void pix_invert :: processGrayMMX(imageStruct &image)
{
int i = (image.xsize * image.ysize) / 8; // 8 pixels at a time
vector64i offset;
vector64i *input = (vector64i*)image.data;
offset.c[0]=255;
offset.c[1]=255;
offset.c[2]=255;
offset.c[3]=255;
offset.c[4]=255;
offset.c[5]=255;
offset.c[6]=255;
offset.c[7]=255;
while (i--) {
//*((unsigned long *)base) = ~*((unsigned long *)base);
input[0].v= _mm_xor_si64(input[0].v, offset.v);
input++;
}
_mm_empty();
}
void pix_invert :: processRGBAMMX(imageStruct &image)
{
int i = (image.xsize * image.ysize) / 2; // 2 pixels at a time
vector64i offset;
vector64i *input = (vector64i*)image.data;
offset.c[0+chRed]=255;
offset.c[0+chGreen]=255;
offset.c[0+chBlue]=255;
offset.c[0+chAlpha]=0;
offset.c[4+chRed]=255;
offset.c[4+chGreen]=255;
offset.c[4+chBlue]=255;
offset.c[4+chAlpha]=0;
while (i--) {
//*((unsigned long *)base) = ~*((unsigned long *)base);
input[0].v= _mm_xor_si64(input[0].v, offset.v);
input++;
}
_mm_empty();
}
__m64
unsigned_add3 (const __m64 * a, const __m64 * b,
__m64 * result, unsigned int count)
{
__m64 _a, _b, one, sum, carry, onesCarry;
unsigned int i;
carry = _mm_setzero_si64 ();
one = _mm_cmpeq_pi8 (carry, carry);
one = _mm_sub_si64 (carry, one);
for (i = 0; i < count; i++)
{
_a = a[i];
_b = b[i];
sum = _mm_add_si64 (_a, _b);
sum = _mm_add_si64 (sum, carry);
result[i] = sum;
onesCarry = _mm_and_si64 (_mm_xor_si64 (_a, _b), carry);
onesCarry = _mm_or_si64 (_mm_and_si64 (_a, _b), onesCarry);
onesCarry = _mm_and_si64 (onesCarry, one);
_a = _mm_srli_si64 (_a, 1);
_b = _mm_srli_si64 (_b, 1);
carry = _mm_add_si64 (_mm_add_si64 (_a, _b), onesCarry);
carry = _mm_srli_si64 (carry, 63);
}
return carry;
}
mlib_status
mlib_m_conv5x5_u16nw_2(
mlib_image *dst,
mlib_image *src,
mlib_s32 *kern,
mlib_s32 scalef_expon)
{
__m64 *pbuff, *buff_arr[20], **pbuff_arr = buff_arr;
__m64 *buff0, *buff1, *buff2, *buff3;
GET_SRC_DST_PARAMETERS(mlib_s16);
__m64 ker[5][5];
__m64 d0, d1, d2, aa, bb, rr, tmpa, tmpb, ker_off, mask8000;
__m64 prev0h, prev1h, prev2h, prev3h, sum0h, sum1h, sum2h, sum3h, sum4h,
tmph;
__m64 prev0l, prev1l, prev2l, prev3l, sum0l, sum1l, sum2l, sum3l, sum4l,
tmpl;
__m64 *sp, *dp;
mlib_s32 shift, ind, ker_sum = 0;
mlib_s32 row, wid4, i, j;
width -= 4;
height -= 4;
width *= NCHAN;
dl += 2 * (dll + NCHAN);
wid4 = (width + 7) / 4;
pbuff = mlib_malloc(sizeof (__m64) * 20 * wid4);
GET_KERN();
for (i = 0; i < 10; i++) {
buff_arr[i] = pbuff + i * 2 * wid4;
}
ind = 0;
for (j = 1; j <= 4; j++) {
buff0 = buff_arr[ind];
buff1 = buff_arr[ind + 1];
buff2 = buff_arr[ind + 2];
