof(uint16_t));
+ if (bs == 4 && right_available)
+ vpx_memcpy(above_row + bs, above_ref + bs, bs * sizeof(uint16_t));
+ else
+ vpx_memset16(above_row + bs, above_row[bs - 1], bs);
+ // TODO(Peter): this value should probably change for high bitdepth
+ above_row[-1] = left_available ? above_ref[-1] : (base+1);
+ }
+ }
+ } else {+ vpx_memset16(above_row, base - 1, bs * 2);
+ // TODO(Peter): this value should probably change for high bitdepth
+ above_row[-1] = base - 1;
+ }
+
+ // predict
+ if (mode == DC_PRED) {+ dc_pred_high[left_available][up_available][tx_size](dst, dst_stride,
+ const_above_row,
+ left_col, xd->bd);
+ } else {+ pred_high[mode][tx_size](dst, dst_stride, const_above_row, left_col,
+ xd->bd);
+ }
+}
+#endif // CONFIG_VP9_HIGHBITDEPTH
+
static void build_intra_predictors(const MACROBLOCKD *xd, const uint8_t *ref,
int ref_stride, uint8_t *dst, int dst_stride,
PREDICTION_MODE mode, TX_SIZE tx_size,
@@ -454,6 +897,14 @@
const int y = loff * 4;
assert(bwl >= 0);
+#if CONFIG_VP9_HIGHBITDEPTH
+ if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {+ build_intra_predictors_high(xd, ref, ref_stride, dst, dst_stride, mode,
+ tx_size, have_top, have_left, have_right,
+ x, y, plane, xd->bd);
+ return;
+ }
+#endif
build_intra_predictors(xd, ref, ref_stride, dst, dst_stride, mode, tx_size,
have_top, have_left, have_right, x, y, plane);
}
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -448,6 +448,165 @@
# High bitdepth functions
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {#
+ # Intra prediction
+ #
+ add_proto qw/void vp9_high_d207_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d207_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d45_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d45_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d63_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d63_predictor_4x4/;
+
+ add_proto qw/void vp9_high_h_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_h_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d117_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d117_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d135_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d135_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d153_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d153_predictor_4x4/;
+
+ add_proto qw/void vp9_high_v_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_v_predictor_4x4 neon/, "$sse_x86inc";
+
+ add_proto qw/void vp9_high_tm_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_tm_predictor_4x4/, "$sse_x86inc";
+
+ add_proto qw/void vp9_high_dc_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_predictor_4x4/, "$sse_x86inc";
+
+ add_proto qw/void vp9_high_dc_top_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_top_predictor_4x4/;
+
+ add_proto qw/void vp9_high_dc_left_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_left_predictor_4x4/;
+
+ add_proto qw/void vp9_high_dc_128_predictor_4x4/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_128_predictor_4x4/;
+
+ add_proto qw/void vp9_high_d207_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d207_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d45_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d45_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d63_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d63_predictor_8x8/;
+
+ add_proto qw/void vp9_high_h_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_h_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d117_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d117_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d135_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d135_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d153_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d153_predictor_8x8/;
+
+ add_proto qw/void vp9_high_v_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_v_predictor_8x8/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_high_tm_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_tm_predictor_8x8/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_high_dc_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_predictor_8x8/, "$sse2_x86inc";;
+
+ add_proto qw/void vp9_high_dc_top_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_top_predictor_8x8/;
+
+ add_proto qw/void vp9_high_dc_left_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_left_predictor_8x8/;
