chromium/third_party/libjpeg_turbo/jcdctmgr.c

/*
 * jcdctmgr.c
 *
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 1999-2006, MIYASAKA Masaru.
 * Copyright 2009 Pierre Ossman <[email protected]> for Cendio AB
 * Copyright (C) 2011, 2014-2015, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
 * This file contains the forward-DCT management logic.
 * This code selects a particular DCT implementation to be used,
 * and it performs related housekeeping chores including coefficient
 * quantization.
 */

#define JPEG_INTERNALS
#include "jinclude.h"
#include "jpeglib.h"
#include "jdct.h"               /* Private declarations for DCT subsystem */
#include "jsimddct.h"


/* Private subobject for this module */

forward_DCT_method_ptr;
float_DCT_method_ptr;

convsamp_method_ptr;
float_convsamp_method_ptr;

quantize_method_ptr;
float_quantize_method_ptr;

METHODDEF(void) quantize(JCOEFPTR, DCTELEM *, DCTELEM *);

my_fdct_controller;

my_fdct_ptr;


#if BITS_IN_JSAMPLE == 8

/*
 * Find the highest bit in an integer through binary search.
 */

LOCAL(int)
flss(UINT16 val)
{ … }


/*
 * Compute values to do a division using reciprocal.
 *
 * This implementation is based on an algorithm described in
 *   "How to optimize for the Pentium family of microprocessors"
 *   (http://www.agner.org/assem/).
 * More information about the basic algorithm can be found in
 * the paper "Integer Division Using Reciprocals" by Robert Alverson.
 *
 * The basic idea is to replace x/d by x * d^-1. In order to store
 * d^-1 with enough precision we shift it left a few places. It turns
 * out that this algoright gives just enough precision, and also fits
 * into DCTELEM:
 *
 *   b = (the number of significant bits in divisor) - 1
 *   r = (word size) + b
 *   f = 2^r / divisor
 *
 * f will not be an integer for most cases, so we need to compensate
 * for the rounding error introduced:
 *
 *   no fractional part:
 *
 *       result = input >> r
 *
 *   fractional part of f < 0.5:
 *
 *       round f down to nearest integer
 *       result = ((input + 1) * f) >> r
 *
 *   fractional part of f > 0.5:
 *
 *       round f up to nearest integer
 *       result = (input * f) >> r
 *
 * This is the original algorithm that gives truncated results. But we
 * want properly rounded results, so we replace "input" with
 * "input + divisor/2".
 *
 * In order to allow SIMD implementations we also tweak the values to
 * allow the same calculation to be made at all times:
 *
 *   dctbl[0] = f rounded to nearest integer
 *   dctbl[1] = divisor / 2 (+ 1 if fractional part of f < 0.5)
 *   dctbl[2] = 1 << ((word size) * 2 - r)
 *   dctbl[3] = r - (word size)
 *
 * dctbl[2] is for stupid instruction sets where the shift operation
 * isn't member wise (e.g. MMX).
 *
 * The reason dctbl[2] and dctbl[3] reduce the shift with (word size)
 * is that most SIMD implementations have a "multiply and store top
 * half" operation.
 *
 * Lastly, we store each of the values in their own table instead
 * of in a consecutive manner, yet again in order to allow SIMD
 * routines.
 */

LOCAL(int)
compute_reciprocal(UINT16 divisor, DCTELEM *dtbl)
{ … }

#endif


/*
 * Initialize for a processing pass.
 * Verify that all referenced Q-tables are present, and set up
 * the divisor table for each one.
 * In the current implementation, DCT of all components is done during
 * the first pass, even if only some components will be output in the
 * first scan.  Hence all components should be examined here.
 */

METHODDEF(void)
start_pass_fdctmgr(j_compress_ptr cinfo)
{ … }


/*
 * Load data into workspace, applying unsigned->signed conversion.
 */

METHODDEF(void)
convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace)
{ … }


/*
 * Quantize/descale the coefficients, and store into coef_blocks[].
 */

METHODDEF(void)
quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
{ … }


/*
 * Perform forward DCT on one or more blocks of a component.
 *
 * The input samples are taken from the sample_data[] array starting at
 * position start_row/start_col, and moving to the right for any additional
 * blocks. The quantized coefficients are returned in coef_blocks[].
 */

METHODDEF(void)
forward_DCT(j_compress_ptr cinfo, jpeg_component_info *compptr,
            JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
            JDIMENSION start_row, JDIMENSION start_col, JDIMENSION num_blocks)
/* This version is used for integer DCT implementations. */
{ … }


#ifdef DCT_FLOAT_SUPPORTED

METHODDEF(void)
convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col,
               FAST_FLOAT *workspace)
{ … }


METHODDEF(void)
quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors,
               FAST_FLOAT *workspace)
{ … }


METHODDEF(void)
forward_DCT_float(j_compress_ptr cinfo, jpeg_component_info *compptr,
                  JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
                  JDIMENSION start_row, JDIMENSION start_col,
                  JDIMENSION num_blocks)
/* This version is used for floating-point DCT implementations. */
{ … }

#endif /* DCT_FLOAT_SUPPORTED */


/*
 * Initialize FDCT manager.
 */

GLOBAL(void)
jinit_forward_dct(j_compress_ptr cinfo)
{ … }