sl@0: /*
sl@0:  * LIBOIL - Library of Optimized Inner Loops
sl@0:  * Copyright (c) 2003 David A. Schleef <ds@schleef.org>
sl@0:  * All rights reserved.
sl@0:  *
sl@0:  * Redistribution and use in source and binary forms, with or without
sl@0:  * modification, are permitted provided that the following conditions
sl@0:  * are met:
sl@0:  * 1. Redistributions of source code must retain the above copyright
sl@0:  *    notice, this list of conditions and the following disclaimer.
sl@0:  * 2. Redistributions in binary form must reproduce the above copyright
sl@0:  *    notice, this list of conditions and the following disclaimer in the
sl@0:  *    documentation and/or other materials provided with the distribution.
sl@0:  * 
sl@0:  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
sl@0:  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
sl@0:  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
sl@0:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
sl@0:  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
sl@0:  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
sl@0:  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
sl@0:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
sl@0:  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
sl@0:  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
sl@0:  * POSSIBILITY OF SUCH DAMAGE.
sl@0:  */
sl@0: //Portions Copyright (c)  2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved. 
sl@0: 
sl@0: #ifdef HAVE_CONFIG_H
sl@0: #include "config.h"
sl@0: #endif
sl@0: 
sl@0: #include <liboil/liboilfunction.h>
sl@0: #include "liboil/dct/dct.h"
sl@0: #include <math.h>
sl@0: 
sl@0: /**
sl@0:  * oil_fdct8_f64:
sl@0:  * @d_8:
sl@0:  * @s_8:
sl@0:  * @dstr:
sl@0:  * @sstr:
sl@0:  *
sl@0:  * Performs a Forward Discrete Cosine Transform on @s_8 and places
sl@0:  * the result in @d_8.
sl@0:  */
sl@0: OIL_DEFINE_CLASS (fdct8_f64, "double *d_8, double *s_8, int dstr, int sstr");
sl@0: 
sl@0: #define C0_9808 0.980785280
sl@0: #define C0_9239 0.923879532
sl@0: #define C0_8315 0.831469612
sl@0: #define C0_7071 0.707106781
sl@0: #define C0_5556 0.555570233
sl@0: #define C0_3827 0.382683432
sl@0: #define C0_1951 0.195090322
sl@0: 
sl@0: static void
sl@0: fdct8_f64_ref (double *dest, const double *src, int dstr, int sstr)
sl@0: {
sl@0: 	static double fdct_coeff[8][8];
sl@0: 	static int fdct_coeff_init = 0;
sl@0: 	int i,j;
sl@0: 	double x;
sl@0: 
sl@0: 	if(!fdct_coeff_init){
sl@0: 		double scale;
sl@0: 
sl@0: 		for(i=0;i<8;i++){
sl@0: 			scale = (i==0) ? sqrt(0.125) : 0.5;
sl@0: 			for(j=0;j<8;j++){
sl@0: 				fdct_coeff[j][i] = scale *
sl@0: 					cos((M_PI/8)*i*(j+0.5));
sl@0: 			}
sl@0: 		}
sl@0: 		fdct_coeff_init = 1;
sl@0: 	}
sl@0: 
