1 /*
2 * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include "./vp9_rtcd.h"
12 #include "./vpx_config.h"
13 #include "./vpx_dsp_rtcd.h"
14 #include "./vpx_scale_rtcd.h"
15 #include "vp9/common/vp9_blockd.h"
16 #include "vp9/encoder/vp9_encoder.h"
17 #include "vpx/vpx_codec.h"
18 #include "vpx_dsp/vpx_filter.h"
19 #include "vpx_scale/yv12config.h"
20
vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG * src,YV12_BUFFER_CONFIG * dst,INTERP_FILTER filter_type,int phase_scaler)21 void vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG *src,
22 YV12_BUFFER_CONFIG *dst,
23 INTERP_FILTER filter_type, int phase_scaler) {
24 const int src_w = src->y_crop_width;
25 const int src_h = src->y_crop_height;
26 const uint8_t *const srcs[3] = { src->y_buffer, src->u_buffer,
27 src->v_buffer };
28 const int src_strides[3] = { src->y_stride, src->uv_stride, src->uv_stride };
29 uint8_t *const dsts[3] = { dst->y_buffer, dst->u_buffer, dst->v_buffer };
30 const int dst_strides[3] = { dst->y_stride, dst->uv_stride, dst->uv_stride };
31 const InterpKernel *const kernel = vp9_filter_kernels[filter_type];
32 int x, y, i;
33
34 #if HAVE_SSSE3 || HAVE_NEON
35 // TODO(linfengz): The 4:3 specialized C code is disabled by default since
36 // it's much slower than the general version which calls vpx_scaled_2d() even
37 // if vpx_scaled_2d() is not optimized. It will only be enabled as a reference
38 // for the platforms which have faster optimization.
39 if (4 * dst->y_crop_width == 3 * src_w &&
40 4 * dst->y_crop_height == 3 * src_h) {
41 // Specialize 4 to 3 scaling.
42 // Example pixel locations.
43 // (O: Original pixel. S: Scaled pixel. X: Overlapped pixel.)
44 // phase_scaler = 0 | phase_scaler = 8
45 // |
46 // X O S O S O X | O O O O O
47 // |
48 // |
49 // | S S S
50 // |
51 // |
52 // O O O O O | O O O O O
53 // |
54 // S S S S |
55 // |
56 // |
57 // | S S S
58 // O O O O O | O O O O O
59 // |
60 // |
61 // |
62 // S S S S |
63 // |
64 // O O O O O | O O O O O
65 // | S S S
66 // |
67 // |
68 // |
69 // |
70 // X O S O S O X | O O O O O
71
72 const int dst_ws[3] = { dst->y_crop_width, dst->uv_crop_width,
73 dst->uv_crop_width };
74 const int dst_hs[3] = { dst->y_crop_height, dst->uv_crop_height,
75 dst->uv_crop_height };
76 for (i = 0; i < MAX_MB_PLANE; ++i) {
77 const int dst_w = dst_ws[i];
78 const int dst_h = dst_hs[i];
79 const int src_stride = src_strides[i];
80 const int dst_stride = dst_strides[i];
81 for (y = 0; y < dst_h; y += 3) {
82 for (x = 0; x < dst_w; x += 3) {
83 const uint8_t *src_ptr = srcs[i] + 4 * y / 3 * src_stride + 4 * x / 3;
84 uint8_t *dst_ptr = dsts[i] + y * dst_stride + x;
85
86 // Must call c function because its optimization doesn't support 3x3.
87 vpx_scaled_2d_c(src_ptr, src_stride, dst_ptr, dst_stride, kernel,
88 phase_scaler, 64 / 3, phase_scaler, 64 / 3, 3, 3);
89 }
90 }
91 }
92 } else
93 #endif
94 {
95 const int dst_w = dst->y_crop_width;
96 const int dst_h = dst->y_crop_height;
97
98 // The issue b/311394513 reveals a corner case bug. vpx_scaled_2d() requires
99 // both x_step_q4 and y_step_q4 are less than or equal to 64. Otherwise, it
100 // needs to call vp9_scale_and_extend_frame_nonnormative() that supports
101 // arbitrary scaling.
102 const int x_step_q4 = 16 * src_w / dst_w;
103 const int y_step_q4 = 16 * src_h / dst_h;
104 if (x_step_q4 > 64 || y_step_q4 > 64) {
105 // This function is only called while cm->bit_depth is VPX_BITS_8.
106 #if CONFIG_VP9_HIGHBITDEPTH
107 vp9_scale_and_extend_frame_nonnormative(src, dst, (int)VPX_BITS_8);
108 #else
109 vp9_scale_and_extend_frame_nonnormative(src, dst);
110 #endif // CONFIG_VP9_HIGHBITDEPTH
111 return;
112 }
113
114 for (i = 0; i < MAX_MB_PLANE; ++i) {
115 const int factor = (i == 0 || i == 3 ? 1 : 2);
116 const int src_stride = src_strides[i];
117 const int dst_stride = dst_strides[i];
118 for (y = 0; y < dst_h; y += 16) {
119 const int y_q4 = y * (16 / factor) * src_h / dst_h + phase_scaler;
120 for (x = 0; x < dst_w; x += 16) {
121 const int x_q4 = x * (16 / factor) * src_w / dst_w + phase_scaler;
122 const uint8_t *src_ptr = srcs[i] +
123 (y / factor) * src_h / dst_h * src_stride +
124 (x / factor) * src_w / dst_w;
125 uint8_t *dst_ptr = dsts[i] + (y / factor) * dst_stride + (x / factor);
126
127 vpx_scaled_2d(src_ptr, src_stride, dst_ptr, dst_stride, kernel,
128 x_q4 & 0xf, 16 * src_w / dst_w, y_q4 & 0xf,
129 16 * src_h / dst_h, 16 / factor, 16 / factor);
130 }
131 }
132 }
133 }
134
135 vpx_extend_frame_borders(dst);
136 }
137