OLD | NEW |
1 /* | 1 /* |
2 * jidctred.c | 2 * jidctred.c |
3 * | 3 * |
| 4 * This file was part of the Independent JPEG Group's software. |
4 * Copyright (C) 1994-1998, Thomas G. Lane. | 5 * Copyright (C) 1994-1998, Thomas G. Lane. |
5 * This file is part of the Independent JPEG Group's software. | 6 * libjpeg-turbo Modifications: |
6 * For conditions of distribution and use, see the accompanying README file. | 7 * Copyright (C) 2015, D. R. Commander |
| 8 * For conditions of distribution and use, see the accompanying README.ijg |
| 9 * file. |
7 * | 10 * |
8 * This file contains inverse-DCT routines that produce reduced-size output: | 11 * This file contains inverse-DCT routines that produce reduced-size output: |
9 * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block. | 12 * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block. |
10 * | 13 * |
11 * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M) | 14 * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M) |
12 * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step | 15 * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step |
13 * with an 8-to-4 step that produces the four averages of two adjacent outputs | 16 * with an 8-to-4 step that produces the four averages of two adjacent outputs |
14 * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output). | 17 * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output). |
15 * These steps were derived by computing the corresponding values at the end | 18 * These steps were derived by computing the corresponding values at the end |
16 * of the normal LL&M code, then simplifying as much as possible. | 19 * of the normal LL&M code, then simplifying as much as possible. |
17 * | 20 * |
18 * 1x1 is trivial: just take the DC coefficient divided by 8. | 21 * 1x1 is trivial: just take the DC coefficient divided by 8. |
19 * | 22 * |
20 * See jidctint.c for additional comments. | 23 * See jidctint.c for additional comments. |
21 */ | 24 */ |
22 | 25 |
23 #define JPEG_INTERNALS | 26 #define JPEG_INTERNALS |
24 #include "jinclude.h" | 27 #include "jinclude.h" |
25 #include "jpeglib.h" | 28 #include "jpeglib.h" |
26 #include "jdct.h"» » /* Private declarations for DCT subsystem */ | 29 #include "jdct.h" /* Private declarations for DCT subsystem */ |
27 | 30 |
28 #ifdef IDCT_SCALING_SUPPORTED | 31 #ifdef IDCT_SCALING_SUPPORTED |
29 | 32 |
30 | 33 |
31 /* | 34 /* |
32 * This module is specialized to the case DCTSIZE = 8. | 35 * This module is specialized to the case DCTSIZE = 8. |
33 */ | 36 */ |
34 | 37 |
35 #if DCTSIZE != 8 | 38 #if DCTSIZE != 8 |
36 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ | 39 Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ |
37 #endif | 40 #endif |
38 | 41 |
39 | 42 |
40 /* Scaling is the same as in jidctint.c. */ | 43 /* Scaling is the same as in jidctint.c. */ |
41 | 44 |
42 #if BITS_IN_JSAMPLE == 8 | 45 #if BITS_IN_JSAMPLE == 8 |
43 #define CONST_BITS 13 | 46 #define CONST_BITS 13 |
44 #define PASS1_BITS 2 | 47 #define PASS1_BITS 2 |
45 #else | 48 #else |
46 #define CONST_BITS 13 | 49 #define CONST_BITS 13 |
47 #define PASS1_BITS 1» » /* lose a little precision to avoid overflow */ | 50 #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ |
48 #endif | 51 #endif |
49 | 52 |
50 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus | 53 /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus |
51 * causing a lot of useless floating-point operations at run time. | 54 * causing a lot of useless floating-point operations at run time. |
52 * To get around this we use the following pre-calculated constants. | 55 * To get around this we use the following pre-calculated constants. |
53 * If you change CONST_BITS you may want to add appropriate values. | 56 * If you change CONST_BITS you may want to add appropriate values. |
54 * (With a reasonable C compiler, you can just rely on the FIX() macro...) | 57 * (With a reasonable C compiler, you can just rely on the FIX() macro...) |
55 */ | 58 */ |
56 | 59 |
57 #if CONST_BITS == 13 | 60 #if CONST_BITS == 13 |
