int main(int argc, char *argv[]) {
printf("Linear Regression\n");
printf("=================\n\n");
// Check for a filename
if (argc == 2) {
FILE *input = fopen(argv[1], "r");
if (input == NULL) {
perror(argv[1]);
exit(1);
} else {
DataSet theData = load_data(input);
fclose(input);
LinRegResult linReg = linear_regression(theData);
clean(theData);
printf("\nSlope \tm = %15.6e\n", DESCALE(linReg.m)); // print slope
printf("y-intercept\tb = %15.6e\n", DESCALE(linReg.b)); // print y-intercept
printf("Correlation\tr = %15.6e\n", DESCALE(linReg.r)); // print correlation
}
} else {
printf("ERROR: Must enter filename after ./linReg\n");
}
return 0;
}
int dotProd(int *a, int *b, int n) {
double dotProd = 0;
int result = 0;
int i;
for (i = 0; i < n; i++) {
dotProd += DESCALE(a[i]) * DESCALE(b[i]);
}
result = dotProd * SCALE;
return result;
}
ff_fdct248_islow (DCTELEM * data)
{
int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
int_fast32_t tmp10, tmp11, tmp12, tmp13;
int_fast32_t z1;
DCTELEM *dataptr;
int ctr;
row_fdct(data);
/* Pass 2: process columns.
* We remove the PASS1_BITS scaling, but leave the results scaled up
* by an overall factor of 8.
*/
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--)
{
tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*1];
tmp1 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*3];
tmp2 = dataptr[DCTSIZE*4] + dataptr[DCTSIZE*5];
tmp3 = dataptr[DCTSIZE*6] + dataptr[DCTSIZE*7];
tmp4 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*1];
tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*3];
tmp6 = dataptr[DCTSIZE*4] - dataptr[DCTSIZE*5];
tmp7 = dataptr[DCTSIZE*6] - dataptr[DCTSIZE*7];
tmp10 = tmp0 + tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
tmp13 = tmp0 - tmp3;
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS+PASS1_BITS);
tmp10 = tmp4 + tmp7;
tmp11 = tmp5 + tmp6;
tmp12 = tmp5 - tmp6;
tmp13 = tmp4 - tmp7;
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS+PASS1_BITS);
dataptr++; /* advance pointer to next column */
}
}
/* Inverse 2-D Discrete Cosine Transform. */
void IDCT(const FBlock * input, PBlock * output)
{
int Y[64];
int k, l;
/* Pass 1: process rows. */
for (k = 0; k < 8; k++) {
/* Prescale k-th row: */
for (l = 0; l < 8; l++)
Y(k, l) = SCALE(input->block[k][l], S_BITS);
/* 1-D IDCT on k-th row: */
idct_1d(&Y(k, 0));
/* Result Y is scaled up by factor sqrt(8)*2^S_BITS. */
}
/* Pass 2: process columns. */
for (l = 0; l < 8; l++) {
int Yc[8];
for (k = 0; k < 8; k++)
Yc[k] = Y(k, l);
/* 1-D IDCT on l-th column: */
idct_1d(Yc);
/* Result is once more scaled up by a factor sqrt(8). */
for (k = 0; k < 8; k++) {
int r = 128 + DESCALE(Yc[k], S_BITS + 3); /* includes level shift */
/* Clip to 8 bits unsigned: */
r = r > 0 ? (r < 255 ? r : 255) : 0;
X(k, l) = r;
}
}
}
static void idct1(int *block)
{
int i, val;
val = RANGE(DESCALE(block[0], PASS1_BITS+3));
for(i=0;i<DCTSIZE2;i++)
block[i]=val;
}
LinRegResult linear_regression(DataSet theData) {
LinRegResult result;
int n = theData.n; // number of data points
double sumx = DESCALE(sum(theData.x, n)); // sum of x
double sumxx = DESCALE(dotProd(theData.x, theData.x, n)); // sum of each x squared
double sumy = DESCALE(sum(theData.y, n)); // sum of y
double sumyy = DESCALE(dotProd(theData.y, theData.y, n)); // sum of each y squared
double sumxy = DESCALE(dotProd(theData.x, theData.y, n)); // sum of each x * y
double m, b, r;
// Compute least-squares best fit straight line
m = (n * sumxy - sumx * sumy) / (n * sumxx - sqr(sumx)); // slope
b = (sumy * sumxx - (sumx * sumxy)) / (n * sumxx - sqr(sumx)); // y-intercept
r = (sumxy - sumx * sumy / n) / sqrt((sumxx - sqr(sumx) / n) * (sumyy - sqr(sumy)/ n)); // correlation
result.m = m * SCALE;
result.b = b * SCALE;
result.r = r * SCALE;
return result;
}
/*
* Perform dequantization and inverse DCT on one block of coefficients,
* producing a reduced-size 1x1 output block.
*/
void JPEG_SWIDCT_1X1(int16 *coef_block, uint8 *output_buf, const int32 *quantptr)
{
int32 dcval;
/* We hardly need an inverse DCT routine for this: just take the
* average pixel value, which is one-eighth of the DC coefficient.
*/
dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
dcval = (int32) DESCALE((int32) dcval, 3);
output_buf[0]/*[0]*/ = s_pClip_table[dcval+128];
}
static void idct(int* tempptr, const jpgd_block_t* dataptr)
{
// will be optimized at compile time to either an array access, or 0
#define ACCESS_COL(x) (((x) < NONZERO_COLS) ? (int)dataptr[x] : 0)
const int z2 = ACCESS_COL(2);
const int z3 = ACCESS_COL(6);
const int z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
const int tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
const int tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
const int tmp0 = (ACCESS_COL(0) + ACCESS_COL(4)) << CONST_BITS;
const int tmp1 = (ACCESS_COL(0) - ACCESS_COL(4)) << CONST_BITS;
const int tmp10 = tmp0 + tmp3;
const int tmp13 = tmp0 - tmp3;
const int tmp11 = tmp1 + tmp2;
const int tmp12 = tmp1 - tmp2;
const int atmp0 = ACCESS_COL(7);
const int atmp1 = ACCESS_COL(5);
const int atmp2 = ACCESS_COL(3);
const int atmp3 = ACCESS_COL(1);
const int bz1 = atmp0 + atmp3;
const int bz2 = atmp1 + atmp2;
const int bz3 = atmp0 + atmp2;
const int bz4 = atmp1 + atmp3;
const int bz5 = MULTIPLY(bz3 + bz4, FIX_1_175875602);
const int az1 = MULTIPLY(bz1, - FIX_0_899976223);
const int az2 = MULTIPLY(bz2, - FIX_2_562915447);
const int az3 = MULTIPLY(bz3, - FIX_1_961570560) + bz5;
const int az4 = MULTIPLY(bz4, - FIX_0_390180644) + bz5;
const int btmp0 = MULTIPLY(atmp0, FIX_0_298631336) + az1 + az3;
const int btmp1 = MULTIPLY(atmp1, FIX_2_053119869) + az2 + az4;
const int btmp2 = MULTIPLY(atmp2, FIX_3_072711026) + az2 + az3;
const int btmp3 = MULTIPLY(atmp3, FIX_1_501321110) + az1 + az4;
tempptr[0] = DESCALE(tmp10 + btmp3, CONST_BITS-PASS1_BITS);
tempptr[7] = DESCALE(tmp10 - btmp3, CONST_BITS-PASS1_BITS);
tempptr[1] = DESCALE(tmp11 + btmp2, CONST_BITS-PASS1_BITS);
tempptr[6] = DESCALE(tmp11 - btmp2, CONST_BITS-PASS1_BITS);
tempptr[2] = DESCALE(tmp12 + btmp1, CONST_BITS-PASS1_BITS);
tempptr[5] = DESCALE(tmp12 - btmp1, CONST_BITS-PASS1_BITS);
tempptr[3] = DESCALE(tmp13 + btmp0, CONST_BITS-PASS1_BITS);
tempptr[4] = DESCALE(tmp13 - btmp0, CONST_BITS-PASS1_BITS);
}
static void dct_1d(dct_data_t *src, dct_data_t *dst)
{
unsigned int k, n;
int tmp;
const dct_data_t dct_coeff_table[DCT_SIZE][DCT_SIZE] = {
#include "dct_coeff_table.txt"
};
for (k = 0; k < DCT_SIZE; k++) {
for(n = 0, tmp = 0; n < DCT_SIZE; n++) {
int coeff = (int)dct_coeff_table[k][n];
tmp += src[n] * coeff;
}
dst[k] = DESCALE(tmp, CONST_BITS);
}
}
openexif_jpeg_idct_1x1 (oe_j_decompress_ptr cinfo, openexif_jpeg_component_info * compptr,
OE_JCOEFPTR coef_block,
OE_JSAMPARRAY output_buf, OE_JDIMENSION output_col)
{
int dcval;
ISLOW_MULT_TYPE * quantptr;
OE_JSAMPLE *range_limit = IDCT_range_limit(cinfo);
SHIFT_TEMPS
/* We hardly need an inverse DCT routine for this: just take the
* average pixel value, which is one-eighth of the DC coefficient.
