3  * Copyright (C) 2012-2014 Erik de Castro Lopo <erikd@mega-nerd.com>
11  *     http://www.apache.org/licenses/LICENSE-2.0
27 	Copyright:	(c) 2004-2011 Apple, Inc.
34 // up to 24-bit "offset" macros for the individual bytes of a 20/24-bit word
46     There is no plain middle-side option ; instead there are various mixing
47     modes including middle-side, each lossless, as embodied in the mix ()
48     and unmix () functions.  These functions exploit a generalized middle-side
51     u := [(rL + (m-r)R)/m] ;
52     v := L - R ;
56     L = u + v - [rV/m] ;
57     R = L - v ;
60 // 16-bit routines
65 	int32_t 	j ;  in unmix16()  local
70 		for (j = 0 ; j < numSamples ; j++)  in unmix16()
74 			l = u [j] + v [j] - ((mixres * v [j]) >> mixbits) ;  in unmix16()
75 			r = l - v [j] ;  in unmix16()
85 		for (j = 0 ; j < numSamples ; j++)  in unmix16()
87 			out [0] = u [j] << 16 ;  in unmix16()
88 			out [1] = v [j] << 16 ;  in unmix16()
94 // 20-bit routines
95 // - the 20 bits of data are left-justified in 3 bytes of storage but right-aligned for input/outpu…
100 	int32_t 	j ;  in unmix20()  local
105 		for (j = 0 ; j < numSamples ; j++)  in unmix20()
109 			l = u [j] + v [j] - ((mixres * v [j]) >> mixbits) ;  in unmix20()
110 			r = l - v [j] ;  in unmix20()
120 		for (j = 0 ; j < numSamples ; j++)  in unmix20()
122 			out [0] = arith_shift_left (u [j], 12) ;  in unmix20()
123 			out [1] = arith_shift_left (v [j], 12) ;  in unmix20()
129 // 24-bit routines
130 // - the 24 bits of data are right-justified in the input/output predictor buffers
138 	int32_t 		j, k ;  in unmix24()  local
145 			for (j = 0, k = 0 ; j < numSamples ; j++, k += 2)  in unmix24()
147 				l = u [j] + v [j] - ((mixres * v [j]) >> mixbits) ;  in unmix24()
148 				r = l - v [j] ;  in unmix24()
160 			for (j = 0 ; j < numSamples ; j++)  in unmix24()
162 				l = u [j] + v [j] - ((mixres * v [j]) >> mixbits) ;  in unmix24()
163 				r = l - v [j] ;  in unmix24()
176 			for (j = 0, k = 0 ; j < numSamples ; j++, k += 2)  in unmix24()
178 				l = u [j] ;  in unmix24()
179 				r = v [j] ;  in unmix24()
191 			for (j = 0 ; j < numSamples ; j++)  in unmix24()
193 				out [0] = u [j] << 8 ;  in unmix24()
194 				out [1] = v [j] << 8 ;  in unmix24()
201 // 32-bit routines
202 // - note that these really expect the internal data width to be < 32 but the arrays are 32-bit
203 // - otherwise, the calculations might overflow into the 33rd bit and be lost
204 // - therefore, these routines deal with the specified "unused lower" bytes in the "shift" buffers
212 	int32_t 	j, k ;  in unmix32()  local
219 		for (j = 0, k = 0 ; j < numSamples ; j++, k += 2)  in unmix32()
223 			lt = u [j] ;  in unmix32()
224 			rt = v [j] ;  in unmix32()
226 			l = lt + rt - ((mixres * rt) >> mixbits) ;  in unmix32()
227 			r = l - rt ;  in unmix32()
239 			for (j = 0 ; j < numSamples ; j++)  in unmix32()
241 				out [0] = u [j] ;  in unmix32()
242 				out [1] = v [j] ;  in unmix32()
249 			for (j = 0, k = 0 ; j < numSamples ; j++, k += 2)  in unmix32()
251 				out [0] = (u [j] << shift) | (uint32_t) shiftUV [k + 0] ;  in unmix32()
252 				out [1] = (v [j] << shift) | (uint32_t) shiftUV [k + 1] ;  in unmix32()
259 // 20/24-bit <-> 32-bit helper routines (not really matrixing but convenient to put here)
264 	int32_t		j ;  in copyPredictorTo24()  local
266 	for (j = 0 ; j < numSamples ; j++)  in copyPredictorTo24()
268 		out [0] = in [j] << 8 ;  in copyPredictorTo24()
277 	int32_t		j ;  in copyPredictorTo24Shift()  local
281 	for (j = 0 ; j < numSamples ; j++)  in copyPredictorTo24Shift()
283 		int32_t		val = in [j] ;  in copyPredictorTo24Shift()
285 		val = arith_shift_left (val, shiftVal) | (uint32_t) shift [j] ;  in copyPredictorTo24Shift()
294 	int32_t		j ;  in copyPredictorTo20()  local
296 	// 32-bit predictor values are right-aligned but 20-bit output values should be left-aligned  in copyPredictorTo20()
297 	// in the 24-bit output buffer  in copyPredictorTo20()
298 	for (j = 0 ; j < numSamples ; j++)  in copyPredictorTo20()
300 		out [0] = arith_shift_left (in [j], 12) ;  in copyPredictorTo20()
308 	int32_t			i, j ;  in copyPredictorTo32()  local
310 	// this is only a subroutine to abstract the "iPod can only output 16-bit data" problem  in copyPredictorTo32()
311 	for (i = 0, j = 0 ; i < numSamples ; i++, j += stride)  in copyPredictorTo32()
312 		out [j] = arith_shift_left (in [i], 8) ;  in copyPredictorTo32()
320 	int32_t			j ;  in copyPredictorTo32Shift()  local
324 	// this is only a subroutine to abstract the "iPod can only output 16-bit data" problem  in copyPredictorTo32Shift()
325 	for (j = 0 ; j < numSamples ; j++)  in copyPredictorTo32Shift()
327 		op [0] = arith_shift_left (in [j], shiftVal) | (uint32_t) shift [j] ;  in copyPredictorTo32Shift()