buff3 = buff_arr[ind + 3];
sp = (__m64 *) sl;
d1 = (*sp++);
d1 = _mm_xor_si64(d1, mask8000);
d2 = (*sp++);
d2 = _mm_xor_si64(d2, mask8000);
for (i = 0; i < wid4; i++) {
PREP_5x5();
}
sl += sll;
ind += j;
}
for (row = 0; row < height; row++) {
sp = (__m64 *) sl;
dp = (__m64 *) dl;
buff0 = pbuff_arr[0];
buff1 = pbuff_arr[2];
buff2 = pbuff_arr[5];
buff3 = pbuff_arr[9];
d1 = (*sp++);
d1 = _mm_xor_si64(d1, mask8000);
d2 = (*sp++);
d2 = _mm_xor_si64(d2, mask8000);
for (i = 0; i < width / 4; i++) {
CONV_5x5(hi, i);
dp[i] = rr;
}
if (width & 3) {
__m64 mask =
((__m64 *) mlib_mask64_arr)[2 * (width & 3)];
CONV_5x5(hi, i);
dp[i] =
_mm_or_si64(_mm_and_si64(mask, rr),
_mm_andnot_si64(mask, dp[i]));
}
ind = (pbuff_arr == buff_arr) ? 10 : -10;
pbuff_arr[ind + 0] = pbuff_arr[1];
pbuff_arr[ind + 1] = pbuff_arr[3];
pbuff_arr[ind + 2] = pbuff_arr[4];
pbuff_arr[ind + 3] = pbuff_arr[6];
pbuff_arr[ind + 4] = pbuff_arr[7];
pbuff_arr[ind + 5] = pbuff_arr[8];
pbuff_arr[ind + 6] = pbuff_arr[0];
pbuff_arr[ind + 7] = pbuff_arr[2];
pbuff_arr[ind + 8] = pbuff_arr[5];
pbuff_arr[ind + 9] = pbuff_arr[9];
pbuff_arr += ind;
sl += sll;
dl += dll;
}
_mm_empty();
mlib_free(pbuff);
return (MLIB_SUCCESS);
}
__m64 test56(__m64 a, __m64 b) {
// CHECK: pxor
return _mm_xor_si64(a, b);
}
static void Skein_512_Process_Block(Skein_512_Ctxt_t *ctx,const uint8 *blkPtr,size_t blkCnt,size_t byteCntAdd)
{
__m64 kw[12]; /* key schedule words : chaining vars + tweak */
__m64 X0,X1,X2,X3,X4,X5,X6,X7; /* local copy of vars, for speed */
__m64 w[8]; /* local copy of input block */
__m64 z1;
__m64 z3;
__m64 z5;
__m64 z7;
ts[0] = ctx->T[0];
ts[1] = ctx->T[1];
do {
ts[0] = _mm_add_si64(ts[0],_mm_set_pi32(0,byteCntAdd));
z1 = SKEIN_KS_PARITY;
ks[0] = ctx->X[0];
z1 = _mm_xor_si64(z1,ks[0]);
ks[1] = ctx->X[1];
z1 = _mm_xor_si64(z1,ks[1]);
ks[2] = ctx->X[2];
z1 = _mm_xor_si64(z1,ks[2]);
ks[3] = ctx->X[3];
z1 = _mm_xor_si64(z1,ks[3]);
ks[4] = ctx->X[4];
z1 = _mm_xor_si64(z1,ks[4]);
ks[5] = ctx->X[5];
z1 = _mm_xor_si64(z1,ks[5]);
ks[6] = ctx->X[6];
z1 = _mm_xor_si64(z1,ks[6]);
ks[7] = ctx->X[7];
z1 = _mm_xor_si64(z1,ks[7]);
ks[8] = z1;
ts[2] = _mm_xor_si64(ts[0],ts[1]);
X0 = 0[(__m64 *) blkPtr];
X1 = 1[(__m64 *) blkPtr];
X2 = 2[(__m64 *) blkPtr];
X3 = 3[(__m64 *) blkPtr];
X4 = 4[(__m64 *) blkPtr];
X5 = 5[(__m64 *) blkPtr];
X6 = 6[(__m64 *) blkPtr];
X7 = 7[(__m64 *) blkPtr];
w[0] = X0;
w[1] = X1;
w[2] = X2;
w[3] = X3;
w[4] = X4;
w[5] = X5;
w[6] = X6;
w[7] = X7;
X0 = _mm_add_si64(X0,ks[0]);
X1 = _mm_add_si64(X1,ks[1]);
X2 = _mm_add_si64(X2,ks[2]);
X3 = _mm_add_si64(X3,ks[3]);