*/
quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
dcval = (int) DESCALE((INT32) dcval, 3);
output_buf[0][output_col] = range_limit[dcval & RANGE_MASK];
}
METHODDEF void
start_pass_fdctmgr (j_compress_ptr cinfo)
{
my_fdct_ptr fdct = (my_fdct_ptr) cinfo->fdct;
int ci, qtblno, i;
jpeg_component_info *compptr;
JQUANT_TBL * qtbl;
#ifdef DCT_ISLOW_SUPPORTED
DCTELEM * dtbl;
#endif
for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
ci++, compptr++) {
qtblno = compptr->quant_tbl_no;
/* Make sure specified quantization table is present */
if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
cinfo->quant_tbl_ptrs[qtblno] == NULL)
ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
qtbl = cinfo->quant_tbl_ptrs[qtblno];
/* Compute divisors for this quant table */
/* We may do this more than once for same table, but it's not a big deal */
switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
case JDCT_ISLOW:
/* For LL&M IDCT method, divisors are equal to raw quantization
* coefficients multiplied by 8 (to counteract scaling).
*/
if (fdct->divisors[qtblno] == NULL) {
fdct->divisors[qtblno] = (DCTELEM *)
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(DCTELEM));
}
dtbl = fdct->divisors[qtblno];
for (i = 0; i < DCTSIZE2; i++) {
dtbl[i] = ((DCTELEM) qtbl->quantval[jpeg_zigzag_order[i]]) << 3;
}
break;
#endif
#ifdef DCT_IFAST_SUPPORTED
case JDCT_IFAST:
{
/* For AA&N IDCT method, divisors are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* We apply a further scale factor of 8.
*/
#define CONST_BITS 14
static const INT16 aanscales[DCTSIZE2] = {
/* precomputed values scaled up by 14 bits: in natural order */
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
};
SHIFT_TEMPS
if (fdct->divisors[qtblno] == NULL) {
fdct->divisors[qtblno] = (DCTELEM *)
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(DCTELEM));
}
dtbl = fdct->divisors[qtblno];
for (i = 0; i < DCTSIZE2; i++) {
dtbl[i] = (DCTELEM)
DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[jpeg_zigzag_order[i]],
(INT32) aanscales[i]),
CONST_BITS-3);
}
}
break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
case JDCT_FLOAT:
{
/* For float AA&N IDCT method, divisors are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* We apply a further scale factor of 8.
* What's actually stored is 1/divisor so that the inner loop can
* use a multiplication rather than a division.
*/
FAST_FLOAT * fdtbl;
int row, col;
static const double aanscalefactor[DCTSIZE] = {
1.0, 1.387039845, 1.306562965, 1.175875602,
1.0, 0.785694958, 0.541196100, 0.275899379
};
if (fdct->float_divisors[qtblno] == NULL) {
fdct->float_divisors[qtblno] = (FAST_FLOAT *)
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(FAST_FLOAT));
}
fdtbl = fdct->float_divisors[qtblno];
i = 0;
for (row = 0; row < DCTSIZE; row++) {
for (col = 0; col < DCTSIZE; col++) {
fdtbl[i] = (FAST_FLOAT)
(1.0 / (((double) qtbl->quantval[jpeg_zigzag_order[i]] *
aanscalefactor[row] * aanscalefactor[col] * 8.0)));
i++;
}
}
}
break;
#endif
default:
ERREXIT(cinfo, JERR_NOT_COMPILED);
break;
}
}
}
static av_always_inline void row_fdct(DCTELEM * data)
{
int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
int_fast32_t tmp10, tmp11, tmp12, tmp13;
int_fast32_t z1, z2, z3, z4, z5;
DCTELEM *dataptr;
int ctr;
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true DCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--)
{
tmp0 = dataptr[0] + dataptr[7];
tmp7 = dataptr[0] - dataptr[7];
tmp1 = dataptr[1] + dataptr[6];
tmp6 = dataptr[1] - dataptr[6];
tmp2 = dataptr[2] + dataptr[5];
tmp5 = dataptr[2] - dataptr[5];
tmp3 = dataptr[3] + dataptr[4];
tmp4 = dataptr[3] - dataptr[4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS-PASS1_BITS);
dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS-PASS1_BITS);
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
dataptr += DCTSIZE; /* advance pointer to next row */
}
}
void idct_sh4(DCTELEM *block)
{
DEFREG;
int i;
float tblock[8*8],*fblock;
int ofs1,ofs2,ofs3;
#if defined(__SH4__)
#error "FIXME!! change to single float"
#endif
/* row */
/* even part */
load_matrix(even_table);
fblock = tblock+4;
i = 8;
do {
fr0 = block[0];
fr1 = block[2];
fr2 = block[4];
fr3 = block[6];
block+=8;
ftrv();
*--fblock = fr3;
*--fblock = fr2;
*--fblock = fr1;
*--fblock = fr0;
fblock+=8+4;
} while(--i);
block-=8*8;
fblock-=8*8+4;
load_matrix(odd_table);
i = 8;
// ofs1 = sizeof(float)*1;
// ofs2 = sizeof(float)*2;
// ofs3 = sizeof(float)*3;
do {
float t0,t1,t2,t3;
fr0 = block[1];
fr1 = block[3];
fr2 = block[5];
fr3 = block[7];
block+=8;
ftrv();
t0 = *fblock++;
t1 = *fblock++;
t2 = *fblock++;
t3 = *fblock++;
fblock+=4;
*--fblock = t0 - fr0;
*--fblock = t1 - fr1;
*--fblock = t2 - fr2;
*--fblock = t3 - fr3;
*--fblock = t3 + fr3;
*--fblock = t2 + fr2;
*--fblock = t1 + fr1;
*--fblock = t0 + fr0;
fblock+=8;
} while(--i);
block-=8*8;
fblock-=8*8;
/* col */
/* even part */
load_matrix(even_table);
ofs1 = sizeof(float)*2*8;
ofs2 = sizeof(float)*4*8;
ofs3 = sizeof(float)*6*8;
i = 8;
#define OA(fblock,ofs) *(float*)((char*)fblock + ofs)
do {
fr0 = OA(fblock, 0);
fr1 = OA(fblock,ofs1);
fr2 = OA(fblock,ofs2);
fr3 = OA(fblock,ofs3);
ftrv();
OA(fblock,0 ) = fr0;
OA(fblock,ofs1) = fr1;
OA(fblock,ofs2) = fr2;
OA(fblock,ofs3) = fr3;
fblock++;
} while(--i);
fblock-=8;
load_matrix(odd_table);
i=8;
do {
float t0,t1,t2,t3;
t0 = OA(fblock, 0); /* [8*0] */
t1 = OA(fblock,ofs1); /* [8*2] */
t2 = OA(fblock,ofs2); /* [8*4] */
t3 = OA(fblock,ofs3); /* [8*6] */
fblock+=8;
fr0 = OA(fblock, 0); /* [8*1] */
fr1 = OA(fblock,ofs1); /* [8*3] */
fr2 = OA(fblock,ofs2); /* [8*5] */
fr3 = OA(fblock,ofs3); /* [8*7] */
fblock+=-8+1;
ftrv();
block[8*0] = DESCALE(t0 + fr0,3);
block[8*7] = DESCALE(t0 - fr0,3);
block[8*1] = DESCALE(t1 + fr1,3);
block[8*6] = DESCALE(t1 - fr1,3);
block[8*2] = DESCALE(t2 + fr2,3);
block[8*5] = DESCALE(t2 - fr2,3);
block[8*3] = DESCALE(t3 + fr3,3);
block[8*4] = DESCALE(t3 - fr3,3);
block++;
} while(--i);
#if defined(__SH4__)
#error "FIXME!! change to double"
#endif
}
void j_rev_dct(DCTBLOCK data)
{
INT32 tmp0, tmp1, tmp2, tmp3;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
register DCTELEM *dataptr;
int rowctr;
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = data;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--)
{
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (INT32) dataptr[2];
z3 = (INT32) dataptr[6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
tmp0 = ((INT32) dataptr[0] + (INT32) dataptr[4]) << CONST_BITS;
tmp1 = ((INT32) dataptr[0] - (INT32) dataptr[4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (INT32) dataptr[7];
tmp1 = (INT32) dataptr[5];
tmp2 = (INT32) dataptr[3];
tmp3 = (INT32) dataptr[1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[0] = (DCTELEM) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
dataptr[7] = (DCTELEM) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
dataptr[1] = (DCTELEM) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
dataptr[6] = (DCTELEM) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
dataptr[2] = (DCTELEM) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
dataptr[5] = (DCTELEM) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
dataptr[3] = (DCTELEM) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
dataptr[4] = (DCTELEM) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
dataptr += DCTSIZE; /* advance pointer to next row */
}
/* Pass 2: process columns. */
/* Note that we must descale the results by a factor of 8 == 2**3, */
/* and also undo the PASS1_BITS scaling. */
dataptr = data;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (INT32) dataptr[DCTSIZE*2];
z3 = (INT32) dataptr[DCTSIZE*6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
tmp0 = ((INT32) dataptr[DCTSIZE*0] + (INT32) dataptr[DCTSIZE*4]) << CONST_BITS;
tmp1 = ((INT32) dataptr[DCTSIZE*0] - (INT32) dataptr[DCTSIZE*4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (INT32) dataptr[DCTSIZE*7];
tmp1 = (INT32) dataptr[DCTSIZE*5];
tmp2 = (INT32) dataptr[DCTSIZE*3];
tmp3 = (INT32) dataptr[DCTSIZE*1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp3, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp10 - tmp3, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp11 + tmp2, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(tmp11 - tmp2, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp12 + tmp1, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12 - tmp1, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp13 + tmp0, CONST_BITS+PASS1_BITS+1);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp13 - tmp0, CONST_BITS+PASS1_BITS+1);
dataptr++; /* advance pointer to next column */
}
}
void ff_j_rev_dct4(DCTBLOCK data)
{
int32_t tmp0, tmp1, tmp2, tmp3;
int32_t tmp10, tmp11, tmp12, tmp13;
int32_t z1;
int32_t d0, d2, d4, d6;
register DCTELEM *dataptr;
int rowctr;
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
data[0] += 4;
dataptr = data;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
/* Due to quantization, we will usually find that many of the input
* coefficients are zero, especially the AC terms. We can exploit this
* by short-circuiting the IDCT calculation for any row in which all
* the AC terms are zero. In that case each output is equal to the
* DC coefficient (with scale factor as needed).
