Sweden-Number/libs/jxr/image/decode/strInvTransform.c

1889 lines
65 KiB
C

//*@@@+++@@@@******************************************************************
//
// Copyright © Microsoft Corp.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
//
// • Redistributions of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
// • Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
// POSSIBILITY OF SUCH DAMAGE.
//
//*@@@---@@@@******************************************************************
#include "strTransform.h"
#include "strcodec.h"
#include "decode.h"
/** rotation by -pi/8 **/
#define IROTATE1(a, b) (a) -= (((b) + 1) >> 1), (b) += (((a) + 1) >> 1) // this works well too
#define IROTATE2(a, b) (a) -= (((b)*3 + 4) >> 3), (b) += (((a)*3 + 4) >> 3) // this works well too
/** local functions **/
static Void invOddOdd(PixelI *, PixelI *, PixelI *, PixelI *);
static Void invOddOddPost(PixelI *, PixelI *, PixelI *, PixelI *);
static Void invOdd(PixelI *, PixelI *, PixelI *, PixelI *);
static Void strHSTdec(PixelI *, PixelI *, PixelI *, PixelI *);
static Void strHSTdec1(PixelI *, PixelI *);
static Void strHSTdec1_alternate(PixelI *, PixelI *);
static Void strHSTdec1_edge(PixelI *pa, PixelI *pd);
/** IDCT stuff **/
/** reordering should be combined with zigzag scan **/
/** data order before IDCT **/
/** 0 8 4 6 **/
/** 2 10 14 12 **/
/** 1 11 15 13 **/
/** 9 3 7 5 **/
/** data order after IDCT **/
/** 0 1 2 3 **/
/** 4 5 6 7 **/
/** 8 9 10 11 **/
/** 12 13 14 15 **/
Void strIDCT4x4Stage1(PixelI* p)
{
/** top left corner, butterfly => butterfly **/
strDCT2x2up(p + 0, p + 1, p + 2, p + 3);
/** top right corner, -pi/8 rotation => butterfly **/
invOdd(p + 5, p + 4, p + 7, p + 6);
/** bottom left corner, butterfly => -pi/8 rotation **/
invOdd(p + 10, p + 8, p + 11, p + 9);
/** bottom right corner, -pi/8 rotation => -pi/8 rotation **/
invOddOdd(p + 15, p + 14, p + 13, p + 12);
/** butterfly **/
//FOURBUTTERFLY(p, 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15);
FOURBUTTERFLY_HARDCODED1(p);
}
Void strIDCT4x4Stage2(PixelI* p)
{
/** bottom left corner, butterfly => -pi/8 rotation **/
invOdd(p + 32, p + 48, p + 96, p + 112);
/** top right corner, -pi/8 rotation => butterfly **/
invOdd(p + 128, p + 192, p + 144, p + 208);
/** bottom right corner, -pi/8 rotation => -pi/8 rotation **/
invOddOdd(p + 160, p + 224, p + 176, p + 240);
/** top left corner, butterfly => butterfly **/
strDCT2x2up(p + 0, p + 64, p + 16, p + 80);
/** butterfly **/
FOURBUTTERFLY(p, 0, 192, 48, 240, 64, 128, 112, 176, 16, 208, 32, 224, 80, 144, 96, 160);
}
Void strNormalizeDec(PixelI* p, Bool bChroma)
{
int i;
if (!bChroma) {
//for (i = 0; i < 256; i += 16) {
// p[i] <<= 2;
//}
}
else {
for (i = 0; i < 256; i += 16) {
p[i] += p[i];
}
}
}
/** 2x2 DCT with post-scaling - for use on decoder side **/
Void strDCT2x2dnDec(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d, C, t;
a = *pa;
b = *pb;
C = *pc;
d = *pd;
a += d;
b -= C;
t = ((a - b) >> 1);
c = t - d;
d = t - C;
a -= d;
b += c;
*pa = a * 2;
*pb = b * 2;
*pc = c * 2;
*pd = d * 2;
}
/** post filter stuff **/
/** 2-point post for boundaries **/
Void strPost2(PixelI * a, PixelI * b)
{
*b += ((*a + 4) >> 3);
*a += ((*b + 2) >> 2);
*b += ((*a + 4) >> 3);
}
Void strPost2_alternate(PixelI * pa, PixelI * pb)
{
PixelI a, b;
a = *pa;
b = *pb;
/** rotate **/
b += ((a + 2) >> 2);
a += ((b + 1) >> 1);
a += (b >> 5);
a += (b >> 9);
a += (b >> 13);
b += ((a + 2) >> 2);
*pa = a;
*pb = b;
}
Void strPost2x2(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
/** butterflies **/
a += d;
b += c;
d -= (a + 1) >> 1;
c -= (b + 1) >> 1;
/** rotate **/
b += ((a + 2) >> 2);
a += ((b + 1) >> 1);
b += ((a + 2) >> 2);
/** butterflies **/
d += (a + 1) >> 1;
c += (b + 1) >> 1;
a -= d;
b -= c;
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
Void strPost2x2_alternate(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
/** butterflies **/
a += d;
b += c;
d -= (a + 1) >> 1;
c -= (b + 1) >> 1;
/** rotate **/
b += ((a + 2) >> 2);
a += ((b + 1) >> 1);
a += (b >> 5);
a += (b >> 9);
a += (b >> 13);
b += ((a + 2) >> 2);
/** butterflies **/
d += (a + 1) >> 1;
c += (b + 1) >> 1;
a -= d;
b -= c;
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
/** 4-point post for boundaries **/
Void strPost4(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
a += d, b += c;
d -= ((a + 1) >> 1), c -= ((b + 1) >> 1);
IROTATE1(c, d);
d += ((a + 1) >> 1), c += ((b + 1) >> 1);
a -= d - ((d * 3 + 16) >> 5), b -= c - ((c * 3 + 16) >> 5);
d += ((a * 3 + 8) >> 4), c += ((b * 3 + 8) >> 4);
a += ((d * 3 + 16) >> 5), b += ((c * 3 + 16) >> 5);
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
Void strPost4_alternate(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
a += d, b += c;
d -= ((a + 1) >> 1), c -= ((b + 1) >> 1);
strHSTdec1_edge(&a, &d); strHSTdec1_edge(&b, &c);
IROTATE1(c, d);
d += ((a + 1) >> 1), c += ((b + 1) >> 1);
a -= d, b -= c;
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
/*****************************************************************************************
Input data offsets:
(15)(14)|(10+64)(11+64) p0 (15)(14)|(74)(75)
(13)(12)|( 8+64)( 9+64) (13)(12)|(72)(73)
--------+-------------- --------+--------
( 5)( 4)|( 0+64) (1+64) p1 ( 5)( 4)|(64)(65)
( 7)( 6)|( 2+64) (3+64) ( 7)( 6)|(66)(67)
*****************************************************************************************/
Void DCCompensate (PixelI *a, PixelI *b, PixelI *c, PixelI *d, int iDC)
{
iDC = iDC>>1;
*a -= iDC;
*d -= iDC;
*b += iDC;
*c += iDC;
}
#ifndef max
#define max(a,b) (((a) > (b)) ? (a) : (b))
#endif
#ifndef min
#define min(a,b) (((a) < (b)) ? (a) : (b))
#endif
int ClipDCL(int iDCL, int iAltDCL)
{
int iClipDCL = 0;
if (iDCL > 0) {
if (iAltDCL > 0)
iClipDCL = min(iDCL, iAltDCL);
else
iClipDCL = 0;
}
else if (iDCL < 0) {
if (iAltDCL < 0)
iClipDCL = max(iDCL, iAltDCL);
else
iClipDCL = 0;
}
return iClipDCL;
}
