1/*
2 *  Copyright 2015 The LibYuv 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 "libyuv/scale.h"
12#include "libyuv/scale_row.h"
13
14#include "libyuv/basic_types.h"
15
16#ifdef __cplusplus
17namespace libyuv {
18extern "C" {
19#endif
20
21// Definition for ScaleFilterCols, ScaleARGBCols and ScaleARGBFilterCols
22#define CANY(NAMEANY, TERP_SIMD, TERP_C, BPP, MASK)                            \
23    void NAMEANY(uint8* dst_ptr, const uint8* src_ptr,                         \
24                 int dst_width, int x, int dx) {                               \
25      int n = dst_width & ~MASK;                                               \
26      if (n > 0) {                                                             \
27        TERP_SIMD(dst_ptr, src_ptr, n, x, dx);                                 \
28      }                                                                        \
29      TERP_C(dst_ptr + n * BPP, src_ptr,                                       \
30             dst_width & MASK, x + n * dx, dx);                                \
31    }
32
33#ifdef HAS_SCALEFILTERCOLS_NEON
34CANY(ScaleFilterCols_Any_NEON, ScaleFilterCols_NEON, ScaleFilterCols_C, 1, 7)
35#endif
36#ifdef HAS_SCALEARGBCOLS_NEON
37CANY(ScaleARGBCols_Any_NEON, ScaleARGBCols_NEON, ScaleARGBCols_C, 4, 7)
38#endif
39#ifdef HAS_SCALEARGBFILTERCOLS_NEON
40CANY(ScaleARGBFilterCols_Any_NEON, ScaleARGBFilterCols_NEON,
41     ScaleARGBFilterCols_C, 4, 3)
42#endif
43#undef CANY
44
45// Fixed scale down.
46#define SDANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK)   \
47    void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride,                   \
48                 uint8* dst_ptr, int dst_width) {                              \
49      int r = (int)((unsigned int)dst_width % (MASK + 1));                     \
50      int n = dst_width - r;                                                   \
51      if (n > 0) {                                                             \
52        SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n);                    \
53      }                                                                        \
54      SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride,                 \
55                     dst_ptr + n * BPP, r);                                    \
56    }
57
58#ifdef HAS_SCALEROWDOWN2_SSE2
59SDANY(ScaleRowDown2_Any_SSE2, ScaleRowDown2_SSE2, ScaleRowDown2_C, 2, 1, 15)
60SDANY(ScaleRowDown2Linear_Any_SSE2, ScaleRowDown2Linear_SSE2,
61      ScaleRowDown2Linear_C, 2, 1, 15)
62SDANY(ScaleRowDown2Box_Any_SSE2, ScaleRowDown2Box_SSE2, ScaleRowDown2Box_C,
63      2, 1, 15)
64#endif
65#ifdef HAS_SCALEROWDOWN2_AVX2
66SDANY(ScaleRowDown2_Any_AVX2, ScaleRowDown2_AVX2, ScaleRowDown2_C, 2, 1, 31)
67SDANY(ScaleRowDown2Linear_Any_AVX2, ScaleRowDown2Linear_AVX2,
68      ScaleRowDown2Linear_C, 2, 1, 31)
69SDANY(ScaleRowDown2Box_Any_AVX2, ScaleRowDown2Box_AVX2, ScaleRowDown2Box_C,
70      2, 1, 31)
71#endif
72#ifdef HAS_SCALEROWDOWN2_NEON
73SDANY(ScaleRowDown2_Any_NEON, ScaleRowDown2_NEON, ScaleRowDown2_C, 2, 1, 15)
74SDANY(ScaleRowDown2Linear_Any_NEON, ScaleRowDown2Linear_NEON,
75      ScaleRowDown2Linear_C, 2, 1, 15)
76SDANY(ScaleRowDown2Box_Any_NEON, ScaleRowDown2Box_NEON,
77      ScaleRowDown2Box_C, 2, 1, 15)
78#endif
79#ifdef HAS_SCALEROWDOWN4_SSE2
80SDANY(ScaleRowDown4_Any_SSE2, ScaleRowDown4_SSE2, ScaleRowDown4_C, 4, 1, 7)
81SDANY(ScaleRowDown4Box_Any_SSE2, ScaleRowDown4Box_SSE2, ScaleRowDown4Box_C,
82      4, 1, 7)
83#endif
84#ifdef HAS_SCALEROWDOWN4_AVX2
85SDANY(ScaleRowDown4_Any_AVX2, ScaleRowDown4_AVX2, ScaleRowDown4_C, 4, 1, 15)
86SDANY(ScaleRowDown4Box_Any_AVX2, ScaleRowDown4Box_AVX2, ScaleRowDown4Box_C,
87      4, 1, 15)
88#endif
89#ifdef HAS_SCALEROWDOWN4_NEON
90SDANY(ScaleRowDown4_Any_NEON, ScaleRowDown4_NEON, ScaleRowDown4_C, 4, 1, 7)
91SDANY(ScaleRowDown4Box_Any_NEON, ScaleRowDown4Box_NEON, ScaleRowDown4Box_C,
92      4, 1, 7)
93#endif
94#ifdef HAS_SCALEROWDOWN34_SSSE3
95SDANY(ScaleRowDown34_Any_SSSE3, ScaleRowDown34_SSSE3,
96      ScaleRowDown34_C, 4 / 3, 1, 23)
97SDANY(ScaleRowDown34_0_Box_Any_SSSE3, ScaleRowDown34_0_Box_SSSE3,
98      ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
99SDANY(ScaleRowDown34_1_Box_Any_SSSE3, ScaleRowDown34_1_Box_SSSE3,
100      ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
101#endif
102#ifdef HAS_SCALEROWDOWN34_NEON
