michael@0: /*
michael@0:  * jddctmgr.c
michael@0:  *
michael@0:  * This file was part of the Independent JPEG Group's software:
michael@0:  * Copyright (C) 1994-1996, Thomas G. Lane.
michael@0:  * Modified 2002-2010 by Guido Vollbeding.
michael@0:  * libjpeg-turbo Modifications:
michael@0:  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0:  * Copyright (C) 2010, D. R. Commander.
michael@0:  * For conditions of distribution and use, see the accompanying README file.
michael@0:  *
michael@0:  * This file contains the inverse-DCT management logic.
michael@0:  * This code selects a particular IDCT implementation to be used,
michael@0:  * and it performs related housekeeping chores.  No code in this file
michael@0:  * is executed per IDCT step, only during output pass setup.
michael@0:  *
michael@0:  * Note that the IDCT routines are responsible for performing coefficient
michael@0:  * dequantization as well as the IDCT proper.  This module sets up the
michael@0:  * dequantization multiplier table needed by the IDCT routine.
michael@0:  */
michael@0: 
michael@0: #define JPEG_INTERNALS
michael@0: #include "jinclude.h"
michael@0: #include "jpeglib.h"
michael@0: #include "jdct.h"		/* Private declarations for DCT subsystem */
michael@0: #include "jsimddct.h"
michael@0: #include "jpegcomp.h"
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * The decompressor input side (jdinput.c) saves away the appropriate
michael@0:  * quantization table for each component at the start of the first scan
michael@0:  * involving that component.  (This is necessary in order to correctly
michael@0:  * decode files that reuse Q-table slots.)
michael@0:  * When we are ready to make an output pass, the saved Q-table is converted
michael@0:  * to a multiplier table that will actually be used by the IDCT routine.
michael@0:  * The multiplier table contents are IDCT-method-dependent.  To support
michael@0:  * application changes in IDCT method between scans, we can remake the
michael@0:  * multiplier tables if necessary.
michael@0:  * In buffered-image mode, the first output pass may occur before any data
michael@0:  * has been seen for some components, and thus before their Q-tables have
michael@0:  * been saved away.  To handle this case, multiplier tables are preset
michael@0:  * to zeroes; the result of the IDCT will be a neutral gray level.
michael@0:  */
michael@0: 
michael@0: 
michael@0: /* Private subobject for this module */
michael@0: 
michael@0: typedef struct {
michael@0:   struct jpeg_inverse_dct pub;	/* public fields */
michael@0: 
michael@0:   /* This array contains the IDCT method code that each multiplier table
michael@0:    * is currently set up for, or -1 if it's not yet set up.
michael@0:    * The actual multiplier tables are pointed to by dct_table in the
michael@0:    * per-component comp_info structures.
michael@0:    */
michael@0:   int cur_method[MAX_COMPONENTS];
michael@0: } my_idct_controller;
michael@0: 
michael@0: typedef my_idct_controller * my_idct_ptr;
michael@0: 
michael@0: 
michael@0: /* Allocated multiplier tables: big enough for any supported variant */
michael@0: 
michael@0: typedef union {
michael@0:   ISLOW_MULT_TYPE islow_array[DCTSIZE2];
michael@0: #ifdef DCT_IFAST_SUPPORTED
michael@0:   IFAST_MULT_TYPE ifast_array[DCTSIZE2];
michael@0: #endif
michael@0: #ifdef DCT_FLOAT_SUPPORTED
michael@0:   FLOAT_MULT_TYPE float_array[DCTSIZE2];
michael@0: #endif
michael@0: } multiplier_table;
michael@0: 
michael@0: 
michael@0: /* The current scaled-IDCT routines require ISLOW-style multiplier tables,
michael@0:  * so be sure to compile that code if either ISLOW or SCALING is requested.
michael@0:  */
michael@0: #ifdef DCT_ISLOW_SUPPORTED
michael@0: #define PROVIDE_ISLOW_TABLES
michael@0: #else
michael@0: #ifdef IDCT_SCALING_SUPPORTED
michael@0: #define PROVIDE_ISLOW_TABLES
michael@0: #endif
michael@0: #endif
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Prepare for an output pass.
michael@0:  * Here we select the proper IDCT routine for each component and build
michael@0:  * a matching multiplier table.
