template<class _mm> void TimgFilterNoiseMplayer::Tprocess::lineNoiseAvg_simd(uint8_t *dst,const uint8_t *src, int len, int8_t **shift_)
{
if (_mm::align && (intptr_t(src)&15 || intptr_t(dst)&15)) {
lineNoiseAvg_simd<typename _mm::T64>(dst,src,len,shift_);
return;
}
const int mmx_len=len&(~(_mm::size-1));
int8_t *shift2[3]= {shift_[0]+mmx_len, shift_[1]+mmx_len, shift_[2]+mmx_len};
for (int x=-mmx_len; x<0; x+=_mm::size) {
//".balign 16 \n\t"
typename _mm::__m mm0,mm1,mm2,mm3;
movdqu (mm1,shift2[0]+mmx_len+x);
movq (mm0,src+mmx_len+x);
typename _mm::__m shift1_8;
movVqu(shift1_8, shift2[1]+mmx_len+x);
paddb (mm1,shift1_8);
typename _mm::__m shift2_8;
movVqu(shift2_8, shift2[2]+mmx_len+x);
paddb (mm1,shift2_8);
movq (mm2,mm0);
movq (mm3,mm1);
punpcklbw (mm0,mm0);
punpckhbw (mm2,mm2);
punpcklbw (mm1,mm1);
punpckhbw (mm3,mm3);
pmulhw (mm1,mm0);
pmulhw (mm3,mm2);
paddw (mm1,mm1);
paddw (mm3,mm3);
paddw (mm1,mm0);
paddw (mm3,mm2);
psrlw (mm1,8);
psrlw (mm3,8);
packuswb (mm1,mm3);
movq (dst+mmx_len+x,mm1);
}
if (mmx_len!=len) {
lineNoiseAvg_C(dst+mmx_len, src+mmx_len, len-mmx_len, shift2);
}
}
int main()
{
int rval;
mmx_t ma;
mmx_t mb;
movq_r2r(mm0, mm1);
rval = mmx_ok();
/* Announce return value of mmx_ok() */
// printf("Value returned from init was %x.", rval);
// printf(" (Indicates MMX %s available)\n\n",(rval)? "is" : "not");
// fflush(stdout); fflush(stderr);
// if(rval)
{
/* PADD *****************************************************/
ma.q = 0x1111111180000000LL;
mb.q = 0x7fffffff00000001LL;
paddd(ma, mb);
fprintf(stdout, "paddd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddd: mb.q is 9111111080000001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x0001000100010001LL;
mb.q = 0x80007fffffff0001LL;
paddw(ma, mb);
fprintf(stdout, "paddw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddw: mb.q is 8001800000000002\n");
fflush(stdout); fflush(stderr);
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010000LL;
paddw(ma, mb);
fprintf(stdout, "paddw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddw: mb.q is 8001800000000001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x01010101807fff01LL;
mb.q = 0x807fff0101010101LL;
paddb(ma, mb);
fprintf(stdout, "paddb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddb: mb.q is 8180000281800002\n");
fflush(stdout); fflush(stderr);
/* PADDS ****************************************************/
ma.q = 0x0001000100010001LL;
mb.q = 0x80007fffffff0001LL;
paddsw(ma, mb);
fprintf(stdout, "paddsw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddsw: mb.q is 80017fff00000002\n");
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010000LL;
paddsw(ma, mb);
fprintf(stdout, "paddsw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddsw: mb.q is 80017fff00000001\n");
ma.q = 0x01010101807fff01LL;
mb.q = 0x807fff0101010101LL;
paddsb(ma, mb);
fprintf(stdout, "paddsb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddsb: mb.q is 817f0002817f0002\n");
fflush(stdout); fflush(stderr);
/* PADDUS ***************************************************/
ma.q = 0x0001000100010001LL;
mb.q = 0x80007fffffff0001LL;
paddusw(ma, mb);
fprintf(stdout, "paddusw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddusw: mb.q is 80018000ffff0002\n");
fflush(stdout); fflush(stderr);
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010000LL;
paddusw(ma, mb);
fprintf(stdout, "paddusw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddusw: mb.q is 80018000ffff0001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x01010101807fff01LL;
mb.q = 0x807fff0101010101LL;
paddusb(ma, mb);
fprintf(stdout, "paddusb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "paddusb: mb.q is 8180ff028180ff02\n");
