0N/A/*
0N/A * reserved comment block
0N/A * DO NOT REMOVE OR ALTER!
0N/A */
0N/A/*
0N/A * jidctred.c
0N/A *
0N/A * Copyright (C) 1994-1998, 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 *
0N/A * This file contains inverse-DCT routines that produce reduced-size output:
0N/A * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block.
0N/A *
0N/A * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M)
0N/A * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step
0N/A * with an 8-to-4 step that produces the four averages of two adjacent outputs
0N/A * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output).
0N/A * These steps were derived by computing the corresponding values at the end
0N/A * of the normal LL&M code, then simplifying as much as possible.
0N/A *
0N/A * 1x1 is trivial: just take the DC coefficient divided by 8.
0N/A *
0N/A * See jidctint.c for additional comments.
0N/A */
0N/A
0N/A#define JPEG_INTERNALS
0N/A#include "jinclude.h"
0N/A#include "jpeglib.h"
0N/A#include "jdct.h" /* Private declarations for DCT subsystem */
0N/A
0N/A#ifdef IDCT_SCALING_SUPPORTED
0N/A
0N/A
0N/A/*
0N/A * This module is specialized to the case DCTSIZE = 8.
0N/A */
0N/A
0N/A#if DCTSIZE != 8
0N/A Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
0N/A#endif
0N/A
0N/A
0N/A/* Scaling is the same as in jidctint.c. */
0N/A
0N/A#if BITS_IN_JSAMPLE == 8
0N/A#define CONST_BITS 13
0N/A#define PASS1_BITS 2
0N/A#else
0N/A#define CONST_BITS 13
0N/A#define PASS1_BITS 1 /* lose a little precision to avoid overflow */
0N/A#endif
0N/A
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 */
0N/A
0N/A#if CONST_BITS == 13
0N/A#define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */
0N/A#define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */
0N/A#define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */
0N/A#define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */
0N/A#define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */
0N/A#define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */
0N/A#define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */
0N/A#define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */
0N/A#define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */
0N/A#define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */
0N/A#define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */
0N/A#define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */
0N/A#define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */
0N/A#define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */
0N/A#else
0N/A#define FIX_0_211164243 FIX(0.211164243)
0N/A#define FIX_0_509795579 FIX(0.509795579)
0N/A#define FIX_0_601344887 FIX(0.601344887)
0N/A#define FIX_0_720959822 FIX(0.720959822)
0N/A#define FIX_0_765366865 FIX(0.765366865)
0N/A#define FIX_0_850430095 FIX(0.850430095)
0N/A#define FIX_0_899976223 FIX(0.899976223)
0N/A#define FIX_1_061594337 FIX(1.061594337)
0N/A#define FIX_1_272758580 FIX(1.272758580)
0N/A#define FIX_1_451774981 FIX(1.451774981)
0N/A#define FIX_1_847759065 FIX(1.847759065)
0N/A#define FIX_2_172734803 FIX(2.172734803)
0N/A#define FIX_2_562915447 FIX(2.562915447)
0N/A#define FIX_3_624509785 FIX(3.624509785)
0N/A#endif
0N/A
0N/A
0N/A/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
0N/A * For 8-bit samples with the recommended scaling, all the variable
0N/A * and constant values involved are no more than 16 bits wide, so a
0N/A * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
0N/A * For 12-bit samples, a full 32-bit multiplication will be needed.
0N/A */
0N/A
0N/A#if BITS_IN_JSAMPLE == 8
0N/A#define MULTIPLY(var,const) MULTIPLY16C16(var,const)
0N/A#else
0N/A#define MULTIPLY(var,const) ((var) * (const))
0N/A#endif
0N/A
0N/A
0N/A/* Dequantize a coefficient by multiplying it by the multiplier-table
0N/A * entry; produce an int result. In this module, both inputs and result
0N/A * are 16 bits or less, so either int or short multiply will work.
0N/A */
0N/A
0N/A#define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval))
0N/A
0N/A
0N/A/*
0N/A * Perform dequantization and inverse DCT on one block of coefficients,
0N/A * producing a reduced-size 4x4 output block.
