4 * Copyright (C) 1994-1996, Thomas G. Lane.
5 * This file is part of the Independent JPEG Group's software.
6 * For conditions of distribution and use, see the accompanying README file.
8 * This file contains the inverse-DCT management logic.
9 * This code selects a particular IDCT implementation to be used,
10 * and it performs related housekeeping chores. No code in this file
11 * is executed per IDCT step, only during output pass setup.
13 * Note that the IDCT routines are responsible for performing coefficient
14 * dequantization as well as the IDCT proper. This module sets up the
15 * dequantization multiplier table needed by the IDCT routine.
18 #define JPEG_INTERNALS
21 #include "jdct.h" /* Private declarations for DCT subsystem */
22 extern int SSE2Available
;
25 * The decompressor input side (jdinput.c) saves away the appropriate
26 * quantization table for each component at the start of the first scan
27 * involving that component. (This is necessary in order to correctly
28 * decode files that reuse Q-table slots.)
29 * When we are ready to make an output pass, the saved Q-table is converted
30 * to a multiplier table that will actually be used by the IDCT routine.
31 * The multiplier table contents are IDCT-method-dependent. To support
32 * application changes in IDCT method between scans, we can remake the
33 * multiplier tables if necessary.
34 * In buffered-image mode, the first output pass may occur before any data
35 * has been seen for some components, and thus before their Q-tables have
36 * been saved away. To handle this case, multiplier tables are preset
37 * to zeroes; the result of the IDCT will be a neutral gray level.
41 /* Private subobject for this module */
44 struct jpeg_inverse_dct pub
; /* public fields */
46 /* This array contains the IDCT method code that each multiplier table
47 * is currently set up for, or -1 if it's not yet set up.
48 * The actual multiplier tables are pointed to by dct_table in the
49 * per-component comp_info structures.
51 int cur_method
[MAX_COMPONENTS
];
54 typedef my_idct_controller
* my_idct_ptr
;
57 /* Allocated multiplier tables: big enough for any supported variant */
60 ISLOW_MULT_TYPE islow_array
[DCTSIZE2
];
61 #ifdef DCT_IFAST_SUPPORTED
62 IFAST_MULT_TYPE ifast_array
[DCTSIZE2
];
64 #ifdef DCT_FLOAT_SUPPORTED
65 FLOAT_MULT_TYPE float_array
[DCTSIZE2
];
70 /* The current scaled-IDCT routines require ISLOW-style multiplier tables,
71 * so be sure to compile that code if either ISLOW or SCALING is requested.
73 #ifdef DCT_ISLOW_SUPPORTED
74 #define PROVIDE_ISLOW_TABLES
76 #ifdef IDCT_SCALING_SUPPORTED
77 #define PROVIDE_ISLOW_TABLES
82 jpeg_idct_islow_sse2 (
83 j_decompress_ptr cinfo
,
84 jpeg_component_info
* compptr
,
86 JSAMPARRAY output_buf
,
87 JDIMENSION output_col
);
91 * Prepare for an output pass.
92 * Here we select the proper IDCT routine for each component and build
93 * a matching multiplier table.
97 start_pass (j_decompress_ptr cinfo
)
99 my_idct_ptr idct
= (my_idct_ptr
) cinfo
->idct
;
101 jpeg_component_info
*compptr
;
103 inverse_DCT_method_ptr method_ptr
= NULL
;
106 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
108 /* Select the proper IDCT routine for this component's scaling */
109 switch (compptr
->DCT_scaled_size
) {
110 #ifdef IDCT_SCALING_SUPPORTED
112 method_ptr
= jpeg_idct_1x1
;
113 method
= JDCT_ISLOW
; /* jidctred uses islow-style table */
116 method_ptr
= jpeg_idct_2x2
;
117 method
= JDCT_ISLOW
; /* jidctred uses islow-style table */
120 method_ptr
= jpeg_idct_4x4
;
121 method
= JDCT_ISLOW
; /* jidctred uses islow-style table */
125 switch (cinfo
->dct_method
) {
126 #ifdef DCT_ISLOW_SUPPORTED
128 #ifdef HAVE_SSE2_INTEL_MNEMONICS
129 if(SSE2Available
== 1)
131 method_ptr
= jpeg_idct_islow_sse2
;
136 method_ptr
= jpeg_idct_islow
;
140 method_ptr
= jpeg_idct_islow
;
143 #endif /* HAVE_SSE2_INTEL_MNEMONICS */
146 #ifdef DCT_IFAST_SUPPORTED
148 #ifdef HAVE_SSE2_INTEL_MNEMONICS
149 if (SSE2Available
==1)
151 method_ptr
= jpeg_idct_islow_sse2
;
156 method_ptr
= jpeg_idct_ifast
;
160 method_ptr
= jpeg_idct_ifast
;
162 #endif /* HAVE_SSE2_INTEL_MNEMONICS */
166 #ifdef DCT_FLOAT_SUPPORTED
168 method_ptr
= jpeg_idct_float
;
173 ERREXIT(cinfo
, JERR_NOT_COMPILED
);
178 ERREXIT1(cinfo
, JERR_BAD_DCTSIZE
, compptr
->DCT_scaled_size
);
181 idct
->pub
.inverse_DCT
[ci
] = method_ptr
;
182 /* Create multiplier table from quant table.
183 * However, we can skip this if the component is uninteresting
184 * or if we already built the table. Also, if no quant table
185 * has yet been saved for the component, we leave the
186 * multiplier table all-zero; we'll be reading zeroes from the
187 * coefficient controller's buffer anyway.
189 if (! compptr
->component_needed
|| idct
->cur_method
[ci
] == method
)
191 qtbl
= compptr
->quant_table
;
192 if (qtbl
== NULL
) /* happens if no data yet for component */
194 idct
->cur_method
[ci
] = method
;
196 #ifdef PROVIDE_ISLOW_TABLES
199 /* For LL&M IDCT method, multipliers are equal to raw quantization
200 * coefficients, but are stored as ints to ensure access efficiency.
202 ISLOW_MULT_TYPE
* ismtbl
= (ISLOW_MULT_TYPE
*) compptr
->dct_table
;
203 for (i
= 0; i
< DCTSIZE2
; i
++) {
204 ismtbl
[i
] = (ISLOW_MULT_TYPE
) qtbl
->quantval
[i
];
209 #ifdef DCT_IFAST_SUPPORTED
212 /* For AA&N IDCT method, multipliers are equal to quantization
213 * coefficients scaled by scalefactor[row]*scalefactor[col], where
215 * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
216 * For integer operation, the multiplier table is to be scaled by
219 IFAST_MULT_TYPE
* ifmtbl
= (IFAST_MULT_TYPE
*) compptr
->dct_table
;
220 #define CONST_BITS 14
221 static const INT16 aanscales
[DCTSIZE2
] = {
222 /* precomputed values scaled up by 14 bits */
223 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
224 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
225 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
226 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
227 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
228 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
229 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
230 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
234 for (i
= 0; i
< DCTSIZE2
; i
++) {
235 ifmtbl
[i
] = (IFAST_MULT_TYPE
)
236 DESCALE(MULTIPLY16V16((INT32
) qtbl
->quantval
[i
],
237 (INT32
) aanscales
[i
]),
238 CONST_BITS
-IFAST_SCALE_BITS
);
243 #ifdef DCT_FLOAT_SUPPORTED
246 /* For float AA&N IDCT method, multipliers are equal to quantization
247 * coefficients scaled by scalefactor[row]*scalefactor[col], where
249 * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
251 FLOAT_MULT_TYPE
* fmtbl
= (FLOAT_MULT_TYPE
*) compptr
->dct_table
;
253 static const double aanscalefactor
[DCTSIZE
] = {
254 1.0, 1.387039845, 1.306562965, 1.175875602,
255 1.0, 0.785694958, 0.541196100, 0.275899379
259 for (row
= 0; row
< DCTSIZE
; row
++) {
260 for (col
= 0; col
< DCTSIZE
; col
++) {
261 fmtbl
[i
] = (FLOAT_MULT_TYPE
)
262 ((double) qtbl
->quantval
[i
] *
263 aanscalefactor
[row
] * aanscalefactor
[col
]);
271 ERREXIT(cinfo
, JERR_NOT_COMPILED
);
279 * Initialize IDCT manager.
283 jinit_inverse_dct (j_decompress_ptr cinfo
)
287 jpeg_component_info
*compptr
;
290 (*cinfo
->mem
->alloc_small
) ((j_common_ptr
) cinfo
, JPOOL_IMAGE
,
291 SIZEOF(my_idct_controller
));
292 cinfo
->idct
= (struct jpeg_inverse_dct
*) idct
;
293 idct
->pub
.start_pass
= start_pass
;
295 for (ci
= 0, compptr
= cinfo
->comp_info
; ci
< cinfo
->num_components
;
297 /* Allocate and pre-zero a multiplier table for each component */
299 (*cinfo
->mem
->alloc_small
) ((j_common_ptr
) cinfo
, JPOOL_IMAGE
,
300 SIZEOF(multiplier_table
));
301 MEMZERO(compptr
->dct_table
, SIZEOF(multiplier_table
));
302 /* Mark multiplier table not yet set up for any method */
303 idct
->cur_method
[ci
] = -1;