103SDANY(ScaleRowDown34_Any_NEON, ScaleRowDown34_NEON,
104      ScaleRowDown34_C, 4 / 3, 1, 23)
105SDANY(ScaleRowDown34_0_Box_Any_NEON, ScaleRowDown34_0_Box_NEON,
106      ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
107SDANY(ScaleRowDown34_1_Box_Any_NEON, ScaleRowDown34_1_Box_NEON,
108      ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
109#endif
110#ifdef HAS_SCALEROWDOWN38_SSSE3
111SDANY(ScaleRowDown38_Any_SSSE3, ScaleRowDown38_SSSE3,
112      ScaleRowDown38_C, 8 / 3, 1, 11)
113SDANY(ScaleRowDown38_3_Box_Any_SSSE3, ScaleRowDown38_3_Box_SSSE3,
114      ScaleRowDown38_3_Box_C, 8 / 3, 1, 5)
115SDANY(ScaleRowDown38_2_Box_Any_SSSE3, ScaleRowDown38_2_Box_SSSE3,
116      ScaleRowDown38_2_Box_C, 8 / 3, 1, 5)
117#endif
118#ifdef HAS_SCALEROWDOWN38_NEON
119SDANY(ScaleRowDown38_Any_NEON, ScaleRowDown38_NEON,
120      ScaleRowDown38_C, 8 / 3, 1, 11)
121SDANY(ScaleRowDown38_3_Box_Any_NEON, ScaleRowDown38_3_Box_NEON,
122      ScaleRowDown38_3_Box_C, 8 / 3, 1, 11)
123SDANY(ScaleRowDown38_2_Box_Any_NEON, ScaleRowDown38_2_Box_NEON,
124      ScaleRowDown38_2_Box_C, 8 / 3, 1, 11)
125#endif
126
127#ifdef HAS_SCALEARGBROWDOWN2_SSE2
128SDANY(ScaleARGBRowDown2_Any_SSE2, ScaleARGBRowDown2_SSE2,
129      ScaleARGBRowDown2_C, 2, 4, 3)
130SDANY(ScaleARGBRowDown2Linear_Any_SSE2, ScaleARGBRowDown2Linear_SSE2,
131      ScaleARGBRowDown2Linear_C, 2, 4, 3)
132SDANY(ScaleARGBRowDown2Box_Any_SSE2, ScaleARGBRowDown2Box_SSE2,
133      ScaleARGBRowDown2Box_C, 2, 4, 3)
134#endif
135#ifdef HAS_SCALEARGBROWDOWN2_NEON
136SDANY(ScaleARGBRowDown2_Any_NEON, ScaleARGBRowDown2_NEON,
137      ScaleARGBRowDown2_C, 2, 4, 7)
138SDANY(ScaleARGBRowDown2Linear_Any_NEON, ScaleARGBRowDown2Linear_NEON,
139      ScaleARGBRowDown2Linear_C, 2, 4, 7)
140SDANY(ScaleARGBRowDown2Box_Any_NEON, ScaleARGBRowDown2Box_NEON,
141      ScaleARGBRowDown2Box_C, 2, 4, 7)
142#endif
143#undef SDANY
144
145// Scale down by even scale factor.
146#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK)          \
147    void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, int src_stepx,    \
148                 uint8* dst_ptr, int dst_width) {                              \
149      int r = (int)((unsigned int)dst_width % (MASK + 1));                     \
150      int n = dst_width - r;                                                   \
151      if (n > 0) {                                                             \
152        SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n);         \
153      }                                                                        \
154      SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride,              \
155                     src_stepx, dst_ptr + n * BPP, r);                         \
156    }
157
158#ifdef HAS_SCALEARGBROWDOWNEVEN_SSE2
159SDAANY(ScaleARGBRowDownEven_Any_SSE2, ScaleARGBRowDownEven_SSE2,
160       ScaleARGBRowDownEven_C, 4, 3)
161SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2, ScaleARGBRowDownEvenBox_SSE2,
162       ScaleARGBRowDownEvenBox_C, 4, 3)
163#endif
164#ifdef HAS_SCALEARGBROWDOWNEVEN_NEON
165SDAANY(ScaleARGBRowDownEven_Any_NEON, ScaleARGBRowDownEven_NEON,
166       ScaleARGBRowDownEven_C, 4, 3)
167SDAANY(ScaleARGBRowDownEvenBox_Any_NEON, ScaleARGBRowDownEvenBox_NEON,
168       ScaleARGBRowDownEvenBox_C, 4, 3)
169#endif
170
171// Add rows box filter scale down.
172#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK)                  \
173  void NAMEANY(const uint8* src_ptr, uint16* dst_ptr, int src_width) {         \
174      int n = src_width & ~MASK;                                               \
175      if (n > 0) {                                                             \
176        SCALEADDROW_SIMD(src_ptr, dst_ptr, n);                                 \
177      }                                                                        \
178      SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK);               \
179    }
180
181#ifdef HAS_SCALEADDROW_SSE2
182SAANY(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, ScaleAddRow_C, 15)
183#endif
184#ifdef HAS_SCALEADDROW_AVX2
185SAANY(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, ScaleAddRow_C, 31)
186#endif
187#ifdef HAS_SCALEADDROW_NEON
188SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15)
189#endif
190#undef SAANY
191
192#ifdef __cplusplus
193}  // extern "C"
194}  // namespace libyuv
195#endif
196
197
198
199
200
201