nuclear@0: /* nuclear@0: * jidctred.c nuclear@0: * nuclear@0: * Copyright (C) 1994-1998, Thomas G. Lane. nuclear@0: * This file is part of the Independent JPEG Group's software. nuclear@0: * For conditions of distribution and use, see the accompanying README file. nuclear@0: * nuclear@0: * This file contains inverse-DCT routines that produce reduced-size output: nuclear@0: * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block. nuclear@0: * nuclear@0: * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M) nuclear@0: * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step nuclear@0: * with an 8-to-4 step that produces the four averages of two adjacent outputs nuclear@0: * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output). nuclear@0: * These steps were derived by computing the corresponding values at the end nuclear@0: * of the normal LL&M code, then simplifying as much as possible. nuclear@0: * nuclear@0: * 1x1 is trivial: just take the DC coefficient divided by 8. nuclear@0: * nuclear@0: * See jidctint.c for additional comments. nuclear@0: */ nuclear@0: nuclear@0: #define JPEG_INTERNALS nuclear@0: #include "jinclude.h" nuclear@0: #include "jpeglib.h" nuclear@0: #include "jdct.h" /* Private declarations for DCT subsystem */ nuclear@0: nuclear@0: #ifdef IDCT_SCALING_SUPPORTED nuclear@0: nuclear@0: nuclear@0: /* nuclear@0: * This module is specialized to the case DCTSIZE = 8. nuclear@0: */ nuclear@0: nuclear@0: #if DCTSIZE != 8 nuclear@0: Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ nuclear@0: #endif nuclear@0: nuclear@0: nuclear@0: /* Scaling is the same as in jidctint.c. */ nuclear@0: nuclear@0: #if BITS_IN_JSAMPLE == 8 nuclear@0: #define CONST_BITS 13 nuclear@0: #define PASS1_BITS 2 nuclear@0: #else nuclear@0: #define CONST_BITS 13 nuclear@0: #define PASS1_BITS 1 /* lose a little precision to avoid overflow */ nuclear@0: #endif nuclear@0: nuclear@0: /* Some C compilers fail to reduce "FIX(constant)" at compile time, thus nuclear@0: * causing a lot of useless floating-point operations at run time. nuclear@0: * To get around this we use the following pre-calculated constants. nuclear@0: * If you change CONST_BITS you may want to add appropriate values. nuclear@0: * (With a reasonable C compiler, you can just rely on the FIX() macro...) nuclear@0: */ nuclear@0: nuclear@0: #if CONST_BITS == 13 nuclear@0: #define FIX_0_211164243 ((INT32) 1730) /* FIX(0.211164243) */ nuclear@0: #define FIX_0_509795579 ((INT32) 4176) /* FIX(0.509795579) */ nuclear@0: #define FIX_0_601344887 ((INT32) 4926) /* FIX(0.601344887) */ nuclear@0: #define FIX_0_720959822 ((INT32) 5906) /* FIX(0.720959822) */ nuclear@0: #define FIX_0_765366865 ((INT32) 6270) /* FIX(0.765366865) */ nuclear@0: #define FIX_0_850430095 ((INT32) 6967) /* FIX(0.850430095) */ nuclear@0: #define FIX_0_899976223 ((INT32) 7373) /* FIX(0.899976223) */ nuclear@0: #define FIX_1_061594337 ((INT32) 8697) /* FIX(1.061594337) */ nuclear@0: #define FIX_1_272758580 ((INT32) 10426) /* FIX(1.272758580) */ nuclear@0: #define FIX_1_451774981 ((INT32) 11893) /* FIX(1.451774981) */ nuclear@0: #define FIX_1_847759065 ((INT32) 15137) /* FIX(1.847759065) */ nuclear@0: #define FIX_2_172734803 ((INT32) 17799) /* FIX(2.172734803) */ nuclear@0: #define FIX_2_562915447 ((INT32) 20995) /* FIX(2.562915447) */ nuclear@0: #define FIX_3_624509785 ((INT32) 29692) /* FIX(3.624509785) */ nuclear@0: #else nuclear@0: #define FIX_0_211164243 FIX(0.211164243) nuclear@0: #define FIX_0_509795579 FIX(0.509795579) nuclear@0: #define FIX_0_601344887 FIX(0.601344887) nuclear@0: #define FIX_0_720959822 FIX(0.720959822) nuclear@0: #define FIX_0_765366865 FIX(0.765366865) nuclear@0: #define FIX_0_850430095 FIX(0.850430095) nuclear@0: #define FIX_0_899976223 FIX(0.899976223) nuclear@0: #define FIX_1_061594337 FIX(1.061594337) nuclear@0: #define FIX_1_272758580 FIX(1.272758580) nuclear@0: #define FIX_1_451774981 FIX(1.451774981) nuclear@0: #define FIX_1_847759065 FIX(1.847759065) nuclear@0: #define FIX_2_172734803 FIX(2.172734803) nuclear@0: #define FIX_2_562915447 FIX(2.562915447) nuclear@0: #define