X4 = _mm_add_si64(X4,ks[4]);
X5 = _mm_add_si64(X5,_mm_add_si64(ks[5],ts[0]));
X6 = _mm_add_si64(X6,_mm_add_si64(ks[6],ts[1]));
X7 = _mm_add_si64(X7,ks[7]);
blkPtr += 64;
#define R512(p0,p1,p2,p3,p4,p5,p6,p7,ROT,rNum) \
X##p0 = _mm_add_si64(X##p0,X##p1); \
X##p2 = _mm_add_si64(X##p2,X##p3); \
X##p4 = _mm_add_si64(X##p4,X##p5); \
X##p6 = _mm_add_si64(X##p6,X##p7); \
z1 = X##p1; \
X##p1 = _m_psrlqi(X##p1,64-ROT##_0); \
z1 = _m_psllqi(z1,ROT##_0); \
X##p1 = _mm_or_si64(X##p1,z1); \
z3 = X##p3; \
X##p3 = _m_psrlqi(X##p3,64-ROT##_1); \
z3 = _m_psllqi(z3,ROT##_1); \
X##p3 = _mm_or_si64(X##p3,z3); \
z5 = X##p5; \
z5 = _m_psllqi(z5,ROT##_2); \
X##p5 = _m_psrlqi(X##p5,64-ROT##_2); \
X##p5 = _mm_or_si64(X##p5,z5); \
z7 = X##p7; \
X##p7 = _m_psrlqi(X##p7,64-ROT##_3); \
z7 = _m_psllqi(z7,ROT##_3); \
X##p7 = _mm_or_si64(X##p7,z7); \
X##p1 = _mm_xor_si64(X##p1,X##p0); \
X##p3 = _mm_xor_si64(X##p3,X##p2); \
X##p5 = _mm_xor_si64(X##p5,X##p4); \
X##p7 = _mm_xor_si64(X##p7,X##p6); \
#define I512(R) \
X0 = _mm_add_si64(X0,ks[((R)+1) % 9]); /* inject the key schedule value */ \
X1 = _mm_add_si64(X1,ks[((R)+2) % 9]); \
X2 = _mm_add_si64(X2,ks[((R)+3) % 9]); \
X3 = _mm_add_si64(X3,ks[((R)+4) % 9]); \
X4 = _mm_add_si64(X4,ks[((R)+5) % 9]); \
X5 = _mm_add_si64(X5,_mm_add_si64(ks[((R)+6) % 9],ts[((R)+1) % 3])); \
X6 = _mm_add_si64(X6,_mm_add_si64(ks[((R)+7) % 9],ts[((R)+2) % 3])); \
X7 = _mm_add_si64(X7,_mm_add_si64(ks[((R)+8) % 9],_mm_set_pi32(0,(R)+1))); \
#define R512_8_rounds(R) \
R512(0,1,2,3,4,5,6,7,R_512_0,8*(R)+ 1); \
R512(2,1,4,7,6,5,0,3,R_512_1,8*(R)+ 2); \
R512(4,1,6,3,0,5,2,7,R_512_2,8*(R)+ 3); \
R512(6,1,0,7,2,5,4,3,R_512_3,8*(R)+ 4); \
I512(2*(R)); \
R512(0,1,2,3,4,5,6,7,R_512_4,8*(R)+ 5); \
R512(2,1,4,7,6,5,0,3,R_512_5,8*(R)+ 6); \
R512(4,1,6,3,0,5,2,7,R_512_6,8*(R)+ 7); \
R512(6,1,0,7,2,5,4,3,R_512_7,8*(R)+ 8); \
I512(2*(R)+1);
R512_8_rounds( 0);
R512_8_rounds( 1);
R512_8_rounds( 2);
R512_8_rounds( 3);
R512_8_rounds( 4);
R512_8_rounds( 5);
R512_8_rounds( 6);
R512_8_rounds( 7);
R512_8_rounds( 8);
ctx->X[0] = _mm_xor_si64(X0,w[0]);
ctx->X[1] = _mm_xor_si64(X1,w[1]);
ctx->X[2] = _mm_xor_si64(X2,w[2]);
ctx->X[3] = _mm_xor_si64(X3,w[3]);
ctx->X[4] = _mm_xor_si64(X4,w[4]);
ctx->X[5] = _mm_xor_si64(X5,w[5]);
ctx->X[6] = _mm_xor_si64(X6,w[6]);
ctx->X[7] = _mm_xor_si64(X7,w[7]);
ts[1] = _mm_and_si64(ts[1],_mm_set_pi32(~(((uint32) 64 ) << 24),~0));
} while (--blkCnt);
ctx->T[0] = ts[0];
ctx->T[1] = ts[1];
}
__m64 test_mm_xor_si64(__m64 a, __m64 b) {
// CHECK-LABEL: test_mm_xor_si64
// CHECK: call x86_mmx @llvm.x86.mmx.pxor
return _mm_xor_si64(a, b);
}
