static void memzero_align8(void *dst,size_t size)
{
int fpscr;
fp_single_enter(fpscr);
dst = (char *)dst + size;
size /= 32;
__asm__ volatile (
" fldi0 fr0\n"
" fldi0 fr1\n"
" fschg\n" // double
"1: \n" \
" dt %1\n"
" fmov dr0,@-%0\n"
" fmov dr0,@-%0\n"
" fmov dr0,@-%0\n"
" bf.s 1b\n"
" fmov dr0,@-%0\n"
" fschg" //back to single
: "+r"(dst),"+r"(size) :: "memory" );
fp_single_leave(fpscr);
}
void idct_sh4(DCTELEM *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);
}
