• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2011 Apple Inc. All rights reserved.
3  * Copyright (C) 2013-2014 Erik de Castro Lopo <erikd@mega-nerd.com>
4  *
5  * @APPLE_APACHE_LICENSE_HEADER_START@
6  *
7  * Licensed under the Apache License, Version 2.0 (the "License") ;
8  * you may not use this file except in compliance with the License.
9  * You may obtain a copy of the License at
10  *
11  *	 http://www.apache.org/licenses/LICENSE-2.0
12  *
13  * Unless required by applicable law or agreed to in writing, software
14  * distributed under the License is distributed on an "AS IS" BASIS,
15  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
16  * See the License for the specific language governing permissions and
17  * limitations under the License.
18  *
19  * @APPLE_APACHE_LICENSE_HEADER_END@
20  */
21 
22 /*
23 	File:		ag_enc.c
24 
25 	Contains:   Adaptive Golomb encode routines.
26 
27 	Copyright:	(c) 2001-2011 Apple, Inc.
28 */
29 
30 #include "aglib.h"
31 #include "ALACBitUtilities.h"
32 #include "EndianPortable.h"
33 #include "ALACAudioTypes.h"
34 
35 #include <math.h>
36 #include <stdio.h>
37 #include <stdlib.h>
38 #include <string.h>
39 
40 #define CODE_TO_LONG_MAXBITS	32
41 #define N_MAX_MEAN_CLAMP		0xffff
42 #define N_MEAN_CLAMP_VAL		0xffff
43 #define REPORT_VAL				40
44 
45 #if __GNUC__
46 #define ALWAYS_INLINE		__attribute__ ((always_inline))
47 #else
48 #define ALWAYS_INLINE
49 #endif
50 
51 
52 /*	And on the subject of the CodeWarrior x86 compiler and inlining, I reworked a lot of this
53 	to help the compiler out.   In many cases this required manual inlining or a macro.  Sorry
54 	if it is ugly but the performance gains are well worth it.
55 	- WSK 5/19/04
56 */
57 
58 // note: implementing this with some kind of "count leading zeros" assembly is a big performance win
lead(int32_t m)59 static inline int32_t lead (int32_t m)
60 {
61 	long j ;
62 	unsigned long c = (1ul << 31) ;
63 
64 	for (j = 0 ; j < 32 ; j++)
65 	{
66 		if ((c & m) != 0)
67 			break ;
68 		c >>= 1 ;
69 	}
70 	return j ;
71 }
72 
73 #define arithmin (a, b) ((a) < (b) ? (a) : (b))
74 
lg3a(int32_t x)75 static inline int32_t ALWAYS_INLINE lg3a (int32_t x)
76 {
77 	int32_t result ;
78 
79 	x += 3 ;
80 	result = lead (x) ;
81 
82 	return 31 - result ;
83 }
84 
abs_func(int32_t a)85 static inline int32_t ALWAYS_INLINE abs_func (int32_t a)
86 {
87 	// note: the CW PPC intrinsic __abs () turns into these instructions so no need to try and use it
88 	int32_t isneg = a >> 31 ;
89 	int32_t xorval = a ^ isneg ;
90 	int32_t result = xorval-isneg ;
91 
92 	return result ;
93 }
94 
95 #if PRAGMA_MARK
96 #pragma mark -
97 #endif
98 
dyn_code(int32_t m,int32_t k,int32_t n,uint32_t * outNumBits)99 static inline int32_t dyn_code (int32_t m, int32_t k, int32_t n, uint32_t *outNumBits)
100 {
101 	uint32_t 	divx, mod, de ;
102 	uint32_t	numBits ;
103 	uint32_t	value ;
104 
105 	// Assert (n >= 0) ;
106 
107 	divx = n / m ;
108 
109 	if (divx >= MAX_PREFIX_16)
110 	{
111 		numBits = MAX_PREFIX_16 + MAX_DATATYPE_BITS_16 ;
112 		value = (((1 << MAX_PREFIX_16) - 1) << MAX_DATATYPE_BITS_16) + n ;
113 	}
114 	else
115 	{
116 		mod = n%m ;
117 		de = (mod == 0) ;
118 		numBits = divx + k + 1 - de ;