0N/A */
0N/A
0N/AGLOBAL(void)
0N/Ajpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
0N/A JCOEFPTR coef_block,
0N/A JSAMPARRAY output_buf, JDIMENSION output_col)
0N/A{
0N/A INT32 tmp0, tmp2, tmp10, tmp12;
0N/A INT32 z1, z2, z3, z4;
0N/A JCOEFPTR inptr;
0N/A ISLOW_MULT_TYPE * quantptr;
0N/A int * wsptr;
0N/A JSAMPROW outptr;
0N/A JSAMPLE *range_limit = IDCT_range_limit(cinfo);
0N/A int ctr;
0N/A int workspace[DCTSIZE*4]; /* buffers data between passes */
0N/A SHIFT_TEMPS
0N/A
0N/A /* Pass 1: process columns from input, store into work array. */
0N/A
0N/A inptr = coef_block;
0N/A quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
0N/A wsptr = workspace;
0N/A for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
0N/A /* Don't bother to process column 4, because second pass won't use it */
0N/A if (ctr == DCTSIZE-4)
0N/A continue;
0N/A if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
0N/A inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 &&
0N/A inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) {
0N/A /* AC terms all zero; we need not examine term 4 for 4x4 output */
0N/A int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
0N/A
0N/A wsptr[DCTSIZE*0] = dcval;
0N/A wsptr[DCTSIZE*1] = dcval;
0N/A wsptr[DCTSIZE*2] = dcval;
0N/A wsptr[DCTSIZE*3] = dcval;
0N/A
0N/A continue;
0N/A }
0N/A
0N/A /* Even part */
0N/A
0N/A tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
0N/A tmp0 <<= (CONST_BITS+1);
0N/A
0N/A z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
0N/A z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
0N/A
0N/A tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865);
0N/A
0N/A tmp10 = tmp0 + tmp2;
0N/A tmp12 = tmp0 - tmp2;
0N/A
0N/A /* Odd part */
0N/A
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
0N/A z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
0N/A z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
0N/A z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
0N/A
0N/A tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
0N/A + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
0N/A + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
0N/A + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
0N/A
0N/A tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
0N/A + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
0N/A + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
0N/A + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
0N/A
0N/A /* Final output stage */
0N/A
0N/A wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1);
0N/A wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1);
0N/A wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1);
0N/A wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1);
0N/A }
0N/A
0N/A /* Pass 2: process 4 rows from work array, store into output array. */
0N/A
0N/A wsptr = workspace;
0N/A for (ctr = 0; ctr < 4; ctr++) {
0N/A outptr = output_buf[ctr] + output_col;
0N/A /* It's not clear whether a zero row test is worthwhile here ... */
0N/A
0N/A#ifndef NO_ZERO_ROW_TEST
0N/A if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 &&
0N/A wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
0N/A /* AC terms all zero */
0N/A JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
0N/A & RANGE_MASK];
0N/A
0N/A outptr[0] = dcval;
0N/A outptr[1] = dcval;
0N/A outptr[2] = dcval;
0N/A outptr[3] = dcval;
0N/A
0N/A wsptr += DCTSIZE; /* advance pointer to next row */
0N/A continue;
0N/A }
0N/A#endif
0N/A
0N/A /* Even part */
0N/A
0N/A tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);
0N/A
0N/A tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)
0N/A + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865);
0N/A
0N/A tmp10 = tmp0 + tmp2;
0N/A tmp12 = tmp0 - tmp2;
0N/A
0N/A /* Odd part */
0N/A
0N/A z1 = (INT32) wsptr[7];
0N/A z2 = (INT32) wsptr[5];
0N/A z3 = (INT32) wsptr[3];
0N/A z4 = (INT32) wsptr[1];
0N/A
0N/A tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
0N/A + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
0N/A + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
0N/A + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
0N/A
0N/A tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
0N/A + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
0N/A + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
0N/A + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
0N/A
0N/A /* Final output stage */
0N/A
0N/A outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,
0N/A CONST_BITS+PASS1_BITS+3+1)
0N/A & RANGE_MASK];
0N/A outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2,
0N/A CONST_BITS+PASS1_BITS+3+1)
0N/A & RANGE_MASK];
0N/A outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,
0N/A CONST_BITS+PASS1_BITS+3+1)
0N/A & RANGE_MASK];
0N/A outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0,
0N/A CONST_BITS+PASS1_BITS+3+1)
0N/A & RANGE_MASK];
0N/A
0N/A wsptr += DCTSIZE; /* advance pointer to next row */
0N/A }
0N/A}
0N/A
0N/A
0N/A/*
0N/A * Perform dequantization and inverse DCT on one block of coefficients,
0N/A * producing a reduced-size 2x2 output block.
