/src/libvpx/vpx_dsp/x86/convolve.h
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2015 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 | | #ifndef VPX_VPX_DSP_X86_CONVOLVE_H_ |
11 | | #define VPX_VPX_DSP_X86_CONVOLVE_H_ |
12 | | |
13 | | #include <assert.h> |
14 | | |
15 | | #include "./vpx_config.h" |
16 | | #include "vpx/vpx_integer.h" |
17 | | #include "vpx_ports/compiler_attributes.h" |
18 | | |
19 | | // TODO(chiyotsai@google.com): Refactor the code here. Currently this is pretty |
20 | | // hacky and awful to read. Note that there is a filter_x[3] == 128 check in |
21 | | // HIGHBD_FUN_CONV_2D to avoid seg fault due to the fact that the c function |
22 | | // assumes the filter is always 8 tap. |
23 | | typedef void filter8_1dfunction(const uint8_t *src_ptr, ptrdiff_t src_pitch, |
24 | | uint8_t *output_ptr, ptrdiff_t out_pitch, |
25 | | uint32_t output_height, const int16_t *filter); |
26 | | |
27 | | // TODO(chiyotsai@google.com): Remove the is_avg argument to the MACROS once we |
28 | | // have 4-tap vert avg filter. |
29 | | #define FUN_CONV_1D(name, offset, step_q4, dir, src_start, avg, opt, is_avg) \ |
30 | | void vpx_convolve8_##name##_##opt( \ |
31 | | const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \ |
32 | | ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \ |
33 | 261M | int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \ |
34 | 261M | const int16_t *filter_row = filter[offset]; \ |
35 | 261M | (void)x0_q4; \ |
36 | 261M | (void)x_step_q4; \ |
37 | 261M | (void)y0_q4; \ |
38 | 261M | (void)y_step_q4; \ |
39 | 261M | assert(filter_row[3] != 128); \ |
40 | 261M | assert(step_q4 == 16); \ |
41 | 261M | if (filter_row[0] | filter_row[1] | filter_row[6] | filter_row[7]) { \ |
42 | 103M | const int num_taps = 8; \ |
43 | 114M | while (w >= 16) { \ |
44 | 10.9M | vpx_filter_block1d16_##dir##8_##avg##opt(src_start, src_stride, dst, \ |
45 | 10.9M | dst_stride, h, filter_row); \ |
46 | 10.9M | src += 16; \ |
47 | 10.9M | dst += 16; \ |
48 | 10.9M | w -= 16; \ |
49 | 10.9M | } \ |
50 | 103M | if (w == 8) { \ |
51 | 26.9M | vpx_filter_block1d8_##dir##8_##avg##opt(src_start, src_stride, dst, \ |
52 | 26.9M | dst_stride, h, filter_row); \ |
53 | 76.8M | } else if (w == 4) { \ |
54 | 67.9M | vpx_filter_block1d4_##dir##8_##avg##opt(src_start, src_stride, dst, \ |
55 | 67.9M | dst_stride, h, filter_row); \ |
56 | 67.9M | } \ |
57 | 103M | (void)num_taps; \ |
58 | 157M | } else if (filter_row[2] | filter_row[5]) { \ |
59 | 157M | const int num_taps = is_avg ? 8 : 4; \ |
60 | 171M | while (w >= 16) { \ |
61 | 13.6M | vpx_filter_block1d16_##dir##4_##avg##opt(src_start, src_stride, dst, \ |
62 | 13.6M | dst_stride, h, filter_row); \ |
63 | 13.6M | src += 16; \ |
64 | 13.6M | dst += 16; \ |
65 | 13.6M | w -= 16; \ |
66 | 13.6M | } \ |
67 | 157M | if (w == 8) { \ |
68 | 32.1M | vpx_filter_block1d8_##dir##4_##avg##opt(src_start, src_stride, dst, \ |
69 | 32.1M | dst_stride, h, filter_row); \ |