58 #define FIX_0_211164243 ((INT32) 1730)» /* FIX(0.211164243) */ | 61 #define FIX_0_211164243 ((JLONG) 1730) /* FIX(0.211164243) */ |
59 #define FIX_0_509795579 ((INT32) 4176)» /* FIX(0.509795579) */ | 62 #define FIX_0_509795579 ((JLONG) 4176) /* FIX(0.509795579) */ |
60 #define FIX_0_601344887 ((INT32) 4926)» /* FIX(0.601344887) */ | 63 #define FIX_0_601344887 ((JLONG) 4926) /* FIX(0.601344887) */ |
61 #define FIX_0_720959822 ((INT32) 5906)» /* FIX(0.720959822) */ | 64 #define FIX_0_720959822 ((JLONG) 5906) /* FIX(0.720959822) */ |
62 #define FIX_0_765366865 ((INT32) 6270)» /* FIX(0.765366865) */ | 65 #define FIX_0_765366865 ((JLONG) 6270) /* FIX(0.765366865) */ |
63 #define FIX_0_850430095 ((INT32) 6967)» /* FIX(0.850430095) */ | 66 #define FIX_0_850430095 ((JLONG) 6967) /* FIX(0.850430095) */ |
64 #define FIX_0_899976223 ((INT32) 7373)» /* FIX(0.899976223) */ | 67 #define FIX_0_899976223 ((JLONG) 7373) /* FIX(0.899976223) */ |
65 #define FIX_1_061594337 ((INT32) 8697)» /* FIX(1.061594337) */ | 68 #define FIX_1_061594337 ((JLONG) 8697) /* FIX(1.061594337) */ |
66 #define FIX_1_272758580 ((INT32) 10426)» /* FIX(1.272758580) */ | 69 #define FIX_1_272758580 ((JLONG) 10426) /* FIX(1.272758580) */ |
67 #define FIX_1_451774981 ((INT32) 11893)» /* FIX(1.451774981) */ | 70 #define FIX_1_451774981 ((JLONG) 11893) /* FIX(1.451774981) */ |
68 #define FIX_1_847759065 ((INT32) 15137)» /* FIX(1.847759065) */ | 71 #define FIX_1_847759065 ((JLONG) 15137) /* FIX(1.847759065) */ |
69 #define FIX_2_172734803 ((INT32) 17799)» /* FIX(2.172734803) */ | 72 #define FIX_2_172734803 ((JLONG) 17799) /* FIX(2.172734803) */ |
70 #define FIX_2_562915447 ((INT32) 20995)» /* FIX(2.562915447) */ | 73 #define FIX_2_562915447 ((JLONG) 20995) /* FIX(2.562915447) */ |
71 #define FIX_3_624509785 ((INT32) 29692)» /* FIX(3.624509785) */ | 74 #define FIX_3_624509785 ((JLONG) 29692) /* FIX(3.624509785) */ |
72 #else | 75 #else |
73 #define FIX_0_211164243 FIX(0.211164243) | 76 #define FIX_0_211164243 FIX(0.211164243) |
74 #define FIX_0_509795579 FIX(0.509795579) | 77 #define FIX_0_509795579 FIX(0.509795579) |
75 #define FIX_0_601344887 FIX(0.601344887) | 78 #define FIX_0_601344887 FIX(0.601344887) |
76 #define FIX_0_720959822 FIX(0.720959822) | 79 #define FIX_0_720959822 FIX(0.720959822) |
77 #define FIX_0_765366865 FIX(0.765366865) | 80 #define FIX_0_765366865 FIX(0.765366865) |
78 #define FIX_0_850430095 FIX(0.850430095) | 81 #define FIX_0_850430095 FIX(0.850430095) |
79 #define FIX_0_899976223 FIX(0.899976223) | 82 #define FIX_0_899976223 FIX(0.899976223) |
80 #define FIX_1_061594337 FIX(1.061594337) | 83 #define FIX_1_061594337 FIX(1.061594337) |
81 #define FIX_1_272758580 FIX(1.272758580) | 84 #define FIX_1_272758580 FIX(1.272758580) |
82 #define FIX_1_451774981 FIX(1.451774981) | 85 #define FIX_1_451774981 FIX(1.451774981) |
83 #define FIX_1_847759065 FIX(1.847759065) | 86 #define FIX_1_847759065 FIX(1.847759065) |
84 #define FIX_2_172734803 FIX(2.172734803) | 87 #define FIX_2_172734803 FIX(2.172734803) |
85 #define FIX_2_562915447 FIX(2.562915447) | 88 #define FIX_2_562915447 FIX(2.562915447) |
86 #define FIX_3_624509785 FIX(3.624509785) | 89 #define FIX_3_624509785 FIX(3.624509785) |
87 #endif | 90 #endif |
88 | 91 |
89 | 92 |
90 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. | 93 /* Multiply a JLONG variable by a JLONG constant to yield a JLONG result. |
91 * For 8-bit samples with the recommended scaling, all the variable | 94 * For 8-bit samples with the recommended scaling, all the variable |
92 * and constant values involved are no more than 16 bits wide, so a | 95 * and constant values involved are no more than 16 bits wide, so a |
93 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. | 96 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. |
94 * For 12-bit samples, a full 32-bit multiplication will be needed. | 97 * For 12-bit samples, a full 32-bit multiplication will be needed. |
95 */ | 98 */ |
96 | 99 |
97 #if BITS_IN_JSAMPLE == 8 | 100 #if BITS_IN_JSAMPLE == 8 |
98 #define MULTIPLY(var,const) MULTIPLY16C16(var,const) | 101 #define MULTIPLY(var,const) MULTIPLY16C16(var,const) |
99 #else | 102 #else |
