• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * jidctflt.c
3  *
4  * Copyright (C) 1994-1998, Thomas G. Lane.
5  * This file is part of the Independent JPEG Group's software.
6  *
7  * The authors make NO WARRANTY or representation, either express or implied,
8  * with respect to this software, its quality, accuracy, merchantability, or
9  * fitness for a particular purpose.  This software is provided "AS IS", and you,
10  * its user, assume the entire risk as to its quality and accuracy.
11  *
12  * This software is copyright (C) 1991-1998, Thomas G. Lane.
13  * All Rights Reserved except as specified below.
14  *
15  * Permission is hereby granted to use, copy, modify, and distribute this
16  * software (or portions thereof) for any purpose, without fee, subject to these
17  * conditions:
18  * (1) If any part of the source code for this software is distributed, then this
19  * README file must be included, with this copyright and no-warranty notice
20  * unaltered; and any additions, deletions, or changes to the original files
21  * must be clearly indicated in accompanying documentation.
22  * (2) If only executable code is distributed, then the accompanying
23  * documentation must state that "this software is based in part on the work of
24  * the Independent JPEG Group".
25  * (3) Permission for use of this software is granted only if the user accepts
26  * full responsibility for any undesirable consequences; the authors accept
27  * NO LIABILITY for damages of any kind.
28  *
29  * These conditions apply to any software derived from or based on the IJG code,
30  * not just to the unmodified library.  If you use our work, you ought to
31  * acknowledge us.
32  *
33  * Permission is NOT granted for the use of any IJG author's name or company name
34  * in advertising or publicity relating to this software or products derived from
35  * it.  This software may be referred to only as "the Independent JPEG Group's
36  * software".
37  *
38  * We specifically permit and encourage the use of this software as the basis of
39  * commercial products, provided that all warranty or liability claims are
40  * assumed by the product vendor.
41  *
42  *
43  * This file contains a floating-point implementation of the
44  * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
45  * must also perform dequantization of the input coefficients.
46  *
47  * This implementation should be more accurate than either of the integer
48  * IDCT implementations.  However, it may not give the same results on all
49  * machines because of differences in roundoff behavior.  Speed will depend
50  * on the hardware's floating point capacity.
51  *
52  * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
53  * on each row (or vice versa, but it's more convenient to emit a row at
54  * a time).  Direct algorithms are also available, but they are much more
55  * complex and seem not to be any faster when reduced to code.
56  *
57  * This implementation is based on Arai, Agui, and Nakajima's algorithm for
58  * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
59  * Japanese, but the algorithm is described in the Pennebaker & Mitchell
60  * JPEG textbook (see REFERENCES section in file README).  The following code
61  * is based directly on figure 4-8 in P&M.
62  * While an 8-point DCT cannot be done in less than 11 multiplies, it is
63  * possible to arrange the computation so that many of the multiplies are
64  * simple scalings of the final outputs.  These multiplies can then be
65  * folded into the multiplications or divisions by the JPEG quantization
66  * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
67  * to be done in the DCT itself.
68  * The primary disadvantage of this method is that with a fixed-point
69  * implementation, accuracy is lost due to imprecise representation of the
70  * scaled quantization values.  However, that problem does not arise if
71  * we use floating point arithmetic.
72 */
73 
74 #include <stdint.h>
75 #include "tinyjpeg-internal.h"
76 
77 #define FAST_FLOAT float
78 #define DCTSIZE	   8
79 #define DCTSIZE2   (DCTSIZE * DCTSIZE)
80 
81 #define DEQUANTIZE(coef, quantval)  (((FAST_FLOAT) (coef)) * (quantval))
82 
83 #if defined(__GNUC__) && (defined(__i686__) || defined(__x86_64__))
84 
descale_and_clamp(int x,int shift)85 static inline unsigned char descale_and_clamp(int x, int shift)
86 {
87 	__asm__ (
88 		"add %3,%1\n"
89 		"\tsar %2,%1\n"
90 		"\tsub $-128,%1\n"
91 		"\tcmovl %5,%1\n"	/* Use the sub to compare to 0 */
92 		"\tcmpl %4,%1\n"
93 		"\tcmovg %4,%1\n"
94 		: "=r"(x)
95 		: "0"(x), "Ic"((unsigned char)shift), "ir" (1U << (shift - 1)), "r" (0xff), "r" (0)
96 		);
97 	return x;
98 }
99 
100 #else
descale_and_clamp(int x,int shift)101 static inline unsigned char descale_and_clamp(int x, int shift)
102 {
103 	x += 1UL << (shift - 1);
104 	if (x < 0)
105 		x = (x >> shift) | ((~(0UL)) << (32 - (shift)));
106 	else
107 		x >>= shift;
108 	x += 128;
109 	if (x > 255)
110 		return 255;
111 	if (x < 0)
112 		return 0;
113 	return x;
114 }
115 #endif
116 
117 /*
118  * Perform dequantization and inverse DCT on one block of coefficients.
119  */
120 
tinyjpeg_idct_float(struct component * compptr,uint8_t * output_buf,int stride)121 void tinyjpeg_idct_float(struct component *compptr, uint8_t *output_buf, int stride)
122 {
123 	FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
124 	FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
125 	FAST_FLOAT z5, z10, z11, z12, z13;
126 	int16_t *inptr;
127 	FAST_FLOAT *quantptr;
128 	FAST_FLOAT *wsptr;
129 	uint8_t *outptr;
130 	int ctr;
131 	FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
132 
133 	/* Pass 1: process columns from input, store into work array. */
134 
135 	inptr = compptr->DCT;
136 	quantptr = compptr->Q_table;
137 	wsptr = workspace;
138 	for (ctr = DCTSIZE; ctr > 0; ctr--) {
139 		/* Due to quantization, we will usually find that many of the input
140 		 * coefficients are zero, especially the AC terms.  We can exploit this
141 		 * by short-circuiting the IDCT calculation for any column in which all
142 		 * the AC terms are zero.  In that case each output is equal to the
143 		 * DC coefficient (with scale factor as needed).