Void strPost4x4Stage1Split(PixelI *p0, PixelI *p1, Int iOffset, Int iHPQP, Bool bHPAbsent)
{
int iDCLAlt1, iDCLAlt2, iDCLAlt3, iDCLAlt0;
int iDCL1, iDCL2, iDCL3, iDCL0;
int iTmp1, iTmp2, iTmp3, iTmp0;
PixelI *p2 = p0 + 72 - iOffset;
PixelI *p3 = p1 + 64 - iOffset;
p0 += 12;
p1 += 4;
/** buttefly **/
strDCT2x2dn(p0 + 0, p2 + 0, p1 + 0, p3 + 0);
strDCT2x2dn(p0 + 1, p2 + 1, p1 + 1, p3 + 1);
strDCT2x2dn(p0 + 2, p2 + 2, p1 + 2, p3 + 2);
strDCT2x2dn(p0 + 3, p2 + 3, p1 + 3, p3 + 3);
/** bottom right corner: -pi/8 rotation => -pi/8 rotation **/
invOddOddPost(p3 + 0, p3 + 1, p3 + 2, p3 + 3);
/** anti diagonal corners: rotation by -pi/8 **/
IROTATE1(p1[2], p1[3]);
IROTATE1(p1[0], p1[1]);
IROTATE1(p2[1], p2[3]);
IROTATE1(p2[0], p2[2]);
/** butterfly **/
strHSTdec1(p0 + 0, p3 + 0);
strHSTdec1(p0 + 1, p3 + 1);
strHSTdec1(p0 + 2, p3 + 2);
strHSTdec1(p0 + 3, p3 + 3);
strHSTdec(p0 + 0, p2 + 0, p1 + 0, p3 + 0);
strHSTdec(p0 + 1, p2 + 1, p1 + 1, p3 + 1);
strHSTdec(p0 + 2, p2 + 2, p1 + 2, p3 + 2);
strHSTdec(p0 + 3, p2 + 3, p1 + 3, p3 + 3);
iTmp0 = (*(p0 +0) + *(p1 +0) + *(p2 +0) + *(p3 +0))>>1;
iTmp1 = (*(p0 +1) + *(p1 +1) + *(p2 +1) + *(p3 +1))>>1;
iTmp2 = (*(p0 +2) + *(p1 +2) + *(p2 +2) + *(p3 +2))>>1;
iTmp3 = (*(p0 +3) + *(p1 +3) + *(p2 +3) + *(p3 +3))>>1;
iDCL0 = (iTmp0 * 595 + 65536)>>17; //Approximating 27/5947
iDCL1 = (iTmp1 * 595 + 65536)>>17;
iDCL2 = (iTmp2 * 595 + 65536)>>17;
iDCL3 = (iTmp3 * 595 + 65536)>>17;
if ((abs(iDCL0) < iHPQP && iHPQP > 20) || bHPAbsent) {
iDCLAlt0 = (*(p0 +0) - *(p1 +0) - *(p2 +0) + *(p3 +0))>>1;
iDCL0 = ClipDCL (iDCL0, iDCLAlt0);
DCCompensate (p0 + 0, p2 + 0, p1 + 0, p3 + 0, iDCL0);
}
if ((abs(iDCL1) < iHPQP && iHPQP > 20) || bHPAbsent) {
iDCLAlt1 = (*(p0 +1) - *(p1 +1) - *(p2 +1) + *(p3 +1))>>1;
iDCL1 = ClipDCL (iDCL1, iDCLAlt1);
DCCompensate (p0 + 1, p2 + 1, p1 + 1, p3 + 1, iDCL1);
}
if ((abs(iDCL2) < iHPQP && iHPQP > 20) || bHPAbsent) {
iDCLAlt2 = (*(p0 +2) - *(p1 +2) - *(p2 +2) + *(p3 +2))>>1;
iDCL2 = ClipDCL (iDCL2, iDCLAlt2);
DCCompensate (p0 + 2, p2 + 2, p1 + 2, p3 + 2, iDCL2);
}
if ((abs(iDCL3) < iHPQP && iHPQP > 20) || bHPAbsent) {
iDCLAlt3 = (*(p0 +3) - *(p1 +3) - *(p2 +3) + *(p3 +3))>>1;
iDCL3 = ClipDCL (iDCL3, iDCLAlt3);
DCCompensate (p0 + 3, p2 + 3, p1 + 3, p3 + 3, iDCL3);
}
}
Void strPost4x4Stage1(PixelI* p, Int iOffset, Int iHPQP, Bool bHPAbsent)
{
strPost4x4Stage1Split(p, p + 16, iOffset, iHPQP, bHPAbsent);
}
Void strPost4x4Stage1Split_alternate(PixelI *p0, PixelI *p1, Int iOffset)
{
PixelI *p2 = p0 + 72 - iOffset;
PixelI *p3 = p1 + 64 - iOffset;
p0 += 12;
p1 += 4;
/** buttefly **/
strDCT2x2dn(p0 + 0, p2 + 0, p1 + 0, p3 + 0);
strDCT2x2dn(p0 + 1, p2 + 1, p1 + 1, p3 + 1);
strDCT2x2dn(p0 + 2, p2 + 2, p1 + 2, p3 + 2);
strDCT2x2dn(p0 + 3, p2 + 3, p1 + 3, p3 + 3);
/** bottom right corner: -pi/8 rotation => -pi/8 rotation **/
invOddOddPost(p3 + 0, p3 + 1, p3 + 2, p3 + 3);
/** anti diagonal corners: rotation by -pi/8 **/
IROTATE1(p1[2], p1[3]);
IROTATE1(p1[0], p1[1]);
IROTATE1(p2[1], p2[3]);
IROTATE1(p2[0], p2[2]);
/** butterfly **/
strHSTdec1_alternate(p0 + 0, p3 + 0);
strHSTdec1_alternate(p0 + 1, p3 + 1);
strHSTdec1_alternate(p0 + 2, p3 + 2);
strHSTdec1_alternate(p0 + 3, p3 + 3);
strHSTdec(p0 + 0, p2 + 0, p1 + 0, p3 + 0);
strHSTdec(p0 + 1, p2 + 1, p1 + 1, p3 + 1);
strHSTdec(p0 + 2, p2 + 2, p1 + 2, p3 + 2);
strHSTdec(p0 + 3, p2 + 3, p1 + 3, p3 + 3);
}
Void strPost4x4Stage1_alternate(PixelI* p, Int iOffset)
{
strPost4x4Stage1Split_alternate(p, p + 16, iOffset);
}
/*****************************************************************************************
Input data offsets:
(15)(14)|(10+32)(11+32) p0 (15)(14)|(42)(43)
(13)(12)|( 8+32)( 9+32) (13)(12)|(40)(41)
--------+-------------- --------+--------
( 5)( 4)|( 0+32) (1+32) p1 ( 5)( 4)|(32)(33)
( 7)( 6)|( 2+32) (3+32) ( 7)( 6)|(34)(35)
*****************************************************************************************/
/*****************************************************************************************
Input data offsets:
( -96)(-32)|(32)( 96) p0
( -80)(-16)|(48)(112)
-----------+------------
(-128)(-64)|( 0)( 64) p1
(-112)(-48)|(16)( 80)
*****************************************************************************************/
Void strPost4x4Stage2Split(PixelI* p0, PixelI* p1)
{
/** buttefly **/
strDCT2x2dn(p0 - 96, p0 + 96, p1 - 112, p1 + 80);
strDCT2x2dn(p0 - 32, p0 + 32, p1 - 48, p1 + 16);
strDCT2x2dn(p0 - 80, p0 + 112, p1 - 128, p1 + 64);
strDCT2x2dn(p0 - 16, p0 + 48, p1 - 64, p1 + 0);
/** bottom right corner: -pi/8 rotation => -pi/8 rotation **/
invOddOddPost(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
/** anti diagonal corners: rotation by -pi/8 **/
IROTATE1(p0[ 48], p0[ 32]);
IROTATE1(p0[112], p0[ 96]);
IROTATE1(p1[-64], p1[-128]);
IROTATE1(p1[-48], p1[-112]);
/** butterfly **/
strHSTdec1(p0 - 96, p1 + 80);
strHSTdec1(p0 - 32, p1 + 16);
strHSTdec1(p0 - 80, p1 + 64);
strHSTdec1(p0 - 16, p1 + 0);
strHSTdec(p0 - 96, p1 - 112, p0 + 96, p1 + 80);
strHSTdec(p0 - 32, p1 - 48, p0 + 32, p1 + 16);
strHSTdec(p0 - 80, p1 - 128, p0 + 112, p1 + 64);
strHSTdec(p0 - 16, p1 - 64, p0 + 48, p1 + 0);
}
Void strPost4x4Stage2Split_alternate(PixelI* p0, PixelI* p1)
{
/** buttefly **/
strDCT2x2dn(p0 - 96, p0 + 96, p1 - 112, p1 + 80);
strDCT2x2dn(p0 - 32, p0 + 32, p1 - 48, p1 + 16);
strDCT2x2dn(p0 - 80, p0 + 112, p1 - 128, p1 + 64);
strDCT2x2dn(p0 - 16, p0 + 48, p1 - 64, p1 + 0);
/** bottom right corner: -pi/8 rotation => -pi/8 rotation **/
invOddOddPost(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
/** anti diagonal corners: rotation by -pi/8 **/
IROTATE1(p0[ 48], p0[ 32]);
IROTATE1(p0[112], p0[ 96]);
IROTATE1(p1[-64], p1[-128]);
IROTATE1(p1[-48], p1[-112]);
/** butterfly **/
strHSTdec1_alternate(p0 - 96, p1 + 80);
strHSTdec1_alternate(p0 - 32, p1 + 16);
strHSTdec1_alternate(p0 - 80, p1 + 64);
strHSTdec1_alternate(p0 - 16, p1 + 0);
strHSTdec(p0 - 96, p1 - 112, p0 + 96, p1 + 80);
strHSTdec(p0 - 32, p1 - 48, p0 + 32, p1 + 16);
strHSTdec(p0 - 80, p1 - 128, p0 + 112, p1 + 64);