100 #define MULTIPLY(var,const) ((var) * (const)) | 103 #define MULTIPLY(var,const) ((var) * (const)) |
101 #endif | 104 #endif |
102 | 105 |
103 | 106 |
104 /* Dequantize a coefficient by multiplying it by the multiplier-table | 107 /* Dequantize a coefficient by multiplying it by the multiplier-table |
105 * entry; produce an int result. In this module, both inputs and result | 108 * entry; produce an int result. In this module, both inputs and result |
106 * are 16 bits or less, so either int or short multiply will work. | 109 * are 16 bits or less, so either int or short multiply will work. |
107 */ | 110 */ |
108 | 111 |
109 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval)) | 112 #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval)) |
110 | 113 |
111 | 114 |
112 /* | 115 /* |
113 * Perform dequantization and inverse DCT on one block of coefficients, | 116 * Perform dequantization and inverse DCT on one block of coefficients, |
114 * producing a reduced-size 4x4 output block. | 117 * producing a reduced-size 4x4 output block. |
115 */ | 118 */ |
116 | 119 |
117 GLOBAL(void) | 120 GLOBAL(void) |
118 jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr, | 121 jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info *compptr, |
119 » JCOEFPTR coef_block, | 122 JCOEFPTR coef_block, |
120 » JSAMPARRAY output_buf, JDIMENSION output_col) | 123 JSAMPARRAY output_buf, JDIMENSION output_col) |
121 { | 124 { |
122 INT32 tmp0, tmp2, tmp10, tmp12; | 125 JLONG tmp0, tmp2, tmp10, tmp12; |
123 INT32 z1, z2, z3, z4; | 126 JLONG z1, z2, z3, z4; |
124 JCOEFPTR inptr; | 127 JCOEFPTR inptr; |
125 ISLOW_MULT_TYPE * quantptr; | 128 ISLOW_MULT_TYPE *quantptr; |
126 int * wsptr; | 129 int *wsptr; |
127 JSAMPROW outptr; | 130 JSAMPROW outptr; |
128 JSAMPLE *range_limit = IDCT_range_limit(cinfo); | 131 JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
129 int ctr; | 132 int ctr; |
130 int workspace[DCTSIZE*4];» /* buffers data between passes */ | 133 int workspace[DCTSIZE*4]; /* buffers data between passes */ |
131 SHIFT_TEMPS | 134 SHIFT_TEMPS |
132 | 135 |
133 /* Pass 1: process columns from input, store into work array. */ | 136 /* Pass 1: process columns from input, store into work array. */ |
134 | 137 |
135 inptr = coef_block; | 138 inptr = coef_block; |
136 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; | 139 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
137 wsptr = workspace; | 140 wsptr = workspace; |
138 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { | 141 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { |
139 /* Don't bother to process column 4, because second pass won't use it */ | 142 /* Don't bother to process column 4, because second pass won't use it */ |
140 if (ctr == DCTSIZE-4) | 143 if (ctr == DCTSIZE-4) |
141 continue; | 144 continue; |
142 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 && | 145 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 && |
143 » inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 && | 146 inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 && |
144 » inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) { | 147 inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) { |
145 /* AC terms all zero; we need not examine term 4 for 4x4 output */ | 148 /* AC terms all zero; we need not examine term 4 for 4x4 output */ |
146 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BIT
S; | 149 int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]), |
147 | 150 PASS1_BITS); |
| 151 |
148 wsptr[DCTSIZE*0] = dcval; | 152 wsptr[DCTSIZE*0] = dcval; |
149 wsptr[DCTSIZE*1] = dcval; | 153 wsptr[DCTSIZE*1] = dcval; |
150 wsptr[DCTSIZE*2] = dcval; | 154 wsptr[DCTSIZE*2] = dcval; |
151 wsptr[DCTSIZE*3] = dcval; | 155 wsptr[DCTSIZE*3] = dcval; |
152 | 156 |
153 continue; | 157 continue; |
154 } | 158 } |
155 | 159 |
156 /* Even part */ | 160 /* Even part */ |
157 | 161 |
158 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); | 162 tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); |
159 tmp0 <<= (CONST_BITS+1); | 163 tmp0 = LEFT_SHIFT(tmp0, CONST_BITS+1); |
160 | 164 |