119 		value = (((1 << divx) - 1) << (numBits - divx)) + mod + 1 - de ;
120 
121 		// if coding this way is bigger than doing escape, then do escape
122 		if (numBits > MAX_PREFIX_16 + MAX_DATATYPE_BITS_16)
123 		{
124 			numBits = MAX_PREFIX_16 + MAX_DATATYPE_BITS_16 ;
125 			value = (((1 << MAX_PREFIX_16) - 1) << MAX_DATATYPE_BITS_16) + n ;
126 		}
127 	}
128 
129 	*outNumBits = numBits ;
130 
131 	return (int32_t) value ;
132 }
133 
134 
dyn_code_32bit(int32_t maxbits,uint32_t m,uint32_t k,uint32_t n,uint32_t * outNumBits,uint32_t * outValue,uint32_t * overflow,uint32_t * overflowbits)135 static inline int32_t dyn_code_32bit (int32_t maxbits, uint32_t m, uint32_t k, uint32_t n, uint32_t *outNumBits, uint32_t *outValue, uint32_t *overflow, uint32_t *overflowbits)
136 {
137 	uint32_t 	divx, mod, de ;
138 	uint32_t	numBits ;
139 	uint32_t	value ;
140 	int32_t			didOverflow = 0 ;
141 
142 	divx = n / m ;
143 
144 	if (divx < MAX_PREFIX_32)
145 	{
146 		mod = n - (m * divx) ;
147 
148 		de = (mod == 0) ;
149 		numBits = divx + k + 1 - de ;
150 		value = (((1 << divx) - 1) << (numBits - divx)) + mod + 1 - de ;
151 		if (numBits > 25)
152 			goto codeasescape ;
153 	}
154 	else
155 	{
156 codeasescape:
157 		numBits = MAX_PREFIX_32 ;
158 		value = (((1 << MAX_PREFIX_32) - 1)) ;
159 		*overflow = n ;
160 		*overflowbits = maxbits ;
161 		didOverflow = 1 ;
162 	}
163 
164 	*outNumBits = numBits ;
165 	*outValue = value ;
166 
167 	return didOverflow ;
168 }
169 
170 
dyn_jam_noDeref(unsigned char * out,uint32_t bitPos,uint32_t numBits,uint32_t value)171 static inline void ALWAYS_INLINE dyn_jam_noDeref (unsigned char *out, uint32_t bitPos, uint32_t numBits, uint32_t value)
172 {
173 	uint32_t	mask ;
174 	uint32_t	curr ;
175 	uint32_t	shift ;
176 
177 	//Assert (numBits <= 32) ;
178 
179 	curr = psf_get_be32 (out, bitPos >> 3) ;
180 
181 	shift = 32 - (bitPos & 7) - numBits ;
182 
183 	mask = ~0u >> (32 - numBits) ;		// mask must be created in two steps to avoid compiler sequencing ambiguity
184 	mask <<= shift ;
185 
186 	value = (value << shift) & mask ;
187 	value |= curr & ~mask ;
188 
189 	psf_put_be32 (out, bitPos >> 3, value) ;
190 }
191 
192 
dyn_jam_noDeref_large(unsigned char * out,uint32_t bitPos,uint32_t numBits,uint32_t value)193 static inline void ALWAYS_INLINE dyn_jam_noDeref_large (unsigned char *out, uint32_t bitPos, uint32_t numBits, uint32_t value)
194 {
195 	uint32_t	w ;
196 	uint32_t	curr ;
197 	uint32_t	mask ;
198 	int32_t		shiftvalue = (32 - (bitPos & 7) - numBits) ;
199 
200 	//Assert (numBits <= 32) ;
201 
202 	curr = psf_get_be32 (out, bitPos >> 3) ;
203 
204 	if (shiftvalue < 0)
205 	{
206 		uint8_t 	tailbyte ;
207 		uint8_t 	*tailptr ;
208 
209 		w = value >> -shiftvalue ;
210 		mask = ~0u >> -shiftvalue ;
211 		w |= (curr & ~mask) ;
212 
213 		tailptr = out + (bitPos >> 3) + 4 ;
214 		tailbyte = (value << ((8+shiftvalue))) & 0xff ;
215 		*tailptr = (uint8_t) tailbyte ;
216 	}
217 	else
218 	{
219 		mask = ~0u >> (32 - numBits) ;
220 		mask <<= shiftvalue ;			// mask must be created in two steps to avoid compiler sequencing ambiguity
221 
222 		w = (value << shiftvalue) & mask ;
223 		w |= curr & ~mask ;
224 	}
225 
226 	psf_put_be32 (out, bitPos >> 3, w) ;
