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