70 | 125M | } else if (w == 4) { \ |
71 | 115M | vpx_filter_block1d4_##dir##4_##avg##opt(src_start, src_stride, dst, \ |
72 | 115M | dst_stride, h, filter_row); \ |
73 | 115M | } \ |
74 | 157M | (void)num_taps; \ |
75 | 157M | } else { \ |
76 | 0 | const int num_taps = 2; \ |
77 | 0 | while (w >= 16) { \ |
78 | 0 | vpx_filter_block1d16_##dir##2_##avg##opt(src_start, src_stride, dst, \ |
79 | 0 | dst_stride, h, filter_row); \ |
80 | 0 | src += 16; \ |
81 | 0 | dst += 16; \ |
82 | 0 | w -= 16; \ |
83 | 0 | } \ |
84 | 0 | if (w == 8) { \ |
85 | 0 | vpx_filter_block1d8_##dir##2_##avg##opt(src_start, src_stride, dst, \ |
86 | 0 | dst_stride, h, filter_row); \ |
87 | 0 | } else if (w == 4) { \ |
88 | 0 | vpx_filter_block1d4_##dir##2_##avg##opt(src_start, src_stride, dst, \ |
89 | 0 | dst_stride, h, filter_row); \ |
90 | 0 | } \ |
91 | 0 | (void)num_taps; \ |
92 | 0 | } \ |
93 | 261M | } Unexecuted instantiation: vpx_convolve8_horiz_sse2 Unexecuted instantiation: vpx_convolve8_vert_sse2 Unexecuted instantiation: vpx_convolve8_avg_horiz_sse2 Unexecuted instantiation: vpx_convolve8_avg_vert_sse2 Line | Count | Source | 33 | 131M | int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \ | 34 | 131M | const int16_t *filter_row = filter[offset]; \ | 35 | 131M | (void)x0_q4; \ | 36 | 131M | (void)x_step_q4; \ | 37 | 131M | (void)y0_q4; \ | 38 | 131M | (void)y_step_q4; \ | 39 | 131M | assert(filter_row[3] != 128); \ | 40 | 131M | assert(step_q4 == 16); \ | 41 | 131M | if (filter_row[0] | filter_row[1] | filter_row[6] | filter_row[7]) { \ | 42 | 52.1M | const int num_taps = 8; \ | 43 | 57.7M | while (w >= 16) { \ | 44 | 5.56M | vpx_filter_block1d16_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 45 | 5.56M | dst_stride, h, filter_row); \ | 46 | 5.56M | src += 16; \ | 47 | 5.56M | dst += 16; \ | 48 | 5.56M | w -= 16; \ | 49 | 5.56M | } \ | 50 | 52.1M | if (w == 8) { \ | 51 | 13.6M | vpx_filter_block1d8_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 52 | 13.6M | dst_stride, h, filter_row); \ | 53 | 38.4M | } else if (w == 4) { \ | 54 | 33.9M | vpx_filter_block1d4_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 55 | 33.9M | dst_stride, h, filter_row); \ | 56 | 33.9M | } \ | 57 | 52.1M | (void)num_taps; \ | 58 | 79.3M | } else if (filter_row[2] | filter_row[5]) { \ | 59 | 79.3M | const int num_taps = is_avg ? 8 : 4; \ | 60 | 86.1M | while (w >= 16) { \ | 61 | 6.82M | vpx_filter_block1d16_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 62 | 6.82M | dst_stride, h, filter_row); \ | 63 | 6.82M | src += 16; \ | 64 | 6.82M | dst += 16; \ | 65 | 6.82M | w -= 16; \ | 66 | 6.82M | } \ | 67 | 79.3M | if (w == 8) { \ | 68 | 16.1M | vpx_filter_block1d8_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 69 | 16.1M | dst_stride, h, filter_row); \ | 70 | 63.1M | } else if (w == 4) { \ | 71 | 57.8M | vpx_filter_block1d4_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 72 | 57.8M | dst_stride, h, filter_row); \ | 73 | 