* With typical images and quantization tables, half or more of the
* row DCT calculations can be simplified this way.
*/
register int *idataptr = (int*)dataptr;
d0 = dataptr[0];
d2 = dataptr[1];
d4 = dataptr[2];
d6 = dataptr[3];
if ((d2 | d4 | d6) == 0) {
/* AC terms all zero */
if (d0) {
/* Compute a 32 bit value to assign. */
DCTELEM dcval = (DCTELEM) (d0 << PASS1_BITS);
register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000);
idataptr[0] = v;
idataptr[1] = v;
}
dataptr += DCTSTRIDE; /* advance pointer to next row */
continue;
}
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
if (d6) {
if (d2) {
/* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
} else {
/* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
tmp2 = MULTIPLY(-d6, FIX_1_306562965);
tmp3 = MULTIPLY(d6, FIX_0_541196100);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
}
} else {
if (d2) {
/* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
tmp2 = MULTIPLY(d2, FIX_0_541196100);
tmp3 = MULTIPLY(d2, FIX_1_306562965);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
} else {
/* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
}
}
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[0] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
dataptr[1] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
dataptr[2] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
dataptr[3] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
dataptr += DCTSTRIDE; /* advance pointer to next row */
}
/* Pass 2: process columns. */
/* Note that we must descale the results by a factor of 8 == 2**3, */
/* and also undo the PASS1_BITS scaling. */
dataptr = data;
for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
/* Columns of zeroes can be exploited in the same way as we did with rows.
* However, the row calculation has created many nonzero AC terms, so the
* simplification applies less often (typically 5% to 10% of the time).
* On machines with very fast multiplication, it's possible that the
* test takes more time than it's worth. In that case this section
* may be commented out.
*/
d0 = dataptr[DCTSTRIDE*0];
d2 = dataptr[DCTSTRIDE*1];
d4 = dataptr[DCTSTRIDE*2];
d6 = dataptr[DCTSTRIDE*3];
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
if (d6) {
if (d2) {
/* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
} else {
/* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
tmp2 = MULTIPLY(-d6, FIX_1_306562965);
tmp3 = MULTIPLY(d6, FIX_0_541196100);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
}
} else {
if (d2) {
/* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
tmp2 = MULTIPLY(d2, FIX_0_541196100);
tmp3 = MULTIPLY(d2, FIX_1_306562965);
tmp0 = (d0 + d4) << CONST_BITS;
tmp1 = (d0 - d4) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
} else {
/* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
}
}
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
dataptr[DCTSTRIDE*0] = tmp10 >> (CONST_BITS+PASS1_BITS+3);
dataptr[DCTSTRIDE*1] = tmp11 >> (CONST_BITS+PASS1_BITS+3);
dataptr[DCTSTRIDE*2] = tmp12 >> (CONST_BITS+PASS1_BITS+3);
dataptr[DCTSTRIDE*3] = tmp13 >> (CONST_BITS+PASS1_BITS+3);
dataptr++; /* advance pointer to next column */
}
}
void ff_idct_sh4(int16_t *block)
{
DEFREG;
int i;
float tblock[8*8],*fblock;
int ofs1,ofs2,ofs3;
int fpscr;
fp_single_enter(fpscr);
/* row */
/* even part */
load_matrix(even_table);
fblock = tblock+4;
i = 8;
do {
fr0 = block[0];
fr1 = block[2];
fr2 = block[4];
fr3 = block[6];
block+=8;
ftrv();
*--fblock = fr3;
*--fblock = fr2;
*--fblock = fr1;
*--fblock = fr0;
fblock+=8+4;
} while(--i);
block-=8*8;
fblock-=8*8+4;
load_matrix(odd_table);
i = 8;
do {
float t0,t1,t2,t3;
fr0 = block[1];
fr1 = block[3];
fr2 = block[5];
fr3 = block[7];
block+=8;
ftrv();
t0 = *fblock++;
t1 = *fblock++;
t2 = *fblock++;
t3 = *fblock++;
fblock+=4;
*--fblock = t0 - fr0;
*--fblock = t1 - fr1;
*--fblock = t2 - fr2;
*--fblock = t3 - fr3;
*--fblock = t3 + fr3;
*--fblock = t2 + fr2;
*--fblock = t1 + fr1;
*--fblock = t0 + fr0;
fblock+=8;
} while(--i);
block-=8*8;
fblock-=8*8;
/* col */
/* even part */
load_matrix(even_table);
ofs1 = sizeof(float)*2*8;
ofs2 = sizeof(float)*4*8;
ofs3 = sizeof(float)*6*8;
i = 8;
#define OA(fblock,ofs) *(float*)((char*)fblock + ofs)
do {
fr0 = OA(fblock, 0);
fr1 = OA(fblock,ofs1);
fr2 = OA(fblock,ofs2);
fr3 = OA(fblock,ofs3);
ftrv();
OA(fblock,0 ) = fr0;
OA(fblock,ofs1) = fr1;
OA(fblock,ofs2) = fr2;
OA(fblock,ofs3) = fr3;
fblock++;
} while(--i);
fblock-=8;
load_matrix(odd_table);
i=8;
do {
float t0,t1,t2,t3;
t0 = OA(fblock, 0); /* [8*0] */
t1 = OA(fblock,ofs1); /* [8*2] */
t2 = OA(fblock,ofs2); /* [8*4] */
t3 = OA(fblock,ofs3); /* [8*6] */
fblock+=8;
fr0 = OA(fblock, 0); /* [8*1] */
fr1 = OA(fblock,ofs1); /* [8*3] */
fr2 = OA(fblock,ofs2); /* [8*5] */
fr3 = OA(fblock,ofs3); /* [8*7] */
fblock+=-8+1;
ftrv();
block[8*0] = DESCALE(t0 + fr0,3);
block[8*7] = DESCALE(t0 - fr0,3);
block[8*1] = DESCALE(t1 + fr1,3);
block[8*6] = DESCALE(t1 - fr1,3);
block[8*2] = DESCALE(t2 + fr2,3);
block[8*5] = DESCALE(t2 - fr2,3);
block[8*3] = DESCALE(t3 + fr3,3);
block[8*4] = DESCALE(t3 - fr3,3);
block++;
} while(--i);
fp_single_leave(fpscr);
}
GLOBAL void jpeg_fdct_islow( void )
{
#ifdef PROFILING
/* Profiling variables. Remove for actual WCET analyses. */
int iters_ctr1 = 0, min_ctr1 = 100000, max_ctr1 = 0;
int iters_ctr2 = 0, min_ctr2 = 100000, max_ctr2 = 0;
#endif
INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
DCTELEM *dataptr;
int ctr;
SHIFT_TEMPS
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true DCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = data;
#ifdef PROFILING
iters_ctr1 = 0;
#endif
_Pragma("loopbound min 8 max 8")
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
#ifdef PROFILING
iters_ctr1++;
#endif
tmp0 = dataptr[0] + dataptr[7];
tmp7 = dataptr[0] - dataptr[7];
tmp1 = dataptr[1] + dataptr[6];
tmp6 = dataptr[1] - dataptr[6];
tmp2 = dataptr[2] + dataptr[5];
tmp5 = dataptr[2] - dataptr[5];
tmp3 = dataptr[3] + dataptr[4];
tmp4 = dataptr[3] - dataptr[4];
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS-PASS1_BITS);
dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS-PASS1_BITS);
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
dataptr += DCTSIZE; /* advance pointer to next row */
}
#ifdef PROFILING
if ( iters_ctr1 < min_ctr1 )
min_ctr1 = iters_ctr1;
if ( iters_ctr1 > max_ctr1 )
max_ctr1 = iters_ctr1;
#endif
dataptr = data;
#ifdef PROFILING
iters_ctr2 = 0;
#endif
_Pragma("loopbound min 8 max 8")
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
#ifdef PROFILING
iters_ctr2++;
#endif
tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS+PASS1_BITS);
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4,
CONST_BITS+PASS1_BITS);
dataptr++; /* advance pointer to next column */
}
#ifdef PROFILING
if ( iters_ctr2 < min_ctr2 )
min_ctr2 = iters_ctr2;
if ( iters_ctr2 > max_ctr2 )
max_ctr2 = iters_ctr2;
#endif
#ifdef PROFILING
printf( "ctr1-loop: [%d, %d]\n", min_ctr1, max_ctr1 );
printf( "ctr2-loop: [%d, %d]\n", min_ctr2, max_ctr2 );
#endif
}
static CvStatus CV_STDCALL
icvImgToObs_DCT_32f_C1R( float * img, int imgStep, CvSize roi,
float *obs, CvSize dctSize,
CvSize obsSize, CvSize delta )
{
/* dct transform matrices: horizontal and vertical */
work_t tab_x[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
work_t tab_y[MAX_DCT_SIZE * MAX_DCT_SIZE / 2 + 2];
/* temporary buffers for dct */
work_t temp0[MAX_DCT_SIZE * 4];
work_t temp1[MAX_DCT_SIZE * 4];
work_t *buffer = 0;
work_t *buf_limit;
double s;
int y;
