libavcodec/faandct.c - third_party/ffmpeg - Git at Google

 /*
  * Floating point AAN DCT
  * this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
  *
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  * Copyright (c) 2003 Roman Shaposhnik
  *
  * Permission to use, copy, modify, and/or distribute this software for any
  * purpose with or without fee is hereby granted, provided that the above
  * copyright notice and this permission notice appear in all copies.
  *
  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
  */

 /**
  * @file
  * @brief
  *     Floating point AAN DCT
  * @author Michael Niedermayer <michaelni@gmx.at>
  */

 #include "faandct.h"
 #include "libavutil/internal.h"
 #include "libavutil/libm.h"

 typedef float FLOAT;

 /* numbers generated by arbitrary precision arithmetic followed by truncation
 to 36 fractional digits (enough for a 128-bit IEEE quad, see /usr/include/math.h
 for this approach). Unfortunately, long double is not always available correctly,
 e.g ppc has issues.
 TODO: add L suffixes when ppc and toolchains sort out their stuff.
 */
 #define B0 1.000000000000000000000000000000000000
 #define B1 0.720959822006947913789091890943021267 // (cos(pi*1/16)sqrt(2))^-1
 #define B2 0.765366864730179543456919968060797734 // (cos(pi*2/16)sqrt(2))^-1
 #define B3 0.850430094767256448766702844371412325 // (cos(pi*3/16)sqrt(2))^-1
 #define B4 1.000000000000000000000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
 #define B5 1.272758580572833938461007018281767032 // (cos(pi*5/16)sqrt(2))^-1
 #define B6 1.847759065022573512256366378793576574 // (cos(pi*6/16)sqrt(2))^-1
 #define B7 3.624509785411551372409941227504289587 // (cos(pi*7/16)sqrt(2))^-1

 #define A1 M_SQRT1_2              // cos(pi*4/16)
 #define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
 #define A5 0.38268343236508977170 // cos(pi*6/16)
 #define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)

 static const FLOAT postscale[64]={
 B0*B0, B0*B1, B0*B2, B0*B3, B0*B4, B0*B5, B0*B6, B0*B7,
 B1*B0, B1*B1, B1*B2, B1*B3, B1*B4, B1*B5, B1*B6, B1*B7,
 B2*B0, B2*B1, B2*B2, B2*B3, B2*B4, B2*B5, B2*B6, B2*B7,
 B3*B0, B3*B1, B3*B2, B3*B3, B3*B4, B3*B5, B3*B6, B3*B7,
 B4*B0, B4*B1, B4*B2, B4*B3, B4*B4, B4*B5, B4*B6, B4*B7,
 B5*B0, B5*B1, B5*B2, B5*B3, B5*B4, B5*B5, B5*B6, B5*B7,
 B6*B0, B6*B1, B6*B2, B6*B3, B6*B4, B6*B5, B6*B6, B6*B7,
 B7*B0, B7*B1, B7*B2, B7*B3, B7*B4, B7*B5, B7*B6, B7*B7,
 };

 static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
 {
     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
     FLOAT tmp10, tmp11, tmp12, tmp13;
     FLOAT z2, z4, z11, z13;
     int i;

     for (i=0; i<8*8; i+=8) {
         tmp0= data[0 + i] + data[7 + i];
         tmp7= data[0 + i] - data[7 + i];
         tmp1= data[1 + i] + data[6 + i];
         tmp6= data[1 + i] - data[6 + i];
         tmp2= data[2 + i] + data[5 + i];
         tmp5= data[2 + i] - data[5 + i];
         tmp3= data[3 + i] + data[4 + i];
         tmp4= data[3 + i] - data[4 + i];

         tmp10= tmp0 + tmp3;
         tmp13= tmp0 - tmp3;
         tmp11= tmp1 + tmp2;
         tmp12= tmp1 - tmp2;

         temp[0 + i]= tmp10 + tmp11;
         temp[4 + i]= tmp10 - tmp11;

         tmp12 += tmp13;
         tmp12 *= A1;
         temp[2 + i]= tmp13 + tmp12;
         temp[6 + i]= tmp13 - tmp12;

         tmp4 += tmp5;
         tmp5 += tmp6;
         tmp6 += tmp7;

         z2= tmp4*(A2+A5) - tmp6*A5;
         z4= tmp6*(A4-A5) + tmp4*A5;

         tmp5*=A1;