strHSTdec(p0 - 16, p1 - 64, p0 + 48, p1 + 0);
}
/**
Hadamard+Scale transform
for some strange reason, breaking up the function into two blocks, strHSTdec1 and strHSTdec
seems to work faster
**/
static Void strHSTdec1(PixelI *pa, PixelI *pd)
{
/** different realization : does rescaling as well! **/
PixelI a, d;
a = *pa;
d = *pd;
a += d;
d = (a >> 1) - d;
a += (d * 3 + 0) >> 3;
d += (a * 3 + 0) >> 4;
//a += (d * 3 + 4) >> 3;
*pa = a;
*pd = d;
}
static Void strHSTdec1_alternate(PixelI *pa, PixelI *pd)
{
/** different realization : does rescaling as well! **/
PixelI a, d;
a = *pa;
d = *pd;
a += d;
d = (a >> 1) - d;
a += (d * 3 + 0) >> 3;
d += (a * 3 + 0) >> 4;
//a += (d * 3 + 4) >> 3;
d += (a >> 7);
d -= (a >> 10);
*pa = a;
*pd = d;
}
static Void strHSTdec1_edge (PixelI *pa, PixelI *pd)
{
/** different realization as compared to scaling operator for 2D case **/
PixelI a, d;
a = *pa;
d = *pd;
a += d;
d = (a >> 1) - d;
a += (d * 3 + 0) >> 3;
d += (a * 3 + 0) >> 4;
//Scaling modification of adding 7/1024 in 2 steps (without multiplication by 7).
d += (a >> 7);
d -= (a >> 10);
a += (d * 3 + 4) >> 3;
d -= (a >> 1);
a += d;
// End new operations
*pa = a;
*pd = -d; // Negative sign needed here for 1D scaling case to ensure correct scaling.
}
static Void strHSTdec(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
/** different realization : does rescaling as well! **/
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
b -= c;
a += (d * 3 + 4) >> 3;
d -= (b >> 1);
c = ((a - b) >> 1) - c;
*pc = d;
*pd = c;
*pa = a - c, *pb = b + d;
}
/** Kron(Rotate(pi/8), Rotate(pi/8)) **/
static Void invOddOdd(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d, t1, t2;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
/** butterflies **/
d += a;
c -= b;
a -= (t1 = d >> 1);
b += (t2 = c >> 1);
/** rotate pi/4 **/
a -= (b * 3 + 3) >> 3;
b += (a * 3 + 3) >> 2;
a -= (b * 3 + 4) >> 3;
/** butterflies **/
b -= t2;
a += t1;
c += b;
d -= a;
/** sign flips **/
*pa = a;
*pb = -b;
*pc = -c;
*pd = d;
}
/** Kron(Rotate(pi/8), Rotate(pi/8)) **/
static Void invOddOddPost(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d, t1, t2;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
/** butterflies **/
d += a;
c -= b;
a -= (t1 = d >> 1);
b += (t2 = c >> 1);
/** rotate pi/4 **/
a -= (b * 3 + 6) >> 3;
b += (a * 3 + 2) >> 2;
a -= (b * 3 + 4) >> 3;
/** butterflies **/
b -= t2;
a += t1;
c += b;
d -= a;
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
/** Kron(Rotate(-pi/8), [1 1; 1 -1]/sqrt(2)) **/
/** [D C A B] => [a b c d] **/
Void invOdd(PixelI *pa, PixelI *pb, PixelI *pc, PixelI *pd)
{
PixelI a, b, c, d;
a = *pa;
b = *pb;
c = *pc;
d = *pd;
/** butterflies **/
b += d;
a -= c;
d -= (b) >> 1;
c += (a + 1) >> 1;
/** rotate pi/8 **/
IROTATE2(a, b);
IROTATE2(c, d);
/** butterflies **/
c -= (b + 1) >> 1;
d = ((a + 1) >> 1) - d;
b += c;
a -= d;
*pa = a;
*pb = b;
*pc = c;
*pd = d;
}
/*************************************************************************
Top-level function to inverse tranform possible part of a macroblock
*************************************************************************/
Int invTransformMacroblock(CWMImageStrCodec * pSC)
{
const OVERLAP olOverlap = pSC->WMISCP.olOverlap;
const COLORFORMAT cfColorFormat = pSC->m_param.cfColorFormat;
// const BITDEPTH_BITS bdBitDepth = pSC->WMII.bdBitDepth;
const Bool left = (pSC->cColumn == 0), right = (pSC->cColumn == pSC->cmbWidth);
const Bool top = (pSC->cRow == 0), bottom = (pSC->cRow == pSC->cmbHeight);
const Bool topORbottom = (top || bottom), leftORright = (left || right);
const Bool topORleft = (top || left), bottomORright = (bottom || right);
const size_t mbWidth = pSC->cmbWidth, mbX = pSC->cColumn;
PixelI * p = NULL;// * pt = NULL;
size_t i;
const size_t iChannels = (cfColorFormat == YUV_420 || cfColorFormat == YUV_422) ? 1 : pSC->m_param.cNumChannels;
const size_t tScale = pSC->m_Dparam->cThumbnailScale;
Int j = 0;
Int qp[MAX_CHANNELS], dcqp[MAX_CHANNELS], iStrength = (1 << pSC->WMII.cPostProcStrength);
// ERR_CODE result = ICERR_OK;
Bool bHPAbsent = (pSC->WMISCP.sbSubband == SB_NO_HIGHPASS || pSC->WMISCP.sbSubband == SB_DC_ONLY);
if(pSC->WMII.cPostProcStrength > 0){
// threshold for post processing
for(i = 0; i < iChannels; i ++){
qp[i] = pSC->pTile[pSC->cTileColumn].pQuantizerLP[i][pSC->MBInfo.iQIndexLP].iQP * iStrength * (olOverlap == OL_NONE ? 2 : 1);
dcqp[i] = pSC->pTile[pSC->cTileColumn].pQuantizerDC[i][0].iQP * iStrength;
}
if(left) // a new MB row
slideOneMBRow(pSC->pPostProcInfo, pSC->m_param.cNumChannels, mbWidth, top, bottom); // previous current row becomes previous row
}
//================================================================
// 400_Y, 444_YUV
for (i = 0; i < iChannels && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[i];
PixelI* const p1 = pSC->p1MBbuffer[i];
Int iHPQP = 255;
if (!bHPAbsent)
iHPQP = pSC->pTile[pSC->cTileColumn].pQuantizerHP[i][pSC->MBInfo.iQIndexHP].iQP;
//================================
// second level inverse transform
if (!bottomORright)
{
if(pSC->WMII.cPostProcStrength > 0)
updatePostProcInfo(pSC->pPostProcInfo, p1, mbX, i); // update postproc info before IDCT
strIDCT4x4Stage2(p1);
if (pSC->m_param.bScaledArith) {
strNormalizeDec(p1, (i != 0));
}
}
//================================
// second level inverse overlap
if (OL_TWO == olOverlap)
{
if (leftORright && (!topORbottom))
{
j = left ? 0 : -128;
strPost4(p0 + j + 32, p0 + j + 48, p1 + j + 0, p1 + j + 16);
strPost4(p0 + j + 96, p0 + j + 112, p1 + j + 64, p1 + j + 80);
}
if (!leftORright)
{
if (topORbottom)
{
p = top ? p1 : p0 + 32;
strPost4(p - 128, p - 64, p + 0, p + 64);
strPost4(p - 112, p - 48, p + 16, p + 80);
p = NULL;
}
else
{
strPost4x4Stage2Split(p0, p1);
}
}
}
if(pSC->WMII.cPostProcStrength > 0)
postProcMB(pSC->pPostProcInfo, p0, p1, mbX, i, dcqp[i]); // second stage deblocking
//================================
// first level inverse transform