161 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]); | 165 z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]); |
162 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]); | 166 z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]); |
163 | 167 |
164 tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865); | 168 tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865); |
165 | 169 |
166 tmp10 = tmp0 + tmp2; | 170 tmp10 = tmp0 + tmp2; |
167 tmp12 = tmp0 - tmp2; | 171 tmp12 = tmp0 - tmp2; |
168 | 172 |
169 /* Odd part */ | 173 /* Odd part */ |
170 | 174 |
171 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); | 175 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); |
172 z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); | 176 z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); |
173 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); | 177 z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); |
174 z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); | 178 z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); |
175 | 179 |
176 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ | 180 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ |
177 » + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ | 181 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ |
178 » + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ | 182 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ |
179 » + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ | 183 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ |
180 | 184 |
181 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ | 185 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ |
182 » + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ | 186 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ |
183 » + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ | 187 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ |
184 » + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ | 188 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ |
185 | 189 |
186 /* Final output stage */ | 190 /* Final output stage */ |
187 | 191 |
188 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1); | 192 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1); |
189 wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1); | 193 wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1); |
190 wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1); | 194 wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1); |
191 wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1); | 195 wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1); |
192 } | 196 } |
193 | 197 |
194 /* Pass 2: process 4 rows from work array, store into output array. */ | 198 /* Pass 2: process 4 rows from work array, store into output array. */ |
195 | 199 |
196 wsptr = workspace; | 200 wsptr = workspace; |
197 for (ctr = 0; ctr < 4; ctr++) { | 201 for (ctr = 0; ctr < 4; ctr++) { |
198 outptr = output_buf[ctr] + output_col; | 202 outptr = output_buf[ctr] + output_col; |
199 /* It's not clear whether a zero row test is worthwhile here ... */ | 203 /* It's not clear whether a zero row test is worthwhile here ... */ |
200 | 204 |
201 #ifndef NO_ZERO_ROW_TEST | 205 #ifndef NO_ZERO_ROW_TEST |
202 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && | 206 if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && |
203 » wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { | 207 wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { |
204 /* AC terms all zero */ | 208 /* AC terms all zero */ |
205 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) | 209 JSAMPLE dcval = range_limit[(int) DESCALE((JLONG) wsptr[0], PASS1_BITS+3) |
206 » » » » & RANGE_MASK]; | 210 & RANGE_MASK]; |
207 | 211 |
208 outptr[0] = dcval; | 212 outptr[0] = dcval; |
209 outptr[1] = dcval; | 213 outptr[1] = dcval; |
210 outptr[2] = dcval; | 214 outptr[2] = dcval; |
211 outptr[3] = dcval; | 215 outptr[3] = dcval; |
212 | 216 |
213 wsptr += DCTSIZE;»» /* advance pointer to next row */ | 217 wsptr += DCTSIZE; /* advance pointer to next row */ |
214 continue; | 218 continue; |
