void diff(spinor * const Q,spinor * const R,spinor * const S, const int N)
{
int ix = 1;
double *s ALIGN;
double *sp ALIGN;
double *r ALIGN;
double *rp ALIGN;
double *q ALIGN;
double _Complex x00, x01, x02, x03, x04, x05, x06, x07,
x08, x09, x10, x11;
double _Complex y00, y01, y02, y03, y04, y05, y06, y07,
y08, y09, y10, y11;
#pragma disjoint(*R, *S)
__alignx(16, Q);
__alignx(16, R);
__alignx(16, S);
r = (double*) R;
s = (double*) S;
q = (double*) Q;
rp = r + 24;
sp = s + 24;
_prefetch_spinor(rp);
_prefetch_spinor(sp);
x00 = __lfpd(r);
x01 = __lfpd(r+2);
x02 = __lfpd(r+4);
x03 = __lfpd(r+6);
x04 = __lfpd(r+8);
x05 = __lfpd(r+10);
x06 = __lfpd(r+12);
x07 = __lfpd(r+14);
x08 = __lfpd(r+16);
x09 = __lfpd(r+18);
x10 = __lfpd(r+20);
x11 = __lfpd(r+22);
y00 = __lfpd(s);
y01 = __lfpd(s+2);
y02 = __lfpd(s+4);
y03 = __lfpd(s+6);
y04 = __lfpd(s+8);
y05 = __lfpd(s+10);
y06 = __lfpd(s+12);
y07 = __lfpd(s+14);
y08 = __lfpd(s+16);
y09 = __lfpd(s+18);
y10 = __lfpd(s+20);
y11 = __lfpd(s+22);
__stfpd(q, __fpsub(x00, y00));
__stfpd(q+2, __fpsub(x01, y01));
__stfpd(q+4, __fpsub(x02, y02));
__stfpd(q+6, __fpsub(x03, y03));
__stfpd(q+8, __fpsub(x04, y04));
__stfpd(q+10, __fpsub(x05, y05));
__stfpd(q+12, __fpsub(x06, y06));
__stfpd(q+14, __fpsub(x07, y07));
__stfpd(q+16, __fpsub(x08, y08));
__stfpd(q+18, __fpsub(x09, y09));
__stfpd(q+20, __fpsub(x10, y10));
__stfpd(q+22, __fpsub(x11, y11));
s = sp;
r = rp;
q+=24;
#pragma unroll(12)
for(ix = 1; ix < N-1; ix++) {
rp+=24;
sp+=24;
_prefetch_spinor(rp);
_prefetch_spinor(sp);
x00 = __lfpd(r);
x01 = __lfpd(r+2);
x02 = __lfpd(r+4);
x03 = __lfpd(r+6);
x04 = __lfpd(r+8);
x05 = __lfpd(r+10);
x06 = __lfpd(r+12);
x07 = __lfpd(r+14);
x08 = __lfpd(r+16);
x09 = __lfpd(r+18);
x10 = __lfpd(r+20);
x11 = __lfpd(r+22);
y00 = __lfpd(s);
y01 = __lfpd(s+2);
y02 = __lfpd(s+4);
y03 = __lfpd(s+6);
y04 = __lfpd(s+8);
y05 = __lfpd(s+10);
y06 = __lfpd(s+12);
y07 = __lfpd(s+14);
y08 = __lfpd(s+16);
y09 = __lfpd(s+18);
y10 = __lfpd(s+20);
y11 = __lfpd(s+22);
__stfpd(q, __fpsub(x00, y00));
__stfpd(q+2, __fpsub(x01, y01));
__stfpd(q+4, __fpsub(x02, y02));
__stfpd(q+6, __fpsub(x03, y03));
__stfpd(q+8, __fpsub(x04, y04));
__stfpd(q+10, __fpsub(x05, y05));
__stfpd(q+12, __fpsub(x06, y06));
__stfpd(q+14, __fpsub(x07, y07));
__stfpd(q+16, __fpsub(x08, y08));
__stfpd(q+18, __fpsub(x09, y09));
__stfpd(q+20, __fpsub(x10, y10));
__stfpd(q+22, __fpsub(x11, y11));
s = sp;