227 }
228 
229 
dyn_comp(AGParamRecPtr params,int32_t * pc,BitBuffer * bitstream,int32_t numSamples,int32_t bitSize,uint32_t * outNumBits)230 int32_t dyn_comp (AGParamRecPtr params, int32_t * pc, BitBuffer * bitstream, int32_t numSamples, int32_t bitSize, uint32_t * outNumBits)
231 {
232 	unsigned char *		out ;
233 	uint32_t		bitPos, startPos ;
234 	uint32_t			m, k, n, c, mz, nz ;
235 	uint32_t		numBits ;
236 	uint32_t			value ;
237 	int32_t				del, zmode ;
238 	uint32_t		overflow, overflowbits ;
239 	int32_t					status ;
240 
241 	// shadow the variables in params so there's not the dereferencing overhead
242 	uint32_t		mb, pb, kb, wb ;
243 	int32_t					rowPos = 0 ;
244 	int32_t					rowSize = params->sw ;
245 	int32_t					rowJump = (params->fw) - rowSize ;
246 	int32_t *			inPtr = pc ;
247 
248 	*outNumBits = 0 ;
249 	RequireAction ((bitSize >= 1) && (bitSize <= 32), return kALAC_ParamError ;) ;
250 
251 	out = bitstream->cur ;
252 	startPos = bitstream->bitIndex ;
253 	bitPos = startPos ;
254 
255 	mb = params->mb = params->mb0 ;
256 	pb = params->pb ;
257 	kb = params->kb ;
258 	wb = params->wb ;
259 	zmode = 0 ;
260 
261 	c = 0 ;
262 	status = ALAC_noErr ;
263 
264 	while (c < (uint32_t) numSamples)
265 	{
266 		m = mb >> QBSHIFT ;
267 		k = lg3a (m) ;
268 		if (k > kb)
269 		{
270 			k = kb ;
271 		}
272 		m = (1 << k) - 1 ;
273 
274 		del = *inPtr++ ;
275 		rowPos++ ;
276 
277 		n = (abs_func (del) << 1) - ((del >> 31) & 1) - zmode ;
278 		//Assert (32-lead (n) <= bitSize) ;
279 
280 		if (dyn_code_32bit (bitSize, m, k, n, &numBits, &value, &overflow, &overflowbits))
281 		{
282 			dyn_jam_noDeref (out, bitPos, numBits, value) ;
283 			bitPos += numBits ;
284 			dyn_jam_noDeref_large (out, bitPos, overflowbits, overflow) ;
285 			bitPos += overflowbits ;
286 		}
287 		else
288 		{
289 			dyn_jam_noDeref (out, bitPos, numBits, value) ;
290 			bitPos += numBits ;
291 		}
292 
293 		c++ ;
294 		if (rowPos >= rowSize)
295 		{
296 			rowPos = 0 ;
297 			inPtr += rowJump ;
298 		}
299 
300 		mb = pb * (n + zmode) + mb - ((pb * mb) >> QBSHIFT) ;
301 
302 		// update mean tracking if it's overflowed
303 		if (n > N_MAX_MEAN_CLAMP)
304 			mb = N_MEAN_CLAMP_VAL ;
305 
306 		zmode = 0 ;
307 
308 		RequireAction (c <= (uint32_t) numSamples, status = kALAC_ParamError ; goto Exit ;) ;
309 
310 		if (((mb << MMULSHIFT) < QB) && (c < (uint32_t) numSamples))
311 		{
312 			zmode = 1 ;
313 			nz = 0 ;
314 
315 			while (c < (uint32_t) numSamples && *inPtr == 0)
316 			{
317 				/* Take care of wrap-around globals. */
318 				++inPtr ;
319 				++nz ;
320 				++c ;
321 				if (++rowPos >= rowSize)
322 				{
323 					rowPos = 0 ;
324 					inPtr += rowJump ;
325 				}
326 
327 				if (nz >= 65535)
328 				{
329 					zmode = 0 ;
330 					break ;
331 				}
332 			}
333 
334 			k = lead (mb) - BITOFF + ((mb + MOFF) >> MDENSHIFT) ;
335 			mz = ((1 << k) - 1) & wb ;
336 
337 			value = dyn_code (mz, k, nz, &numBits) ;
338 			dyn_jam_noDeref (out, bitPos, numBits, value) ;
339 			bitPos += numBits ;
340 
341 			mb = 0 ;
342 		}
343 	}
344 
345 	*outNumBits = (bitPos - startPos) ;
346 	BitBufferAdvance (bitstream, *outNumBits) ;
347 
348 Exit:
349 	return status ;
350 }
351