fflush(stdout); fflush(stderr);
/* PSUB *****************************************************/
ma.q = 0x7fffffff00000001LL;
mb.q = 0x1111111180000000LL;
psubd(ma, mb);
fprintf(stdout, "psubd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubd: mb.q is 911111127fffffff\n");
fflush(stdout); fflush(stderr);
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010001LL;
psubw(ma, mb);
fprintf(stdout, "psubw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubw: mb.q is 8001800200020000\n");
fflush(stdout); fflush(stderr);
ma.q = 0x0001000100010000LL;
mb.q = 0x80007fffffff0001LL;
psubw(ma, mb);
fprintf(stdout, "psubw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubw: mb.q is 7fff7ffefffe0001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x807fff0101010101LL;
mb.q = 0x01010101807fff01LL;
psubb(ma, mb);
fprintf(stdout, "psubb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubb: mb.q is 818202007f7efe00\n");
fflush(stdout); fflush(stderr);
/* PSUBS ****************************************************/
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010001LL;
psubsw(ma, mb);
fprintf(stdout, "psubsw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubsw: mb.q is 7fff800200020000\n");
fflush(stdout); fflush(stderr);
ma.q = 0x0001000100010000LL;
mb.q = 0x80007fffffff0001LL;
psubsw(ma, mb);
fprintf(stdout, "psubsw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubsw: mb.q is 80007ffefffe0001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x807fff0101010101LL;
mb.q = 0x01010101807fff01LL;
psubsb(ma, mb);
fprintf(stdout, "psubsb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubsb: mb.q is 7f820200807efe00\n");
fflush(stdout); fflush(stderr);
/* PSUBUS ***************************************************/
ma.q = 0x80007fffffff0001LL;
mb.q = 0x0001000100010001LL;
psubusw(ma, mb);
fprintf(stdout, "psubusw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubusw: mb.q is 0000000000000000\n");
fflush(stdout); fflush(stderr);
ma.q = 0x0001000100010000LL;
mb.q = 0x80007fffffff0001LL;
psubusw(ma, mb);
fprintf(stdout, "psubusw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubusw: mb.q is 7fff7ffefffe0001\n");
fflush(stdout); fflush(stderr);
ma.q = 0x807fff0101010101LL;
mb.q = 0x01010101807fff01LL;
psubusb(ma, mb);
fprintf(stdout, "psubusb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psubusb: mb.q is 000000007f7efe00\n");
fflush(stdout); fflush(stderr);
/* PMUL *****************************************************/
ma.q = 0x8000ffff00ff0000LL;
mb.q = 0x0200ffff00ffffffLL;
pmulhw(ma, mb);
fprintf(stdout, "pmulhw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pmulhw: mb.q is ff00000000000000\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0200ffff00ffffffLL;
pmullw(ma, mb);
fprintf(stdout, "pmullw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pmullw: mb.q is 00000001fe010000\n");
fflush(stdout); fflush(stderr);
/* PMADD ****************************************************/
ma.q = 0x8000345680007f34LL;
mb.q = 0x93234a27ffff1707LL;
pmaddwd(ma, mb);
fprintf(stdout, "pmaddwd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pmaddwd: mb.q is 4597551a0b71a66c\n");
fflush(stdout); fflush(stderr);
/* PCMPEQ ***************************************************/
ma.q = 0x800034568f237f34LL;
mb.q = 0x93009a568f237f34LL;
pcmpeqd(ma, mb);
fprintf(stdout, "pcmpeqd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpeqd: mb.q is 00000000ffffffff\n");
fflush(stdout); fflush(stderr);
mb.q = 0x93009a568f237f34LL;
pcmpeqw(ma, mb);
fprintf(stdout, "pcmpeqw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpeqw: mb.q is 00000000ffffffff\n");
fflush(stdout); fflush(stderr);