         z11= tmp7 + tmp5;
         z13= tmp7 - tmp5;

         temp[5 + i]= z13 + z2;
         temp[3 + i]= z13 - z2;
         temp[1 + i]= z11 + z4;
         temp[7 + i]= z11 - z4;
     }
 }

 void ff_faandct(int16_t *data)
 {
     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
     FLOAT tmp10, tmp11, tmp12, tmp13;
     FLOAT z2, z4, z11, z13;
     FLOAT temp[64];
     int i;

     emms_c();

     row_fdct(temp, data);

     for (i=0; i<8; i++) {
         tmp0= temp[8*0 + i] + temp[8*7 + i];
         tmp7= temp[8*0 + i] - temp[8*7 + i];
         tmp1= temp[8*1 + i] + temp[8*6 + i];
         tmp6= temp[8*1 + i] - temp[8*6 + i];
         tmp2= temp[8*2 + i] + temp[8*5 + i];
         tmp5= temp[8*2 + i] - temp[8*5 + i];
         tmp3= temp[8*3 + i] + temp[8*4 + i];
         tmp4= temp[8*3 + i] - temp[8*4 + i];

         tmp10= tmp0 + tmp3;
         tmp13= tmp0 - tmp3;
         tmp11= tmp1 + tmp2;
         tmp12= tmp1 - tmp2;

         data[8*0 + i]= lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
         data[8*4 + i]= lrintf(postscale[8*4 + i] * (tmp10 - tmp11));

         tmp12 += tmp13;
         tmp12 *= A1;
         data[8*2 + i]= lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
         data[8*6 + i]= lrintf(postscale[8*6 + i] * (tmp13 - tmp12));

         tmp4 += tmp5;
         tmp5 += tmp6;
         tmp6 += tmp7;

         z2= tmp4*(A2+A5) - tmp6*A5;
         z4= tmp6*(A4-A5) + tmp4*A5;

         tmp5*=A1;

         z11= tmp7 + tmp5;
         z13= tmp7 - tmp5;

         data[8*5 + i]= lrintf(postscale[8*5 + i] * (z13 + z2));
         data[8*3 + i]= lrintf(postscale[8*3 + i] * (z13 - z2));
         data[8*1 + i]= lrintf(postscale[8*1 + i] * (z11 + z4));
         data[8*7 + i]= lrintf(postscale[8*7 + i] * (z11 - z4));
     }
 }

 void ff_faandct248(int16_t *data)
 {
     FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
     FLOAT tmp10, tmp11, tmp12, tmp13;
     FLOAT temp[64];
     int i;

     emms_c();

     row_fdct(temp, data);

     for (i=0; i<8; i++) {
         tmp0 = temp[8*0 + i] + temp[8*1 + i];
         tmp1 = temp[8*2 + i] + temp[8*3 + i];
         tmp2 = temp[8*4 + i] + temp[8*5 + i];
         tmp3 = temp[8*6 + i] + temp[8*7 + i];
         tmp4 = temp[8*0 + i] - temp[8*1 + i];
         tmp5 = temp[8*2 + i] - temp[8*3 + i];
         tmp6 = temp[8*4 + i] - temp[8*5 + i];
         tmp7 = temp[8*6 + i] - temp[8*7 + i];

         tmp10 = tmp0 + tmp3;
         tmp11 = tmp1 + tmp2;
         tmp12 = tmp1 - tmp2;
         tmp13 = tmp0 - tmp3;

         data[8*0 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
         data[8*4 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));

         tmp12 += tmp13;
         tmp12 *= A1;
         data[8*2 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
         data[8*6 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));

         tmp10 = tmp4 + tmp7;
         tmp11 = tmp5 + tmp6;
         tmp12 = tmp5 - tmp6;
         tmp13 = tmp4 - tmp7;

         data[8*1 + i] = lrintf(postscale[8*0 + i] * (tmp10 + tmp11));
         data[8*5 + i] = lrintf(postscale[8*4 + i] * (tmp10 - tmp11));

         tmp12 += tmp13;
         tmp12 *= A1;
         data[8*3 + i] = lrintf(postscale[8*2 + i] * (tmp13 + tmp12));
         data[8*7 + i] = lrintf(postscale[8*6 + i] * (tmp13 - tmp12));
     }
 }
	/*
	* Floating point AAN DCT
	* this implementation is based upon the IJG integer AAN DCT (see jfdctfst.c)
	*
	* Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
	* Copyright (c) 2003 Roman Shaposhnik
	*
	* Permission to use, copy, modify, and/or distribute this software for any
	* purpose with or without fee is hereby granted, provided that the above
	* copyright notice and this permission notice appear in all copies.
	*
	* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
	* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
	* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
	* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
	* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
	* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
	* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
	*/