r = rp;
q+=24;
}
x00 = __lfpd(r);
x01 = __lfpd(r+2);
x02 = __lfpd(r+4);
x03 = __lfpd(r+6);
x04 = __lfpd(r+8);
x05 = __lfpd(r+10);
x06 = __lfpd(r+12);
x07 = __lfpd(r+14);
x08 = __lfpd(r+16);
x09 = __lfpd(r+18);
x10 = __lfpd(r+20);
x11 = __lfpd(r+22);
y00 = __lfpd(s);
y01 = __lfpd(s+2);
y02 = __lfpd(s+4);
y03 = __lfpd(s+6);
y04 = __lfpd(s+8);
y05 = __lfpd(s+10);
y06 = __lfpd(s+12);
y07 = __lfpd(s+14);
y08 = __lfpd(s+16);
y09 = __lfpd(s+18);
y10 = __lfpd(s+20);
y11 = __lfpd(s+22);
__stfpd(q, __fpsub(x00, y00));
__stfpd(q+2, __fpsub(x01, y01));
__stfpd(q+4, __fpsub(x02, y02));
__stfpd(q+6, __fpsub(x03, y03));
__stfpd(q+8, __fpsub(x04, y04));
__stfpd(q+10, __fpsub(x05, y05));
__stfpd(q+12, __fpsub(x06, y06));
__stfpd(q+14, __fpsub(x07, y07));
__stfpd(q+16, __fpsub(x08, y08));
__stfpd(q+18, __fpsub(x09, y09));
__stfpd(q+20, __fpsub(x10, y10));
__stfpd(q+22, __fpsub(x11, y11));
return;
}
int main(int argc, char* argv[])
{
fprintf(stderr,"BEGIN TESTING OF DP ACCUMULATE\n");
printf("%18s %18s %18s %18s %18s %18s\n","dim","basic","1-hummer","2-hummers","4-hummers","8-hummers");
int k;
for (k=6;k<20;k++)
{
int dim = pow(2,k);
int count = 10;
int i,j;
unsigned long long t0, t1;
unsigned long long dt0, dt1, dt2, dt3, dt4;
double* a;
double* b;
double* c;
double scale = 0.1;
posix_memalign((void**)&a, 16*sizeof(double), dim*sizeof(double));
posix_memalign((void**)&b, 16*sizeof(double), dim*sizeof(double));
posix_memalign((void**)&c, 16*sizeof(double), dim*sizeof(double));
for (i=0;i<dim;i++) a[i] = 1.0 - 2*(double)rand()/(double)RAND_MAX;
fprintf(stderr,"BASIC VERSION\n");
// WARM-UP
for (i=0;i<dim;i++) b[i] = 0.0;
for (i=0;i<dim;i++) b[i] += scale*a[i];
// TIMING
for (i=0;i<dim;i++) b[i] = 0.0;
t0 = getticks();
for (j=0;j<count;j++)
{
for (i=0;i<dim;i++) b[i] += scale*a[i];
}
t1 = getticks();
dt0 = t1 - t0;
fprintf(stderr,"INTRINSICS VERSION 1\n");
// WARM-UP
for (i=0;i<dim;i++) c[i] = 0.0;
for (i=0;i<dim;i+=2)
{
__stfpd(&c[i], __fxcpmadd( __lfpd(&c[i]), __lfpd(&a[i]), scale) );
}
// TIMING
for (i=0;i<dim;i++) c[i] = 0.0;
t0 = getticks();
for (j=0;j<count;j++)
{
for (i=0;i<dim;i+=2)
{
__stfpd(&c[i], __fxcpmadd( __lfpd(&c[i]), __lfpd(&a[i]), scale) );
}
}
t1 = getticks();
dt1 = t1 - t0;
// VERIFICATION