57.8M | } \ | 74 | 79.3M | (void)num_taps; \ | 75 | 79.3M | } else { \ | 76 | 0 | const int num_taps = 2; \ | 77 | 0 | while (w >= 16) { \ | 78 | 0 | vpx_filter_block1d16_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 79 | 0 | dst_stride, h, filter_row); \ | 80 | 0 | src += 16; \ | 81 | 0 | dst += 16; \ | 82 | 0 | w -= 16; \ | 83 | 0 | } \ | 84 | 0 | if (w == 8) { \ | 85 | 0 | vpx_filter_block1d8_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 86 | 0 | dst_stride, h, filter_row); \ | 87 | 0 | } else if (w == 4) { \ | 88 | 0 | vpx_filter_block1d4_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 89 | 0 | dst_stride, h, filter_row); \ | 90 | 0 | } \ | 91 | 0 | (void)num_taps; \ | 92 | 0 | } \ | 93 | 131M | } |
Line | Count | Source | 33 | 130M | int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \ | 34 | 130M | const int16_t *filter_row = filter[offset]; \ | 35 | 130M | (void)x0_q4; \ | 36 | 130M | (void)x_step_q4; \ | 37 | 130M | (void)y0_q4; \ | 38 | 130M | (void)y_step_q4; \ | 39 | 130M | assert(filter_row[3] != 128); \ | 40 | 130M | assert(step_q4 == 16); \ | 41 | 130M | if (filter_row[0] | filter_row[1] | filter_row[6] | filter_row[7]) { \ | 42 | 51.6M | const int num_taps = 8; \ | 43 | 56.9M | while (w >= 16) { \ | 44 | 5.34M | vpx_filter_block1d16_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 45 | 5.34M | dst_stride, h, filter_row); \ | 46 | 5.34M | src += 16; \ | 47 | 5.34M | dst += 16; \ | 48 | 5.34M | w -= 16; \ | 49 | 5.34M | } \ | 50 | 51.6M | if (w == 8) { \ | 51 | 13.2M | vpx_filter_block1d8_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 52 | 13.2M | dst_stride, h, filter_row); \ | 53 | 38.3M | } else if (w == 4) { \ | 54 | 33.9M | vpx_filter_block1d4_##dir##8_##avg##opt(src_start, src_stride, dst, \ | 55 | 33.9M | dst_stride, h, filter_row); \ | 56 | 33.9M | } \ | 57 | 51.6M | (void)num_taps; \ | 58 | 78.5M | } else if (filter_row[2] | filter_row[5]) { \ | 59 | 78.5M | const int num_taps = is_avg ? 8 : 4; \ | 60 | 85.3M | while (w >= 16) { \ | 61 | 6.81M | vpx_filter_block1d16_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 62 | 6.81M | dst_stride, h, filter_row); \ | 63 | 6.81M | src += 16; \ | 64 | 6.81M | dst += 16; \ | 65 | 6.81M | w -= 16; \ | 66 | 6.81M | } \ | 67 | 78.5M | if (w == 8) { \ | 68 | 15.9M | vpx_filter_block1d8_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 69 | 15.9M | dst_stride, h, filter_row); \ | 70 | 62.5M | } else if (w == 4) { \ | 71 | 57.2M | vpx_filter_block1d4_##dir##4_##avg##opt(src_start, src_stride, dst, \ | 72 | 57.2M | dst_stride, h, filter_row); \ | 73 | 57.2M | } \ | 74 | 78.5M | (void)num_taps; \ | 75 | 78.5M | } else { \ | 76 | 0 | const int num_taps = 2; \ | 77 | 0 | while (w >= 16) { \ | 78 | 0 | vpx_filter_block1d16_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 79 | 0 | dst_stride, h, filter_row); \ | 80 | 0 | src += 16; \ | 81 | 0 | dst += 16; \ | 82 | 0 | w -= 16; \ | 83 | 0 | } \ | 84 | 0 | if (w == 8) { \ | 85 | 0 | vpx_filter_block1d8_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 86 | 0 | dst_stride, h, filter_row); \ | 87 | 0 | } else if (w == 4) { \ | 88 | 0 | vpx_filter_block1d4_##dir##2_##avg##opt(src_start, src_stride, dst, \ | 89 | 0 | dst_stride, h, filter_row); \ | 90 | 0 | } \ | 91 | 0 | (void)num_taps; \ | 92 | 0 | } \ | 93 | 130M | } |