int Nx, Ny;
int n1 = dctSize.height, m1 = n1 / 2;
int n2 = dctSize.width, m2 = n2 / 2;
if( !img || !obs )
return CV_NULLPTR_ERR;
if( roi.width <= 0 || roi.height <= 0 )
return CV_BADSIZE_ERR;
if( delta.width <= 0 || delta.height <= 0 )
return CV_BADRANGE_ERR;
if( obsSize.width <= 0 || dctSize.width < obsSize.width ||
obsSize.height <= 0 || dctSize.height < obsSize.height )
return CV_BADRANGE_ERR;
if( dctSize.width > MAX_DCT_SIZE || dctSize.height > MAX_DCT_SIZE )
return CV_BADRANGE_ERR;
Nx = (roi.width - dctSize.width + delta.width) / delta.width;
Ny = (roi.height - dctSize.height + delta.height) / delta.height;
if( Nx <= 0 || Ny <= 0 )
return CV_BADRANGE_ERR;
buffer = (work_t *)cvAlloc( roi.width * obsSize.height * sizeof( buffer[0] ));
if( !buffer )
return CV_OUTOFMEM_ERR;
icvCalcDCTMatrix( tab_x, dctSize.width );
icvCalcDCTMatrix( tab_y, dctSize.height );
buf_limit = buffer + obsSize.height * roi.width;
imgStep /= sizeof(img[0]);
for( y = 0; y < Ny; y++, img += delta.height * imgStep )
{
int x, i, j, k;
work_t k0 = 0;
/* do transfroms for each column. Calc only first obsSize.height DCT coefficients */
for( x = 0; x < roi.width; x++ )
{
float is = 0;
work_t *buf = buffer + x;
work_t *tab = tab_y + 2;
if( n1 & 1 )
{
is = img[x + m1 * imgStep];
k0 = ((work_t) is) * tab[-1];
}
/* first coefficient */
for( j = 0; j < m1; j++ )
{
float t0 = img[x + j * imgStep];
float t1 = img[x + (n1 - 1 - j) * imgStep];
float t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
is += t2;
temp1[j] = (work_t) t0;
}
buf[0] = DESCALE( is * tab[-2], PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m1; k++ )
s += temp1[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
if( (buf += roi.width) >= buf_limit )
break;
tab += m1;
s = 0;
if( n1 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m1; k++ )
s += temp0[k] * tab[k];
buf[0] = DESCALE( s, PASS1_SHIFT );
tab += m1;
if( (buf += roi.width) >= buf_limit )
break;
}
}
k0 = 0;
/* do transforms for rows. */
for( x = 0; x + dctSize.width <= roi.width; x += delta.width )
{
for( i = 0; i < obsSize.height; i++ )
{
work_t *buf = buffer + x + roi.width * i;
work_t *tab = tab_x + 2;
float *obs_limit = obs + obsSize.width;
s = 0;
if( n2 & 1 )
{
s = buf[m2];
k0 = (work_t) (s * tab[-1]);
}
/* first coefficient */
for( j = 0; j < m2; j++ )
{
work_t t0 = buf[j];
work_t t1 = buf[n2 - 1 - j];
work_t t2 = t0 + t1;
t0 -= t1;
temp0[j] = (work_t) t2;
s += t2;
temp1[j] = (work_t) t0;
}
*obs++ = (float) DESCALE( s * tab[-2], PASS2_SHIFT );
if( obs == obs_limit )
continue;
/* other coefficients */
for( ;; )
{
s = 0;
for( k = 0; k < m2; k++ )
s += temp1[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
if( ++obs == obs_limit )
break;
tab += m2;
s = 0;
if( n2 & 1 )
{
k0 = -k0;
s = k0;
}
for( k = 0; k < m2; k++ )
s += temp0[k] * tab[k];
obs[0] = (float) DESCALE( s, PASS2_SHIFT );
tab += m2;
if( ++obs == obs_limit )
break;
}
}
}
}
cvFree( &buffer );
return CV_NO_ERR;
}
METHODDEF void
start_pass (j_decompress_ptr cinfo)
{
my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
int ci, i;
jpeg_component_info *compptr;
int method = 0;
inverse_DCT_method_ptr method_ptr = NULL;
JQUANT_TBL * qtbl;
for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
ci++, compptr++) {
/* Select the proper IDCT routine for this component's scaling */
switch (compptr->DCT_scaled_size) {
#ifdef IDCT_SCALING_SUPPORTED
case 1:
method_ptr = jpeg_idct_1x1;
method = JDCT_ISLOW; /* jidctred uses islow-style table */
break;
case 2:
method_ptr = jpeg_idct_2x2;
method = JDCT_ISLOW; /* jidctred uses islow-style table */
break;
case 4:
method_ptr = jpeg_idct_4x4;
method = JDCT_ISLOW; /* jidctred uses islow-style table */
break;
#endif
case DCTSIZE:
switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
case JDCT_ISLOW:
method_ptr = jpeg_idct_islow;
method = JDCT_ISLOW;
break;
#endif
#ifdef DCT_IFAST_SUPPORTED
case JDCT_IFAST:
method_ptr = jpeg_idct_ifast;
method = JDCT_IFAST;
break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
case JDCT_FLOAT:
method_ptr = jpeg_idct_float;
method = JDCT_FLOAT;
break;
#endif
default:
ERREXIT(cinfo, JERR_NOT_COMPILED);
break;
}
break;
default:
ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->DCT_scaled_size);
break;
}
idct->pub.inverse_DCT[ci] = method_ptr;
/* Create multiplier table from quant table.
* However, we can skip this if the component is uninteresting
* or if we already built the table. Also, if no quant table
* has yet been saved for the component, we leave the
* multiplier table all-zero; we'll be reading zeroes from the
* coefficient controller's buffer anyway.
*/
if (! compptr->component_needed || idct->cur_method[ci] == method)
continue;
qtbl = compptr->quant_table;
if (qtbl == NULL) /* happens if no data yet for component */
continue;
idct->cur_method[ci] = method;
switch (method) {
#ifdef PROVIDE_ISLOW_TABLES
case JDCT_ISLOW:
{
/* For LL&M IDCT method, multipliers are equal to raw quantization
* coefficients, but are stored in natural order as ints.
*/
ISLOW_MULT_TYPE * ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
for (i = 0; i < DCTSIZE2; i++) {
ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[jpeg_zigzag_order[i]];
}
}
break;
#endif
#ifdef DCT_IFAST_SUPPORTED
case JDCT_IFAST:
{
/* For AA&N IDCT method, multipliers are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* For integer operation, the multiplier table is to be scaled by
* IFAST_SCALE_BITS. The multipliers are stored in natural order.
*/
IFAST_MULT_TYPE * ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
#define CONST_BITS 14
static const INT16 aanscales[DCTSIZE2] = {
/* precomputed values scaled up by 14 bits */
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
};
SHIFT_TEMPS
for (i = 0; i < DCTSIZE2; i++) {
ifmtbl[i] = (IFAST_MULT_TYPE)
DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[jpeg_zigzag_order[i]],
(INT32) aanscales[i]),
CONST_BITS-IFAST_SCALE_BITS);
}
}
break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
case JDCT_FLOAT:
{
/* For float AA&N IDCT method, multipliers are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* The multipliers are stored in natural order.
*/
FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
int row, col;
static const double aanscalefactor[DCTSIZE] = {
1.0, 1.387039845, 1.306562965, 1.175875602,
1.0, 0.785694958, 0.541196100, 0.275899379
};
i = 0;
for (row = 0; row < DCTSIZE; row++) {
for (col = 0; col < DCTSIZE; col++) {
fmtbl[i] = (FLOAT_MULT_TYPE)
((double) qtbl->quantval[jpeg_zigzag_order[i]] *
aanscalefactor[row] * aanscalefactor[col]);
i++;
}
}
}
break;
#endif
default:
ERREXIT(cinfo, JERR_NOT_COMPILED);
break;
}
}
}
METHODDEF void
start_input_pass (j_decompress_ptr cinfo)
{
my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
int ci, qtblno, i;
jpeg_component_info *compptr;
JQUANT_TBL * qtbl;
for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
compptr = cinfo->cur_comp_info[ci];
qtblno = compptr->quant_tbl_no;
/* Make sure specified quantization table is present */
if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
cinfo->quant_tbl_ptrs[qtblno] == NULL)
ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
qtbl = cinfo->quant_tbl_ptrs[qtblno];
/* Create multiplier table from quant table, unless we already did so. */
if (compptr->dct_table != NULL)
continue;
switch (idct->real_method[compptr->component_index]) {
#ifdef PROVIDE_ISLOW_TABLES
case JDCT_ISLOW:
{
/* For LL&M IDCT method, multipliers are equal to raw quantization
* coefficients, but are stored in natural order as ints.