+
+ add_proto qw/void vp9_high_dc_128_predictor_8x8/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_128_predictor_8x8/;
+
+ add_proto qw/void vp9_high_d207_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d207_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d45_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d45_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d63_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d63_predictor_16x16/;
+
+ add_proto qw/void vp9_high_h_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_h_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d117_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d117_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d135_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d135_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d153_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d153_predictor_16x16/;
+
+ add_proto qw/void vp9_high_v_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_v_predictor_16x16 neon/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_high_tm_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_tm_predictor_16x16/, "$sse2_x86_64";
+
+ add_proto qw/void vp9_high_dc_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_predictor_16x16/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_high_dc_top_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_top_predictor_16x16/;
+
+ add_proto qw/void vp9_high_dc_left_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_left_predictor_16x16/;
+
+ add_proto qw/void vp9_high_dc_128_predictor_16x16/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_128_predictor_16x16/;
+
+ add_proto qw/void vp9_high_d207_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d207_predictor_32x32/;
+
+ add_proto qw/void vp9_high_d45_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d45_predictor_32x32/;
+
+ add_proto qw/void vp9_high_d63_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d63_predictor_32x32/;
+
+ add_proto qw/void vp9_high_h_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_h_predictor_32x32/;
+
+ add_proto qw/void vp9_high_d117_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d117_predictor_32x32/;
+
+ add_proto qw/void vp9_high_d135_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d135_predictor_32x32/;
+
+ add_proto qw/void vp9_high_d153_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_d153_predictor_32x32/;
+
+ add_proto qw/void vp9_high_v_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_v_predictor_32x32/, "$sse2_x86inc";
+
+ add_proto qw/void vp9_high_tm_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_tm_predictor_32x32/, "$sse2_x86_64";
+
+ add_proto qw/void vp9_high_dc_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_predictor_32x32/, "$sse2_x86_64";
+
+ add_proto qw/void vp9_high_dc_top_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_top_predictor_32x32/;
+
+ add_proto qw/void vp9_high_dc_left_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_left_predictor_32x32/;
+
+ add_proto qw/void vp9_high_dc_128_predictor_32x32/, "uint16_t *dst, ptrdiff_t y_stride, const uint16_t *above, const uint16_t *left, int bps";
+ specialize qw/vp9_high_dc_128_predictor_32x32/;
+
+ #
# dct
#
add_proto qw/void vp9_high_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
--- /dev/null
+++ b/vp9/common/x86/vp9_high_intrapred_sse2.asm
@@ -1,0 +1,476 @@
+;
+; Copyright (c) 2014 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 "third_party/x86inc/x86inc.asm"
+
+SECTION_RODATA
+pw_4: times 8 dw 4
+pw_8: times 8 dw 8
+pw_16: times 4 dd 16
+pw_32: times 4 dd 32
+
+SECTION .text
+INIT_MMX sse
+cglobal high_dc_predictor_4x4, 4, 5, 4, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ movq m0, [aboveq]
+ movq m2, [leftq]
+ DEFINE_ARGS dst, stride, one
+ mov oned, 0x0001
+ pxor m1, m1
+ movd m3, oned
+ pshufw m3, m3, 0x0
+ paddw m0, m2
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ paddw m0, [GLOBAL(pw_4)]
+ psraw m0, 3
+ pshufw m0, m0, 0x0
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM sse2
+cglobal high_dc_predictor_8x8, 4, 5, 4, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [leftq]
+ DEFINE_ARGS dst, stride, stride3, one
+ mov oned, 0x00010001
+ lea stride3q, [strideq*3]
+ movd m3, oned
+ pshufd m3, m3, 0x0
+ paddw m0, m2
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ packssdw m0, m1
