__m64 test69(__m64 a) {
// CHECK: psllq
return _mm_slli_si64(a, 3);
}
void uyvy_to_yuv422(int width, int height, int shift_picture_down, const uint8_t *input, uint8_t *output)
{
__m64 chroma_mask = _mm_set_pi8(255, 0, 255, 0, 255, 0, 255, 0);
__m64 luma_mask = _mm_set_pi8(0, 255, 0, 255, 0, 255, 0, 255);
const uint8_t *orig_input = input;
uint8_t *y_comp = output;
uint8_t *u_comp = output + width * height;
uint8_t *v_comp = u_comp + (int)((width * height)/2); // 4:2:2
int i, j;
// When preparing video for PAL DV50 encoding, the video must be shifted
// down by one line to change the field order to be bottom-field-first
int start_line = 0;
if (shift_picture_down) {
memset(y_comp, 0x10, width); // write one line of black Y
y_comp += width;
memset(u_comp, 0x80, width/2); // write one line of black U,V
u_comp += width/2;
memset(v_comp, 0x80, width/2); // write one line of black U,V
v_comp += width/2;
start_line = 1;
}
/* Do the y component */
for (j = start_line; j < height; j++)
{
// Consume 16 bytes of UYVY data per iteration (8 pixels worth)
for (i = 0; i < width*2; i += 16)
{
//__m64 m1 = _mm_and_si64 (*(__m64 *)input, luma_mask);
//__m64 m2 = _mm_and_si64 (*(__m64 *)(input+8), luma_mask);
//__m64 m2 = _mm_set_pi8 (0, 0, 0, 0, 0, 0, 0, 0);
//*(__m64 *)y_comp = _mm_packs_pu16 (m2, m1);
__m64 m0 = *(__m64 *)input;
__m64 m2 = _mm_srli_si64(m0, 8);
__m64 m3 = _mm_slli_si64(m0, 8);
m3 = _mm_and_si64 (m3, chroma_mask);
m2 = _mm_and_si64 (m2, luma_mask);
m2 = _mm_or_si64 (m2, m3);
m2= _mm_and_si64 (m2, luma_mask);
m0 = m2;
__m64 m1 = *(__m64 *)(input+8);
m2 = _mm_srli_si64(m1, 8);
m3 = _mm_slli_si64(m1, 8);
m3 = _mm_and_si64 (m3, chroma_mask);
m2 = _mm_and_si64 (m2, luma_mask);
m2 = _mm_or_si64 (m2, m3);
m2= _mm_and_si64 (m2, luma_mask);
m1 = m2;
*(__m64 *)y_comp = _mm_packs_pu16 (m0, m1);
y_comp += 8;
input += 16;
}
}
/* Do the chroma components */
input = orig_input;
for (j = start_line; j < height; j++)
{
/* Process every line for yuv 4:2:2 */
for (i = 0; i < width*2; i += 16)
{
__m64 m1 = _mm_unpacklo_pi8 (*(__m64 *)input, *(__m64 *)(input+8));
__m64 m2 = _mm_unpackhi_pi8 (*(__m64 *)input, *(__m64 *)(input+8));
__m64 m3 = _mm_unpacklo_pi8 (m1, m2);
__m64 m4 = _mm_unpackhi_pi8 (m1, m2);
//*(__m64 *)u_comp = _mm_unpacklo_pi8 (m1, m2);