*/
ISLOW_MULT_TYPE * ismtbl;
compptr->dct_table =
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(ISLOW_MULT_TYPE));
ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
for (i = 0; i < DCTSIZE2; i++) {
ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[ZIG[i]];
}
}
break;
#endif
#ifdef DCT_IFAST_SUPPORTED
case JDCT_IFAST:
{
/* For AA&N IDCT method, multipliers are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* For integer operation, the multiplier table is to be scaled by
* IFAST_SCALE_BITS. The multipliers are stored in natural order.
*/
IFAST_MULT_TYPE * ifmtbl;
#define CONST_BITS 14
static const INT16 aanscales[DCTSIZE2] = {
/* precomputed values scaled up by 14 bits */
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
};
SHIFT_TEMPS
compptr->dct_table =
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(IFAST_MULT_TYPE));
ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
for (i = 0; i < DCTSIZE2; i++) {
ifmtbl[i] = (IFAST_MULT_TYPE)
DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[ZIG[i]],
(INT32) aanscales[i]),
CONST_BITS-IFAST_SCALE_BITS);
}
}
break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
case JDCT_FLOAT:
{
/* For float AA&N IDCT method, multipliers are equal to quantization
* coefficients scaled by scalefactor[row]*scalefactor[col], where
* scalefactor[0] = 1
* scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7
* The multipliers are stored in natural order.
*/
FLOAT_MULT_TYPE * fmtbl;
int row, col;
static const double aanscalefactor[DCTSIZE] = {
1.0, 1.387039845, 1.306562965, 1.175875602,
1.0, 0.785694958, 0.541196100, 0.275899379
};
compptr->dct_table =
(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
DCTSIZE2 * SIZEOF(FLOAT_MULT_TYPE));
fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
i = 0;
for (row = 0; row < DCTSIZE; row++) {
for (col = 0; col < DCTSIZE; col++) {
fmtbl[i] = (FLOAT_MULT_TYPE)
((double) qtbl->quantval[ZIG[i]] *
aanscalefactor[row] * aanscalefactor[col]);
i++;
}
}
}
break;
#endif
default:
ERREXIT(cinfo, JERR_NOT_COMPILED);
break;
}
}
}
jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
JCOEFPTR coef_block,
JSAMPARRAY output_buf, JDIMENSION output_col)
{
INT32 tmp0, tmp1, tmp2, tmp3;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
JCOEFPTR inptr;
ISLOW_MULT_TYPE * quantptr;
int * wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
int workspace[DCTSIZE2]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
inptr = coef_block;
quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
wsptr = workspace;
for (ctr = DCTSIZE; ctr > 0; ctr--) {
/* Due to quantization, we will usually find that many of the input
* coefficients are zero, especially the AC terms. We can exploit this
* by short-circuiting the IDCT calculation for any column in which all
* the AC terms are zero. In that case each output is equal to the
* DC coefficient (with scale factor as needed).
* With typical images and quantization tables, half or more of the
* column DCT calculations can be simplified this way.
*/
if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
inptr[DCTSIZE*7] == 0) {
/* AC terms all zero */
int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
wsptr[DCTSIZE*2] = dcval;
wsptr[DCTSIZE*3] = dcval;
wsptr[DCTSIZE*4] = dcval;
wsptr[DCTSIZE*5] = dcval;
wsptr[DCTSIZE*6] = dcval;
wsptr[DCTSIZE*7] = dcval;
inptr++; /* advance pointers to next column */
quantptr++;
wsptr++;
continue;
}
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
tmp0 = (z2 + z3) << CONST_BITS;
tmp1 = (z2 - z3) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
inptr++; /* advance pointers to next column */
quantptr++;
wsptr++;
}
/* Pass 2: process rows from work array, store into output array. */
/* Note that we must descale the results by a factor of 8 == 2**3, */
/* and also undo the PASS1_BITS scaling. */
wsptr = workspace;
for (ctr = 0; ctr < DCTSIZE; ctr++) {
outptr = output_buf[ctr] + output_col;
/* Rows of zeroes can be exploited in the same way as we did with columns.
* However, the column calculation has created many nonzero AC terms, so
* the simplification applies less often (typically 5% to 10% of the time).
* On machines with very fast multiplication, it's possible that the
* test takes more time than it's worth. In that case this section
* may be commented out.
*/
#ifndef NO_ZERO_ROW_TEST
if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
/* AC terms all zero */
JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
& RANGE_MASK];
outptr[0] = dcval;
outptr[1] = dcval;
outptr[2] = dcval;
outptr[3] = dcval;
outptr[4] = dcval;
outptr[5] = dcval;
outptr[6] = dcval;
outptr[7] = dcval;
wsptr += DCTSIZE; /* advance pointer to next row */
continue;
}
#endif
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (INT32) wsptr[2];
z3 = (INT32) wsptr[6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (INT32) wsptr[7];
tmp1 = (INT32) wsptr[5];
tmp2 = (INT32) wsptr[3];
tmp3 = (INT32) wsptr[1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
CONST_BITS+PASS1_BITS+3)
& RANGE_MASK];
wsptr += DCTSIZE; /* advance pointer to next row */
}
}
GLOBAL void
jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
JCOEFPTR coef_block,
JSAMPARRAY output_buf, JDIMENSION output_col)
{
INT32 tmp0, tmp2, tmp10, tmp12;
INT32 z1, z2, z3, z4;
JCOEFPTR inptr;
ISLOW_MULT_TYPE * quantptr;
int * wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
int workspace[DCTSIZE*4]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
inptr = coef_block;
quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
wsptr = workspace;
for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
/* Don't bother to process column 4, because second pass won't use it */
if (ctr == DCTSIZE-4)
continue;
if ((inptr[DCTSIZE*1] | inptr[DCTSIZE*2] | inptr[DCTSIZE*3] |
inptr[DCTSIZE*5] | inptr[DCTSIZE*6] | inptr[DCTSIZE*7]) == 0) {
/* AC terms all zero; we need not examine term 4 for 4x4 output */
int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
wsptr[DCTSIZE*2] = dcval;
wsptr[DCTSIZE*3] = dcval;
continue;
}
/* Even part */
tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
tmp0 <<= (CONST_BITS+1);
z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, - FIX_0_765366865);
tmp10 = tmp0 + tmp2;
tmp12 = tmp0 - tmp2;
/* Odd part */
z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
z2 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
z4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
+ MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
+ MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
+ MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
+ MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
+ MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
+ MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
/* Final output stage */
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp2, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*3] = (int) DESCALE(tmp10 - tmp2, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp12 + tmp0, CONST_BITS-PASS1_BITS+1);
wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 - tmp0, CONST_BITS-PASS1_BITS+1);
}
/* Pass 2: process 4 rows from work array, store into output array. */
wsptr = workspace;
for (ctr = 0; ctr < 4; ctr++) {
outptr = output_buf[ctr] + output_col;
/* It's not clear whether a zero row test is worthwhile here ... */
#ifndef NO_ZERO_ROW_TEST
if ((wsptr[1] | wsptr[2] | wsptr[3] | wsptr[5] | wsptr[6] |
wsptr[7]) == 0) {
/* AC terms all zero */
JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
& RANGE_MASK];
outptr[0] = dcval;
outptr[1] = dcval;
outptr[2] = dcval;
outptr[3] = dcval;
wsptr += DCTSIZE; /* advance pointer to next row */
continue;
}
#endif
/* Even part */
tmp0 = ((INT32) wsptr[0]) << (CONST_BITS+1);
tmp2 = MULTIPLY((INT32) wsptr[2], FIX_1_847759065)
+ MULTIPLY((INT32) wsptr[6], - FIX_0_765366865);
tmp10 = tmp0 + tmp2;
tmp12 = tmp0 - tmp2;
/* Odd part */
z1 = (INT32) wsptr[7];
z2 = (INT32) wsptr[5];
z3 = (INT32) wsptr[3];
z4 = (INT32) wsptr[1];
tmp0 = MULTIPLY(z1, - FIX_0_211164243) /* sqrt(2) * (c3-c1) */
+ MULTIPLY(z2, FIX_1_451774981) /* sqrt(2) * (c3+c7) */
+ MULTIPLY(z3, - FIX_2_172734803) /* sqrt(2) * (-c1-c5) */
+ MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * (c5+c7) */
tmp2 = MULTIPLY(z1, - FIX_0_509795579) /* sqrt(2) * (c7-c5) */
+ MULTIPLY(z2, - FIX_0_601344887) /* sqrt(2) * (c5-c1) */
+ MULTIPLY(z3, FIX_0_899976223) /* sqrt(2) * (c3-c7) */
+ MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */
/* Final output stage */
outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp2,
CONST_BITS+PASS1_BITS+3+1)
& RANGE_MASK];
outptr[3] = range_limit[(int) DESCALE(tmp10 - tmp2,
CONST_BITS+PASS1_BITS+3+1)
& RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE(tmp12 + tmp0,
CONST_BITS+PASS1_BITS+3+1)
& RANGE_MASK];
outptr[2] = range_limit[(int) DESCALE(tmp12 - tmp0,
CONST_BITS+PASS1_BITS+3+1)
& RANGE_MASK];
wsptr += DCTSIZE; /* advance pointer to next row */
}
}
void idct_sh4(DCTELEM *block)
{
DEFREG;
int i;
float tblock[8*8],*fblock;
/* row */
/* even part */
load_matrix(even_table);
fblock = tblock;
i = 8;
do {
fr0 = block[0];
fr1 = block[2];
fr2 = block[4];
fr3 = block[6];
block+=8;
ftrv();
fblock[0] = fr0;
fblock[2] = fr1;
fblock[4] = fr2;
fblock[6] = fr3;
fblock+=8;
} while(--i);
block-=8*8;
fblock-=8*8;
load_matrix(odd_table);
i = 8;
do {
float t0,t1,t2,t3;
fr0 = block[1];
fr1 = block[3];
fr2 = block[5];
fr3 = block[7];
block+=8;
ftrv();
t0 = fblock[0];
t1 = fblock[2];
t2 = fblock[4];
t3 = fblock[6];
fblock[0] = t0 + fr0;
fblock[7] = t0 - fr0;
fblock[1] = t1 + fr1;
fblock[6] = t1 - fr1;
fblock[2] = t2 + fr2;
fblock[5] = t2 - fr2;
fblock[3] = t3 + fr3;
fblock[4] = t3 - fr3;
fblock+=8;
} while(--i);
block-=8*8;
fblock-=8*8;
/* col */
/* even part */
load_matrix(even_table);
i = 8;
do {
fr0 = fblock[8*0];
fr1 = fblock[8*2];
fr2 = fblock[8*4];
fr3 = fblock[8*6];
ftrv();
fblock[8*0] = fr0;
fblock[8*2] = fr1;
fblock[8*4] = fr2;
fblock[8*6] = fr3;
fblock++;
} while(--i);
fblock-=8;
load_matrix(odd_table);