	/**
	* @file
	* @brief
	* Floating point AAN DCT
	* @author Michael Niedermayer <michaelni@gmx.at>
	*/

	#include "faandct.h"
	#include "libavutil/internal.h"
	#include "libavutil/libm.h"

	typedef float FLOAT;

	/* numbers generated by arbitrary precision arithmetic followed by truncation
	to 36 fractional digits (enough for a 128-bit IEEE quad, see /usr/include/math.h
	for this approach). Unfortunately, long double is not always available correctly,
	e.g ppc has issues.
	TODO: add L suffixes when ppc and toolchains sort out their stuff.
	*/
	#define B0 1.000000000000000000000000000000000000
	#define B1 0.720959822006947913789091890943021267 // (cos(pi*1/16)sqrt(2))^-1
	#define B2 0.765366864730179543456919968060797734 // (cos(pi*2/16)sqrt(2))^-1
	#define B3 0.850430094767256448766702844371412325 // (cos(pi*3/16)sqrt(2))^-1
	#define B4 1.000000000000000000000000000000000000 // (cos(pi*4/16)sqrt(2))^-1
	#define B5 1.272758580572833938461007018281767032 // (cos(pi*5/16)sqrt(2))^-1
	#define B6 1.847759065022573512256366378793576574 // (cos(pi*6/16)sqrt(2))^-1
	#define B7 3.624509785411551372409941227504289587 // (cos(pi*7/16)sqrt(2))^-1

	#define A1 M_SQRT1_2 // cos(pi*4/16)
	#define A2 0.54119610014619698435 // cos(pi*6/16)sqrt(2)
	#define A5 0.38268343236508977170 // cos(pi*6/16)
	#define A4 1.30656296487637652774 // cos(pi*2/16)sqrt(2)

	static const FLOAT postscale[64]={
	B0B0, B0B1, B0B2, B0B3, B0B4, B0B5, B0B6, B0B7,
	B1B0, B1B1, B1B2, B1B3, B1B4, B1B5, B1B6, B1B7,
	B2B0, B2B1, B2B2, B2B3, B2B4, B2B5, B2B6, B2B7,
	B3B0, B3B1, B3B2, B3B3, B3B4, B3B5, B3B6, B3B7,
	B4B0, B4B1, B4B2, B4B3, B4B4, B4B5, B4B6, B4B7,
	B5B0, B5B1, B5B2, B5B3, B5B4, B5B5, B5B6, B5B7,
	B6B0, B6B1, B6B2, B6B3, B6B4, B6B5, B6B6, B6B7,
	B7B0, B7B1, B7B2, B7B3, B7B4, B7B5, B7B6, B7B7,
	};

	static av_always_inline void row_fdct(FLOAT temp[64], int16_t *data)
	{
	FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
	FLOAT tmp10, tmp11, tmp12, tmp13;
	FLOAT z2, z4, z11, z13;
	int i;

	for (i=0; i<8*8; i+=8) {
	tmp0= data[0 + i] + data[7 + i];
	tmp7= data[0 + i] - data[7 + i];
	tmp1= data[1 + i] + data[6 + i];
	tmp6= data[1 + i] - data[6 + i];
	tmp2= data[2 + i] + data[5 + i];
	tmp5= data[2 + i] - data[5 + i];
	tmp3= data[3 + i] + data[4 + i];
	tmp4= data[3 + i] - data[4 + i];

	tmp10= tmp0 + tmp3;
	tmp13= tmp0 - tmp3;
	tmp11= tmp1 + tmp2;
	tmp12= tmp1 - tmp2;

	temp[0 + i]= tmp10 + tmp11;
	temp[4 + i]= tmp10 - tmp11;

	tmp12 += tmp13;
	tmp12 *= A1;
	temp[2 + i]= tmp13 + tmp12;
	temp[6 + i]= tmp13 - tmp12;

	tmp4 += tmp5;
	tmp5 += tmp6;
	tmp6 += tmp7;

	z2= tmp4(A2+A5) - tmp6A5;
	z4= tmp6(A4-A5) + tmp4A5;

	tmp5*=A1;