+ pmaddwd m0, m3
+ paddw m0, [GLOBAL(pw_8)]
+ psrlw m0, 4
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ lea dstq, [dstq+strideq*8]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+
+ RESTORE_GOT
+ RET
+
+INIT_XMM sse2
+cglobal high_dc_predictor_16x16, 4, 5, 5, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m3, [aboveq+16]
+ mova m2, [leftq]
+ mova m4, [leftq+16]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 4
+ paddw m0, m2
+ paddw m0, m3
+ paddw m0, m4
+ movhlps m2, m0
+ paddw m0, m2
+ punpcklwd m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ punpckldq m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ paddd m0, [GLOBAL(pw_16)]
+ psrad m0, 5
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m0
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2+16], m0
+ lea dstq, [dstq+strideq*8]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal high_dc_predictor_32x32, 4, 5, 9, dst, stride, above, left, goffset
+ GET_GOT goffsetq
+
+ pxor m1, m1
+ mova m0, [aboveq]
+ mova m2, [aboveq+16]
+ mova m3, [aboveq+32]
+ mova m4, [aboveq+48]
+ mova m5, [leftq]
+ mova m6, [leftq+16]
+ mova m7, [leftq+32]
+ mova m8, [leftq+48]
+ DEFINE_ARGS dst, stride, stride3, lines4
+ lea stride3q, [strideq*3]
+ mov lines4d, 8
+ paddw m0, m2
+ paddw m0, m3
+ paddw m0, m4
+ paddw m0, m5
+ paddw m0, m6
+ paddw m0, m7
+ paddw m0, m8
+ movhlps m2, m0
+ paddw m0, m2
+ punpcklwd m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ punpckldq m0, m1
+ movhlps m2, m0
+ paddd m0, m2
+ paddd m0, [GLOBAL(pw_32)]
+ psrad m0, 6
+ pshuflw m0, m0, 0x0
+ punpcklqdq m0, m0
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16 ], m0
+ mova [dstq +32 ], m0
+ mova [dstq +48 ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2+16 ], m0
+ mova [dstq+strideq*2+32 ], m0
+ mova [dstq+strideq*2+48 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4+16 ], m0
+ mova [dstq+strideq*4+32 ], m0
+ mova [dstq+strideq*4+48 ], m0
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2 +16], m0
+ mova [dstq+stride3q*2 +32], m0
+ mova [dstq+stride3q*2 +48], m0
+ lea dstq, [dstq+strideq*8]
+ dec lines4d
+ jnz .loop
+
+ RESTORE_GOT
+ REP_RET
+%endif
+
+INIT_MMX sse
+cglobal high_v_predictor_4x4, 3, 3, 1, dst, stride, above
+ movq m0, [aboveq]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ lea dstq, [dstq+strideq*4]
+ movq [dstq ], m0
+ movq [dstq+strideq*2], m0
+ RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_8x8, 3, 3, 1, dst, stride, above
+ mova m0, [aboveq]
+ DEFINE_ARGS dst, stride, stride3
+ lea stride3q, [strideq*3]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ lea dstq, [dstq+strideq*8]
+ mova [dstq ], m0
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+stride3q*2], m0
+ RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_16x16, 3, 4, 2, dst, stride, above
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 4
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m1
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m1
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2+16], m1
+ lea dstq, [dstq+strideq*8]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal high_v_predictor_32x32, 3, 4, 4, dst, stride, above
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ mova m2, [aboveq+32]
+ mova m3, [aboveq+48]
+ DEFINE_ARGS dst, stride, stride3, nlines4
+ lea stride3q, [strideq*3]
+ mov nlines4d, 8
+.loop:
+ mova [dstq ], m0
+ mova [dstq +16], m1
+ mova [dstq +32], m2
+ mova [dstq +48], m3
+ mova [dstq+strideq*2 ], m0
+ mova [dstq+strideq*2 +16], m1
+ mova [dstq+strideq*2 +32], m2
+ mova [dstq+strideq*2 +48], m3
+ mova [dstq+strideq*4 ], m0
+ mova [dstq+strideq*4 +16], m1
+ mova [dstq+strideq*4 +32], m2
+ mova [dstq+strideq*4 +48], m3
+ mova [dstq+stride3q*2 ], m0
+ mova [dstq+stride3q*2 +16], m1
+ mova [dstq+stride3q*2 +32], m2
+ mova [dstq+stride3q*2 +48], m3
+ lea dstq, [dstq+strideq*8]
+ dec nlines4d
+ jnz .loop
+ REP_RET
+
+INIT_MMX sse
+cglobal high_tm_predictor_4x4, 5, 6, 5, dst, stride, above, left, bps, one
+ movd m1, [aboveq-2]
+ movq m0, [aboveq]
+ pshufw m1, m1, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ movd m3, oned
+ movd m4, bpsd
+ pshufw m3, m3, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ mov lineq, -2
+ mova m2, m3
+ psllw m3, m4
+ add leftq, 8
+ psubw m3, m2 ; max possible value
+ pxor m4, m4 ; min possible value
+ psubw m0, m1
+.loop:
+ movq m1, [leftq+lineq*4]
+ movq m2, [leftq+lineq*4+2]
+ pshufw m1, m1, 0x0
+ pshufw m2, m2, 0x0
+ paddw m1, m0
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m1, m3
+ pminsw m2, m3
+ pmaxsw m1, m4
+ pmaxsw m2, m4
+ ;Store the values
+ movq [dstq ], m1
+ movq [dstq+strideq*2], m2
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal high_tm_predictor_8x8, 5, 6, 5, dst, stride, above, left, bps, one
+ movd m1, [aboveq-2]
+ mova m0, [aboveq]
+ pshuflw m1, m1, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m3, m3
+ pxor m4, m4
+ pinsrw m3, oned, 0
+ pinsrw m4, bpsd, 0
+ pshuflw m3, m3, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m3, m3
+ mov lineq, -4
+ mova m2, m3
+ punpcklqdq m1, m1
+ psllw m3, m4
+ add leftq, 16
+ psubw m3, m2 ; max possible value
+ pxor m4, m4 ; min possible value
+ psubw m0, m1
+.loop:
+ movd m1, [leftq+lineq*4]
+ movd m2, [leftq+lineq*4+2]
+ pshuflw m1, m1, 0x0
+ pshuflw m2, m2, 0x0
+ punpcklqdq m1, m1
+ punpcklqdq m2, m2
+ paddw m1, m0
+ paddw m2, m0
+ ;Clamp to the bit-depth
+ pminsw m1, m3
+ pminsw m2, m3
+ pmaxsw m1, m4
+ pmaxsw m2, m4
+ ;Store the values
+ mova [dstq ], m1
+ mova [dstq+strideq*2], m2
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+%if ARCH_X86_64
+INIT_XMM sse2
+cglobal high_tm_predictor_16x16, 5, 6, 8, dst, stride, above, left, bps, one
+ movd m2, [aboveq-2]
+ mova m0, [aboveq]
+ mova m1, [aboveq+16]
+ pshuflw m2, m2, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m7, m7
+ pxor m8, m8
+ pinsrw m7, oned, 0
+ pinsrw m8, bpsd, 0
+ pshuflw m7, m7, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m7, m7
+ mov lineq, -8
+ mova m5, m7
+ punpcklqdq m2, m2
+ psllw m7, m8
+ add leftq, 32
+ psubw m7, m5 ; max possible value
+ pxor m8, m8 ; min possible value
+ psubw m0, m2
+ psubw m1, m2
+.loop:
+ movd m2, [leftq+lineq*4]
+ movd m3, [leftq+lineq*4+2]
+ pshuflw m2, m2, 0x0
+ pshuflw m3, m3, 0x0
+ punpcklqdq m2, m2
+ punpcklqdq m3, m3
+ paddw m4, m2, m0
+ paddw m5, m3, m0
+ paddw m2, m1
+ paddw m3, m1
+ ;Clamp to the bit-depth
+ pminsw m4, m7
+ pminsw m5, m7
+ pminsw m2, m7
+ pminsw m3, m7
+ pmaxsw m4, m8
+ pmaxsw m5, m8
+ pmaxsw m2, m8
+ pmaxsw m3, m8
+ ;Store the values
+ mova [dstq ], m4
+ mova [dstq+strideq*2 ], m5
+ mova [dstq +16], m2
+ mova [dstq+strideq*2+16], m3
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+
+INIT_XMM sse2
+cglobal high_tm_predictor_32x32, 5, 6, 12, dst, stride, above, left, bps, one
+ movd m0, [aboveq-2]
+ mova m1, [aboveq]
+ mova m2, [aboveq+16]
+ mova m3, [aboveq+32]
+ mova m4, [aboveq+48]
+ pshuflw m0, m0, 0x0
+ ; Get the values to compute the maximum value at this bit depth
+ mov oned, 1
+ pxor m10, m10
+ pxor m11, m11
+ pinsrw m10, oned, 0
+ pinsrw m11, bpsd, 0
+ pshuflw m10, m10, 0x0
+ DEFINE_ARGS dst, stride, line, left
+ punpcklqdq m10, m10
+ mov lineq, -16
+ mova m5, m10
+ punpcklqdq m0, m0
+ psllw m10, m11
+ add leftq, 64
+ psubw m10, m5 ; max possible value
+ pxor m11, m11 ; min possible value
+ psubw m1, m0
+ psubw m2, m0
+ psubw m3, m0
+ psubw m4, m0
+.loop:
+ movd m5, [leftq+lineq*4]
+ movd m6, [leftq+lineq*4+2]
+ pshuflw m5, m5, 0x0
+ pshuflw m6, m6, 0x0
+ punpcklqdq m5, m5
+ punpcklqdq m6, m6
+ paddw m7, m5, m1
+ paddw m8, m5, m2
+ paddw m9, m5, m3
+ paddw m5, m4
+ ;Clamp these values to the bit-depth
+ pminsw m7, m10
+ pminsw m8, m10
+ pminsw m9, m10
+ pminsw m5, m10
+ pmaxsw m7, m11
+ pmaxsw m8, m11
+ pmaxsw m9, m11
+ pmaxsw m5, m11
+ ;Store these values
+ mova [dstq ], m7
+ mova [dstq +16], m8
+ mova [dstq +32], m9
+ mova [dstq +48], m5
+ paddw m7, m6, m1
+ paddw m8, m6, m2
+ paddw m9, m6, m3
+ paddw m6, m4
+ ;Clamp these values to the bit-depth
+ pminsw m7, m10
+ pminsw m8, m10
+ pminsw m9, m10
+ pminsw m6, m10
+ pmaxsw m7, m11
+ pmaxsw m8, m11
+ pmaxsw m9, m11
+ pmaxsw m6, m11
+ ;Store these values
+ mova [dstq+strideq*2 ], m7
+ mova [dstq+strideq*2+16], m8
+ mova [dstq+strideq*2+32], m9
+ mova [dstq+strideq*2+48], m6
+ lea dstq, [dstq+strideq*4]
+ inc lineq
+ jnz .loop
+ REP_RET
+%endif
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -89,6 +89,10 @@
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_intrapred_ssse3.asm
endif
+ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
+VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_high_intrapred_sse2.asm
+endif
+
# common (c)
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_common_dspr2.h
VP9_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/vp9_convolve2_avg_dspr2.c
--
⑨