i=8;
do {
float t0,t1,t2,t3;
fr0 = fblock[8*1];
fr1 = fblock[8*3];
fr2 = fblock[8*5];
fr3 = fblock[8*7];
ftrv();
t0 = fblock[8*0];
t1 = fblock[8*2];
t2 = fblock[8*4];
t3 = fblock[8*6];
fblock++;
block[8*0] = DESCALE(t0 + fr0,3);
block[8*7] = DESCALE(t0 - fr0,3);
block[8*1] = DESCALE(t1 + fr1,3);
block[8*6] = DESCALE(t1 - fr1,3);
block[8*2] = DESCALE(t2 + fr2,3);
block[8*5] = DESCALE(t2 - fr2,3);
block[8*3] = DESCALE(t3 + fr3,3);
block[8*4] = DESCALE(t3 - fr3,3);
block++;
} while(--i);
}
GLOBAL void
jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
JCOEFPTR coef_block,
JSAMPARRAY output_buf, JDIMENSION output_col)
{
INT32 tmp0, tmp10, z1;
JCOEFPTR inptr;
ISLOW_MULT_TYPE * quantptr;
int * wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
int workspace[DCTSIZE*2]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
inptr = coef_block;
quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
wsptr = workspace;
for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) {
/* Don't bother to process columns 2,4,6 */
if (ctr == DCTSIZE-2 || ctr == DCTSIZE-4 || ctr == DCTSIZE-6)
continue;
if ((inptr[DCTSIZE*1] | inptr[DCTSIZE*3] |
inptr[DCTSIZE*5] | inptr[DCTSIZE*7]) == 0) {
/* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */
int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
continue;
}
/* Even part */
z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
tmp10 = z1 << (CONST_BITS+2);
/* Odd part */
z1 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
tmp0 = MULTIPLY(z1, - FIX_0_720959822); /* sqrt(2) * (c7-c5+c3-c1) */
z1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */
z1 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
tmp0 += MULTIPLY(z1, - FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */
z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
/* Final output stage */
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp0, CONST_BITS-PASS1_BITS+2);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp10 - tmp0, CONST_BITS-PASS1_BITS+2);
}
/* Pass 2: process 2 rows from work array, store into output array. */
wsptr = workspace;
for (ctr = 0; ctr < 2; ctr++) {
outptr = output_buf[ctr] + output_col;
/* It's not clear whether a zero row test is worthwhile here ... */
#ifndef NO_ZERO_ROW_TEST
if ((wsptr[1] | wsptr[3] | wsptr[5] | wsptr[7]) == 0) {
/* AC terms all zero */
JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
& RANGE_MASK];
outptr[0] = dcval;
outptr[1] = dcval;
wsptr += DCTSIZE; /* advance pointer to next row */
continue;
}
#endif
/* Even part */
tmp10 = ((INT32) wsptr[0]) << (CONST_BITS+2);
/* Odd part */
tmp0 = MULTIPLY((INT32) wsptr[7], - FIX_0_720959822) /* sqrt(2) * (c7-c5+c3-c1) */
+ MULTIPLY((INT32) wsptr[5], FIX_0_850430095) /* sqrt(2) * (-c1+c3+c5+c7) */
+ MULTIPLY((INT32) wsptr[3], - FIX_1_272758580) /* sqrt(2) * (-c1+c3-c5-c7) */
+ MULTIPLY((INT32) wsptr[1], FIX_3_624509785); /* sqrt(2) * (c1+c3+c5+c7) */
/* Final output stage */
outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp0,
CONST_BITS+PASS1_BITS+3+2)
& RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE(tmp10 - tmp0,
CONST_BITS+PASS1_BITS+3+2)
& RANGE_MASK];
wsptr += DCTSIZE; /* advance pointer to next row */
}
}
ff_jpeg_fdct_islow (DCTELEM * data)
{
int_fast32_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
int_fast32_t tmp10, tmp11, tmp12, tmp13;
int_fast32_t z1, z2, z3, z4, z5;
DCTELEM *dataptr;
int ctr;
row_fdct(data);
/* Pass 2: process columns.
* We remove the PASS1_BITS scaling, but leave the results scaled up
* by an overall factor of 8.
*/
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--)
{
tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865),
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065),
CONST_BITS+PASS1_BITS);
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3,
CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4,
CONST_BITS+PASS1_BITS);
dataptr++; /* advance pointer to next column */
}
}
jpeg_fdct_islow (DCTELEM * data, boolean jpeg8)
{
INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
DCTELEM *dataptr;
int ctr, pass1_bits;
/* set pass1_bits */
if (jpeg8) pass1_bits = 2;
else pass1_bits = 1;
SHIFT_TEMPS
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true DCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
tmp0 = dataptr[0] + dataptr[7];
tmp7 = dataptr[0] - dataptr[7];
tmp1 = dataptr[1] + dataptr[6];
tmp6 = dataptr[1] - dataptr[6];
tmp2 = dataptr[2] + dataptr[5];
tmp5 = dataptr[2] - dataptr[5];
tmp3 = dataptr[3] + dataptr[4];
tmp4 = dataptr[3] - dataptr[4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << pass1_bits);
dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << pass1_bits);
if (jpeg8){
z1 = MULTIPLY16C16(tmp12 + tmp13, FIX_0_541196100);
dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY16C16(tmp13, FIX_0_765366865),
CONST_BITS-pass1_bits);
dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY16C16(tmp12, - FIX_1_847759065),
CONST_BITS-pass1_bits);
} else {
z1 = (tmp12 + tmp13) * FIX_0_541196100;
dataptr[2] = (DCTELEM) DESCALE(z1 + ( tmp13 * FIX_0_765366865),
CONST_BITS-pass1_bits);
dataptr[6] = (DCTELEM) DESCALE(z1 + ( tmp12 * ( - FIX_1_847759065)),
CONST_BITS-pass1_bits); }
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
if (jpeg8) {
z5 = MULTIPLY16C16(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY16C16(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY16C16(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY16C16(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY16C16(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY16C16(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY16C16(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY16C16(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY16C16(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
} else {
z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
tmp4 = tmp4 * FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = tmp5 * FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = tmp6 * FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = tmp7 * FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = z1 * (- FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = z2 * (- FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = z3 * (- FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = z4 * (- FIX_0_390180644); /* sqrt(2) * (c5-c3) */
}
z3 += z5;
z4 += z5;
dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-pass1_bits);
dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-pass1_bits);
dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-pass1_bits);
dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-pass1_bits);
dataptr += DCTSIZE; /* advance pointer to next row */
}
/* Pass 2: process columns.
* We remove the PASS1_BITS scaling, but leave the results scaled up
* by an overall factor of 8.
*/
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, pass1_bits);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, pass1_bits);
if (jpeg8) {
z1 = MULTIPLY16C16(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY16C16(tmp13, FIX_0_765366865),
CONST_BITS+pass1_bits);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY16C16(tmp12, - FIX_1_847759065),
CONST_BITS+pass1_bits);
} else {
z1 = (tmp12 + tmp13) * FIX_0_541196100;
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + (tmp13 * FIX_0_765366865),
CONST_BITS+pass1_bits);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + (tmp12 * (- FIX_1_847759065)),
CONST_BITS+pass1_bits);
}
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
if (jpeg8) {
z5 = MULTIPLY16C16(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY16C16(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY16C16(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY16C16(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY16C16(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY16C16(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY16C16(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY16C16(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY16C16(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
} else {
z5 = (z3 + z4) * FIX_1_175875602; /* sqrt(2) * c3 */
tmp4 = tmp4 * FIX_0_298631336; /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = tmp5 * FIX_2_053119869; /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = tmp6 * FIX_3_072711026; /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = tmp7 * FIX_1_501321110; /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = z1 * ( - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = z2 * ( - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = z3 * ( - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = z4 * ( - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
}
z3 += z5;
z4 += z5;
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3,
CONST_BITS+pass1_bits);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4,
CONST_BITS+pass1_bits);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3,
CONST_BITS+pass1_bits);
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4,
CONST_BITS+pass1_bits);
dataptr++; /* advance pointer to next column */
}
}
jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
JCOEFPTR coef_block,
JSAMPARRAY output_buf, JDIMENSION output_col)
{
FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
FAST_FLOAT z5, z10, z11, z12, z13;
JCOEFPTR inptr;
FLOAT_MULT_TYPE * quantptr;
FAST_FLOAT * wsptr;
JSAMPROW outptr;
JSAMPLE *range_limit = IDCT_range_limit(cinfo);
int ctr;
FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
inptr = coef_block;
quantptr = (FLOAT_MULT_TYPE *) compptr->dct_table;
wsptr = workspace;
for (ctr = DCTSIZE; ctr > 0; ctr--) {
/* Due to quantization, we will usually find that many of the input
* coefficients are zero, especially the AC terms. We can exploit this
* by short-circuiting the IDCT calculation for any column in which all
* the AC terms are zero. In that case each output is equal to the
* DC coefficient (with scale factor as needed).