0N/A */
0N/A
0N/AGLOBAL(void)
0N/Ajpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
0N/A JCOEFPTR coef_block,
0N/A JSAMPARRAY output_buf, JDIMENSION output_col)
0N/A{
0N/A INT32 tmp0, tmp10, z1;
0N/A JCOEFPTR inptr;
0N/A ISLOW_MULT_TYPE * quantptr;
0N/A int * wsptr;
0N/A JSAMPROW outptr;
0N/A JSAMPLE *range_limit = IDCT_range_limit(cinfo);
0N/A int ctr;
0N/A int workspace[DCTSIZE*2]; /* buffers data between passes */
0N/A SHIFT_TEMPS
0N/A
0N/A /* Pass 1: process columns from input, store into work array. */
0N/A
0N/A inptr = coef_block;
0N/A quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
0N/A wsptr = workspace;
0N/A for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
0N/A /* Don't bother to process columns 2,4,6 */
0N/A if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6)
0N/A continue;
0N/A if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 &&
0N/A inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) {
0N/A /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
0N/A int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
0N/A
0N/A wsptr[DCTSIZE*0] = dcval;
0N/A wsptr[DCTSIZE*1] = dcval;
0N/A
0N/A continue;
0N/A }
0N/A
0N/A /* Even part */
0N/A
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
0N/A tmp10 = z1 << (CONST_BITS+2);
0N/A
0N/A /* Odd part */
0N/A
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
0N/A tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
0N/A tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
0N/A tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */
0N/A z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
0N/A tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
0N/A
0N/A /* Final output stage */
0N/A
0N/A wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2);
0N/A wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2);
0N/A }
0N/A
0N/A /* Pass 2: process 2 rows from work array, store into output array. */
0N/A
0N/A wsptr = workspace;
0N/A for (ctr = 0; ctr < 2; ctr++) {
0N/A outptr = output_buf[ctr] + output_col;
0N/A /* It's not clear whether a zero row test is worthwhile here ... */
0N/A
0N/A#ifndef NO_ZERO_ROW_TEST
0N/A if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) {
0N/A /* AC terms all zero */
0N/A JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
0N/A & RANGE_MASK];
0N/A
0N/A outptr[0] = dcval;
0N/A outptr[1] = dcval;
0N/A
0N/A wsptr += DCTSIZE; /* advance pointer to next row */
0N/A continue;
0N/A }
0N/A#endif
0N/A
0N/A /* Even part */
0N/A
0N/A tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);
0N/A
0N/A /* Odd part */
0N/A
0N/A tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
0N/A + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
0N/A + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
0N/A + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
0N/A
0N/A /* Final output stage */
0N/A
0N/A outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,
0N/A CONST_BITS+PASS1_BITS+3+2)
0N/A & RANGE_MASK];
0N/A outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0,
0N/A CONST_BITS+PASS1_BITS+3+2)
0N/A & RANGE_MASK];
0N/A
0N/A wsptr += DCTSIZE; /* advance pointer to next row */
0N/A }
0N/A}
0N/A
0N/A
0N/A/*
0N/A * Perform dequantization and inverse DCT on one block of coefficients,
0N/A * producing a reduced-size 1x1 output block.
0N/A */
0N/A
0N/AGLOBAL(void)
0N/Ajpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
0N/A JCOEFPTR coef_block,
0N/A JSAMPARRAY output_buf, JDIMENSION output_col)
0N/A{
0N/A int dcval;
0N/A ISLOW_MULT_TYPE * quantptr;
0N/A JSAMPLE *range_limit = IDCT_range_limit(cinfo);
0N/A SHIFT_TEMPS
0N/A
0N/A /* We hardly need an inverse DCT routine for this: just take the
0N/A * average pixel value, which is one-eighth of the DC coefficient.
0N/A */
0N/A quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
0N/A dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
0N/A dcval = (int) DESCALE((INT32) dcval, 3);
0N/A
0N/A output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
0N/A}
0N/A
0N/A#endif /* IDCT_SCALING_SUPPORTED */