//*(__m64 *)v_comp = _mm_unpackhi_pi8 (m1, m2);
memcpy (u_comp, &m3, 4);
memcpy (v_comp, &m4, 4);
u_comp += 4;
v_comp += 4;
input += 16;
}
}
_mm_empty(); // Clear aliased fp register state
}
void sha384Process(register sha384Param* sp)
{
#ifdef OPTIMIZE_SSE2
# if defined(_MSC_VER) || defined (__INTEL_COMPILER)
static const __m64 MASK = { 0x00FF00FF00FF00FF00 };
# elif defined(__GNUC__)
static const __m64 MASK = { 0x00FF00FF, 0x00FF00FF };
# else
# error
# endif
__m64 a, b, c, d, e, f, g, h, temp;
register __m64 *w;
register const __m64 *k;
register byte t;
w = (__m64*) sp->data;
t = 16;
while (t--)
{
temp = *w;
*(w++) = _m_pxor(
_mm_slli_si64(_m_pshufw(_m_pand(temp, MASK), 27), 8),
_m_pshufw(_m_pand(_mm_srli_si64(temp, 8), MASK), 27)
);
}
t = 64;
while (t--)
{
temp = _mm_add_si64(_mm_add_si64(sig1(w[-2]), w[-7]), _mm_add_si64(sig0(w[-15]), w[-16]));
*(w++) = temp;
}
w = (__m64*) sp->h;
a = w[0]; b = w[1]; c = w[2]; d = w[3];
e = w[4]; f = w[5]; g = w[6]; h = w[7];
w = (__m64*) sp->data;
k = (__m64*) SHA2_64BIT_K;
#else
register uint64_t a, b, c, d, e, f, g, h, temp;
register uint64_t *w;
register const uint64_t *k;
register byte t;
# if WORDS_BIGENDIAN
w = sp->data + 16;
# else
w = sp->data;
t = 16;
while (t--)
{
temp = swapu64(*w);
*(w++) = temp;
}
# endif
t = 64;
while (t--)
{
temp = sig1(w[-2]) + w[-7] + sig0(w[-15]) + w[-16];
*(w++) = temp;
}
w = sp->data;
a = sp->h[0]; b = sp->h[1]; c = sp->h[2]; d = sp->h[3];
e = sp->h[4]; f = sp->h[5]; g = sp->h[6]; h = sp->h[7];
k = SHA2_64BIT_K;
#endif
ROUND(a,b,c,d,e,f,g,h,w[ 0],k[ 0]);
ROUND(h,a,b,c,d,e,f,g,w[ 1],k[ 1]);
ROUND(g,h,a,b,c,d,e,f,w[ 2],k[ 2]);
ROUND(f,g,h,a,b,c,d,e,w[ 3],k[ 3]);
ROUND(e,f,g,h,a,b,c,d,w[ 4],k[ 4]);
ROUND(d,e,f,g,h,a,b,c,w[ 5],k[ 5]);
ROUND(c,d,e,f,g,h,a,b,w[ 6],k[ 6]);
ROUND(b,c,d,e,f,g,h,a,w[ 7],k[ 7]);
ROUND(a,b,c,d,e,f,g,h,w[ 8],k[ 8]);
ROUND(h,a,b,c,d,e,f,g,w[ 9],k[ 9]);
ROUND(g,h,a,b,c,d,e,f,w[10],k[10]);
ROUND(f,g,h,a,b,c,d,e,w[11],k[11]);
ROUND(e,f,g,h,a,b,c,d,w[12],k[12]);
ROUND(d,e,f,g,h,a,b,c,w[13],k[13]);
ROUND(c,d,e,f,g,h,a,b,w[14],k[14]);
ROUND(b,c,d,e,f,g,h,a,w[15],k[15]);
ROUND(a,b,c,d,e,f,g,h,w[16],k[16]);
ROUND(h,a,b,c,d,e,f,g,w[17],k[17]);
ROUND(g,h,a,b,c,d,e,f,w[18],k[18]);
ROUND(f,g,h,a,b,c,d,e,w[19],k[19]);
ROUND(e,f,g,h,a,b,c,d,w[20],k[20]);