* With typical images and quantization tables, half or more of the
* column DCT calculations can be simplified this way.
*/
if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
inptr[DCTSIZE*7] == 0) {
/* AC terms all zero */
FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
wsptr[DCTSIZE*2] = dcval;
wsptr[DCTSIZE*3] = dcval;
wsptr[DCTSIZE*4] = dcval;
wsptr[DCTSIZE*5] = dcval;
wsptr[DCTSIZE*6] = dcval;
wsptr[DCTSIZE*7] = dcval;
inptr++; /* advance pointers to next column */
quantptr++;
wsptr++;
continue;
}
/* Even part */
tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
tmp10 = tmp0 + tmp2; /* phase 3 */
tmp11 = tmp0 - tmp2;
tmp13 = tmp1 + tmp3; /* phases 5-3 */
tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */
tmp0 = tmp10 + tmp13; /* phase 2 */
tmp3 = tmp10 - tmp13;
tmp1 = tmp11 + tmp12;
tmp2 = tmp11 - tmp12;
/* Odd part */
tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
z13 = tmp6 + tmp5; /* phase 6 */
z10 = tmp6 - tmp5;
z11 = tmp4 + tmp7;
z12 = tmp4 - tmp7;
tmp7 = z11 + z13; /* phase 5 */
tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
tmp6 = tmp12 - tmp7; /* phase 2 */
tmp5 = tmp11 - tmp6;
tmp4 = tmp10 + tmp5;
wsptr[DCTSIZE*0] = tmp0 + tmp7;
wsptr[DCTSIZE*7] = tmp0 - tmp7;
wsptr[DCTSIZE*1] = tmp1 + tmp6;
wsptr[DCTSIZE*6] = tmp1 - tmp6;
wsptr[DCTSIZE*2] = tmp2 + tmp5;
wsptr[DCTSIZE*5] = tmp2 - tmp5;
wsptr[DCTSIZE*4] = tmp3 + tmp4;
wsptr[DCTSIZE*3] = tmp3 - tmp4;
inptr++; /* advance pointers to next column */
quantptr++;
wsptr++;
}
/* Pass 2: process rows from work array, store into output array. */
/* Note that we must descale the results by a factor of 8 == 2**3. */
wsptr = workspace;
for (ctr = 0; ctr < DCTSIZE; ctr++) {
outptr = output_buf[ctr] + output_col;
/* Rows of zeroes can be exploited in the same way as we did with columns.
* However, the column calculation has created many nonzero AC terms, so
* the simplification applies less often (typically 5% to 10% of the time).
* And testing floats for zero is relatively expensive, so we don't bother.
*/
/* Even part */
tmp10 = wsptr[0] + wsptr[4];
tmp11 = wsptr[0] - wsptr[4];
tmp13 = wsptr[2] + wsptr[6];
tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
tmp0 = tmp10 + tmp13;
tmp3 = tmp10 - tmp13;
tmp1 = tmp11 + tmp12;
tmp2 = tmp11 - tmp12;
/* Odd part */
z13 = wsptr[5] + wsptr[3];
z10 = wsptr[5] - wsptr[3];
z11 = wsptr[1] + wsptr[7];
z12 = wsptr[1] - wsptr[7];
tmp7 = z11 + z13;
tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
tmp6 = tmp12 - tmp7;
tmp5 = tmp11 - tmp6;
tmp4 = tmp10 + tmp5;
/* Final output stage: scale down by a factor of 8 and range-limit */
outptr[0] = range_limit[(int) DESCALE((IJG_INT32) (tmp0 + tmp7), 3)
& RANGE_MASK];
outptr[7] = range_limit[(int) DESCALE((IJG_INT32) (tmp0 - tmp7), 3)
& RANGE_MASK];
outptr[1] = range_limit[(int) DESCALE((IJG_INT32) (tmp1 + tmp6), 3)
& RANGE_MASK];
outptr[6] = range_limit[(int) DESCALE((IJG_INT32) (tmp1 - tmp6), 3)
& RANGE_MASK];
outptr[2] = range_limit[(int) DESCALE((IJG_INT32) (tmp2 + tmp5), 3)
& RANGE_MASK];
outptr[5] = range_limit[(int) DESCALE((IJG_INT32) (tmp2 - tmp5), 3)
& RANGE_MASK];
outptr[4] = range_limit[(int) DESCALE((IJG_INT32) (tmp3 + tmp4), 3)
& RANGE_MASK];
outptr[3] = range_limit[(int) DESCALE((IJG_INT32) (tmp3 - tmp4), 3)
& RANGE_MASK];
wsptr += DCTSIZE; /* advance pointer to next row */
}
}
static void fdct(const unsigned short* in, short* out)
{
INT32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
int ctr;
// SHIFT_TEMPS
int data[DCTSIZE * DCTSIZE];
int i, j;
int* dataptr = data;
for( i = 0; i < DCTSIZE; i++ )
{
for( j = 0; j < DCTSIZE; j++ )
{
int temp;
temp = in[i*DCTSIZE + j];
dataptr[i*DCTSIZE + j] = temp;
}
}
/* Pass 1: process rows. */
/* Note results are scaled up by sqrt(8) compared to a true DCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
tmp0 = dataptr[0] + dataptr[7];
tmp7 = dataptr[0] - dataptr[7];
tmp1 = dataptr[1] + dataptr[6];
tmp6 = dataptr[1] - dataptr[6];
tmp2 = dataptr[2] + dataptr[5];
tmp5 = dataptr[2] - dataptr[5];
tmp3 = dataptr[3] + dataptr[4];
tmp4 = dataptr[3] - dataptr[4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[0] = (DCTELEM) ((tmp10 + tmp11) << PASS1_BITS);
dataptr[4] = (DCTELEM) ((tmp10 - tmp11) << PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), CONST_BITS-PASS1_BITS);
dataptr[6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), CONST_BITS-PASS1_BITS);
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS-PASS1_BITS);
dataptr[5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS-PASS1_BITS);
dataptr[3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS-PASS1_BITS);
dataptr[1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS-PASS1_BITS);
dataptr += DCTSIZE; /* advance pointer to next row */
}
/* Pass 2: process columns.
* We remove the PASS1_BITS scaling, but leave the results scaled up
* by an overall factor of 8.
*/
dataptr = data;
for (ctr = DCTSIZE-1; ctr >= 0; ctr--) {
tmp0 = dataptr[DCTSIZE*0] + dataptr[DCTSIZE*7];
tmp7 = dataptr[DCTSIZE*0] - dataptr[DCTSIZE*7];
tmp1 = dataptr[DCTSIZE*1] + dataptr[DCTSIZE*6];
tmp6 = dataptr[DCTSIZE*1] - dataptr[DCTSIZE*6];
tmp2 = dataptr[DCTSIZE*2] + dataptr[DCTSIZE*5];
tmp5 = dataptr[DCTSIZE*2] - dataptr[DCTSIZE*5];
tmp3 = dataptr[DCTSIZE*3] + dataptr[DCTSIZE*4];
tmp4 = dataptr[DCTSIZE*3] - dataptr[DCTSIZE*4];
/* Even part per LL&M figure 1 --- note that published figure is faulty;
* rotator "sqrt(2)*c1" should be "sqrt(2)*c6".
*/
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp11, PASS1_BITS);
dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp10 - tmp11, PASS1_BITS);
z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100);
dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(z1 + MULTIPLY(tmp12, - FIX_1_847759065), CONST_BITS+PASS1_BITS);
/* Odd part per figure 8 --- note paper omits factor of sqrt(2).
* cK represents cos(K*pi/16).
* i0..i3 in the paper are tmp4..tmp7 here.