mb.q = 0x93009a568f237f34LL;
pcmpeqb(ma, mb);
fprintf(stdout, "pcmpeqb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpeqb: mb.q is 00ff00ffffffffff\n");
fflush(stdout); fflush(stderr);
/* PCMPGT ***************************************************/
ma.q = 0x666688884477aaffLL;
mb.q = 0x1234567890abcdefLL;
pcmpgtd(ma, mb);
fprintf(stdout, "pcmpgtd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpgtd: mb.q is 0000000000000000\n");
fflush(stdout); fflush(stderr);
mb.q = 0x1234567890abcdefLL;
pcmpgtw(ma, mb);
fprintf(stdout, "pcmpgtw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpgtw: mb.q is 0000ffff0000ffff\n");
fflush(stdout); fflush(stderr);
mb.q = 0x1234567890abcdefLL;
pcmpgtb(ma, mb);
fprintf(stdout, "pcmpgtb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pcmpgtb: mb.q is 0000ffff0000ff00\n");
fflush(stdout); fflush(stderr);
/* PACKSS ***************************************************/
ma.q = 0x00012222000abbbbLL;
mb.q = 0x0000888800003333LL;
packssdw(ma, mb);
fprintf(stdout, "packssdw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "packssdw: mb.q is 7fff7fff7fff3333\n");
fflush(stdout); fflush(stderr);
ma.q = 0x00aa00dd01009999LL;
mb.q = 0x0011002200330044LL;
packsswb(ma, mb);
fprintf(stdout, "packsswb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "packsswb: mb.q is 7f7f7f8011223344\n");
fflush(stdout); fflush(stderr);
/* PACKUS ***************************************************/
ma.q = 0x00aa00dd01009999LL;
mb.q = 0x0011002200330044LL;
packuswb(ma, mb);
fprintf(stdout, "packuswb: mb.q is %016llx\n", mb.q);
fprintf(stderr, "packuswb: mb.q is aaddff0011223344\n");
fflush(stdout); fflush(stderr);
/* PUNPCKH **************************************************/
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
punpckhdq(ma, mb);
fprintf(stdout, "punpckhdq: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpckhdq: mb.q is 090a0b0c01020304\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0102030405060708LL;
punpckhwd(ma, mb);
fprintf(stdout, "punpckhwd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpckhwd: mb.q is 090a01020b0c0304\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0102030405060708LL;
punpckhbw(ma, mb);
fprintf(stdout, "punpckhbw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpckhbw: mb.q is 09010a020b030c04\n");
fflush(stdout); fflush(stderr);
/* PUNPCKL **************************************************/
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
punpckldq(ma, mb);
fprintf(stdout, "punpckldq: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpckldq: mb.q is 0d0e0f0005060708\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0102030405060708LL;
punpcklwd(ma, mb);
fprintf(stdout, "punpcklwd: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpcklwd: mb.q is 0d0e05060f000708\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0102030405060708LL;
punpcklbw(ma, mb);
fprintf(stdout, "punpcklbw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "punpcklbw: mb.q is 0d050e060f070008\n");
fflush(stdout); fflush(stderr);
/* PAND, PANDN, POR, PXOR ***********************************/
ma.q = 0x5555555555555555LL;
mb.q = 0x3333333333333333LL;
pand(ma, mb);
fprintf(stdout, "pand: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pand: mb.q is 1111111111111111\n");
fflush(stdout); fflush(stderr);
mb.q = 0x3333333333333333LL;
pandn(ma, mb);
fprintf(stdout, "pandn: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pandn: mb.q is 4444444444444444\n");
fflush(stdout); fflush(stderr);
mb.q = 0x3333333333333333LL;
por(ma, mb);
fprintf(stdout, "por: mb.q is %016llx\n", mb.q);
fprintf(stderr, "por: mb.q is 7777777777777777\n");
fflush(stdout); fflush(stderr);
