jidctred.c

M8手机图片查看器

/*
 * jidctred.c
 *
 * Copyright (C) 1994-1998, Thomas G. Lane.
 * Modification developed 2002-2008 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
 * This file contains inverse-DCT routines that produce reduced-size or
 * enlarged-size output: NxN (N=1...16, standard case N=8 kept separate
 * in jidctint.c), 2NxN, or Nx2N (N=1...8) pixels from an 8x8 DCT block.
 * This implementation is based on NxN point IDCTs (N=1...16).
 *
 * For N<8 we simply take the corresponding low-frequency coefficients of
 * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
 * to yield the downscaled outputs.
 * This can be seen as direct low-pass downsampling from the DCT domain
 * point of view rather than the usual spatial domain point of view,
 * yielding significant computational savings and results at least
 * as good as common bilinear (averaging) spatial downsampling.
 *
 * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
 * lower frequencies and higher frequencies assumed to be zero.
 * It turns out that the computational effort is similar to the 8x8 IDCT
 * regarding the output size.
 * Furthermore, the scaling and descaling is the same for all IDCT sizes.
 *
 * The 2NxN and Nx2N routines are provided for direct resolving the common
 * 2x1 or 1x2 subsampling cases without additional resampling.
 *
 * See jidctint.c for additional comments.
 *
 * CAUTION: We rely on the FIX() macro here except for the N=1,2,4,8 cases
 * since there would be too many additional constants to pre-calculate.
 */

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

#ifdef IDCT_SCALING_SUPPORTED


/*
 * This module is specialized to the case DCTSIZE = 8.
 */

#if DCTSIZE != 8
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
#endif


/* Scaling is the same as in jidctint.c. */

#if BITS_IN_JSAMPLE == 8
#define CONST_BITS  13
#define PASS1_BITS  2
#else
#define CONST_BITS  13
#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
#endif

/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
 * causing a lot of useless floating-point operations at run time.
 * To get around this we use the following pre-calculated constants.
 * If you change CONST_BITS you may want to add appropriate values.
 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
 */

#if CONST_BITS == 13
#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
#else
#define FIX_0_298631336  FIX(0.298631336)
#define FIX_0_390180644  FIX(0.390180644)
#define FIX_0_541196100  FIX(0.541196100)
#define FIX_0_765366865  FIX(0.765366865)
#define FIX_0_899976223  FIX(0.899976223)
#define FIX_1_175875602  FIX(1.175875602)
#define FIX_1_501321110  FIX(1.501321110)
#define FIX_1_847759065  FIX(1.847759065)
#define FIX_1_961570560  FIX(1.961570560)
#define FIX_2_053119869  FIX(2.053119869)
#define FIX_2_562915447  FIX(2.562915447)
#define FIX_3_072711026  FIX(3.072711026)
#endif


/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
 * For 8-bit samples with the recommended scaling, all the variable
 * and constant values involved are no more than 16 bits wide, so a
 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
 * For 12-bit samples, a full 32-bit multiplication will be needed.
 */

#if BITS_IN_JSAMPLE == 8
#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
#else
#define MULTIPLY(var,const)  ((var) * (const))
#endif


/* Dequantize a coefficient by multiplying it by the multiplier-table
 * entry; produce an int result.  In this module, both inputs and result
 * are 16 bits or less, so either int or short multiply will work.
 */


#define DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 7x7 output block.
 *
 * Optimized algorithm with 12 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/14).
 */

GLOBAL(void)
jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[7*7];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp13 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp0 = z1 + z3;
    z2 -= tmp0;
    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);

    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
    tmp1 += tmp2;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
    tmp0 += z2;
    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */

    /* Final output stage */

    wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 7 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add fudge factor here for final descale. */
    tmp13 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
    tmp13 <<= CONST_BITS;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[4];
    z3 = (INT32) wsptr[6];

    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp0 = z1 + z3;
    z2 -= tmp0;
    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];

    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
    tmp1 += tmp2;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
    tmp0 += z2;
    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 7;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 6x6 output block.
 *
 * Optimized algorithm with 3 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/12).
 */

GLOBAL(void)
jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[6*6];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
    tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
    tmp1 = (z1 - z2 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*1] = (int) (tmp11 + tmp1);
    wsptr[6*4] = (int) (tmp11 - tmp1);
    wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 6 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add fudge factor here for final descale. */
    tmp0 = (INT32) wsptr[0] + (ONE << (PASS1_BITS+2));
    tmp0 <<= CONST_BITS;
    tmp2 = (INT32) wsptr[4];
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = tmp0 - tmp10 - tmp10;
    tmp10 = (INT32) wsptr[2];
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    z1 <<= CONST_BITS;
    z2 <<= CONST_BITS;
    z3 <<= CONST_BITS;