michael@0:  */
michael@0: 
michael@0: METHODDEF(void)
michael@0: start_pass (j_decompress_ptr cinfo)
michael@0: {
michael@0:   my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
michael@0:   int ci, i;
michael@0:   jpeg_component_info *compptr;
michael@0:   int method = 0;
michael@0:   inverse_DCT_method_ptr method_ptr = NULL;
michael@0:   JQUANT_TBL * qtbl;
michael@0: 
michael@0:   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
michael@0:        ci++, compptr++) {
michael@0:     /* Select the proper IDCT routine for this component's scaling */
michael@0:     switch (compptr->_DCT_scaled_size) {
michael@0: #ifdef IDCT_SCALING_SUPPORTED
michael@0:     case 1:
michael@0:       method_ptr = jpeg_idct_1x1;
michael@0:       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
michael@0:       break;
michael@0:     case 2:
michael@0:       if (jsimd_can_idct_2x2())
michael@0:         method_ptr = jsimd_idct_2x2;
michael@0:       else
michael@0:         method_ptr = jpeg_idct_2x2;
michael@0:       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
michael@0:       break;
michael@0:     case 3:
michael@0:       method_ptr = jpeg_idct_3x3;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 4:
michael@0:       if (jsimd_can_idct_4x4())
michael@0:         method_ptr = jsimd_idct_4x4;
michael@0:       else
michael@0:         method_ptr = jpeg_idct_4x4;
michael@0:       method = JDCT_ISLOW;	/* jidctred uses islow-style table */
michael@0:       break;
michael@0:     case 5:
michael@0:       method_ptr = jpeg_idct_5x5;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 6:
michael@0:       method_ptr = jpeg_idct_6x6;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 7:
michael@0:       method_ptr = jpeg_idct_7x7;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0: #endif
michael@0:     case DCTSIZE:
michael@0:       switch (cinfo->dct_method) {
michael@0: #ifdef DCT_ISLOW_SUPPORTED
michael@0:       case JDCT_ISLOW:
michael@0: 	if (jsimd_can_idct_islow())
michael@0: 	  method_ptr = jsimd_idct_islow;
michael@0: 	else
michael@0: 	  method_ptr = jpeg_idct_islow;
michael@0: 	method = JDCT_ISLOW;
michael@0: 	break;
michael@0: #endif
michael@0: #ifdef DCT_IFAST_SUPPORTED
michael@0:       case JDCT_IFAST:
michael@0: 	if (jsimd_can_idct_ifast())
michael@0: 	  method_ptr = jsimd_idct_ifast;
michael@0: 	else
michael@0: 	  method_ptr = jpeg_idct_ifast;
michael@0: 	method = JDCT_IFAST;
michael@0: 	break;
michael@0: #endif
michael@0: #ifdef DCT_FLOAT_SUPPORTED
michael@0:       case JDCT_FLOAT:
michael@0: 	if (jsimd_can_idct_float())
michael@0: 	  method_ptr = jsimd_idct_float;
michael@0: 	else
michael@0: 	  method_ptr = jpeg_idct_float;
michael@0: 	method = JDCT_FLOAT;
michael@0: 	break;
michael@0: #endif
michael@0:       default:
michael@0: 	ERREXIT(cinfo, JERR_NOT_COMPILED);
michael@0: 	break;
michael@0:       }
michael@0:       break;
michael@0:     case 9:
michael@0:       method_ptr = jpeg_idct_9x9;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 10:
michael@0:       method_ptr = jpeg_idct_10x10;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 11:
michael@0:       method_ptr = jpeg_idct_11x11;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 12:
michael@0:       method_ptr = jpeg_idct_12x12;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 13:
michael@0:       method_ptr = jpeg_idct_13x13;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 14:
michael@0:       method_ptr = jpeg_idct_14x14;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 15:
michael@0:       method_ptr = jpeg_idct_15x15;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     case 16:
michael@0:       method_ptr = jpeg_idct_16x16;
michael@0:       method = JDCT_ISLOW;	/* jidctint uses islow-style table */
michael@0:       break;
michael@0:     default:
michael@0:       ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->_DCT_scaled_size);
michael@0:       break;
michael@0:     }
michael@0:     idct->pub.inverse_DCT[ci] = method_ptr;
michael@0:     /* Create multiplier table from quant table.
michael@0:      * However, we can skip this if the component is uninteresting
michael@0:      * or if we already built the table.  Also, if no quant table
michael@0:      * has yet been saved for the component, we leave the
michael@0:      * multiplier table all-zero; we'll be reading zeroes from the
michael@0:      * coefficient controller's buffer anyway.
michael@0:      */
michael@0:     if (! compptr->component_needed || idct->cur_method[ci] == method)
michael@0:       continue;
michael@0:     qtbl = compptr->quant_table;
michael@0:     if (qtbl == NULL)		/* happens if no data yet for component */
michael@0:       continue;
michael@0:     idct->cur_method[ci] = method;
michael@0:     switch (method) {
michael@0: #ifdef PROVIDE_ISLOW_TABLES
michael@0:     case JDCT_ISLOW:
michael@0:       {
michael@0: 	/* For LL&M IDCT method, multipliers are equal to raw quantization
michael@0: 	 * coefficients, but are stored as ints to ensure access efficiency.