Unexecuted instantiation: vpx_convolve8_avg_horiz_avx2 Unexecuted instantiation: vpx_convolve8_avg_vert_avx2 Unexecuted instantiation: vpx_convolve8_horiz_ssse3 Unexecuted instantiation: vpx_convolve8_vert_ssse3 Unexecuted instantiation: vpx_convolve8_avg_horiz_ssse3 Unexecuted instantiation: vpx_convolve8_avg_vert_ssse3 |
94 | | |
95 | | #define FUN_CONV_2D(avg, opt, is_avg) \ |
96 | | void vpx_convolve8_##avg##opt( \ |
97 | | const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \ |
98 | | ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \ |
99 | 77.0M | int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \ |
100 | 77.0M | const int16_t *filter_x = filter[x0_q4]; \ |
101 | 77.0M | const int16_t *filter_y = filter[y0_q4]; \ |
102 | 77.0M | (void)filter_y; \ |
103 | 77.0M | assert(filter_x[3] != 128); \ |
104 | 77.0M | assert(filter_y[3] != 128); \ |
105 | 77.0M | assert(w <= 64); \ |
106 | 77.0M | assert(h <= 64); \ |
107 | 77.0M | assert(x_step_q4 == 16); \ |
108 | 77.0M | assert(y_step_q4 == 16); \ |
109 | 77.0M | if (filter_x[0] | filter_x[1] | filter_x[6] | filter_x[7]) { \ |
110 | 31.4M | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ |
111 | 31.4M | vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \ |
112 | 31.4M | filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, \ |
113 | 31.4M | h + 7); \ |
114 | 31.4M | vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \ |
115 | 31.4M | filter, x0_q4, x_step_q4, y0_q4, \ |
116 | 31.4M | y_step_q4, w, h); \ |
117 | 45.6M | } else if (filter_x[2] | filter_x[5]) { \ |
118 | 45.6M | const int num_taps = is_avg ? 8 : 4; \ |
119 | 45.6M | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ |
120 | 45.6M | vpx_convolve8_horiz_##opt( \ |
121 | 45.6M | src - (num_taps / 2 - 1) * src_stride, src_stride, fdata2, 64, \ |
122 | 45.6M | filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, h + num_taps - 1); \ |
123 | 45.6M | vpx_convolve8_##avg##vert_##opt(fdata2 + 64 * (num_taps / 2 - 1), 64, \ |
124 | 45.6M | dst, dst_stride, filter, x0_q4, \ |
125 | 45.6M | x_step_q4, y0_q4, y_step_q4, w, h); \ |
126 | 45.6M | } else { \ |
127 | 0 | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65] VPX_UNINITIALIZED); \ |
128 | 0 | vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter, x0_q4, \ |
129 | 0 | x_step_q4, y0_q4, y_step_q4, w, h + 1); \ |
130 | 0 | vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, filter, \ |
131 | 0 | x0_q4, x_step_q4, y0_q4, y_step_q4, w, \ |
132 | 0 | h); \ |
133 | 0 | } \ |