sl@0: 	for(i=0;i<8;i++){
sl@0: 		x = 0;
sl@0: 		for(j=0;j<8;j++){
sl@0: 			x += fdct_coeff[j][i] * OIL_GET(src,sstr*j, double);
sl@0: 		}
sl@0: 		OIL_GET(dest,dstr*i, double) = x;
sl@0: 	}
sl@0: }
sl@0: 
sl@0: OIL_DEFINE_IMPL_REF (fdct8_f64_ref, fdct8_f64);
sl@0: 
sl@0: /*
sl@0:  * This algorithm is roughly similar to a Fast-Fourier transform,
sl@0:  * taking advantage of the symmeties of the base vectors.  For
sl@0:  * reference, the base vectors are (horizontally):
sl@0:  *
sl@0:  * 0: 1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000  1.0000 
sl@0:  * 1: 0.9808  0.8315  0.5556  0.1951 -0.1951 -0.5556 -0.8315 -0.9808 
sl@0:  * 2: 0.9239  0.3827 -0.3827 -0.9239 -0.9239 -0.3827  0.3827  0.9239 
sl@0:  * 3: 0.8315 -0.1951 -0.9808 -0.5556  0.5556  0.9808  0.1951 -0.8315 
sl@0:  * 4: 0.7071 -0.7071 -0.7071  0.7071  0.7071 -0.7071 -0.7071  0.7071 
sl@0:  * 5: 0.5556 -0.9808  0.1951  0.8315 -0.8315 -0.1951  0.9808 -0.5556 
sl@0:  * 6: 0.3827 -0.9239  0.9239 -0.3827 -0.3827  0.9239 -0.9239  0.3827 
sl@0:  * 7: 0.1951 -0.5556  0.8315 -0.9808  0.9808 -0.8315  0.5556 -0.1951 
sl@0:  *
sl@0:  * The symmetries of note: 
sl@0:  *  - even vectors are symmetric around 4 (the middle)
sl@0:  *  - odd vectors are antisymmetric around 4
sl@0:  *  - 0,4 are symmertic around 2 and 6
sl@0:  *  - 2,6 are antisymmetic around 2 and 6
sl@0:  *
sl@0:  * f0 = (x0 + x7) + (x1 + x6) + (x2 + x5) + (x3 + x4);
sl@0:  * f1 = 0.9808*(x0 - x7) + 0.8315*(x1 - x6) + 0.5556*(x2 - x5) + 0.1951*(x3 - x4)
sl@0:  * f2 = 0.9239*(x0 + x7) + 0.3827*(x1 + x6) - 0.3827*(x2 + x5) - 0.9239*(x3 + x4)
sl@0:  * f3 = 0.8315*(x0 - x7) - 0.1951*(x1 - x6) - 0.9808*(x2 - x5) - 0.5556*(x3 - x4)
sl@0:  * f4 = 0.7071*((x0 + x7) - (x1 + x6) - (x2 + x5) + (x3 + x4))
sl@0:  * f5 = 0.5556*(x0 - x7) - 0.9808*(x1 - x6) + 0.1951*(x2 - x5) + 0.8315*(x3 - x4)
sl@0:  * f6 = 0.3827*(x0 + x7) - 0.9239*(x1 + x6) + 0.9239*(x2 + x5) - 0.3827*(x3 + x4)
sl@0:  * f7 = 0.1951*(x0 - x7) - 0.5556*(x1 - x6) + 0.8315*(x2 - x5) - 0.9808*(x3 - x4)
sl@0:  *
sl@0:  * The even vectors can be further simplified:
sl@0:  *
sl@0:  * f0 = ((x0 + x7) + (x3 + x4)) + ((x1 + x6) + (x2 + x5));
sl@0:  * f2 = 0.9239*((x0 + x7) - (x3 + x4)) + 0.3827*((x1 + x6) - (x2 + x5))
sl@0:  * f4 = 0.7071*((x0 + x7) + (x3 + x4)) - 0.7071*((x1 + x6) + (x2 + x5))
sl@0:  * f6 = 0.3827*((x0 + x7) - (x3 + x4)) - 0.9239*((x1 + x6) - (x2 + x5))
sl@0:  *
sl@0:  * Some implementations move some of the normalization to a later
sl@0:  * stage of processing, saving a few multiplies which get absorbed
sl@0:  * into later multiplies.  However, this incurs a bit of error in
sl@0:  * the integer versions of this function.  Also, if the CPU has a
sl@0:  * multiply-and-add function, you don't gain anything.