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
for (j = (left ? 32 : -96); j < (right ? 32 : 160); j += 64)
{
strIDCT4x4Stage1(p0 + j + 0);
strIDCT4x4Stage1(p0 + j + 16);
}
}
if (!bottom)
{
for (j = (left ? 0 : -128); j < (right ? 0 : 128); j += 64)
{
strIDCT4x4Stage1(p1 + j + 0);
strIDCT4x4Stage1(p1 + j + 16);
}
}
//================================
// first level inverse overlap
if (OL_NONE != olOverlap)
{
if (leftORright)
{
j = left ? 0 + 10 : -64 + 14;
if (!top)
{
p = p0 + 16 + j;
strPost4(p + 0, p - 2, p + 6, p + 8);
strPost4(p + 1, p - 1, p + 7, p + 9);
strPost4(p + 16, p + 14, p + 22, p + 24);
strPost4(p + 17, p + 15, p + 23, p + 25);
p = NULL;
}
if (!bottom)
{
p = p1 + j;
strPost4(p + 0, p - 2, p + 6, p + 8);
strPost4(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
if (!topORbottom)
{
strPost4(p0 + 48 + j + 0, p0 + 48 + j - 2, p1 - 10 + j, p1 - 8 + j);
strPost4(p0 + 48 + j + 1, p0 + 48 + j - 1, p1 - 9 + j, p1 - 7 + j);
}
}
if (top)
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
p = p1 + j;
strPost4(p + 5, p + 4, p + 64, p + 65);
strPost4(p + 7, p + 6, p + 66, p + 67);
p = NULL;
strPost4x4Stage1(p1 + j, 0, iHPQP, bHPAbsent);
}
}
else if (bottom)
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
strPost4x4Stage1(p0 + 16 + j, 0, iHPQP, bHPAbsent);
strPost4x4Stage1(p0 + 32 + j, 0, iHPQP, bHPAbsent);
p = p0 + 48 + j;
strPost4(p + 15, p + 14, p + 74, p + 75);
strPost4(p + 13, p + 12, p + 72, p + 73);
p = NULL;
}
}
else
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
strPost4x4Stage1(p0 + 16 + j, 0, iHPQP, bHPAbsent);
strPost4x4Stage1(p0 + 32 + j, 0, iHPQP, bHPAbsent);
strPost4x4Stage1Split(p0 + 48 + j, p1 + j, 0, iHPQP, bHPAbsent);
strPost4x4Stage1(p1 + j, 0, iHPQP, bHPAbsent);
}
}
}
if(pSC->WMII.cPostProcStrength > 0 && (!topORleft))
postProcBlock(pSC->pPostProcInfo, p0, p1, mbX, i, qp[i]); // destairing and first stage deblocking
}
//================================================================
// 420_UV
for (i = 0; i < (YUV_420 == cfColorFormat? 2U : 0U) && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[1 + i];//(0 == i ? pSC->pU0 : pSC->pV0);
PixelI* const p1 = pSC->p1MBbuffer[1 + i];//(0 == i ? pSC->pU1 : pSC->pV1);
Int iHPQP = 255;
if (!bHPAbsent)
iHPQP = pSC->pTile[pSC->cTileColumn].pQuantizerHP[i][pSC->MBInfo.iQIndexHP].iQP;
//========================================
// second level inverse transform (420_UV)
if (!bottomORright)
{
if (!pSC->m_param.bScaledArith) {
strDCT2x2dn(p1, p1 + 32, p1 + 16, p1 + 48);
}
else {
strDCT2x2dnDec(p1, p1 + 32, p1 + 16, p1 + 48);
}
}
//========================================
// second level inverse overlap (420_UV)
if (OL_TWO == olOverlap)
{
if (leftORright && !topORbottom)
{
j = (left ? 0 : -32);
strPost2(p0 + j + 16, p1 + j);
}
if (!leftORright)
{
if (topORbottom)
{
p = (top ? p1 : p0 + 16);
strPost2(p - 32, p);
p = NULL;
}
else{
strPost2x2(p0 - 16, p0 + 16, p1 - 32, p1);
}
}
}
//========================================
// first level inverse transform (420_UV)
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
for (j = (left ? 16 : -16); j < (right ? 16 : 48); j += 32)
{
strIDCT4x4Stage1(p0 + j);
}
}
if (!bottom)
{
for (j = (left ? 0 : -32); j < (right ? 0 : 32); j += 32)
{
strIDCT4x4Stage1(p1 + j);
}
}
//========================================
// first level inverse overlap (420_UV)
if (OL_NONE != olOverlap)
{
if(!left && !top)
{
if (bottom)
{
for (j = -48; j < (right ? -16 : 16); j += 32)
{
p = p0 + j;
strPost4(p + 15, p + 14, p + 42, p + 43);
strPost4(p + 13, p + 12, p + 40, p + 41);
p = NULL;
}
}
else
{
for (j = -48; j < (right ? -16 : 16); j += 32)
{
strPost4x4Stage1Split(p0 + j, p1 - 16 + j, 32, iHPQP, bHPAbsent);
}
}
if (right)
{
if (!bottom)
{
strPost4(p0 - 2 , p0 - 4 , p1 - 28, p1 - 26);
strPost4(p0 - 1 , p0 - 3 , p1 - 27, p1 - 25);
}
strPost4(p0 - 18, p0 - 20, p0 - 12, p0 - 10);
strPost4(p0 - 17, p0 - 19, p0 - 11, p0 - 9);
}
else
{
strPost4x4Stage1(p0 - 32, 32, iHPQP, bHPAbsent);
}
strPost4x4Stage1(p0 - 64, 32, iHPQP, bHPAbsent);
}
else if (top)
{
for (j = (left ? 0: -64); j < (right ? -32: 0); j += 32)
{
p = p1 + j + 4;
strPost4(p + 1, p + 0, p + 28, p + 29);
strPost4(p + 3, p + 2, p + 30, p + 31);
p = NULL;
}
}
else if (left)
{
if (!bottom)
{
strPost4(p0 + 26, p0 + 24, p1 + 0, p1 + 2);
strPost4(p0 + 27, p0 + 25, p1 + 1, p1 + 3);
}
strPost4(p0 + 10, p0 + 8, p0 + 16, p0 + 18);
strPost4(p0 + 11, p0 + 9, p0 + 17, p0 + 19);
}
}
}
//================================================================
// 422_UV
for (i = 0; i < (YUV_422 == cfColorFormat? 2U : 0U) && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[1 + i];//(0 == i ? pSC->pU0 : pSC->pV0);
PixelI* const p1 = pSC->p1MBbuffer[1 + i];//(0 == i ? pSC->pU1 : pSC->pV1);
Int iHPQP = 255;
if (!bHPAbsent)
iHPQP = pSC->pTile[pSC->cTileColumn].pQuantizerHP[i][pSC->MBInfo.iQIndexHP].iQP;
//========================================
// second level inverse transform (422_UV)
if ((!bottomORright) && pSC->m_Dparam->cThumbnailScale < 16)
{
// 1D lossless HT
p1[0] -= ((p1[32] + 1) >> 1);
p1[32] += p1[0];
if (!pSC->m_param.bScaledArith) {
strDCT2x2dn(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
strDCT2x2dn(p1 + 32, p1 + 96, p1 + 48, p1 + 112);
}
else {
strDCT2x2dnDec(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
strDCT2x2dnDec(p1 + 32, p1 + 96, p1 + 48, p1 + 112);
}
}
//========================================
// second level inverse overlap (422_UV)
if (OL_TWO == olOverlap)
{
if (!bottom)
{
if (leftORright)
{
if (!top)
{
j = (left ? 0 : -64);
strPost2(p0 + 48 + j, p1 + j);
}
j = (left ? 16 : -48);
strPost2(p1 + j, p1 + j + 16);
}
else
{
if (top)
{
strPost2(p1 - 64, p1);
}
else
{
strPost2x2(p0 - 16, p0 + 48, p1 - 64, p1);
}
strPost2x2(p1 - 48, p1 + 16, p1 - 32, p1 + 32);
}
}
else if (!leftORright)
{
strPost2(p0 - 16, p0 + 48);
}
}
//========================================
// first level inverse transform (422_UV)
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