FIX_3_624509785 FIX(3.624509785) nuclear@0: #endif nuclear@0: nuclear@0: nuclear@0: /* Multiply an INT32 variable by an INT32 constant to yield an INT32 result. nuclear@0: * For 8-bit samples with the recommended scaling, all the variable nuclear@0: * and constant values involved are no more than 16 bits wide, so a nuclear@0: * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. nuclear@0: * For 12-bit samples, a full 32-bit multiplication will be needed. nuclear@0: */ nuclear@0: nuclear@0: #if BITS_IN_JSAMPLE == 8 nuclear@0: #define MULTIPLY(var,const) MULTIPLY16C16(var,const) nuclear@0: #else nuclear@0: #define MULTIPLY(var,const) ((var) * (const)) nuclear@0: #endif nuclear@0: nuclear@0: nuclear@0: /* Dequantize a coefficient by multiplying it by the multiplier-table nuclear@0: * entry; produce an int result. In this module, both inputs and result nuclear@0: * are 16 bits or less, so either int or short multiply will work. nuclear@0: */ nuclear@0: nuclear@0: #define DEQUANTIZE(coef,quantval) (((ISLOW_MULT_TYPE) (coef)) * (quantval)) nuclear@0: nuclear@0: nuclear@0: /* nuclear@0: * Perform dequantization and inverse DCT on one block of coefficients, nuclear@0: * producing a reduced-size 4x4 output block. nuclear@0: */ nuclear@0: nuclear@0: GLOBAL(void) nuclear@0: jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr, nuclear@0: JCOEFPTR coef_block, nuclear@0: JSAMPARRAY output_buf, JDIMENSION output_col) nuclear@0: { nuclear@0: INT32 tmp0, tmp2, tmp10, tmp12; nuclear@0: INT32 z1, z2, z3, z4; nuclear@0: JCOEFPTR inptr; nuclear@0: ISLOW_MULT_TYPE * quantptr; nuclear@0: int * wsptr; nuclear@0: JSAMPROW outptr; nuclear@0: JSAMPLE *range_limit = IDCT_range_limit(cinfo); nuclear@0: int ctr; nuclear@0: int workspace[DCTSIZE*4]; /* buffers data between passes */ nuclear@0: SHIFT_TEMPS nuclear@0: nuclear@0: /* Pass 1: process columns from input, store into work array. */ nuclear@0: nuclear@0: inptr = coef_block; nuclear@0: quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; nuclear@0: wsptr = workspace; nuclear@0: for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { nuclear@0: /* Don't bother to process column 4, because second pass won't use it */ nuclear@0: if (ctr == DCTSIZE-4) nuclear@0: continue; nuclear@0: if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 && nuclear@0: inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*5] == 0 && nuclear@0: inptr[DCTSIZE*6] == 0 && inptr[DCTSIZE*7] == 0) { nuclear@0: /* AC terms all zero; we need not examine term 4 for 4x4 output */ nuclear@0: int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS; nuclear@0: nuclear@0: wsptr[DCTSIZE*0] = dcval; nuclear@0: wsptr[DCTSIZE*1] = dcval; nuclear@0: wsptr[DCTSIZE*2] = dcval; nuclear@0: wsptr[DCTSIZE*3] = dcval; nuclear@0: nuclear@0: continue; nuclear@0: } nuclear@0: nuclear@0: /* Even part */ nuclear@0: nuclear@0: tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); nuclear@0: tmp0 <<= (CONST_BITS+1); nuclear@0: nuclear@0: z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]); nuclear@0: z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]); nuclear@0: nuclear@0: tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865); nuclear@0: nuclear@0: tmp10 = tmp0 + tmp2; nuclear@0: tmp12 = tmp0 - tmp2; nuclear@0: nuclear@0: /* Odd part */ nuclear@0: nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); nuclear@0: z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); nuclear@0: z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); nuclear@0: z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); nuclear@0: nuclear@0: tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ nuclear@0: + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ nuclear@0: + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ nuclear@0: + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ nuclear@0: nuclear@0: tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ nuclear@0: + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ nuclear@0: + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ nuclear@0: + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ nuclear@0: nuclear@0: /* Final output stage */ nuclear@0: nuclear@0: wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1); nuclear@0: wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1); nuclear@0: wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1); nuclear@0: wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1); nuclear@0: } nuclear@0: nuclear@0: /* Pass 2: process 4 rows from work array, store into output array. */ nuclear@0: nuclear@0: wsptr = workspace; nuclear@0: for (ctr = 0; ctr < 4; ctr++) { nuclear@0: outptr = output_buf[ctr] + output_col; nuclear@0: /* It's not clear whether a zero row test is worthwhile here ... */ nuclear@0: nuclear@0: #ifndef NO_ZERO_ROW_TEST nuclear@0: if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && nuclear@0: wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { nuclear@0: /* AC terms all zero */ nuclear@0: JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) nuclear@0: & RANGE_MASK]; nuclear@0: nuclear@0: outptr[0] = dcval; nuclear@0: outptr[1] = dcval; nuclear@0: outptr[2] = dcval; nuclear@0: outptr[3] = dcval; nuclear@0: nuclear@0: wsptr += DCTSIZE; /* advance pointer to next row */ nuclear@0: continue; nuclear@0: } nuclear@0: #endif nuclear@0: nuclear@0: /* Even part */ nuclear@0: nuclear@0: tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1); nuclear@0: nuclear@0: tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065) nuclear@0: + MULTIPLY((INT32) wsptr[6], - FIX_0_765366865); nuclear@0: nuclear@0: tmp10 = tmp0 + tmp2; nuclear@0: tmp12 = tmp0 - tmp2; nuclear@0: nuclear@0: /* Odd part */ nuclear@0: nuclear@0: z1 = (INT32) wsptr[7]; nuclear@0: z2 = (INT32) wsptr[5]; nuclear@0: z3 = (INT32) wsptr[3]; nuclear@0: z4 = (INT32) wsptr[1]; nuclear@0: nuclear@0: tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */ nuclear@0: + MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */ nuclear@0: + MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */ nuclear@0: + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */ nuclear@0: nuclear@0: tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */ nuclear@0: + MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */ nuclear@0: + MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */ nuclear@0: + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ nuclear@0: nuclear@0: /* Final output stage */ nuclear@0: nuclear@0: outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2, nuclear@0: CONST_BITS+PASS1_BITS+3+1) nuclear@0: & RANGE_MASK]; nuclear@0: outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2, nuclear@0: CONST_BITS+PASS1_BITS+3+1) nuclear@0: & RANGE_MASK]; nuclear@0: outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0, nuclear@0: CONST_BITS+PASS1_BITS+3+1) nuclear@0: & RANGE_MASK]; nuclear@0: outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0, nuclear@0: CONST_BITS+PASS1_BITS+3+1) nuclear@0: & RANGE_MASK]; nuclear@0: nuclear@0: wsptr += DCTSIZE; /* advance pointer to next row */ nuclear@0: } nuclear@0: } nuclear@0: nuclear@0: nuclear@0: /* nuclear@0: * Perform dequantization and inverse DCT on one block of coefficients, nuclear@0: * producing a reduced-size 2x2 output block. nuclear@0: */ nuclear@0: nuclear@0: GLOBAL(void) nuclear@0: jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr, nuclear@0: JCOEFPTR coef_block, nuclear@0: JSAMPARRAY output_buf, JDIMENSION output_col) nuclear@0: { nuclear@0: INT32 tmp0, tmp10, z1; nuclear@0: JCOEFPTR inptr; nuclear@0: ISLOW_MULT_TYPE * quantptr; nuclear@0: int * wsptr; nuclear@0: JSAMPROW outptr; nuclear@0: JSAMPLE *range_limit = IDCT_range_limit(cinfo); nuclear@0: int ctr; nuclear@0: int workspace[DCTSIZE*2]; /* buffers data between passes */ nuclear@0: SHIFT_TEMPS nuclear@0: nuclear@0: /* Pass 1: process columns from input, store into work array. */ nuclear@0: nuclear@0: inptr = coef_block; nuclear@0: quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; nuclear@0: wsptr = workspace; nuclear@0: for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { nuclear@0: /* Don't bother to process columns 2,4,6 */ nuclear@0: if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6) nuclear@0: continue; nuclear@0: if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*3] == 0 && nuclear@0: inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*7] == 0) { nuclear@0: /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */ nuclear@0: int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS; nuclear@0: nuclear@0: wsptr[DCTSIZE*0] = dcval; nuclear@0: wsptr[DCTSIZE*1] = dcval; nuclear@0: nuclear@0: continue; nuclear@0: } nuclear@0: nuclear@0: /* Even part */ nuclear@0: nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]); nuclear@0: tmp10 = z1 << (CONST_BITS+2); nuclear@0: nuclear@0: /* Odd part */ nuclear@0: nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]); nuclear@0: tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */ nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]); nuclear@0: tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */ nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]); nuclear@0: tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */ nuclear@0: z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]); nuclear@0: tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ nuclear@0: nuclear@0: /* Final output stage */ nuclear@0: nuclear@0: wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2); nuclear@0: wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2); nuclear@0: } nuclear@0: nuclear@0: /* Pass 2: process 2 rows from work array, store into output array. */ nuclear@0: nuclear@0: wsptr = workspace; nuclear@0: for (ctr = 0; ctr < 2; ctr++) { nuclear@0: outptr = output_buf[ctr] + output_col; nuclear@0: /* It's not clear whether a zero row test is worthwhile here ... */ nuclear@0: nuclear@0: #ifndef NO_ZERO_ROW_TEST nuclear@0: if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) { nuclear@0: /* AC terms all zero */ nuclear@0: JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3) nuclear@0: & RANGE_MASK]; nuclear@0: nuclear@0: outptr[0] = dcval; nuclear@0: outptr[1] = dcval; nuclear@0: nuclear@0: wsptr += DCTSIZE; /* advance pointer to next row */ nuclear@0: continue; nuclear@0: } nuclear@0: #endif nuclear@0: nuclear@0: /* Even part */ nuclear@0: nuclear@0: tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2); nuclear@0: nuclear@0: /* Odd part */ nuclear@0: nuclear@0: tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */ nuclear@0: + MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */ nuclear@0: + MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */ nuclear@0: + MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */ nuclear@0: nuclear@0: /* Final output stage */ nuclear@0: nuclear@0: outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0, nuclear@0: CONST_BITS+PASS1_BITS+3+2) nuclear@0: & RANGE_MASK]; nuclear@0: outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0, nuclear@0: CONST_BITS+PASS1_BITS+3+2) nuclear@0: & RANGE_MASK]; nuclear@0: nuclear@0: wsptr += DCTSIZE; /* advance pointer to next row */ nuclear@0: } nuclear@0: } nuclear@0: nuclear@0: nuclear@0: /* nuclear@0: * Perform dequantization and inverse DCT on one block of coefficients, nuclear@0: * producing a reduced-size 1x1 output block. nuclear@0: */ nuclear@0: nuclear@0: GLOBAL(void) nuclear@0: jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr, nuclear@0: JCOEFPTR coef_block, nuclear@0: JSAMPARRAY output_buf, JDIMENSION output_col) nuclear@0: { nuclear@0: int dcval; nuclear@0: ISLOW_MULT_TYPE * quantptr; nuclear@0: JSAMPLE *range_limit = IDCT_range_limit(cinfo); nuclear@0: SHIFT_TEMPS nuclear@0: nuclear@0: /* We hardly need an inverse DCT routine for this: just take the nuclear@0: * average pixel value, which is one-eighth of the DC coefficient. nuclear@0: */ nuclear@0: quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table; nuclear@0: dcval = DEQUANTIZE(coef_block[0], quantptr[0]); nuclear@0: dcval = (int) DESCALE((INT32) dcval, 3); nuclear@0: nuclear@0: output_buf[0][output_col] = range_limit[dcval & RANGE_MASK]; nuclear@0: } nuclear@0: nuclear@0: #endif /* IDCT_SCALING_SUPPORTED */