144 		 * With typical images and quantization tables, half or more of the
145 		 * column DCT calculations can be simplified this way.
146 		 */
147 
148 		if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
149 				inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
150 				inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
151 				inptr[DCTSIZE*7] == 0) {
152 			/* AC terms all zero */
153 			FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
154 
155 			wsptr[DCTSIZE*0] = dcval;
156 			wsptr[DCTSIZE*1] = dcval;
157 			wsptr[DCTSIZE*2] = dcval;
158 			wsptr[DCTSIZE*3] = dcval;
159 			wsptr[DCTSIZE*4] = dcval;
160 			wsptr[DCTSIZE*5] = dcval;
161 			wsptr[DCTSIZE*6] = dcval;
162 			wsptr[DCTSIZE*7] = dcval;
163 
164 			inptr++;			/* advance pointers to next column */
165 			quantptr++;
166 			wsptr++;
167 			continue;
168 		}
169 
170 		/* Even part */
171 
172 		tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
173 		tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
174 		tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
175 		tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
176 
177 		tmp10 = tmp0 + tmp2;	/* phase 3 */
178 		tmp11 = tmp0 - tmp2;
179 
180 		tmp13 = tmp1 + tmp3;	/* phases 5-3 */
181 		tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
182 
183 		tmp0 = tmp10 + tmp13;	/* phase 2 */
184 		tmp3 = tmp10 - tmp13;
185 		tmp1 = tmp11 + tmp12;
186 		tmp2 = tmp11 - tmp12;
187 
188 		/* Odd part */
189 
190 		tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
191 		tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
192 		tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
193 		tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
194 
195 		z13 = tmp6 + tmp5;		/* phase 6 */
196 		z10 = tmp6 - tmp5;
197 		z11 = tmp4 + tmp7;
198 		z12 = tmp4 - tmp7;
199 
200 		tmp7 = z11 + z13;		/* phase 5 */
201 		tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
202 
203 		z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
204 		tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
205 		tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
206 
207 		tmp6 = tmp12 - tmp7;	/* phase 2 */
208 		tmp5 = tmp11 - tmp6;
209 		tmp4 = tmp10 + tmp5;
210 
211 		wsptr[DCTSIZE*0] = tmp0 + tmp7;
212 		wsptr[DCTSIZE*7] = tmp0 - tmp7;
213 		wsptr[DCTSIZE*1] = tmp1 + tmp6;
214 		wsptr[DCTSIZE*6] = tmp1 - tmp6;
215 		wsptr[DCTSIZE*2] = tmp2 + tmp5;
216 		wsptr[DCTSIZE*5] = tmp2 - tmp5;
217 		wsptr[DCTSIZE*4] = tmp3 + tmp4;
218 		wsptr[DCTSIZE*3] = tmp3 - tmp4;
219 
220 		inptr++;			/* advance pointers to next column */
221 		quantptr++;
222 		wsptr++;
223 	}
224 
225 	/* Pass 2: process rows from work array, store into output array. */
226 	/* Note that we must descale the results by a factor of 8 == 2**3. */
227 
228 	wsptr = workspace;
229 	outptr = output_buf;
230 	for (ctr = 0; ctr < DCTSIZE; ctr++) {
231 		/* Rows of zeroes can be exploited in the same way as we did with columns.
232 		 * However, the column calculation has created many nonzero AC terms, so
233 		 * the simplification applies less often (typically 5% to 10% of the time).
234 		 * And testing floats for zero is relatively expensive, so we don't bother.
235 		 */
236 
237 		/* Even part */
238 
239 		tmp10 = wsptr[0] + wsptr[4];
240 		tmp11 = wsptr[0] - wsptr[4];
241 
242 		tmp13 = wsptr[2] + wsptr[6];
243 		tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
244 
245 		tmp0 = tmp10 + tmp13;
246 		tmp3 = tmp10 - tmp13;
247 		tmp1 = tmp11 + tmp12;
248 		tmp2 = tmp11 - tmp12;
249 
250 		/* Odd part */
251 
252 		z13 = wsptr[5] + wsptr[3];
253 		z10 = wsptr[5] - wsptr[3];
254 		z11 = wsptr[1] + wsptr[7];
255 		z12 = wsptr[1] - wsptr[7];
256 
257 		tmp7 = z11 + z13;
258 		tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
259 
260 		z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
261 		tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
262 		tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
263 
264 		tmp6 = tmp12 - tmp7;
265 		tmp5 = tmp11 - tmp6;
266 		tmp4 = tmp10 + tmp5;
267 
268 		/* Final output stage: scale down by a factor of 8 and range-limit */
269 
270 		outptr[0] = descale_and_clamp((int)(tmp0 + tmp7), 3);
271 		outptr[7] = descale_and_clamp((int)(tmp0 - tmp7), 3);
272 		outptr[1] = descale_and_clamp((int)(tmp1 + tmp6), 3);
273 		outptr[6] = descale_and_clamp((int)(tmp1 - tmp6), 3);
274 		outptr[2] = descale_and_clamp((int)(tmp2 + tmp5), 3);
275 		outptr[5] = descale_and_clamp((int)(tmp2 - tmp5), 3);
276 		outptr[4] = descale_and_clamp((int)(tmp3 + tmp4), 3);
277 		outptr[3] = descale_and_clamp((int)(tmp3 - tmp4), 3);
278 
279 
280 		wsptr += DCTSIZE;		/* advance pointer to next row */
281 		outptr += stride;
282 	}
283 }
284 
285