for (j = (left ? 48 : -16); j < (right ? 48 : 112); j += 64)
{
strIDCT4x4Stage1(p0 + j);
}
}
if (!bottom)
{
for (j = (left ? 0 : -64); j < (right ? 0 : 64); j += 64)
{
strIDCT4x4Stage1(p1 + j + 0);
strIDCT4x4Stage1(p1 + j + 16);
strIDCT4x4Stage1(p1 + j + 32);
}
}
//========================================
// first level inverse overlap (422_UV)
if (OL_NONE != olOverlap)
{
if (!top)
{
if (leftORright)
{
j = (left ? 32 + 10 : -32 + 14);
p = p0 + j;
strPost4(p + 0, p - 2, p + 6, p + 8);
strPost4(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
for (j = (left ? 0 : -128); j < (right ? -64 : 0); j += 64)
{
strPost4x4Stage1(p0 + j + 32, 0, iHPQP, bHPAbsent);
}
}
if (!bottom)
{
if (leftORright)
{
j = (left ? 0 + 10 : -64 + 14);
p = p1 + j;
strPost4(p + 0, p - 2, p + 6, p + 8);
strPost4(p + 1, p - 1, p + 7, p + 9);
p += 16;
strPost4(p + 0, p - 2, p + 6, p + 8);
strPost4(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
for (j = (left ? 0 : -128); j < (right ? -64 : 0); j += 64)
{
strPost4x4Stage1(p1 + j + 0, 0, iHPQP, bHPAbsent);
strPost4x4Stage1(p1 + j + 16, 0, iHPQP, bHPAbsent);
}
}
if (topORbottom)
{
p = (top ? p1 + 5 : p0 + 48 + 13);
for (j = (left ? 0 : -128); j < (right ? -64 : 0); j += 64)
{
strPost4(p + j + 0, p + j - 1, p + j + 59, p + j + 60);
strPost4(p + j + 2, p + j + 1, p + j + 61, p + j + 62);
}
p = NULL;
}
else
{
if (leftORright)
{
j = (left ? 0 + 0 : -64 + 4);
strPost4(p0 + j + 48 + 10 + 0, p0 + j + 48 + 10 - 2, p1 + j + 0, p1 + j + 2);
strPost4(p0 + j + 48 + 10 + 1, p0 + j + 48 + 10 - 1, p1 + j + 1, p1 + j + 3);
}
for (j = (left ? 0 : -128); j < (right ? -64 : 0); j += 64)
{
strPost4x4Stage1Split(p0 + j + 48, p1 + j + 0, 0, iHPQP, bHPAbsent);
}
}
}
}
return ICERR_OK;
}
Int invTransformMacroblock_alteredOperators_hard(CWMImageStrCodec * pSC)
{
const OVERLAP olOverlap = pSC->WMISCP.olOverlap;
const COLORFORMAT cfColorFormat = pSC->m_param.cfColorFormat;
// const BITDEPTH_BITS bdBitDepth = pSC->WMII.bdBitDepth;
const Bool left = (pSC->cColumn == 0), right = (pSC->cColumn == pSC->cmbWidth);
const Bool top = (pSC->cRow == 0), bottom = (pSC->cRow == pSC->cmbHeight);
const Bool topORbottom = (top || bottom), leftORright = (left || right);
const Bool topORleft = (top || left), bottomORright = (bottom || right);
Bool leftAdjacentColumn = (pSC->cColumn == 1), rightAdjacentColumn = (pSC->cColumn == pSC->cmbWidth - 1);
// Bool topAdjacentRow = (pSC->cRow == 1), bottomAdjacentRow = (pSC->cRow == pSC->cmbHeight - 1);
const size_t mbWidth = pSC->cmbWidth;
PixelI * p = NULL;// * pt = NULL;
size_t i;
const size_t iChannels = (cfColorFormat == YUV_420 || cfColorFormat == YUV_422) ? 1 : pSC->m_param.cNumChannels;
const size_t tScale = pSC->m_Dparam->cThumbnailScale;
Int j = 0;
Int qp[MAX_CHANNELS], dcqp[MAX_CHANNELS], iStrength = (1 << pSC->WMII.cPostProcStrength);
// ERR_CODE result = ICERR_OK;
#define mbX pSC->mbX
#define mbY pSC->mbY
#define tileX pSC->tileX
#define tileY pSC->tileY
#define bVertTileBoundary pSC->bVertTileBoundary
#define bHoriTileBoundary pSC->bHoriTileBoundary
#define bOneMBLeftVertTB pSC->bOneMBLeftVertTB
#define bOneMBRightVertTB pSC->bOneMBRightVertTB
#define iPredBefore pSC->iPredBefore
#define iPredAfter pSC->iPredAfter
if (pSC->WMISCP.bUseHardTileBoundaries) {
//Add tile location information
if (pSC->cColumn == 0) {
bVertTileBoundary = FALSE;
tileY = 0;
}
bOneMBLeftVertTB = bOneMBRightVertTB = FALSE;
if(tileY > 0 && tileY <= pSC->WMISCP.cNumOfSliceMinus1H && (pSC->cColumn - 1) == pSC->WMISCP.uiTileY[tileY])
bOneMBRightVertTB = TRUE;
if(tileY < pSC->WMISCP.cNumOfSliceMinus1H && pSC->cColumn == pSC->WMISCP.uiTileY[tileY + 1]) {
bVertTileBoundary = TRUE;
tileY++;
}
else
bVertTileBoundary = FALSE;
if(tileY < pSC->WMISCP.cNumOfSliceMinus1H && (pSC->cColumn + 1) == pSC->WMISCP.uiTileY[tileY + 1])
bOneMBLeftVertTB = TRUE;
if (pSC->cRow == 0) {
bHoriTileBoundary = FALSE;
tileX = 0;
}
else if(mbY != pSC->cRow && tileX < pSC->WMISCP.cNumOfSliceMinus1V && pSC->cRow == pSC->WMISCP.uiTileX[tileX + 1]) {
bHoriTileBoundary = TRUE;
tileX++;
}
else if(mbY != pSC->cRow)
bHoriTileBoundary = FALSE;
}
else {
bVertTileBoundary = FALSE;
bHoriTileBoundary = FALSE;
bOneMBLeftVertTB = FALSE;
bOneMBRightVertTB = FALSE;
}
mbX = pSC->cColumn, mbY = pSC->cRow;
if(pSC->WMII.cPostProcStrength > 0){
// threshold for post processing
for(i = 0; i < iChannels; i ++){
qp[i] = pSC->pTile[pSC->cTileColumn].pQuantizerLP[i][pSC->MBInfo.iQIndexLP].iQP * iStrength * (olOverlap == OL_NONE ? 2 : 1);
dcqp[i] = pSC->pTile[pSC->cTileColumn].pQuantizerDC[i][0].iQP * iStrength;
}
if(left) // a new MB row
slideOneMBRow(pSC->pPostProcInfo, pSC->m_param.cNumChannels, mbWidth, top, bottom); // previous current row becomes previous row
}
//================================================================
// 400_Y, 444_YUV
for (i = 0; i < iChannels && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[i];
PixelI* const p1 = pSC->p1MBbuffer[i];
//================================
// second level inverse transform
if (!bottomORright)
{
if(pSC->WMII.cPostProcStrength > 0)
updatePostProcInfo(pSC->pPostProcInfo, p1, mbX, i); // update postproc info before IDCT
strIDCT4x4Stage2(p1);
if (pSC->m_param.bScaledArith) {
strNormalizeDec(p1, (i != 0));
}
}
//================================
// second level inverse overlap
if (OL_TWO == olOverlap)
{
/* Corner operations */
if ((top || bHoriTileBoundary) && (left || bVertTileBoundary))
strPost4_alternate(p1 + 0, p1 + 64, p1 + 0 + 16, p1 + 64 + 16);
if ((top || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p1 - 128, p1 - 64, p1 - 128 + 16, p1 - 64 + 16);
if ((bottom || bHoriTileBoundary) && (left || bVertTileBoundary))
strPost4_alternate(p0 + 32, p0 + 96, p0 + 32 + 16, p0 + 96 + 16);
if ((bottom || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p0 - 96, p0 - 32, p0 - 96 + 16, p0 - 32 + 16);
if ((leftORright || bVertTileBoundary) && (!topORbottom && !bHoriTileBoundary))
{
if (left || bVertTileBoundary) {
j = 0;
strPost4_alternate(p0 + j + 32, p0 + j + 48, p1 + j + 0, p1 + j + 16);
strPost4_alternate(p0 + j + 96, p0 + j + 112, p1 + j + 64, p1 + j + 80);
}
if (right || bVertTileBoundary) {
j = -128;
strPost4_alternate(p0 + j + 32, p0 + j + 48, p1 + j + 0, p1 + j + 16);
strPost4_alternate(p0 + j + 96, p0 + j + 112, p1 + j + 64, p1 + j + 80);
}
}
if (!leftORright)
{
if ((topORbottom || bHoriTileBoundary) && !bVertTileBoundary)
{
if (top || bHoriTileBoundary) {
p = p1;
strPost4_alternate(p - 128, p - 64, p + 0, p + 64);
strPost4_alternate(p - 112, p - 48, p + 16, p + 80);
p = NULL;
}
if (bottom || bHoriTileBoundary) {
p = p0 + 32;
strPost4_alternate(p - 128, p - 64, p + 0, p + 64);
strPost4_alternate(p - 112, p - 48, p + 16, p + 80);
p = NULL;
}
}
if (!topORbottom && !bHoriTileBoundary && !bVertTileBoundary)
strPost4x4Stage2Split_alternate(p0, p1);
}
}
if(pSC->WMII.cPostProcStrength > 0)
postProcMB(pSC->pPostProcInfo, p0, p1, mbX, i, dcqp[i]); // second stage deblocking
//================================
// first level inverse transform
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
for (j = (left ? 32 : -96); j < (right ? 32 : 160); j += 64)
{
strIDCT4x4Stage1(p0 + j + 0);
strIDCT4x4Stage1(p0 + j + 16);
}
}
if (!bottom)
{
for (j = (left ? 0 : -128); j < (right ? 0 : 128); j += 64)
{
// if(tScale == 2 && bdBitDepth != BD_1){
// MIPgen(p1 + j + 0);
// MIPgen(p1 + j + 16);
// }
strIDCT4x4Stage1(p1 + j + 0);
strIDCT4x4Stage1(p1 + j + 16);
}
}
//================================
// first level inverse overlap
if (OL_NONE != olOverlap)
{
if (leftORright || bVertTileBoundary)
{
/* Corner operations */
if ((top || bHoriTileBoundary) && (left || bVertTileBoundary))
strPost4_alternate(p1 + 0, p1 + 1, p1 + 2, p1 + 3);
if ((top || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p1 - 59, p1 - 60, p1 - 57, p1 - 58);
if ((bottom || bHoriTileBoundary) && (left || bVertTileBoundary))
strPost4_alternate(p0 + 48 + 10, p0 + 48 + 11, p0 + 48 + 8, p0 + 48 + 9);
if ((bottom || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p0 - 1, p0 - 2, p0 - 3, p0 - 4);
if (left || bVertTileBoundary) {
j = 0 + 10;
if (!top)
{
p = p0 + 16 + j;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
strPost4_alternate(p + 16, p + 14, p + 22, p + 24);
strPost4_alternate(p + 17, p + 15, p + 23, p + 25);
p = NULL;
}
if (!bottom)
{
p = p1 + j;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
if (!topORbottom && !bHoriTileBoundary)
{
strPost4_alternate(p0 + 48 + j + 0, p0 + 48 + j - 2, p1 - 10 + j, p1 - 8 + j);
strPost4_alternate(p0 + 48 + j + 1, p0 + 48 + j - 1, p1 - 9 + j, p1 - 7 + j);
}
}
if (right || bVertTileBoundary) {
j = -64 + 14;
if (!top)
{
p = p0 + 16 + j;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
strPost4_alternate(p + 16, p + 14, p + 22, p + 24);
strPost4_alternate(p + 17, p + 15, p + 23, p + 25);
p = NULL;
}
if (!bottom)
{
p = p1 + j;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
if (!topORbottom && !bHoriTileBoundary)
{
strPost4_alternate(p0 + 48 + j + 0, p0 + 48 + j - 2, p1 - 10 + j, p1 - 8 + j);
strPost4_alternate(p0 + 48 + j + 1, p0 + 48 + j - 1, p1 - 9 + j, p1 - 7 + j);
}
}
}
if (top || bHoriTileBoundary)
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
if (!bVertTileBoundary || j != -64) {
p = p1 + j;
strPost4_alternate(p + 5, p + 4, p + 64, p + 65);
strPost4_alternate(p + 7, p + 6, p + 66, p + 67);
p = NULL;
strPost4x4Stage1_alternate(p1 + j, 0);
}
}
}
if (bottom || bHoriTileBoundary)
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
if (!bVertTileBoundary || j != -64) {
strPost4x4Stage1_alternate(p0 + 16 + j, 0);
strPost4x4Stage1_alternate(p0 + 32 + j, 0);
p = p0 + 48 + j;
strPost4_alternate(p + 15, p + 14, p + 74, p + 75);
strPost4_alternate(p + 13, p + 12, p + 72, p + 73);
p = NULL;
}
}
}
if (!top && !bottom && !bHoriTileBoundary)
{
for (j = (left ? 0 : -192); j < (right ? -64 : 64); j += 64)
{
if (!bVertTileBoundary || j != -64) {
strPost4x4Stage1_alternate(p0 + 16 + j, 0);
strPost4x4Stage1_alternate(p0 + 32 + j, 0);
strPost4x4Stage1Split_alternate(p0 + 48 + j, p1 + j, 0);
strPost4x4Stage1_alternate(p1 + j, 0);
}
}
}
}
if(pSC->WMII.cPostProcStrength > 0 && (!topORleft))
postProcBlock(pSC->pPostProcInfo, p0, p1, mbX, i, qp[i]); // destairing and first stage deblocking
}
//================================================================
// 420_UV
for (i = 0; i < (YUV_420 == cfColorFormat? 2U : 0U) && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[1 + i];//(0 == i ? pSC->pU0 : pSC->pV0);
PixelI* const p1 = pSC->p1MBbuffer[1 + i];//(0 == i ? pSC->pU1 : pSC->pV1);
//========================================
// second level inverse transform (420_UV)
if (!bottomORright)
{
if (!pSC->m_param.bScaledArith) {
strDCT2x2dn(p1, p1 + 32, p1 + 16, p1 + 48);
}
else {
strDCT2x2dnDec(p1, p1 + 32, p1 + 16, p1 + 48);
}
}
//========================================
// second level inverse overlap (420_UV)
if (OL_TWO == olOverlap)
{
if ((leftAdjacentColumn || bOneMBRightVertTB) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p1 - 64 + 0, *(p1 - 64 + 32));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (top || bHoriTileBoundary))
iPredBefore[i][0] = *(p1 + 0);
if ((right || bVertTileBoundary) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p1 - 64 + 32, iPredBefore[i][0]);
if ((leftAdjacentColumn || bOneMBRightVertTB) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p0 - 64 + 16, *(p0 - 64 + 48));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (bottom || bHoriTileBoundary))
iPredBefore[i][1] = *(p0 + 16);
if ((right || bVertTileBoundary) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p0 - 64 + 48, iPredBefore[i][1]);
if ((leftORright || bVertTileBoundary) && !topORbottom && !bHoriTileBoundary)
{
if (left || bVertTileBoundary)
strPost2_alternate(p0 + 0 + 16, p1 + 0);
if (right || bVertTileBoundary)
strPost2_alternate(p0 + -32 + 16, p1 + -32);
}
if (!leftORright)
{