*/
z1 = tmp4 + tmp7;
z2 = tmp5 + tmp6;
z3 = tmp4 + tmp6;
z4 = tmp5 + tmp7;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp4 + z1 + z3, CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp5 + z2 + z4, CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp6 + z2 + z3, CONST_BITS+PASS1_BITS);
dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp7 + z1 + z4, CONST_BITS+PASS1_BITS);
dataptr++; /* advance pointer to next column */
}
for( i = 0; i < DCTSIZE; i++ )
for( j = 0; j < DCTSIZE; j++ )
out[i*DCTSIZE + j] = data[i*DCTSIZE + j] >> 3;
}
void
FeDrone_inverse_dct8 (
int32_t *block
)
{
int32_t tmp0, tmp1, tmp2, tmp3,
tmp10, tmp11, tmp12, tmp13,
z1, z2, z3, z4, z5,
workspace[DCT8_SIZE2];
int32_t *ws_ptr = workspace;
int32_t *out_ptr = block,
*in_ptr = block,
idx;
for (idx = DCT8_SIZE; idx > 0; idx--, in_ptr++, ws_ptr++) {
if (in_ptr[DCT8_SIZE * 1] == 0 && in_ptr[DCT8_SIZE * 2] == 0 &&
in_ptr[DCT8_SIZE * 3] == 0 && in_ptr[DCT8_SIZE * 4] == 0 &&
in_ptr[DCT8_SIZE * 5] == 0 && in_ptr[DCT8_SIZE * 6] == 0 &&
in_ptr[DCT8_SIZE * 7] == 0)
{
int32_t dcval = in_ptr[DCT8_SIZE * 0] << PASS1_BITS;
ws_ptr[DCT8_SIZE * 0] = dcval;
ws_ptr[DCT8_SIZE * 1] = dcval;
ws_ptr[DCT8_SIZE * 2] = dcval;
ws_ptr[DCT8_SIZE * 3] = dcval;
ws_ptr[DCT8_SIZE * 4] = dcval;
ws_ptr[DCT8_SIZE * 5] = dcval;
ws_ptr[DCT8_SIZE * 6] = dcval;
ws_ptr[DCT8_SIZE * 7] = dcval;
continue;
}
z2 = in_ptr[DCT8_SIZE * 2];
z3 = in_ptr[DCT8_SIZE * 6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
z2 = in_ptr[DCT8_SIZE * 0];
z3 = in_ptr[DCT8_SIZE * 4];
tmp0 = (z2 + z3) << CONST_BITS;
tmp1 = (z2 - z3) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
tmp0 = in_ptr[DCT8_SIZE * 7];
tmp1 = in_ptr[DCT8_SIZE * 5];
tmp2 = in_ptr[DCT8_SIZE * 3];
tmp3 = in_ptr[DCT8_SIZE * 1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
tmp0 = MULTIPLY(tmp0, FIX_0_298631336);
tmp1 = MULTIPLY(tmp1, FIX_2_053119869);
tmp2 = MULTIPLY(tmp2, FIX_3_072711026);
tmp3 = MULTIPLY(tmp3, FIX_1_501321110);
z1 = MULTIPLY( z1, -FIX_0_899976223);
z2 = MULTIPLY( z2, -FIX_2_562915447);
z3 = MULTIPLY( z3, -FIX_1_961570560);
z4 = MULTIPLY( z4, -FIX_0_390180644);
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
ws_ptr[DCT8_SIZE * 0] = DESCALE(tmp10 + tmp3, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 7] = DESCALE(tmp10 - tmp3, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 1] = DESCALE(tmp11 + tmp2, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 6] = DESCALE(tmp11 - tmp2, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 2] = DESCALE(tmp12 + tmp1, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 5] = DESCALE(tmp12 - tmp1, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 3] = DESCALE(tmp13 + tmp0, CONST_BITS - PASS1_BITS);
ws_ptr[DCT8_SIZE * 4] = DESCALE(tmp13 - tmp0, CONST_BITS - PASS1_BITS);
}
for (ws_ptr = workspace, idx = 0;
idx < DCT8_SIZE;
idx++, ws_ptr += DCT8_SIZE, out_ptr += DCT8_SIZE)
{
z2 = ws_ptr[2];
z3 = ws_ptr[6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
tmp0 = (ws_ptr[0] + ws_ptr[4]) << CONST_BITS;
tmp1 = (ws_ptr[0] - ws_ptr[4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
tmp0 = ws_ptr[7];
tmp1 = ws_ptr[5];
tmp2 = ws_ptr[3];
tmp3 = ws_ptr[1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
tmp0 = MULTIPLY(tmp0, FIX_0_298631336);
tmp1 = MULTIPLY(tmp1, FIX_2_053119869);
tmp2 = MULTIPLY(tmp2, FIX_3_072711026);
tmp3 = MULTIPLY(tmp3, FIX_1_501321110);
z1 = MULTIPLY(z1, - FIX_0_899976223);
z2 = MULTIPLY(z2, - FIX_2_562915447);
z3 = MULTIPLY(z3, - FIX_1_961570560);
z4 = MULTIPLY(z4, - FIX_0_390180644);
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
out_ptr[0] = (tmp10 + tmp3) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[7] = (tmp10 - tmp3) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[1] = (tmp11 + tmp2) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[6] = (tmp11 - tmp2) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[2] = (tmp12 + tmp1) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[5] = (tmp12 - tmp1) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[3] = (tmp13 + tmp0) >> (CONST_BITS + PASS1_BITS + 3);
out_ptr[4] = (tmp13 - tmp0) >> (CONST_BITS + PASS1_BITS + 3);
}
}
static void idct(const short* in, unsigned short* out)
{
INT32 tmp0, tmp1, tmp2, tmp3;
INT32 tmp10, tmp11, tmp12, tmp13;
INT32 z1, z2, z3, z4, z5;
int* wsptr;
int* outptr;
const short* inptr;
int ctr;
int workspace[DCTSIZE2]; /* buffers data between passes */
int data[DCTSIZE2];
// SHIFT_TEMPS
/* Pass 1: process columns from input, store into work array. */
/* Note results are scaled up by sqrt(8) compared to a true IDCT; */
/* furthermore, we scale the results by 2**PASS1_BITS. */
inptr = in;
wsptr = workspace;
for (ctr = DCTSIZE; ctr > 0; ctr--) {
/* Due to quantization, we will usually find that many of the input
* coefficients are zero, especially the AC terms. We can exploit this
* by short-circuiting the IDCT calculation for any column in which all
* the AC terms are zero. In that case each output is equal to the
* DC coefficient (with scale factor as needed).
* With typical images and quantization tables, half or more of the
* column DCT calculations can be simplified this way.
*/
if( inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
inptr[DCTSIZE*7] == 0 ) {
/* AC terms all zero */
int dcval = inptr[DCTSIZE*0] << PASS1_BITS;
wsptr[DCTSIZE*0] = dcval;
wsptr[DCTSIZE*1] = dcval;
wsptr[DCTSIZE*2] = dcval;
wsptr[DCTSIZE*3] = dcval;
wsptr[DCTSIZE*4] = dcval;
wsptr[DCTSIZE*5] = dcval;
wsptr[DCTSIZE*6] = dcval;
wsptr[DCTSIZE*7] = dcval;
inptr++; /* advance pointers to next column */
wsptr++;
continue;
}
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = inptr[DCTSIZE*2];
z3 = inptr[DCTSIZE*6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
z2 = inptr[DCTSIZE*0];
z3 = inptr[DCTSIZE*4];
tmp0 = (z2 + z3) << CONST_BITS;
tmp1 = (z2 - z3) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = inptr[DCTSIZE*7];
tmp1 = inptr[DCTSIZE*5];
tmp2 = inptr[DCTSIZE*3];
tmp3 = inptr[DCTSIZE*1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
inptr++; /* advance pointers to next column */
wsptr++;
}
/* Pass 2: process rows from work array, store into output array. */
/* Note that we must descale the results by a factor of 8 == 2**3, */
/* and also undo the PASS1_BITS scaling. */
wsptr = workspace;
outptr = data;
for (ctr = 0; ctr < DCTSIZE; ctr++) {
/* Even part: reverse the even part of the forward DCT. */
/* The rotator is sqrt(2)*c(-6). */
z2 = (INT32) wsptr[2];
z3 = (INT32) wsptr[6];
z1 = MULTIPLY(z2 + z3, FIX_0_541196100);
tmp2 = z1 + MULTIPLY(z3, - FIX_1_847759065);
tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865);
tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;
tmp10 = tmp0 + tmp3;
tmp13 = tmp0 - tmp3;
tmp11 = tmp1 + tmp2;
tmp12 = tmp1 - tmp2;
/* Odd part per figure 8; the matrix is unitary and hence its
* transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively.
*/
tmp0 = (INT32) wsptr[7];
tmp1 = (INT32) wsptr[5];
tmp2 = (INT32) wsptr[3];
tmp3 = (INT32) wsptr[1];
z1 = tmp0 + tmp3;
z2 = tmp1 + tmp2;
z3 = tmp0 + tmp2;
z4 = tmp1 + tmp3;
z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */
tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
z1 = MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
z2 = MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
z3 = MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
z4 = MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */
z3 += z5;
z4 += z5;
tmp0 += z1 + z3;
tmp1 += z2 + z4;
tmp2 += z2 + z3;
tmp3 += z1 + z4;
/* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
outptr[0] = (tmp10 + tmp3) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[7] = (tmp10 - tmp3) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[1] = (tmp11 + tmp2) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[6] = (tmp11 - tmp2) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[2] = (tmp12 + tmp1) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[5] = (tmp12 - tmp1) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[3] = (tmp13 + tmp0) >> ( CONST_BITS+PASS1_BITS+3 );
outptr[4] = (tmp13 - tmp0) >> ( CONST_BITS+PASS1_BITS+3 );
wsptr += DCTSIZE; /* advance pointer to next row */
outptr += DCTSIZE;
}
for(ctr = 0; ctr < DCTSIZE2; ctr++)
out[ctr] = data[ctr];
}