ROUND(d,e,f,g,h,a,b,c,w[21],k[21]);
ROUND(c,d,e,f,g,h,a,b,w[22],k[22]);
ROUND(b,c,d,e,f,g,h,a,w[23],k[23]);
ROUND(a,b,c,d,e,f,g,h,w[24],k[24]);
ROUND(h,a,b,c,d,e,f,g,w[25],k[25]);
ROUND(g,h,a,b,c,d,e,f,w[26],k[26]);
ROUND(f,g,h,a,b,c,d,e,w[27],k[27]);
ROUND(e,f,g,h,a,b,c,d,w[28],k[28]);
ROUND(d,e,f,g,h,a,b,c,w[29],k[29]);
ROUND(c,d,e,f,g,h,a,b,w[30],k[30]);
ROUND(b,c,d,e,f,g,h,a,w[31],k[31]);
ROUND(a,b,c,d,e,f,g,h,w[32],k[32]);
ROUND(h,a,b,c,d,e,f,g,w[33],k[33]);
ROUND(g,h,a,b,c,d,e,f,w[34],k[34]);
ROUND(f,g,h,a,b,c,d,e,w[35],k[35]);
ROUND(e,f,g,h,a,b,c,d,w[36],k[36]);
ROUND(d,e,f,g,h,a,b,c,w[37],k[37]);
ROUND(c,d,e,f,g,h,a,b,w[38],k[38]);
ROUND(b,c,d,e,f,g,h,a,w[39],k[39]);
ROUND(a,b,c,d,e,f,g,h,w[40],k[40]);
ROUND(h,a,b,c,d,e,f,g,w[41],k[41]);
ROUND(g,h,a,b,c,d,e,f,w[42],k[42]);
ROUND(f,g,h,a,b,c,d,e,w[43],k[43]);
ROUND(e,f,g,h,a,b,c,d,w[44],k[44]);
ROUND(d,e,f,g,h,a,b,c,w[45],k[45]);
ROUND(c,d,e,f,g,h,a,b,w[46],k[46]);
ROUND(b,c,d,e,f,g,h,a,w[47],k[47]);
ROUND(a,b,c,d,e,f,g,h,w[48],k[48]);
ROUND(h,a,b,c,d,e,f,g,w[49],k[49]);
ROUND(g,h,a,b,c,d,e,f,w[50],k[50]);
ROUND(f,g,h,a,b,c,d,e,w[51],k[51]);
ROUND(e,f,g,h,a,b,c,d,w[52],k[52]);
ROUND(d,e,f,g,h,a,b,c,w[53],k[53]);
ROUND(c,d,e,f,g,h,a,b,w[54],k[54]);
ROUND(b,c,d,e,f,g,h,a,w[55],k[55]);
ROUND(a,b,c,d,e,f,g,h,w[56],k[56]);
ROUND(h,a,b,c,d,e,f,g,w[57],k[57]);
ROUND(g,h,a,b,c,d,e,f,w[58],k[58]);
ROUND(f,g,h,a,b,c,d,e,w[59],k[59]);
ROUND(e,f,g,h,a,b,c,d,w[60],k[60]);
ROUND(d,e,f,g,h,a,b,c,w[61],k[61]);
ROUND(c,d,e,f,g,h,a,b,w[62],k[62]);
ROUND(b,c,d,e,f,g,h,a,w[63],k[63]);
ROUND(a,b,c,d,e,f,g,h,w[64],k[64]);
ROUND(h,a,b,c,d,e,f,g,w[65],k[65]);
ROUND(g,h,a,b,c,d,e,f,w[66],k[66]);
ROUND(f,g,h,a,b,c,d,e,w[67],k[67]);
ROUND(e,f,g,h,a,b,c,d,w[68],k[68]);
ROUND(d,e,f,g,h,a,b,c,w[69],k[69]);
ROUND(c,d,e,f,g,h,a,b,w[70],k[70]);
ROUND(b,c,d,e,f,g,h,a,w[71],k[71]);
ROUND(a,b,c,d,e,f,g,h,w[72],k[72]);
ROUND(h,a,b,c,d,e,f,g,w[73],k[73]);
ROUND(g,h,a,b,c,d,e,f,w[74],k[74]);
ROUND(f,g,h,a,b,c,d,e,w[75],k[75]);
ROUND(e,f,g,h,a,b,c,d,w[76],k[76]);
ROUND(d,e,f,g,h,a,b,c,w[77],k[77]);
ROUND(c,d,e,f,g,h,a,b,w[78],k[78]);
ROUND(b,c,d,e,f,g,h,a,w[79],k[79]);
#ifdef OPTIMIZE_SSE2
w = (__m64*) sp->h;
w[0] = _mm_add_si64(w[0], a);
w[1] = _mm_add_si64(w[1], b);
w[2] = _mm_add_si64(w[2], c);
w[3] = _mm_add_si64(w[3], d);