if ((topORbottom || bHoriTileBoundary) && !bVertTileBoundary)
{
if (top || bHoriTileBoundary)
strPost2_alternate(p1 - 32, p1);
if (bottom || bHoriTileBoundary)
strPost2_alternate(p0 + 16 - 32, p0 + 16);
}
else if (!topORbottom && !bHoriTileBoundary && !bVertTileBoundary) {
strPost2x2_alternate(p0 - 16, p0 + 16, p1 - 32, p1);
}
}
if ((leftAdjacentColumn || bOneMBRightVertTB) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p1 - 64 + 0, *(p1 - 64 + 32));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (top || bHoriTileBoundary))
iPredAfter[i][0] = *(p1 + 0);
if ((right || bVertTileBoundary) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p1 - 64 + 32, iPredAfter[i][0]);
if ((leftAdjacentColumn || bOneMBRightVertTB) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p0 - 64 + 16, *(p0 - 64 + 48));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (bottom || bHoriTileBoundary))
iPredAfter[i][1] = *(p0 + 16);
if ((right || bVertTileBoundary) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p0 - 64 + 48, iPredAfter[i][1]);
}
//========================================
// first level inverse transform (420_UV)
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
// In order to allow correction operation of corner chroma overlap operators (fixed)
// processing of left most MB column must be delayed by one MB
// Thus left MB not processed until leftAdjacentColumn = 1
for (j = ((left) ? 48 : ((leftAdjacentColumn || bOneMBRightVertTB) ? -48 : -16)); j < ((right || bVertTileBoundary) ? 16 : 48); j += 32)
{
strIDCT4x4Stage1(p0 + j);
}
}
if (!bottom)
{
// In order to allow correction operation of corner chroma overlap operators (fixed)
// processing of left most MB column must be delayed by one MB
// Thus left MB not processed until leftAdjacentColumn = 1
for (j = ((left) ? 32 : ((leftAdjacentColumn || bOneMBRightVertTB) ? -64 : -32)); j < ((right || bVertTileBoundary) ? 0 : 32); j += 32)
{
strIDCT4x4Stage1(p1 + j);
}
}
//========================================
// first level inverse overlap (420_UV)
if (OL_NONE != olOverlap)
{
/* Corner operations */
/* Change because the top-left corner ICT will not have happened until leftAdjacentColumn ==1 */
if ((top || bHoriTileBoundary) && (leftAdjacentColumn || bOneMBRightVertTB))
strPost4_alternate(p1 - 64 + 0, p1 - 64 + 1, p1 - 64 + 2, p1 - 64 + 3);
if ((top || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p1 - 27, p1 - 28, p1 - 25, p1 - 26);
/* Change because the bottom-left corner ICT will not have happened until leftAdjacentColumn ==1 */
if ((bottom || bHoriTileBoundary) && (leftAdjacentColumn || bOneMBRightVertTB))
strPost4_alternate(p0 - 64 + 16 + 10, p0 - 64 + 16 + 11, p0 - 64 + 16 + 8, p0 - 64 + 16 + 9);
if ((bottom || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p0 - 1, p0 - 2, p0 - 3, p0 - 4);
if(!left && !top)
{
/* Change because the vertical 1-D overlap operations of the left edge pixels cannot be performed until leftAdjacentColumn ==1 */
if (leftAdjacentColumn || bOneMBRightVertTB)
{
if (!bottom && !bHoriTileBoundary)
{
strPost4_alternate(p0 - 64 + 26, p0 - 64 + 24, p1 - 64 + 0, p1 - 64 + 2);
strPost4_alternate(p0 - 64 + 27, p0 - 64 + 25, p1 - 64 + 1, p1 - 64 + 3);
}
strPost4_alternate(p0 - 64 + 10, p0 - 64 + 8, p0 - 64 + 16, p0 - 64 + 18);
strPost4_alternate(p0 - 64 + 11, p0 - 64 + 9, p0 - 64 + 17, p0 - 64 + 19);
}
if (bottom || bHoriTileBoundary)
{
p = p0 + -48;
strPost4_alternate(p + 15, p + 14, p + 42, p + 43);
strPost4_alternate(p + 13, p + 12, p + 40, p + 41);
p = NULL;
if (!right && !bVertTileBoundary)
{
p = p0 + -16;
strPost4_alternate(p + 15, p + 14, p + 42, p + 43);
strPost4_alternate(p + 13, p + 12, p + 40, p + 41);
p = NULL;
}
}
else
{
strPost4x4Stage1Split_alternate(p0 + -48, p1 - 16 + -48, 32);
if (!right && !bVertTileBoundary)
strPost4x4Stage1Split_alternate(p0 + -16, p1 - 16 + -16, 32);
}
if (right || bVertTileBoundary)
{
if (!bottom && !bHoriTileBoundary)
{
strPost4_alternate(p0 - 2 , p0 - 4 , p1 - 28, p1 - 26);
strPost4_alternate(p0 - 1 , p0 - 3 , p1 - 27, p1 - 25);
}
strPost4_alternate(p0 - 18, p0 - 20, p0 - 12, p0 - 10);
strPost4_alternate(p0 - 17, p0 - 19, p0 - 11, p0 - 9);
}
else
{
strPost4x4Stage1_alternate(p0 - 32, 32);
}
strPost4x4Stage1_alternate(p0 - 64, 32);
}
if (top || bHoriTileBoundary)
{
if (!left)
{
p = p1 + -64 + 4;
strPost4_alternate(p + 1, p + 0, p + 28, p + 29);
strPost4_alternate(p + 3, p + 2, p + 30, p + 31);
p = NULL;
}
if (!left && !right && !bVertTileBoundary)
{
p = p1 + -32 + 4;
strPost4_alternate(p + 1, p + 0, p + 28, p + 29);
strPost4_alternate(p + 3, p + 2, p + 30, p + 31);
p = NULL;
}
}
}
}
//================================================================
// 422_UV
for (i = 0; i < (YUV_422 == cfColorFormat? 2U : 0U) && tScale < 16; ++i)
{
PixelI* const p0 = pSC->p0MBbuffer[1 + i];//(0 == i ? pSC->pU0 : pSC->pV0);
PixelI* const p1 = pSC->p1MBbuffer[1 + i];//(0 == i ? pSC->pU1 : pSC->pV1);
//========================================
// second level inverse transform (422_UV)
if ((!bottomORright) && pSC->m_Dparam->cThumbnailScale < 16)
{
// 1D lossless HT
p1[0] -= ((p1[32] + 1) >> 1);
p1[32] += p1[0];
if (!pSC->m_param.bScaledArith) {
strDCT2x2dn(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
strDCT2x2dn(p1 + 32, p1 + 96, p1 + 48, p1 + 112);
}
else {
strDCT2x2dnDec(p1 + 0, p1 + 64, p1 + 16, p1 + 80);
strDCT2x2dnDec(p1 + 32, p1 + 96, p1 + 48, p1 + 112);
}
}
//========================================
// second level inverse overlap (422_UV)
if (OL_TWO == olOverlap)
{
if ((leftAdjacentColumn || bOneMBRightVertTB) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p1 - 128 + 0, *(p1 - 128 + 64));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (top || bHoriTileBoundary))
iPredBefore[i][0] = *(p1 + 0);
if ((right || bVertTileBoundary) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p1 - 128 + 64, iPredBefore[i][0]);