215 } | 219 } |
216 #endif | 220 #endif |
217 | 221 |
218 /* Even part */ | 222 /* Even part */ |
219 | 223 |
220 tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1); | 224 tmp0 = LEFT_SHIFT((JLONG) wsptr[0], CONST_BITS+1); |
221 | 225 |
222 tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065) | 226 tmp2 = MULTIPLY((JLONG) wsptr[2], FIX_1_847759065) |
223 » + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865); | 227 + MULTIPLY((JLONG) wsptr[6], - FIX_0_765366865); |
224 | 228 |
225 tmp10 = tmp0 + tmp2; | 229 tmp10 = tmp0 + tmp2; |
226 tmp12 = tmp0 - tmp2; | 230 tmp12 = tmp0 - tmp2; |
227 | 231 |
228 /* Odd part */ | 232 /* Odd part */ |
229 | 233 |
230 z1 = (INT32) wsptr[7]; | 234 z1 = (JLONG) wsptr[7]; |
231 z2 = (INT32) wsptr[5]; | 235 z2 = (JLONG) wsptr[5]; |
232 z3 = (INT32) wsptr[3]; | 236 z3 = (JLONG) wsptr[3]; |
233 z4 = (INT32) wsptr[1]; | 237 z4 = (JLONG) wsptr[1]; |
234 | 238 |
235 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ | 239 tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ |
236 » + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ | 240 + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ |
237 » + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ | 241 + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ |
238 » + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ | 242 + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ |
239 | 243 |
240 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ | 244 tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ |
241 » + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ | 245 + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ |
242 » + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ | 246 + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ |
243 » + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ | 247 + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ |
244 | 248 |
245 /* Final output stage */ | 249 /* Final output stage */ |
246 | 250 |
247 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2, | 251 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2, |
248 » » » » » CONST_BITS+PASS1_BITS+3+1) | 252 CONST_BITS+PASS1_BITS+3+1) |
249 » » » & RANGE_MASK]; | 253 & RANGE_MASK]; |
250 outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2, | 254 outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2, |
251 » » » » » CONST_BITS+PASS1_BITS+3+1) | 255 CONST_BITS+PASS1_BITS+3+1) |
252 » » » & RANGE_MASK]; | 256 & RANGE_MASK]; |
253 outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0, | 257 outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0, |
254 » » » » » CONST_BITS+PASS1_BITS+3+1) | 258 CONST_BITS+PASS1_BITS+3+1) |
255 » » » & RANGE_MASK]; | 259 & RANGE_MASK]; |
256 outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0, | 260 outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0, |
257 » » » » » CONST_BITS+PASS1_BITS+3+1) | 261 CONST_BITS+PASS1_BITS+3+1) |
258 » » » & RANGE_MASK]; | 262 & RANGE_MASK]; |
259 | 263 |
260 wsptr += DCTSIZE;» » /* advance pointer to next row */ | 264 wsptr += DCTSIZE; /* advance pointer to next row */ |
261 } | 265 } |
262 } | 266 } |
263 | 267 |
264 | 268 |
265 /* | 269 /* |
266 * Perform dequantization and inverse DCT on one block of coefficients, | 270 * Perform dequantization and inverse DCT on one block of coefficients, |
267 * producing a reduced-size 2x2 output block. | 271 * producing a reduced-size 2x2 output block. |
268 */ | 272 */ |
269 | 273 |
270 GLOBAL(void) | 274 GLOBAL(void) |
271 jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, | 275 jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info *compptr, |
272 » JCOEFPTR coef_block, | 276 JCOEFPTR coef_block, |
273 » JSAMPARRAY output_buf, JDIMENSION output_col) | 277 JSAMPARRAY output_buf, JDIMENSION output_col) |
274 { | 278 { |
275 INT32 tmp0, tmp10, z1; | 279 JLONG tmp0, tmp10, z1; |
276 JCOEFPTR inptr; | 280 JCOEFPTR inptr; |
277 ISLOW_MULT_TYPE * quantptr; | 281 ISLOW_MULT_TYPE *quantptr; |
278 int * wsptr; | 282 int *wsptr; |
279 JSAMPROW outptr; | 283 JSAMPROW outptr; |
