7 * Copyright (C) 1994, Thomas G. Lane.
9 * This file is part of the Independent JPEG Group's software.
11 * For conditions of distribution and use, see the accompanying README file.
15 * This include file contains common declarations for the forward and
17 * inverse DCT modules. These declarations are private to the DCT managers
19 * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms.
21 * The individual DCT algorithms are kept in separate files to ease
23 * machine-dependent tuning (e.g., assembly coding).
33 * A forward DCT routine is given a pointer to a work area of type DCTELEM[];
35 * the DCT is to be performed in-place in that buffer. Type DCTELEM is int
37 * for 8-bit samples, INT32 for 12-bit samples. (NOTE: Floating-point DCT
39 * implementations use an array of type FAST_FLOAT, instead.)
41 * The DCT inputs are expected to be signed (range +-CENTERJSAMPLE).
43 * The DCT outputs are returned scaled up by a factor of 8; they therefore
45 * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This
47 * convention improves accuracy in integer implementations and saves some
49 * work in floating-point ones.
51 * Quantization of the output coefficients is done by jcdctmgr.c.
57 #if BITS_IN_JSAMPLE == 8
59 typedef int DCTELEM; /* 16 or 32 bits is fine */
63 typedef INT32 DCTELEM; /* must have 32 bits */
69 typedef JMETHOD(void, forward_DCT_method_ptr, (DCTELEM * data));
71 typedef JMETHOD(void, float_DCT_method_ptr, (FAST_FLOAT * data));
79 * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer
81 * to an output sample array. The routine must dequantize the input data as
83 * well as perform the IDCT; for dequantization, it uses the multiplier table
85 * pointed to by compptr->dct_table. The output data is to be placed into the
87 * sample array starting at a specified column. (Any row offset needed will
89 * be applied to the array pointer before it is passed to the IDCT code.)
91 * Note that the number of samples emitted by the IDCT routine is
93 * DCT_scaled_size * DCT_scaled_size.
99 /* typedef inverse_DCT_method_ptr is declared in jpegint.h */
105 * Each IDCT routine has its own ideas about the best dct_table element type.
111 typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */
113 #if BITS_IN_JSAMPLE == 8
115 typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */
117 #define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */
121 typedef INT32 IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */
123 #define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */
127 typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */
135 * Each IDCT routine is responsible for range-limiting its results and
137 * converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could
139 * be quite far out of range if the input data is corrupt, so a bulletproof
141 * range-limiting step is required. We use a mask-and-table-lookup method
143 * to do the combined operations quickly. See the comments with
145 * prepare_range_limit_table (in jdmaster.c) for more info.
151 #define IDCT_range_limit(cinfo) ((cinfo)->sample_range_limit + CENTERJSAMPLE)
155 #define RANGE_MASK (MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */
161 /* Short forms of external names for systems with brain-damaged linkers. */
165 #ifdef NEED_SHORT_EXTERNAL_NAMES
167 #define jpeg_fdct_islow jFDislow
169 #define jpeg_fdct_ifast jFDifast
171 #define jpeg_fdct_float jFDfloat
173 #define jpeg_idct_islow jRDislow
175 #define jpeg_idct_ifast jRDifast
177 #define jpeg_idct_float jRDfloat
179 #define jpeg_idct_4x4 jRD4x4
181 #define jpeg_idct_2x2 jRD2x2
183 #define jpeg_idct_1x1 jRD1x1
185 #endif /* NEED_SHORT_EXTERNAL_NAMES */
189 /* Extern declarations for the forward and inverse DCT routines. */
193 EXTERN void jpeg_fdct_islow JPP((DCTELEM * data));
195 EXTERN void jpeg_fdct_ifast JPP((DCTELEM * data));
197 EXTERN void jpeg_fdct_float JPP((FAST_FLOAT * data));
201 EXTERN void jpeg_idct_islow
203 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
205 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
207 EXTERN void jpeg_idct_ifast
209 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
211 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
213 EXTERN void jpeg_idct_float
215 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
217 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
219 EXTERN void jpeg_idct_4x4
221 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
223 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
225 EXTERN void jpeg_idct_2x2
227 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
229 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
231 EXTERN void jpeg_idct_1x1
233 JPP((j_decompress_ptr cinfo, jpeg_component_info * compptr,
235 JCOEFPTR coef_block, JSAMPARRAY output_buf, JDIMENSION output_col));
243 * Macros for handling fixed-point arithmetic; these are used by many
245 * but not all of the DCT/IDCT modules.
249 * All values are expected to be of type INT32.
251 * Fractional constants are scaled left by CONST_BITS bits.
253 * CONST_BITS is defined within each module using these macros,
255 * and may differ from one module to the next.
261 #define ONE ((INT32) 1)
263 #define CONST_SCALE (ONE << CONST_BITS)
267 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
269 * Caution: some C compilers fail to reduce "FIX(constant)" at compile time,
271 * thus causing a lot of useless floating-point operations at run time.
277 #define FIX(x) ((INT32) ((x) * CONST_SCALE + 0.5))
281 /* Descale and correctly round an INT32 value that's scaled by N bits.
283 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
285 * the fudge factor is correct for either sign of X.
291 #define DESCALE(x,n) RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
295 /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
297 * This macro is used only when the two inputs will actually be no more than
299 * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a
301 * full 32x32 multiply. This provides a useful speedup on many machines.
303 * Unfortunately there is no way to specify a 16x16->32 multiply portably
305 * in C, but some C compilers will do the right thing if you provide the
307 * correct combination of casts.
313 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
315 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT16) (const)))
319 #ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */
321 #define MULTIPLY16C16(var,const) (((INT16) (var)) * ((INT32) (const)))
327 #ifndef MULTIPLY16C16 /* default definition */
329 #define MULTIPLY16C16(var,const) ((var) * (const))
335 /* Same except both inputs are variables. */
339 #ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */
341 #define MULTIPLY16V16(var1,var2) (((INT16) (var1)) * ((INT16) (var2)))
347 #ifndef MULTIPLY16V16 /* default definition */
349 #define MULTIPLY16V16(var1,var2) ((var1) * (var2))