mb.q = 0x3333333333333333LL;
pxor(ma, mb);
fprintf(stdout, "pxor: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pxor: mb.q is 6666666666666666\n");
fflush(stdout); fflush(stderr);
/* PSLL *****************************************************/
ma.q = 0x0000000000000018LL;
mb.q = 0x0123456789abcdefLL;
psllq(ma, mb);
fprintf(stdout, "psllq: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psllq: mb.q is 6789abcdef000000\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0123456789abcdefLL;
pslld(ma, mb);
fprintf(stdout, "pslld: mb.q is %016llx\n", mb.q);
fprintf(stderr, "pslld: mb.q is 67000000ef000000\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0123456789abcdefLL;
psllw(ma, mb);
fprintf(stdout, "psllw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psllw: mb.q is 0000000000000000\n");
fflush(stdout); fflush(stderr);
/* PSRL *****************************************************/
ma.q = 0x0000000000000018LL;
mb.q = 0x0123456789abcdefLL;
psrlq(ma, mb);
fprintf(stdout, "psrlq: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psrlq: mb.q is 0000000123456789\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0123456789abcdefLL;
psrld(ma, mb);
fprintf(stdout, "psrld: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psrld: mb.q is 0000000100000089\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0123456789abcdefLL;
psrlw(ma, mb);
fprintf(stdout, "psrlw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psrlw: mb.q is 0000000000000000\n");
fflush(stdout); fflush(stderr);
/* PSRA *****************************************************/
ma.q = 0x0000000000000018LL;
mb.q = 0x0123456789abcdefLL;
psrad(ma, mb);
fprintf(stdout, "psrad: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psrad: mb.q is 00000001ffffff89\n");
fflush(stdout); fflush(stderr);
mb.q = 0x0123456789abcdefLL;
psraw(ma, mb);
fprintf(stdout, "psraw: mb.q is %016llx\n", mb.q);
fprintf(stderr, "psraw: mb.q is 00000000ffffffff\n");
fflush(stdout); fflush(stderr);
/* Exit MXX *************************************************/
emms();
}
/* Clean-up and exit nicely */
exit(0);
}
template <class _mm, int src_aligned, int dst_aligned> void TffdshowConverters2::convert_simd_GBRPtoRGB(
const uint8_t* srcG,
const uint8_t* srcB,
const uint8_t* srcR,
uint8_t* dst,
int dx,
int dy,
stride_t stride_src,
stride_t stride_dst)
{
int xCount = dx / _mm::size;
if (xCount <= 0)
return;
_mm::__m _mm0,_mm1,_mm2,_mm3,_mm4,_mm5,_mm6;
_mm::__m ffff;
pxor(ffff,ffff);
pcmpeqb(ffff,ffff);
for (int y = 0 ; y < dy ; y++) {
const uint8_t *g = srcG + y * stride_src;
const uint8_t *b = srcB + y * stride_src;
const uint8_t *r = srcR + y * stride_src;
uint8_t *dst1 = dst + y * stride_dst;
int x = xCount;
do {
if (src_aligned) {
movVqa(_mm0, r);
movVqa(_mm1, g);
movVqa(_mm2, b);
} else {
movVqu(_mm0, r);
movVqu(_mm1, g);
movVqu(_mm2, b);
}
_mm4 = _mm0;
_mm5 = _mm2;
punpcklbw(_mm0,ffff); // 0xff,R
punpcklbw(_mm2,_mm1); // G,B
punpckhbw(_mm4,ffff); // 0xff,R
punpckhbw(_mm5,_mm1); // G,B
_mm3 = _mm2;
_mm6 = _mm5;
punpckhwd(_mm2,_mm0); // 0xff,RGB * 4 (dst+_mm::size)
punpcklwd(_mm3,_mm0); // 0xff,RGB * 4 (dst)
punpckhwd(_mm5,_mm4); // 0xff,RGB * 4 (dst+_mm::size)
punpcklwd(_mm6,_mm4); // 0xff,RGB * 4 (dst)
if (dst_aligned) {
_mm::movntVq(dst1 , _mm3);
_mm::movntVq(dst1 + _mm::size , _mm2);
_mm::movntVq(dst1 + _mm::size*2, _mm6);
_mm::movntVq(dst1 + _mm::size*3, _mm5);
} else {
movVqu(dst1 , _mm3);
movVqu(dst1 + _mm::size , _mm2);
movVqu(dst1 + _mm::size*2, _mm6);
movVqu(dst1 + _mm::size*3, _mm5);
}
r += _mm::size;
g += _mm::size;
b += _mm::size;
dst1 += _mm::size * 4;
} while(--x);
}
if (xCount * (int)_mm::size < dx && dx > _mm::size) {
// handle non-mod 8 resolution.