w[4] = _mm_add_si64(w[4], e);
w[5] = _mm_add_si64(w[5], f);
w[6] = _mm_add_si64(w[6], g);
w[7] = _mm_add_si64(w[7], h);
_mm_empty();
#else
sp->h[0] += a;
sp->h[1] += b;
sp->h[2] += c;
sp->h[3] += d;
sp->h[4] += e;
sp->h[5] += f;
sp->h[6] += g;
sp->h[7] += h;
#endif
}
void
mlib_m_ImageMaximum_U8_3(
mlib_s32 *res32,
const mlib_image *img)
{
/* src address */
__m64 *sp, *sl;
/* src data */
__m64 sd;
/* max values */
__m64 max0, max1, max2, max3;
/* edge mask */
mlib_s32 emask;
/* loop variables */
mlib_s32 n1;
/* height of image */
mlib_s32 height = mlib_ImageGetHeight(img);
/* elements to next row */
mlib_s32 slb = mlib_ImageGetStride(img);
mlib_s32 width = mlib_ImageGetWidth(img) * 3;
mlib_u8 *dend;
if (slb == width) {
width *= height;
height = 1;
}
sp = sl = (__m64 *) mlib_ImageGetData(img);
max1 = _mm_set1_pi8(MLIB_U8_MIN);
max2 = _mm_set1_pi8(MLIB_U8_MIN);
max3 = _mm_set1_pi8(MLIB_U8_MIN);
for (; height > 0; height--) {
n1 = width;
dend = (mlib_u8 *)sp + width;
for (; n1 > 23; n1 -= 24) {
sd = (*sp++);
MLIB_M_IMAGE_MAXIMUM_U8(max1, max1, sd);
sd = (*sp++);
MLIB_M_IMAGE_MAXIMUM_U8(max2, max2, sd);
sd = (*sp++);
MLIB_M_IMAGE_MAXIMUM_U8(max3, max3, sd);
}
if (n1 > 0) {
emask = (n1 > 7) ? 0xFF : (0xFF << (8 - n1));
sd = (*sp++);
MLIB_M_IMAGE_MAXIMUM_U8_M32(max1, max1, sd, emask);
n1 = ((mlib_u8 *)dend - (mlib_u8 *)sp);
if (n1 > 0) {
emask = (n1 > 7) ? 0xFF : (0xFF << (8 - n1));
sd = (*sp++);
MLIB_M_IMAGE_MAXIMUM_U8_M32(max2, max2, sd,
emask);
n1 = ((mlib_u8 *)dend - (mlib_u8 *)sp);
if (n1 > 0) {
emask = (0xFF << (8 - n1));
sd = *sp;
MLIB_M_IMAGE_MAXIMUM_U8_M32(max3, max3,
sd, emask);
}
}
}
sp = sl = (__m64 *) ((mlib_u8 *)sl + slb);
}
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max1, _mm_srli_si64(max2, 8),
mmx_write_64(0x00ffffffffffffffll));
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max0, _mm_slli_si64(max2, 16),
mmx_write_64(0x0000000000ff0000ll));
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max0, _mm_srli_si64(max3, 16),
mmx_write_64(0x0000ffffffffffffll));
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max0, _mm_slli_si64(max3, 8),
mmx_write_64(0x0000000000ffff00ll));
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max0, _mm_srli_si64(max0, 24),
mmx_write_64(0x000000ffff000000ll));