280 JSAMPLE *range_limit = IDCT_range_limit(cinfo); | 284 JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
281 int ctr; | 285 int ctr; |
282 int workspace[DCTSIZE*2];» /* buffers data between passes */ | 286 int workspace[DCTSIZE*2]; /* buffers data between passes */ |
283 SHIFT_TEMPS | 287 SHIFT_TEMPS |
284 | 288 |
285 /* Pass 1: process columns from input, store into work array. */ | 289 /* Pass 1: process columns from input, store into work array. */ |
286 | 290 |
287 inptr = coef_block; | 291 inptr = coef_block; |
288 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; | 292 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
289 wsptr = workspace; | 293 wsptr = workspace; |
290 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { | 294 for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { |
291 /* Don't bother to process columns 2,4,6 */ | 295 /* Don't bother to process columns 2,4,6 */ |
292 if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6) | 296 if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6) |
293 continue; | 297 continue; |
294 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 && | 298 if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 && |
295 » inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) { | 299 inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) { |
296 /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */ | 300 /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */ |
297 int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BIT
S; | 301 int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]), |
298 | 302 PASS1_BITS); |
| 303 |
299 wsptr[DCTSIZE*0] = dcval; | 304 wsptr[DCTSIZE*0] = dcval; |
300 wsptr[DCTSIZE*1] = dcval; | 305 wsptr[DCTSIZE*1] = dcval; |
301 | 306 |
302 continue; | 307 continue; |
303 } | 308 } |
304 | 309 |
305 /* Even part */ | 310 /* Even part */ |
306 | 311 |
307 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); | 312 z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); |
308 tmp10 = z1 << (CONST_BITS+2); | 313 tmp10 = LEFT_SHIFT(z1, CONST_BITS+2); |
309 | 314 |
310 /* Odd part */ | 315 /* Odd part */ |
311 | 316 |
312 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); | 317 z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); |
313 tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */ | 318 tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */ |
314 z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); | 319 z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); |
315 tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */ | 320 tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */ |
316 z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); | 321 z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); |
317 tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */ | 322 tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */ |
318 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); | 323 z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); |
319 tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ | 324 tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ |
320 | 325 |
321 /* Final output stage */ | 326 /* Final output stage */ |
322 | 327 |
323 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2); | 328 wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2); |
324 wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2); | 329 wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2); |
325 } | 330 } |
326 | 331 |
327 /* Pass 2: process 2 rows from work array, store into output array. */ | 332 /* Pass 2: process 2 rows from work array, store into output array. */ |
328 | 333 |
329 wsptr = workspace; | 334 wsptr = workspace; |
330 for (ctr = 0; ctr < 2; ctr++) { | 335 for (ctr = 0; ctr < 2; ctr++) { |
331 outptr = output_buf[ctr] + output_col; | 336 outptr = output_buf[ctr] + output_col; |
332 /* It's not clear whether a zero row test is worthwhile here ... */ | 337 /* It's not clear whether a zero row test is worthwhile here ... */ |
333 | 338 |
334 #ifndef NO_ZERO_ROW_TEST | 339 #ifndef NO_ZERO_ROW_TEST |