	z11= tmp7 + tmp5;
	z13= tmp7 - tmp5;

	temp[5 + i]= z13 + z2;
	temp[3 + i]= z13 - z2;
	temp[1 + i]= z11 + z4;
	temp[7 + i]= z11 - z4;
	}
	}

	void ff_faandct(int16_t *data)
	{
	FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
	FLOAT tmp10, tmp11, tmp12, tmp13;
	FLOAT z2, z4, z11, z13;
	FLOAT temp[64];
	int i;

	emms_c();

	row_fdct(temp, data);

	for (i=0; i<8; i++) {
	tmp0= temp[80 + i] + temp[87 + i];
	tmp7= temp[80 + i] - temp[87 + i];
	tmp1= temp[81 + i] + temp[86 + i];
	tmp6= temp[81 + i] - temp[86 + i];
	tmp2= temp[82 + i] + temp[85 + i];
	tmp5= temp[82 + i] - temp[85 + i];
	tmp3= temp[83 + i] + temp[84 + i];
	tmp4= temp[83 + i] - temp[84 + i];

	tmp10= tmp0 + tmp3;
	tmp13= tmp0 - tmp3;
	tmp11= tmp1 + tmp2;
	tmp12= tmp1 - tmp2;

	data[80 + i]= lrintf(postscale[80 + i] * (tmp10 + tmp11));
	data[84 + i]= lrintf(postscale[84 + i] * (tmp10 - tmp11));

	tmp12 += tmp13;
	tmp12 *= A1;
	data[82 + i]= lrintf(postscale[82 + i] * (tmp13 + tmp12));
	data[86 + i]= lrintf(postscale[86 + i] * (tmp13 - tmp12));

	tmp4 += tmp5;
	tmp5 += tmp6;
	tmp6 += tmp7;

	z2= tmp4(A2+A5) - tmp6A5;
	z4= tmp6(A4-A5) + tmp4A5;

	tmp5*=A1;

	z11= tmp7 + tmp5;
	z13= tmp7 - tmp5;

	data[85 + i]= lrintf(postscale[85 + i] * (z13 + z2));
	data[83 + i]= lrintf(postscale[83 + i] * (z13 - z2));
	data[81 + i]= lrintf(postscale[81 + i] * (z11 + z4));
	data[87 + i]= lrintf(postscale[87 + i] * (z11 - z4));
	}
	}

	void ff_faandct248(int16_t *data)
	{
	FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
	FLOAT tmp10, tmp11, tmp12, tmp13;
	FLOAT temp[64];
	int i;

	emms_c();

	row_fdct(temp, data);

	for (i=0; i<8; i++) {
	tmp0 = temp[80 + i] + temp[81 + i];
	tmp1 = temp[82 + i] + temp[83 + i];
	tmp2 = temp[84 + i] + temp[85 + i];
	tmp3 = temp[86 + i] + temp[87 + i];
	tmp4 = temp[80 + i] - temp[81 + i];
	tmp5 = temp[82 + i] - temp[83 + i];
	tmp6 = temp[84 + i] - temp[85 + i];
	tmp7 = temp[86 + i] - temp[87 + i];

	tmp10 = tmp0 + tmp3;
	tmp11 = tmp1 + tmp2;
	tmp12 = tmp1 - tmp2;
	tmp13 = tmp0 - tmp3;

	data[80 + i] = lrintf(postscale[80 + i] * (tmp10 + tmp11));
	data[84 + i] = lrintf(postscale[84 + i] * (tmp10 - tmp11));

	tmp12 += tmp13;
	tmp12 *= A1;
	data[82 + i] = lrintf(postscale[82 + i] * (tmp13 + tmp12));
	data[86 + i] = lrintf(postscale[86 + i] * (tmp13 - tmp12));

	tmp10 = tmp4 + tmp7;
	tmp11 = tmp5 + tmp6;
	tmp12 = tmp5 - tmp6;
	tmp13 = tmp4 - tmp7;

	data[81 + i] = lrintf(postscale[80 + i] * (tmp10 + tmp11));
	data[85 + i] = lrintf(postscale[84 + i] * (tmp10 - tmp11));

	tmp12 += tmp13;
	tmp12 *= A1;
	data[83 + i] = lrintf(postscale[82 + i] * (tmp13 + tmp12));
	data[87 + i] = lrintf(postscale[86 + i] * (tmp13 - tmp12));
	}
	}