134 | 77.0M | } Unexecuted instantiation: vpx_convolve8_sse2 Unexecuted instantiation: vpx_convolve8_avg_sse2 Line | Count | Source | 99 | 77.0M | int x_step_q4, int y0_q4, int y_step_q4, int w, int h) { \ | 100 | 77.0M | const int16_t *filter_x = filter[x0_q4]; \ | 101 | 77.0M | const int16_t *filter_y = filter[y0_q4]; \ | 102 | 77.0M | (void)filter_y; \ | 103 | 77.0M | assert(filter_x[3] != 128); \ | 104 | 77.0M | assert(filter_y[3] != 128); \ | 105 | 77.0M | assert(w <= 64); \ | 106 | 77.0M | assert(h <= 64); \ | 107 | 77.0M | assert(x_step_q4 == 16); \ | 108 | 77.0M | assert(y_step_q4 == 16); \ | 109 | 77.0M | if (filter_x[0] | filter_x[1] | filter_x[6] | filter_x[7]) { \ | 110 | 31.4M | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ | 111 | 31.4M | vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \ | 112 | 31.4M | filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, \ | 113 | 31.4M | h + 7); \ | 114 | 31.4M | vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \ | 115 | 31.4M | filter, x0_q4, x_step_q4, y0_q4, \ | 116 | 31.4M | y_step_q4, w, h); \ | 117 | 45.6M | } else if (filter_x[2] | filter_x[5]) { \ | 118 | 45.6M | const int num_taps = is_avg ? 8 : 4; \ | 119 | 45.6M | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ | 120 | 45.6M | vpx_convolve8_horiz_##opt( \ | 121 | 45.6M | src - (num_taps / 2 - 1) * src_stride, src_stride, fdata2, 64, \ | 122 | 45.6M | filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, h + num_taps - 1); \ | 123 | 45.6M | vpx_convolve8_##avg##vert_##opt(fdata2 + 64 * (num_taps / 2 - 1), 64, \ | 124 | 45.6M | dst, dst_stride, filter, x0_q4, \ | 125 | 45.6M | x_step_q4, y0_q4, y_step_q4, w, h); \ | 126 | 45.6M | } else { \ | 127 | 0 | DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65] VPX_UNINITIALIZED); \ | 128 | 0 | vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter, x0_q4, \ | 129 | 0 | x_step_q4, y0_q4, y_step_q4, w, h + 1); \ | 130 | 0 | vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, filter, \ | 131 | 0 | x0_q4, x_step_q4, y0_q4, y_step_q4, w, \ | 132 | 0 | h); \ | 133 | 0 | } \ | 134 | 77.0M | } |
Unexecuted instantiation: vpx_convolve8_avg_avx2 Unexecuted instantiation: vpx_convolve8_ssse3 Unexecuted instantiation: vpx_convolve8_avg_ssse3 |
135 | | |
136 | | #if CONFIG_VP9_HIGHBITDEPTH |
137 | | |
138 | | typedef void highbd_filter8_1dfunction(const uint16_t *src_ptr, |
139 | | const ptrdiff_t src_pitch, |
140 | | uint16_t *output_ptr, |
141 | | ptrdiff_t out_pitch, |
142 | | unsigned int output_height, |
143 | | const int16_t *filter, int bd); |
144 | | |
145 | | #define HIGH_FUN_CONV_1D(name, offset, step_q4, dir, src_start, avg, opt, \ |
146 | | is_avg) \ |
147 | | void vpx_highbd_convolve8_##name##_##opt( \ |
148 | | const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \ |
149 | | ptrdiff_t dst_stride, const InterpKernel *filter_kernel, int x0_q4, \ |
150 | 0 | int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bd) { \ |
151 | 0 | const int16_t *filter_row = filter_kernel[offset]; \ |
152 | 0 | if (step_q4 == 16 && filter_row[3] != 128) { \ |
153 | 0 | if (filter_row[0] | filter_row[1] | filter_row[6] | filter_row[7]) { \ |
154 | 0 | const int num_taps = 8; \ |
155 | 0 | while (w >= 16) { \ |
156 | 0 | vpx_highbd_filter_block1d16_##dir##8_##avg##opt( \ |
157 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
158 | 0 | src += 16; \ |
159 | 0 | dst += 16; \ |
160 | 0 | w -= 16; \ |
161 | 0 | } \ |
162 | 0 | while (w >= 8) { \ |
163 | 0 | vpx_highbd_filter_block1d8_##dir##8_##avg##opt( \ |
164 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
165 | 0 | src += 8; \ |
166 | 0 | dst += 8; \ |
167 | 0 | w -= 8; \ |
168 | 0 | } \ |
169 | 0 | while (w >= 4) { \ |
170 | 0 | vpx_highbd_filter_block1d4_##dir##8_##avg##opt( \ |
171 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
172 | 0 | src += 4; \ |
173 | 0 | dst += 4; \ |
174 | 0 | w -= 4; \ |
175 | 0 | } \ |
176 | 0 | (void)num_taps; \ |
177 | 0 | } else if (filter_row[2] | filter_row[5]) { \ |
178 | 0 | const int num_taps = is_avg ? 8 : 4; \ |
179 | 0 | while (w >= 16) { \ |
180 | 0 | vpx_highbd_filter_block1d16_##dir##4_##avg##opt( \ |
181 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
182 | 0 | src += 16; \ |
183 | 0 | dst += 16; \ |
184 | 0 | w -= 16; \ |
185 | 0 | } \ |
186 | 0 | while (w >= 8) { \ |
187 | 0 | vpx_highbd_filter_block1d8_##dir##4_##avg##opt( \ |
188 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
189 | 0 | src += 8; \ |
190 | 0 | dst += 8; \ |
191 | 0 | w -= 8; \ |
192 | 0 | } \ |
193 | 0 | while (w >= 4) { \ |
194 | 0 | vpx_highbd_filter_block1d4_##dir##4_##avg##opt( \ |
195 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
196 | 0 | src += 4; \ |
197 | 0 | dst += 4; \ |
198 | 0 | w -= 4; \ |
199 | 0 | } \ |
200 | 0 | (void)num_taps; \ |
201 | 0 | } else { \ |
202 | 0 | const int num_taps = 2; \ |
203 | 0 | while (w >= 16) { \ |
204 | 0 | vpx_highbd_filter_block1d16_##dir##2_##avg##opt( \ |
205 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
206 | 0 | src += 16; \ |
207 | 0 | dst += 16; \ |
208 | 0 | w -= 16; \ |
209 | 0 | } \ |
210 | 0 | while (w >= 8) { \ |
211 | 0 | vpx_highbd_filter_block1d8_##dir##2_##avg##opt( \ |
212 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
213 | 0 | src += 8; \ |
214 | 0 | dst += 8; \ |
215 | 0 | w -= 8; \ |
216 | 0 | } \ |
217 | 0 | while (w >= 4) { \ |
218 | 0 | vpx_highbd_filter_block1d4_##dir##2_##avg##opt( \ |
219 | 0 | src_start, src_stride, dst, dst_stride, h, filter_row, bd); \ |
220 | 0 | src += 4; \ |
221 | 0 | dst += 4; \ |
222 | 0 | w -= 4; \ |
223 | 0 | } \ |
224 | 0 | (void)num_taps; \ |
225 | 0 | } \ |
226 | 0 | } \ |
227 | 0 | if (w) { \ |
228 | 0 | vpx_highbd_convolve8_##name##_c(src, src_stride, dst, dst_stride, \ |
229 | 0 | filter_kernel, x0_q4, x_step_q4, y0_q4, \ |
230 | 0 | y_step_q4, w, h, bd); \ |
231 | 0 | } \ |
232 | 0 | } Unexecuted instantiation: vpx_highbd_convolve8_horiz_sse2 Unexecuted instantiation: vpx_highbd_convolve8_vert_sse2 Unexecuted instantiation: vpx_highbd_convolve8_avg_horiz_sse2 Unexecuted instantiation: vpx_highbd_convolve8_avg_vert_sse2 Unexecuted instantiation: vpx_highbd_convolve8_horiz_avx2 Unexecuted instantiation: vpx_highbd_convolve8_vert_avx2 Unexecuted instantiation: vpx_highbd_convolve8_avg_horiz_avx2 Unexecuted instantiation: vpx_highbd_convolve8_avg_vert_avx2 |
233 | | |
234 | | #define HIGH_FUN_CONV_2D(avg, opt, is_avg) \ |
235 | | void vpx_highbd_convolve8_##avg##opt( \ |
236 | | const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \ |
237 | | ptrdiff_t dst_stride, const InterpKernel *filter, int x0_q4, \ |
238 | 0 | int x_step_q4, int y0_q4, int y_step_q4, int w, int h, int bd) { \ |
239 | 0 | const int16_t *filter_x = filter[x0_q4]; \ |
240 | 0 | assert(w <= 64); \ |
241 | 0 | assert(h <= 64); \ |
242 | 0 | if (x_step_q4 == 16 && y_step_q4 == 16) { \ |
243 | 0 | if ((filter_x[0] | filter_x[1] | filter_x[6] | filter_x[7]) || \ |
244 | 0 | filter_x[3] == 128) { \ |
245 | 0 | DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ |
246 | 0 | vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \ |
247 | 0 | fdata2, 64, filter, x0_q4, x_step_q4, \ |
248 | 0 | y0_q4, y_step_q4, w, h + 7, bd); \ |
249 | 0 | vpx_highbd_convolve8_##avg##vert_##opt( \ |
250 | 0 | fdata2 + 192, 64, dst, dst_stride, filter, x0_q4, x_step_q4, \ |
251 | 0 | y0_q4, y_step_q4, w, h, bd); \ |
252 | 0 | } else if (filter_x[2] | filter_x[5]) { \ |
253 | 0 | const int num_taps = is_avg ? 8 : 4; \ |
254 | 0 | DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71] VPX_UNINITIALIZED); \ |
255 | 0 | vpx_highbd_convolve8_horiz_##opt( \ |
256 | 0 | src - (num_taps / 2 - 1) * src_stride, src_stride, fdata2, 64, \ |
257 | 0 | filter, x0_q4, x_step_q4, y0_q4, y_step_q4, w, h + num_taps - 1, \ |
258 | 0 | bd); \ |
259 | 0 | vpx_highbd_convolve8_##avg##vert_##opt( \ |
260 | 0 | fdata2 + 64 * (num_taps / 2 - 1), 64, dst, dst_stride, filter, \ |
261 | 0 | x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, bd); \ |
262 | 0 | } else { \ |
263 | 0 | DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65] VPX_UNINITIALIZED); \ |
264 | 0 | vpx_highbd_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter, \ |
265 | 0 | x0_q4, x_step_q4, y0_q4, y_step_q4, \ |
266 | 0 | w, h + 1, bd); \ |
267 | 0 | vpx_highbd_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \ |
268 | 0 | filter, x0_q4, x_step_q4, \ |
269 | 0 | y0_q4, y_step_q4, w, h, bd); \ |
270 | 0 | } \ |
271 | 0 | } else { \ |
272 | 0 | vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, filter, \ |
273 | 0 | x0_q4, x_step_q4, y0_q4, y_step_q4, w, h, \ |
274 | 0 | bd); \ |
275 | 0 | } \ |
276 | 0 | } Unexecuted instantiation: vpx_highbd_convolve8_sse2 Unexecuted instantiation: vpx_highbd_convolve8_avg_sse2 Unexecuted instantiation: vpx_highbd_convolve8_avx2 Unexecuted instantiation: vpx_highbd_convolve8_avg_avx2 |
277 | | |
278 | | #endif // CONFIG_VP9_HIGHBITDEPTH |
279 | | #endif // VPX_VPX_DSP_X86_CONVOLVE_H_ |