MLIB_M_IMAGE_MAXIMUM_U8_M64(max0, max0, _mm_srli_si64(max0, 24),
mmx_write_64(0x0000000000ffffffll));
res32[0] = _mm_cvtsi64_si32(_mm_and_si64(max0,
mmx_write_64(0x00000000000000ffll)));
res32[1] =
_mm_cvtsi64_si32(_mm_and_si64(_mm_srli_si64(max0, 8),
mmx_write_64(0x00000000000000ffll)));
res32[2] =
_mm_cvtsi64_si32(_mm_and_si64(_mm_srli_si64(max0, 16),
mmx_write_64(0x00000000000000ffll)));
_mm_empty();
}
__m64 test_mm_slli_si64(__m64 a) {
// CHECK-LABEL: test_mm_slli_si64
// CHECK: call x86_mmx @llvm.x86.mmx.pslli.q
return _mm_slli_si64(a, 3);
}
void lines_scale2(const unsigned char *src, unsigned y, unsigned char *dst1, unsigned char *dst2, unsigned nPix)
{
const unsigned char
*u = src + ((y-1) & 7)*sc2lines_width,
*m = src + ((y+0) & 7)*sc2lines_width,
*l = src + ((y+1) & 7)*sc2lines_width;
for (unsigned i = 0; i < nPix; i += 4) {
if (*(unsigned*)(u+i) ^ *(unsigned*)(l+i)) {
__m64 mm = *(__m64*)(m+i-2);
__m64 uu = *(__m64*)(u+i-2);
__m64 ll = *(__m64*)(l+i-2);
__m64 md = _mm_slli_si64(mm,8);
__m64 mf = _mm_srli_si64(mm,8);
__m64 maskall = _mm_or_si64(_mm_cmpeq_pi8(md,mf), _mm_cmpeq_pi8(uu,ll));
__m64 e0, e1, v1, v2;
e0 = _mm_cmpeq_pi8(md,uu);
e0 = _mm_andnot_si64(maskall, e0);
e0 = _mm_srli_si64(e0,16);
e0 = _mm_unpacklo_pi8(e0, _mm_setzero_si64());
e1 = _mm_cmpeq_pi8(mf,uu);
e1 = _mm_andnot_si64(maskall, e1);
e1 = _mm_srli_si64(e1,16);
e1 = _mm_unpacklo_pi8(_mm_setzero_si64(), e1);
e0 = _mm_or_si64(e0, e1);
v1 = _m_from_int(*(unsigned*)(m+i));
v2 = _m_from_int(*(unsigned*)(u+i));
v1 = _mm_unpacklo_pi8(v1,v1);
v2 = _mm_unpacklo_pi8(v2,v2);
*(__m64*)(dst1 + 2*i) = _mm_or_si64( _mm_and_si64(e0,v2), _mm_andnot_si64(e0,v1) );
e0 = _mm_cmpeq_pi8(md,ll);
e0 = _mm_andnot_si64(maskall, e0);
e0 = _mm_srli_si64(e0,16);
e0 = _mm_unpacklo_pi8(e0, _mm_setzero_si64());
e1 = _mm_cmpeq_pi8(mf,ll);
e1 = _mm_andnot_si64(maskall, e1);
e1 = _mm_srli_si64(e1,16);
e1 = _mm_unpacklo_pi8(_mm_setzero_si64(), e1);
e0 = _mm_or_si64(e0, e1);
v1 = _m_from_int(*(unsigned*)(m+i));
v2 = _m_from_int(*(unsigned*)(l+i));
v1 = _mm_unpacklo_pi8(v1,v1);
v2 = _mm_unpacklo_pi8(v2,v2);
*(__m64*)(dst2 + 2*i) = _mm_or_si64( _mm_and_si64(e0,v2), _mm_andnot_si64(e0,v1) );
} else {
__m64 v1 = _m_from_int(*(unsigned*)(m+i));
v1 = _mm_unpacklo_pi8(v1,v1);
*(__m64*)(dst1 + 2*i) = v1;
*(__m64*)(dst2 + 2*i) = v1;
}
}
}