michael@0: 	 */
michael@0: 	ISLOW_MULT_TYPE * ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
michael@0: 	for (i = 0; i < DCTSIZE2; i++) {
michael@0: 	  ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[i];
michael@0: 	}
michael@0:       }
michael@0:       break;
michael@0: #endif
michael@0: #ifdef DCT_IFAST_SUPPORTED
michael@0:     case JDCT_IFAST:
michael@0:       {
michael@0: 	/* For AA&N IDCT method, multipliers are equal to quantization
michael@0: 	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
michael@0: 	 *   scalefactor[0] = 1
michael@0: 	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
michael@0: 	 * For integer operation, the multiplier table is to be scaled by
michael@0: 	 * IFAST_SCALE_BITS.
michael@0: 	 */
michael@0: 	IFAST_MULT_TYPE * ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
michael@0: #define CONST_BITS 14
michael@0: 	static const INT16 aanscales[DCTSIZE2] = {
michael@0: 	  /* precomputed values scaled up by 14 bits */
michael@0: 	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
michael@0: 	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
michael@0: 	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
michael@0: 	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
michael@0: 	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
michael@0: 	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
michael@0: 	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
michael@0: 	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
michael@0: 	};
michael@0: 	SHIFT_TEMPS
michael@0: 
michael@0: 	for (i = 0; i < DCTSIZE2; i++) {
michael@0: 	  ifmtbl[i] = (IFAST_MULT_TYPE)
michael@0: 	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
michael@0: 				  (INT32) aanscales[i]),
michael@0: 		    CONST_BITS-IFAST_SCALE_BITS);
michael@0: 	}
michael@0:       }
michael@0:       break;
michael@0: #endif
michael@0: #ifdef DCT_FLOAT_SUPPORTED
michael@0:     case JDCT_FLOAT:
michael@0:       {
michael@0: 	/* For float AA&N IDCT method, multipliers are equal to quantization
michael@0: 	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
michael@0: 	 *   scalefactor[0] = 1
michael@0: 	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
michael@0: 	 */
michael@0: 	FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
michael@0: 	int row, col;
michael@0: 	static const double aanscalefactor[DCTSIZE] = {
michael@0: 	  1.0, 1.387039845, 1.306562965, 1.175875602,
michael@0: 	  1.0, 0.785694958, 0.541196100, 0.275899379
michael@0: 	};
michael@0: 
michael@0: 	i = 0;
michael@0: 	for (row = 0; row < DCTSIZE; row++) {
michael@0: 	  for (col = 0; col < DCTSIZE; col++) {
michael@0: 	    fmtbl[i] = (FLOAT_MULT_TYPE)
michael@0: 	      ((double) qtbl->quantval[i] *
michael@0: 	       aanscalefactor[row] * aanscalefactor[col]);
michael@0: 	    i++;
michael@0: 	  }
michael@0: 	}
michael@0:       }
michael@0:       break;
michael@0: #endif
michael@0:     default:
michael@0:       ERREXIT(cinfo, JERR_NOT_COMPILED);
michael@0:       break;
michael@0:     }
michael@0:   }
michael@0: }
michael@0: 
michael@0: 
michael@0: /*
michael@0:  * Initialize IDCT manager.
michael@0:  */
michael@0: 
michael@0: GLOBAL(void)
michael@0: jinit_inverse_dct (j_decompress_ptr cinfo)
michael@0: {
michael@0:   my_idct_ptr idct;
michael@0:   int ci;
michael@0:   jpeg_component_info *compptr;
michael@0: 
michael@0:   idct = (my_idct_ptr)
michael@0:     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
michael@0: 				SIZEOF(my_idct_controller));
michael@0:   cinfo->idct = (struct jpeg_inverse_dct *) idct;
michael@0:   idct->pub.start_pass = start_pass;
michael@0: 
michael@0:   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
michael@0:        ci++, compptr++) {
michael@0:     /* Allocate and pre-zero a multiplier table for each component */
michael@0:     compptr->dct_table =
michael@0:       (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
michael@0: 				  SIZEOF(multiplier_table));
michael@0:     MEMZERO(compptr->dct_table, SIZEOF(multiplier_table));
michael@0:     /* Mark multiplier table not yet set up for any method */
michael@0:     idct->cur_method[ci] = -1;
michael@0:   }
michael@0: }