int dxDone = dx - _mm::size;
srcG += dxDone;
srcB += dxDone;
srcR += dxDone;
dst += dxDone * 4;
convert_simd_GBRPtoRGB<_mm, 0, 0>(srcG, srcB, srcR, dst, _mm::size, dy, stride_src, stride_dst);
}
_mm::empty();
}
template <class _mm, int src_aligned, int dst_aligned> void TffdshowConverters2::convert_simd_AYUV(
const uint8_t* srcY,
const uint8_t* srcCb,
const uint8_t* srcCr,
uint8_t* dst,
int dx,
int dy,
stride_t stride_src,
stride_t stride_dst)
{
int xCount = dx / _mm::size;
if (xCount <= 0)
return;
_mm::__m _mm0,_mm1,_mm2,_mm3,_mm4,_mm5;
_mm::__m ffff;
pxor(ffff,ffff);
pcmpeqb(ffff,ffff);
for (int y = 0 ; y < dy ; y++) {
if (y == dy-1)
int a=0;
const uint8_t *Y = srcY + y * stride_src;
uint8_t *dst1 = dst + y * stride_dst;
const uint8_t *Cb = srcCb + y * stride_src;
const uint8_t *Cr = srcCr + y * stride_src;
int x = xCount;
do {
if (src_aligned) {
movVqa(_mm1, Y);
movVqa(_mm4, Cb);
movVqa(_mm2, Cr);
} else {
movVqu(_mm1, Y);
movVqu(_mm4, Cb);
movVqu(_mm2, Cr);
}
Y += _mm::size;
Cb += _mm::size;
Cr += _mm::size;
_mm3 = _mm2;
punpcklbw(_mm2, _mm4);
punpckhbw(_mm3, _mm4);
_mm5 = _mm1;
punpcklbw(_mm1, ffff);
punpckhbw(_mm5, ffff);
_mm4 = _mm2;
punpcklwd(_mm2, _mm1);
punpckhwd(_mm4, _mm1);
_mm0 = _mm3;
punpcklwd(_mm0, _mm5);
punpckhwd(_mm3, _mm5);
if (dst_aligned) {
_mm::movntVq(dst1, _mm2);
_mm::movntVq(dst1 + _mm::size, _mm4);
_mm::movntVq(dst1 + _mm::size * 2, _mm0);
_mm::movntVq(dst1 + _mm::size * 3, _mm3);
} else {
movVqu(dst1, _mm2);
movVqu(dst1 + _mm::size, _mm4);
movVqu(dst1 + _mm::size * 2, _mm0);
movVqu(dst1 + _mm::size * 3, _mm3);
}
dst1 += _mm::size * 4;
} while(--x);
}
if (xCount * (int)_mm::size < dx && dx > _mm::size) {
// handle non-mod 8 resolution.
int dxDone = dx - _mm::size;
srcY += dxDone;
srcCb += dxDone;
srcCr += dxDone;
dst += dxDone * 4;
convert_simd_AYUV<_mm, 0, 0>(srcY, srcCb, srcCr, dst, _mm::size, dy, stride_src, stride_dst);
}
_mm::empty();
}
template <class _mm, int src_aligned, int dst_aligned> void TffdshowConverters2::convert_YV12toNV12(
const uint8_t* srcY,
const uint8_t* srcCb,
const uint8_t* srcCr,
uint8_t* dstY,
uint8_t* dstCbCr,
int dx,
int dy,
stride_t stride_Y,
stride_t stride_CbCr,
stride_t stride_dstY,
stride_t stride_dstCbCr)
{
int xCount = dx / (_mm::size*2);
if (xCount <= 0)
return;
_mm::__m _mm0,_mm1,_mm2,_mm3;
_mm::__m zero;
pxor(zero,zero);
for (int y = 0 ; y < dy ; y++) {
const uint8_t *src = srcY + y * stride_Y;
uint8_t *dst = dstY + y * stride_dstY;
int x = xCount;
do {
if (src_aligned) {
movVqa(_mm0, src);
movVqa(_mm1, src + _mm::size);
} else {
movVqu(_mm0, src);
movVqu(_mm1, src + _mm::size);
}
src += _mm::size * 2;
if (dst_aligned) {
_mm::movntVq(dst, _mm0);
_mm::movntVq(dst + _mm::size, _mm1);
} else {
movVqu(dst, _mm0);
movVqu(dst + _mm::size, _mm1);
}
dst += _mm::size * 2;
} while(--x);
}
int dyCbCr = dy/2;
for (int y = 0 ; y < dyCbCr ; y++) {
const uint8_t *Cb = srcCb + y * stride_CbCr;
const uint8_t *Cr = srcCr + y * stride_CbCr;
uint8_t *dst = dstCbCr + y * stride_dstCbCr;
int x = xCount;
do {
if (src_aligned) {
movVqa(_mm0, Cb);
movVqa(_mm1, Cr);
} else {
movVqu(_mm0, Cb);
movVqu(_mm1, Cr);
}
_mm2 = _mm0;
_mm3 = _mm1;
Cb += _mm::size;
Cr += _mm::size;
punpcklbw(_mm0, _mm1);
punpckhbw(_mm2, _mm3);
if (dst_aligned) {
_mm::movntVq(dst, _mm0);
_mm::movntVq(dst + _mm::size, _mm2);
} else {
movVqu(dst, _mm0);
movVqu(dst + _mm::size, _mm2);
}
dst += _mm::size * 2;
} while(--x);
}
if (xCount * (int)_mm::size * 2 < dx && dx > _mm::size * 2) {
int dxDone = dx - _mm::size * 2;
srcY += dxDone;
srcCb += dxDone/2;
srcCr += dxDone/2;
dstY += dxDone;
dstCbCr += dxDone;
convert_YV12toNV12<_mm, 0, 0>(srcY, srcCb, srcCr, dstY, dstCbCr, _mm::size * 2, dy, stride_Y, stride_CbCr, stride_dstY, stride_dstCbCr);
}
_mm::empty();
}
int main()
{
int rval, i, co, tmp; //, j;
mmx_t ma, mb; //, mm0, mm1, *pm0, *pm1;
tmp = calc_cpu_speed();
printf(" Calculating CPU-Speed....running at %d MHz\n", tmp);
printf(" Calculating CPU-OFFSET...found : ");
co = calc_cpu_ofs();
printf(" %d\n",co);
rval = 1; //mmx_ok(); /* Announce return value of mmx_ok() */
printf(" *** Bogus message since we are emulating so MMX does allways exist ***\n");
printf(" Value returned from init was %x.", rval);
printf(" (Indicates MMX %s available)\n\n",(rval)? "is" : "not");
if(rval) {
/* PADD *****************************************************/
ma.q = 0x1111111180000000LL;
mb.q = 0x7fffffff00000001LL;
mmx_regdump(ma);
mmx_regdump(mb);
paddd( &ma, &mb);
printf("paddd: mb.q is %016llx\n", mb.q);
mmx_regdump(mb);
paddw( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
ma.q = 0x0000000000000008LL;
mb.q = 0x0001000200030004LL;
mmx_regdump(ma);
mmx_regdump(mb);
psllw( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
ma.q = 0x0000000000000000LL;
mb.q = 0x888044a87f06fe80LL;
mmx_regdump(ma);
mmx_regdump(mb);
// packuswb( &ma,&mb);
ma.q = 0x00aa00dd01009999LL;
mb.q = 0x0011002200330044LL;
mmx_regdump(ma);
mmx_regdump(mb);
packuswb( &ma, &mb);
// punpckhdq(&ma,&mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpckhdq */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpckhdq( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpckhwd */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpckhwd( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpckhbw */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpckhbw( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpckldq */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpckldq( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpcklwd */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpcklwd( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* punpcklbw */
ma.q = 0x090a0b0c0d0e0f00LL;
mb.q = 0x0102030405060708LL;
mmx_regdump(ma);
mmx_regdump(mb);
punpcklbw( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* pmaddwd */
ma.q = 0x8000800080008000LL;
mb.q = 0x8000800080008000LL;
mmx_regdump(ma);
mmx_regdump(mb);
pmaddwd( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* packsswb */
ma.q = 0x00aa00dd01009999LL;
mb.q = 0x0011002200330044LL;
mmx_regdump(ma);
mmx_regdump(mb);
packsswb( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
/* packsswb */
ma.q = 0x007e7f00ef9dff88LL;
mb.q = 0xff020085007e81cfLL;
mmx_regdump(ma);
mmx_regdump(mb);
packsswb( &ma, &mb);
mmx_regdump(ma);
mmx_regdump(mb);
for (i=0; i< 0x0F; i++){
printf ("%d. Test running\n", i);
GET_TSC(tsc1);
mb.w[0] = i;
ma.w[0] = i;
GET_TSC(tsc2);
tick_dump(co);
GET_TSC(tsc1); /* Let's test some real asm */
asm("
// pushl %eax\n
// movl $0, %eax\n
// cpuid\n
packsswb %mm0, %mm1\n
pmaddwd %mm0, %mm1\n
punpcklbw %mm0, %mm1\n
punpcklbw %mm0, %mm1\n
pmaddwd %mm0, %mm1\n
punpcklbw %mm0, %mm1\n
// popl %eax\n
");
GET_TSC(tsc2);
tick_dump(co);
ma.q = 0x8000800080008000LL;
mb.q = 0x8000800080008000LL;
GET_TSC(tsc1); /* Here some emulated routines */
pmaddwd( &ma, &mb);
// packsswb( &ma, &mb);
GET_TSC(tsc2);
tick_dump(co);
// mmx_regdump(mb);
}
}
exit(0); /* Clean-up and exit nicely */
}