sl@0:  */
sl@0: 
sl@0: static void
sl@0: fdct8_f64_fast(double *dest, const double *src, int dstr, int sstr)
sl@0: {
sl@0: 	double s07, s16, s25, s34;
sl@0: 	double d07, d16, d25, d34;
sl@0: 	double ss07s34, ds07s34, ss16s25, ds16s25;
sl@0: 
sl@0: 	s07 = OIL_GET(src,sstr*0,double) + OIL_GET(src,sstr*7,double);
sl@0: 	d07 = OIL_GET(src,sstr*0,double) - OIL_GET(src,sstr*7,double);
sl@0: 	s16 = OIL_GET(src,sstr*1,double) + OIL_GET(src,sstr*6,double);
sl@0: 	d16 = OIL_GET(src,sstr*1,double) - OIL_GET(src,sstr*6,double);
sl@0: 	s25 = OIL_GET(src,sstr*2,double) + OIL_GET(src,sstr*5,double);
sl@0: 	d25 = OIL_GET(src,sstr*2,double) - OIL_GET(src,sstr*5,double);
sl@0: 	s34 = OIL_GET(src,sstr*3,double) + OIL_GET(src,sstr*4,double);
sl@0: 	d34 = OIL_GET(src,sstr*3,double) - OIL_GET(src,sstr*4,double);
sl@0: 
sl@0: 	ss07s34 = s07 + s34;
sl@0: 	ds07s34 = s07 - s34;
sl@0: 	ss16s25 = s16 + s25;
sl@0: 	ds16s25 = s16 - s25;
sl@0: 
sl@0: 	OIL_GET(dest,dstr*0,double) = 0.5*C0_7071*(ss07s34 + ss16s25);
sl@0: 	OIL_GET(dest,dstr*2,double) = 0.5*(C0_9239*ds07s34 + C0_3827*ds16s25);
sl@0: 	OIL_GET(dest,dstr*4,double) = 0.5*C0_7071*(ss07s34 - ss16s25);
sl@0: 	OIL_GET(dest,dstr*6,double) = 0.5*(C0_3827*ds07s34 - C0_9239*ds16s25);
sl@0: 
sl@0: 	OIL_GET(dest,dstr*1,double) = 0.5*(C0_9808*d07 + C0_8315*d16
sl@0: 			+ C0_5556*d25 + C0_1951*d34);
sl@0: 	OIL_GET(dest,dstr*3,double) = 0.5*(C0_8315*d07 - C0_1951*d16
sl@0: 			- C0_9808*d25 - C0_5556*d34);
sl@0: 	OIL_GET(dest,dstr*5,double) = 0.5*(C0_5556*d07 - C0_9808*d16
sl@0: 			+ C0_1951*d25 + C0_8315*d34);
sl@0: 	OIL_GET(dest,dstr*7,double) = 0.5*(C0_1951*d07 - C0_5556*d16
sl@0: 			+ C0_8315*d25 - C0_9808*d34);
sl@0:  
sl@0: #if 0
sl@0: 	z1 = (ds1  tmp12 + tmp13) * 0.707106781;
sl@0: 	OIL_GET(dest,dstr*2,double) = (tmp13 + z1)*(0.25*M_SQRT2)*0.765366864;
sl@0: 	OIL_GET(dest,dstr*6,double) = (tmp13 - z1)*(0.25*M_SQRT2)*1.847759066;
sl@0: 
sl@0: 	tmp10 = tmp4 + tmp5;
sl@0: 	tmp11 = tmp5 + tmp6;
sl@0: 	tmp12 = tmp6 + tmp7;
sl@0: 
sl@0: 	z5 = (tmp10 - tmp12) * 0.382683433;
sl@0: 	z2 = 0.541196100 * tmp10 + z5;
sl@0: 	z4 = 1.306562965 * tmp12 + z5;
sl@0: 	z3 = tmp11 * 0.707106781;
sl@0: 
sl@0: 	z11 = tmp7 + z3;
sl@0: 	z13 = tmp7 - z3;
sl@0: 
sl@0: 	OIL_GET(dest,dstr*5,double) = (z13 + z2)*(0.25*M_SQRT2)*1.2728;
sl@0: 	OIL_GET(dest,dstr*3,double) = (z13 - z2)*(0.25*M_SQRT2)*0.8504;
sl@0: 	OIL_GET(dest,dstr*1,double) = (z11 + z4)*(0.25*M_SQRT2)*0.7209;
sl@0: 	OIL_GET(dest,dstr*7,double) = (z11 - z4)*(0.25*M_SQRT2)*3.6245;
sl@0: #endif
sl@0: }
sl@0: OIL_DEFINE_IMPL (fdct8_f64_fast, fdct8_f64);
sl@0: 
sl@0: 
sl@0: 
sl@0: 
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionClass* __oil_function_class_fdct8_f64() {
sl@0: 		return &_oil_function_class_fdct8_f64;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_fdct8_f64_ref() {
sl@0: 		return &_oil_function_impl_fdct8_f64_ref;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_fdct8_f64_fast() {
sl@0: 		return &_oil_function_impl_fdct8_f64_fast;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: EXPORT_C void** _oil_function_class_ptr_fdct8_f64 ()	{
sl@0: 	oil_function_class_ptr_fdct8_f64 = __oil_function_class_fdct8_f64();
sl@0: 	return &oil_function_class_ptr_fdct8_f64->func;
sl@0: 	}
sl@0: #endif
sl@0: