1 /******************************************************************************
2 *
3 * Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at:
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 *
17 ******************************************************************************/
18 /**
19 *******************************************************************************
20 * @file
21 * ihevc_platform_macros.h
22 *
23 * @brief
24 * Platform specific Macro definitions used in the codec
25 *
26 * @author
27 * Ittiam
28 *
29 * @remarks
30 * None
31 *
32 *******************************************************************************
33 */
34
35
36 #ifndef _IHEVC_PLATFORM_MACROS_H_
37 #define _IHEVC_PLATFORM_MACROS_H_
38
39 //#include <immintrin.h>
40
41
42 #define CLIP_U8(x) CLIP3((x), 0, 255)
43 #define CLIP_S8(x) CLIP3((x), -128, 127)
44
45 #define CLIP_U10(x) CLIP3((x), 0, 1023);
46 #define CLIP_S10(x) CLIP3((x), -512, 511);
47
48 #define CLIP_U12(x) CLIP3((x), 0, 4095);
49 #define CLIP_S12(x) CLIP3((x), -2048, 2047);
50
51 #define CLIP_U16(x) CLIP3((x), 0, 65535)
52 #define CLIP_S16(x) CLIP3((x), -32768, 32767)
53
54
55
56 #define SHL(x,y) (((y) < 32) ? ((x) << (y)) : 0)
57 #define SHR(x,y) (((y) < 32) ? ((x) >> (y)) : 0)
58
59 #define SHR_NEG(val,shift) ((shift>0)?(val>>shift):(val<<(-shift)))
60 #define SHL_NEG(val,shift) ((shift<0)?(val>>(-shift)):(val<<shift))
61
62
63 #define ITT_BIG_ENDIAN(x) ((x << 24)) | \
64 ((x & 0x0000ff00) << 8) | \
65 ((x & 0x00ff0000) >> 8) | \
66 ((UWORD32)x >> 24);
67
68
69 #define NOP(nop_cnt) {UWORD32 nop_i; for (nop_i = 0; nop_i < nop_cnt; nop_i++);}
70
71 #define POPCNT_U32(x) __builtin_popcount(x)
72
73 #define PLD(a)
74 #define INLINE inline
75
CLZ(UWORD32 u4_word)76 static INLINE UWORD32 CLZ(UWORD32 u4_word)
77 {
78 if(u4_word)
79 return (__builtin_clz(u4_word));
80 else
81 return 32;
82 }
CLZNZ(UWORD32 u4_word)83 static INLINE UWORD32 CLZNZ(UWORD32 u4_word)
84 {
85 return (__builtin_clz(u4_word));
86 }
CTZ(UWORD32 u4_word)87 static INLINE UWORD32 CTZ(UWORD32 u4_word)
88 {
89 if(0 == u4_word)
90 return 31;
91 else
92 {
93 unsigned int index;
94 index = __builtin_ctz(u4_word);
95 return (UWORD32)index;
96 }
97 }
98
99 /**
100 ******************************************************************************
101 * @brief returns postion of msb bit for 32bit input
102 ******************************************************************************
103 */
104 #define GET_POS_MSB_32(r,word) \
105 { \
106 if(word) \
107 { \
108 r = 31 - __builtin_clz(word); \
109 } \
110 else \
111 { \
112 r = -1; \
113 } \
114 }
115
116 /**
117 ******************************************************************************
118 * @brief returns postion of msb bit for 64bit input
119 ******************************************************************************
120 */
121 #define GET_POS_MSB_64(r,word) \
122 { \
123 if(word) \
124 { \
125 r = 63 - __builtin_clzll(word); \
126 } \
127 else \
128 { \
129 r = -1; \
130 } \
131 }
132
133
134 /**
135 ******************************************************************************
136 * @brief returns max number of bits required to represent input word (max 32bits)
137 ******************************************************************************
138 */
139 #define GETRANGE(r,word) \
140 { \
141 if(word) \
142 { \
143 r = 32 - __builtin_clz(word); \
144 } \
145 else \
146 { \
147 r = 1; \
148 } \
149 }
150 #define GCC_ENABLE 1
151
152 #if GCC_ENABLE
153 #define _mm256_loadu2_m128i(X,Y) _mm256_insertf128_si256(_mm256_castsi128_si256(_mm_loadu_si128((Y))), _mm_loadu_si128((X)),1);
154
155 #define _mm256_storeu2_m128i(X,Y,Z) {_mm_storeu_si128 ((Y), _mm256_castsi256_si128((Z)));_mm_storeu_si128 ((X), _mm256_extracti128_si256((Z),1));}
156
157 #define _mm256_set_m128i(X,Y) _mm256_insertf128_si256(_mm256_castsi128_si256((Y)),(X),1);
158
159 #endif
160
161
162 #define PREFETCH_ENABLE 1
163
164 #if PREFETCH_ENABLE
165 #define PREFETCH(ptr, type) _mm_prefetch(ptr, type);
166 #else
167 #define PREFETCH(ptr, type)
168 #endif
169
170 #define MEM_ALIGN8 __attribute__ ((aligned (8)))
171 #define MEM_ALIGN16 __attribute__ ((aligned (16)))
172 #define MEM_ALIGN32 __attribute__ ((aligned (32)))
173
174 #endif /* _IHEVC_PLATFORM_MACROS_H_ */
175