0N/A * reserved comment block 0N/A * DO NOT REMOVE OR ALTER! 0N/A * Copyright (C) 1994-1996, Thomas G. Lane. 0N/A * This file is part of the Independent JPEG Group's software. 0N/A * For conditions of distribution and use, see the accompanying README file. 0N/A * This file contains a fast, not so accurate integer implementation of the 0N/A * forward DCT (Discrete Cosine Transform). 0N/A * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT 0N/A * on each column. Direct algorithms are also available, but they are 0N/A * much more complex and seem not to be any faster when reduced to code. 0N/A * This implementation is based on Arai, Agui, and Nakajima's algorithm for 0N/A * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in 0N/A * Japanese, but the algorithm is described in the Pennebaker & Mitchell 0N/A * JPEG textbook (see REFERENCES section in file README). The following code 0N/A * is based directly on figure 4-8 in P&M. 0N/A * While an 8-point DCT cannot be done in less than 11 multiplies, it is 0N/A * possible to arrange the computation so that many of the multiplies are 0N/A * simple scalings of the final outputs. These multiplies can then be 0N/A * folded into the multiplications or divisions by the JPEG quantization 0N/A * table entries. The AA&N method leaves only 5 multiplies and 29 adds 0N/A * to be done in the DCT itself. 0N/A * The primary disadvantage of this method is that with fixed-point math, 0N/A * accuracy is lost due to imprecise representation of the scaled 0N/A * quantization values. The smaller the quantization table entry, the less 0N/A * precise the scaled value, so this implementation does worse with high- 0N/A * quality-setting files than with low-quality ones. 0N/A#
include "jdct.h" /* Private declarations for DCT subsystem */ 0N/A * This module is specialized to the case DCTSIZE = 8. 0N/A/* Scaling decisions are generally the same as in the LL&M algorithm; 0N/A * (right shift) multiplication products as soon as they are formed, 0N/A * rather than carrying additional fractional bits into subsequent additions. 0N/A * This compromises accuracy slightly, but it lets us save a few shifts. 0N/A * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples) 0N/A * everywhere except in the multiplications proper; this saves a good deal 0N/A * of work on 16-bit-int machines. 0N/A * Again to save a few shifts, the intermediate results between pass 1 and 0N/A * pass 2 are not upscaled, but are represented only to integral precision. 0N/A * A final compromise is to represent the multiplicative constants to only 0N/A * 8 fractional bits, rather than 13. This saves some shifting work on some 0N/A * machines, and may also reduce the cost of multiplication (since there 0N/A * are fewer one-bits in the constants). 0N/A/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus 0N/A * causing a lot of useless floating-point operations at run time. 0N/A * To get around this we use the following pre-calculated constants. 0N/A * If you change CONST_BITS you may want to add appropriate values. 0N/A * (With a reasonable C compiler, you can just rely on the FIX() macro...) 0N/A/* We can gain a little more speed, with a further compromise in accuracy, 0N/A * by omitting the addition in a descaling shift. This yields an incorrectly 0N/A * rounded result half the time... 0N/A/* Multiply a DCTELEM variable by an INT32 constant, and immediately 0N/A * descale to yield a DCTELEM result. 0N/A * Perform the forward DCT on one block of samples. 0N/A /* Pass 1: process rows. */ 0N/A /* The rotator is modified from fig 4-8 to avoid extra negations. */ 0N/A /* Pass 2: process columns. */ 0N/A /* The rotator is modified from fig 4-8 to avoid extra negations. */ 0N/A#
endif /* DCT_IFAST_SUPPORTED */