for (i=0;i<dim;i++)
{
if (b[i] != c[i])
{
printf("%4d %30.15f %30.15f\n",i,b[i],c[i]);
}
}
fprintf(stderr,"INTRINSICS VERSION 2\n");
// WARM-UP
for (i=0;i<dim;i++) c[i] = 0.0;
for (i=0;i<dim;i+=4)
{
{
double _Complex a0, a2, c0, c2;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+2]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+2]);
c0 = __fxcpmadd(c0,a0,scale);
c2 = __fxcpmadd(c2,a2,scale);
__stfpd(&c[i ],c0);
__stfpd(&c[i+2],c2);
}
}
// TIMING
for (i=0;i<dim;i++) c[i] = 0.0;
t0 = getticks();
for (j=0;j<count;j++)
{
for (i=0;i<dim;i+=4)
{
{
double _Complex a0, a2, c0, c2;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+2]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+2]);
c0 = __fxcpmadd(c0,a0,scale);
c2 = __fxcpmadd(c2,a2,scale);
__stfpd(&c[i ],c0);
__stfpd(&c[i+2],c2);
}
}
}
t1 = getticks();
dt2 = t1 - t0;
// VERIFICATION
for (i=0;i<dim;i++)
{
if (b[i] != c[i])
{
printf("%4d %30.15f %30.15f\n",i,b[i],c[i]);
}
}
fprintf(stderr,"INTRINSICS VERSION 3\n");
// WARM-UP
for (i=0;i<dim;i++) c[i] = 0.0;
for (i=0;i<dim;i+=8)
{
{
double _Complex a0, a2, a4, a6;
double _Complex c0, c2, c4, c6;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+2]);
a4 = __lfpd(&a[i+4]);
a6 = __lfpd(&a[i+6]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+2]);
c4 = __lfpd(&c[i+4]);
c6 = __lfpd(&c[i+6]);
c0 = __fxcpmadd(c0,a0,scale);
c2 = __fxcpmadd(c2,a2,scale);
c4 = __fxcpmadd(c4,a4,scale);
c6 = __fxcpmadd(c6,a6,scale);
__stfpd(&c[i ],c0);
__stfpd(&c[i+2],c2);
__stfpd(&c[i+4],c4);
__stfpd(&c[i+6],c6);
}
}
// TIMING
for (i=0;i<dim;i++) c[i] = 0.0;
t0 = getticks();
for (j=0;j<count;j++)
{
for (i=0;i<dim;i+=8)
{
{
double _Complex a0, a2, a4, a6;
double _Complex c0, c2, c4, c6;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+2]);
a4 = __lfpd(&a[i+4]);
a6 = __lfpd(&a[i+6]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+2]);
c4 = __lfpd(&c[i+4]);
c6 = __lfpd(&c[i+6]);
c0 = __fxcpmadd(c0,a0,scale);
c2 = __fxcpmadd(c2,a2,scale);
c4 = __fxcpmadd(c4,a4,scale);
c6 = __fxcpmadd(c6,a6,scale);
__stfpd(&c[i ],c0);
__stfpd(&c[i+2],c2);
__stfpd(&c[i+4],c4);
__stfpd(&c[i+6],c6);
}
}
}
t1 = getticks();
dt3 = t1 - t0;
// VERIFICATION
for (i=0;i<dim;i++)
{
if (b[i] != c[i])
{
printf("%4d %30.15f %30.15f\n",i,b[i],c[i]);
}
}
fprintf(stderr,"INTRINSICS VERSION 4\n");