335 if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) { | 340 if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) { |
336 /* AC terms all zero */ | 341 /* AC terms all zero */ |
337 JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) | 342 JSAMPLE dcval = range_limit[(int) DESCALE((JLONG) wsptr[0], PASS1_BITS+3) |
338 » » » » & RANGE_MASK]; | 343 & RANGE_MASK]; |
339 | 344 |
340 outptr[0] = dcval; | 345 outptr[0] = dcval; |
341 outptr[1] = dcval; | 346 outptr[1] = dcval; |
342 | 347 |
343 wsptr += DCTSIZE;»» /* advance pointer to next row */ | 348 wsptr += DCTSIZE; /* advance pointer to next row */ |
344 continue; | 349 continue; |
345 } | 350 } |
346 #endif | 351 #endif |
347 | 352 |
348 /* Even part */ | 353 /* Even part */ |
349 | 354 |
350 tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2); | 355 tmp10 = LEFT_SHIFT((JLONG) wsptr[0], CONST_BITS+2); |
351 | 356 |
352 /* Odd part */ | 357 /* Odd part */ |
353 | 358 |
354 tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-
c1) */ | 359 tmp0 = MULTIPLY((JLONG) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-
c1) */ |
355 » + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c
7) */ | 360 + MULTIPLY((JLONG) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c
7) */ |
356 » + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5
-c7) */ | 361 + MULTIPLY((JLONG) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5
-c7) */ |
357 » + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c
7) */ | 362 + MULTIPLY((JLONG) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c
7) */ |
358 | 363 |
359 /* Final output stage */ | 364 /* Final output stage */ |
360 | 365 |
361 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0, | 366 outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0, |
362 » » » » » CONST_BITS+PASS1_BITS+3+2) | 367 CONST_BITS+PASS1_BITS+3+2) |
363 » » » & RANGE_MASK]; | 368 & RANGE_MASK]; |
364 outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0, | 369 outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0, |
365 » » » » » CONST_BITS+PASS1_BITS+3+2) | 370 CONST_BITS+PASS1_BITS+3+2) |
366 » » » & RANGE_MASK]; | 371 & RANGE_MASK]; |
367 | 372 |
368 wsptr += DCTSIZE;» » /* advance pointer to next row */ | 373 wsptr += DCTSIZE; /* advance pointer to next row */ |
369 } | 374 } |
370 } | 375 } |
371 | 376 |
372 | 377 |
373 /* | 378 /* |
374 * Perform dequantization and inverse DCT on one block of coefficients, | 379 * Perform dequantization and inverse DCT on one block of coefficients, |
375 * producing a reduced-size 1x1 output block. | 380 * producing a reduced-size 1x1 output block. |
376 */ | 381 */ |
377 | 382 |
378 GLOBAL(void) | 383 GLOBAL(void) |
379 jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr, | 384 jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info *compptr, |
380 » JCOEFPTR coef_block, | 385 JCOEFPTR coef_block, |
381 » JSAMPARRAY output_buf, JDIMENSION output_col) | 386 JSAMPARRAY output_buf, JDIMENSION output_col) |
382 { | 387 { |
383 int dcval; | 388 int dcval; |
384 ISLOW_MULT_TYPE * quantptr; | 389 ISLOW_MULT_TYPE *quantptr; |
385 JSAMPLE *range_limit = IDCT_range_limit(cinfo); | 390 JSAMPLE *range_limit = IDCT_range_limit(cinfo); |
386 SHIFT_TEMPS | 391 SHIFT_TEMPS |
387 | 392 |
388 /* We hardly need an inverse DCT routine for this: just take the | 393 /* We hardly need an inverse DCT routine for this: just take the |
389 * average pixel value, which is one-eighth of the DC coefficient. | 394 * average pixel value, which is one-eighth of the DC coefficient. |
390 */ | 395 */ |
391 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; | 396 quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; |
392 dcval = DEQUANTIZE(coef_block[0], quantptr[0]); | 397 dcval = DEQUANTIZE(coef_block[0], quantptr[0]); |
393 dcval = (int) DESCALE((INT32) dcval, 3); | 398 dcval = (int) DESCALE((JLONG) dcval, 3); |
394 | 399 |
395 output_buf[0][output_col] = range_limit[dcval & RANGE_MASK]; | 400 output_buf[0][output_col] = range_limit[dcval & RANGE_MASK]; |
396 } | 401 } |
397 | 402 |
398 #endif /* IDCT_SCALING_SUPPORTED */ | 403 #endif /* IDCT_SCALING_SUPPORTED */ |
OLD | NEW |