if ((leftAdjacentColumn || bOneMBRightVertTB) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p0 - 128 + 48, *(p0 - 128 + 112));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (bottom || bHoriTileBoundary))
iPredBefore[i][1] = *(p0 + 48);
if ((right || bVertTileBoundary) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_DIFF(p0 - 128 + 112, iPredBefore[i][1]);
if (!bottom)
{
if (leftORright || bVertTileBoundary)
{
if (!top && !bHoriTileBoundary)
{
if (left || bVertTileBoundary)
strPost2_alternate(p0 + 48 + 0, p1 + 0);
if (right || bVertTileBoundary)
strPost2_alternate(p0 + 48 + -64, p1 + -64);
}
if (left || bVertTileBoundary)
strPost2_alternate(p1 + 16, p1 + 16 + 16);
if (right || bVertTileBoundary)
strPost2_alternate(p1 + -48, p1 + -48 + 16);
}
if (!leftORright && !bVertTileBoundary)
{
if (top || bHoriTileBoundary)
strPost2_alternate(p1 - 64, p1);
else
strPost2x2_alternate(p0 - 16, p0 + 48, p1 - 64, p1);
strPost2x2_alternate(p1 - 48, p1 + 16, p1 - 32, p1 + 32);
}
}
if ((bottom || bHoriTileBoundary) && (!leftORright && !bVertTileBoundary))
strPost2_alternate(p0 - 16, p0 + 48);
if ((leftAdjacentColumn || bOneMBRightVertTB) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p1 - 128 + 0, *(p1 - 128 + 64));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (top || bHoriTileBoundary))
iPredAfter[i][0] = *(p1 + 0);
if ((right || bVertTileBoundary) && (top || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p1 - 128 + 64, iPredAfter[i][0]);
if ((leftAdjacentColumn || bOneMBRightVertTB) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p0 - 128 + 48, *(p0 - 128 + 112));
if ((rightAdjacentColumn || bOneMBLeftVertTB) && (bottom || bHoriTileBoundary))
iPredAfter[i][1] = *(p0 + 48);
if ((right || bVertTileBoundary) && (bottom || bHoriTileBoundary))
COMPUTE_CORNER_PRED_ADD(p0 - 128 + 112, iPredAfter[i][1]);
}
//========================================
// first level inverse transform (422_UV)
if(tScale >= 4) // bypass first level transform for 4:1 and smaller thumbnail
continue;
if (!top)
{
// Need to delay processing of left column until leftAdjacentColumn = 1 for corner overlap operators
// Since 422 has no vertical downsampling, no top MB delay of processing is necessary
for (j = (left ? 112 : ((leftAdjacentColumn || bOneMBRightVertTB) ? -80 : -16)); j < ((right || bVertTileBoundary) ? 48 : 112); j += 64)
{
strIDCT4x4Stage1(p0 + j);
}
}
if (!bottom)
{
// Need to delay processing of left column until leftAdjacentColumn = 1 for corner overlap operators
// Since 422 has no vertical downsampling, no top MB delay of processing is necessary
for (j = (left ? 64 : ((leftAdjacentColumn || bOneMBRightVertTB) ? -128 : -64)); j < ((right || bVertTileBoundary) ? 0 : 64); j += 64)
{
strIDCT4x4Stage1(p1 + j + 0);
strIDCT4x4Stage1(p1 + j + 16);
strIDCT4x4Stage1(p1 + j + 32);
}
}
//========================================
// first level inverse overlap (422_UV)
if (OL_NONE != olOverlap)
{
/* Corner operations */
if ((top || bHoriTileBoundary) && (leftAdjacentColumn || bOneMBRightVertTB))
strPost4_alternate(p1 - 128 + 0, p1 - 128 + 1, p1 - 128 + 2, p1 - 128 + 3);
if ((top || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p1 - 59, p1 - 60, p1 - 57, p1 - 58);
if ((bottom || bHoriTileBoundary) && (leftAdjacentColumn || bOneMBRightVertTB))
strPost4_alternate(p0 - 128 + 48 + 10, p0 - 128 + 48 + 11, p0 - 128 + 48 + 8, p0 - 128 + 48 + 9);
if ((bottom || bHoriTileBoundary) && (right || bVertTileBoundary))
strPost4_alternate(p0 - 1, p0 - 2, p0 - 3, p0 - 4);
if (!top)
{
// Need to delay processing of left column until leftAdjacentColumn = 1 for corner overlap operators
if (leftAdjacentColumn || bOneMBRightVertTB) {
p = p0 + 32 + 10 - 128;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
if (right || bVertTileBoundary) {
p = p0 + -32 + 14;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
for (j = (left ? 0 : -128); j < ((right || bVertTileBoundary) ? -64 : 0); j += 64)
strPost4x4Stage1_alternate(p0 + j + 32, 0);
}
if (!bottom)
{
// Need to delay processing of left column until leftAdjacentColumn = 1 for corner overlap operators
if (leftAdjacentColumn || bOneMBRightVertTB)
{
p = p1 + 0 + 10 - 128;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p += 16;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
if (right || bVertTileBoundary)
{
p = p1 + -64 + 14;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p += 16;
strPost4_alternate(p + 0, p - 2, p + 6, p + 8);
strPost4_alternate(p + 1, p - 1, p + 7, p + 9);
p = NULL;
}
for (j = (left ? 0 : -128); j < ((right || bVertTileBoundary) ? -64 : 0); j += 64)
{
strPost4x4Stage1_alternate(p1 + j + 0, 0);
strPost4x4Stage1_alternate(p1 + j + 16, 0);
}
}
if (topORbottom || bHoriTileBoundary)
{
if (top || bHoriTileBoundary) {
p = p1 + 5;
for (j = (left ? 0 : -128); j < ((right || bVertTileBoundary) ? -64 : 0); j += 64)
{
strPost4_alternate(p + j + 0, p + j - 1, p + j + 59, p + j + 60);
strPost4_alternate(p + j + 2, p + j + 1, p + j + 61, p + j + 62);
}
p = NULL;
}
if (bottom || bHoriTileBoundary) {
p = p0 + 48 + 13;
for (j = (left ? 0 : -128); j < ((right || bVertTileBoundary) ? -64 : 0); j += 64)
{
strPost4_alternate(p + j + 0, p + j - 1, p + j + 59, p + j + 60);
strPost4_alternate(p + j + 2, p + j + 1, p + j + 61, p + j + 62);
}
p = NULL;
}
}
else
{
// Need to delay processing of left column until leftAdjacentColumn = 1 for corner overlap operators
if (leftAdjacentColumn || bOneMBRightVertTB)
{
j = 0 + 0 - 128;
strPost4_alternate(p0 + j + 48 + 10 + 0, p0 + j + 48 + 10 - 2, p1 + j + 0, p1 + j + 2);
strPost4_alternate(p0 + j + 48 + 10 + 1, p0 + j + 48 + 10 - 1, p1 + j + 1, p1 + j + 3);
}
if (right || bVertTileBoundary)
{
j = -64 + 4;
strPost4_alternate(p0 + j + 48 + 10 + 0, p0 + j + 48 + 10 - 2, p1 + j + 0, p1 + j + 2);
strPost4_alternate(p0 + j + 48 + 10 + 1, p0 + j + 48 + 10 - 1, p1 + j + 1, p1 + j + 3);
}
for (j = (left ? 0 : -128); j < ((right || bVertTileBoundary) ? -64 : 0); j += 64)
strPost4x4Stage1Split_alternate(p0 + j + 48, p1 + j + 0, 0);
}
}
}
return ICERR_OK;
}