// WARM-UP
for (i=0;i<dim;i++) c[i] = 0.0;
for (i=0;i<dim;i+=16)
{
{
double _Complex a0, a2, a4, a6, a8, a10, a12, a14;
double _Complex c0, c2, c4, c6, c8, c10, c12, c14;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+ 2]);
a4 = __lfpd(&a[i+ 4]);
a6 = __lfpd(&a[i+ 6]);
a4 = __lfpd(&a[i+ 8]);
a6 = __lfpd(&a[i+10]);
a4 = __lfpd(&a[i+12]);
a6 = __lfpd(&a[i+14]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+ 2]);
c4 = __lfpd(&c[i+ 4]);
c6 = __lfpd(&c[i+ 6]);
c4 = __lfpd(&c[i+ 8]);
c6 = __lfpd(&c[i+10]);
c4 = __lfpd(&c[i+12]);
c6 = __lfpd(&c[i+14]);
c0 = __fxcpmadd( c0, a0,scale);
c2 = __fxcpmadd( c2, a2,scale);
c4 = __fxcpmadd( c4, a4,scale);
c6 = __fxcpmadd( c6, a6,scale);
c4 = __fxcpmadd( c8, a8,scale);
c6 = __fxcpmadd(c10,a10,scale);
c4 = __fxcpmadd(c12,a12,scale);
c6 = __fxcpmadd(c14,a14,scale);
__stfpd(&c[i ],c0);
__stfpd(&c[i+ 2],c2);
__stfpd(&c[i+ 4],c4);
__stfpd(&c[i+ 6],c6);
__stfpd(&c[i+ 8],c4);
__stfpd(&c[i+10],c6);
__stfpd(&c[i+12],c4);
__stfpd(&c[i+14],c6);
}
}
// TIMING
for (i=0;i<dim;i++) c[i] = 0.0;
t0 = getticks();
for (j=0;j<count;j++)
{
for (i=0;i<dim;i+=16)
{
{
double _Complex a0, a2, a4, a6, a8, a10, a12, a14;
double _Complex c0, c2, c4, c6, c8, c10, c12, c14;
a0 = __lfpd(&a[i ]);
a2 = __lfpd(&a[i+ 2]);
a4 = __lfpd(&a[i+ 4]);
a6 = __lfpd(&a[i+ 6]);
a8 = __lfpd(&a[i+ 8]);
a10 = __lfpd(&a[i+10]);
a12 = __lfpd(&a[i+12]);
a14 = __lfpd(&a[i+14]);
c0 = __lfpd(&c[i ]);
c2 = __lfpd(&c[i+ 2]);
c4 = __lfpd(&c[i+ 4]);
c6 = __lfpd(&c[i+ 6]);
c8 = __lfpd(&c[i+ 8]);
c10 = __lfpd(&c[i+10]);
c12 = __lfpd(&c[i+12]);
c14 = __lfpd(&c[i+14]);
c0 = __fxcpmadd( c0, a0,scale);
c2 = __fxcpmadd( c2, a2,scale);
c4 = __fxcpmadd( c4, a4,scale);
c6 = __fxcpmadd( c6, a6,scale);
c8 = __fxcpmadd( c8, a8,scale);
c10 = __fxcpmadd(c10,a10,scale);
c12 = __fxcpmadd(c12,a12,scale);
c14 = __fxcpmadd(c14,a14,scale);
__stfpd(&c[i ], c0);
__stfpd(&c[i+ 2], c2);
__stfpd(&c[i+ 4], c4);
__stfpd(&c[i+ 6], c6);
__stfpd(&c[i+ 8], c8);
__stfpd(&c[i+10],c10);
__stfpd(&c[i+12],c12);
__stfpd(&c[i+14],c14);
}
}
}
t1 = getticks();
dt4 = t1 - t0;
// VERIFICATION
for (i=0;i<dim;i++)
{
if (b[i] != c[i])
{
printf("%4d %30.15f %30.15f\n",i,b[i],c[i]);
}
}
printf("%18d %18llu %18llu %18llu %18llu %18llu\n",dim,dt0,dt1,dt2,dt3,dt4);
free(a);
free(b);
free(c);
}
fprintf(stderr,"ALL DONE\n");
return(0);
}
