void profile_for_type(const T& elem, const std::string& type,
const int add_flops, const int multiply_flops)
{
std::cout << "Profiling for array type " << type << "." << std::endl;
unsigned int n = 100;
pyQCD::LexicoLayout layout(std::vector<unsigned int>{n});
pyQCD::Lattice<T> lattice1(layout, elem);
decltype(lattice1) lattice2(layout, elem);
decltype(lattice1) lattice3(layout, elem);
decltype(lattice1) result(layout, elem);
std::cout << "Profiling f(x, y, z) = x + y + z:" << std::endl;
benchmark([&] () {
result = lattice1 + lattice2 + lattice3;
}, 2 * add_flops * n, 1000000);
std::cout << "Profiling f(x, y) = 5.0 * x + y:" << std::endl;
benchmark([&] () {
result = 5.0 * lattice1 + lattice2;
}, 2 * add_flops * n, 1000000);
std::cout << "Profiling f(x, y, z) = x * y + z:" << std::endl;
benchmark([&] () {
result = lattice1 * lattice2 + lattice3;
}, (add_flops + multiply_flops) * n, 1000000);
std::cout << std::endl;
}
float noise_gen(const struct noise *noise, float *f)
{
int n[MAX_DIMENSIONS];
float r[MAX_DIMENSIONS];
float w[MAX_DIMENSIONS];
for(int i = 0; i < noise->ndim; i++) {
n[i] = floor(f[i]);
r[i] = f[i] - n[i];
w[i] = cubic(r[i]);
}
float value;
switch(noise->ndim) {
case 1:
value = lerp(lattice1(noise, n[0] , r[0] ),
lattice1(noise, n[0]+1, r[0]+1),
w[0]);
break;
case 2:
value = lerp(lerp(lattice2(noise, n[0], r[0], n[1], r[1]),
lattice2(noise, n[0]+1, r[0]-1, n[1], r[1]),
w[0]),
lerp(lattice2(noise, n[0], r[0], n[1]+1, r[1]-1),
lattice2(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1),
w[0]),
w[1]);
break;
case 3:
//printf("noise3 start\n");
value = lerp(lerp(lerp(lattice3(noise, n[0] , r[0] , n[1], r[1], n[2], r[2]),
lattice3(noise, n[0]+1, r[0]-1, n[1], r[1], n[2], r[2]),
w[0]),
lerp(lattice3(noise, n[0] , r[0] , n[1]+1, r[1]-1, n[2], r[2]),
lattice3(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2], r[2]),
w[0]),
w[1]),
lerp(lerp(lattice3(noise, n[0] , r[0] , n[1], r[1], n[2]+1, r[2]-1),
lattice3(noise, n[0]+1, r[0]-1, n[1], r[1], n[2]+1, r[2]-1),
w[0]),
lerp(lattice3(noise, n[0] , r[0] , n[1]+1, r[1]-1, n[2]+1, r[2]-1),
lattice3(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2]+1, r[2]-1),
w[0]),
w[1]),
w[2]);
//printf("noise3=%g\n", value);
break;
case 4:
value = lerp(lerp(lerp(lerp(lattice4(noise, n[0], r[0], n[1], r[1], n[2], r[2], n[3], r[3]),
lattice4(noise, n[0]+1, r[0]-1, n[1], r[1], n[2], r[2], n[3], r[3]),
w[0]),
lerp(lattice4(noise, n[0], r[0], n[1]+1, r[1]-1, n[2], r[2], n[3], r[3]),
lattice4(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2], r[2], n[3], r[3]),
w[0]),
w[1]),
lerp(lerp(lattice4(noise, n[0], r[0], n[1], r[1], n[2]+1, r[2]-1, n[3], r[3]),
lattice4(noise, n[0]+1, r[0]-1, n[1], r[1], n[2]+1, r[2]-1, n[3], r[3]),
w[0]),
lerp(lattice4(noise, n[0], r[0], n[1]+1, r[1]-1, n[2]+1, r[2]-1, n[3], r[3]),
lattice4(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2]+1, r[2]-1, n[3], r[3]),
w[0]),
w[1]),
w[2]),
lerp(lerp(lerp(lattice4(noise, n[0], r[0], n[1], r[1], n[2], r[2], n[3]+1, r[3]-1),
lattice4(noise, n[0]+1, r[0]-1, n[1], r[1], n[2], r[2], n[3]+1, r[3]-1),
w[0]),
lerp(lattice4(noise, n[0], r[0], n[1]+1, r[1]-1, n[2], r[2], n[3]+1, r[3]-1),
lattice4(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2], r[2], n[3]+1, r[3]-1),
w[0]),
w[1]),
lerp(lerp(lattice4(noise, n[0], r[0], n[1], r[1], n[2]+1, r[2]-1, n[3]+1, r[3]-1),
lattice4(noise, n[0]+1, r[0]-1, n[1], r[1], n[2]+1, r[2]-1, n[3]+1, r[3]-1),
w[0]),
lerp(lattice4(noise, n[0], r[0], n[1]+1, r[1]-1, n[2]+1, r[2]-1, n[3]+1, r[3]-1),
lattice4(noise, n[0]+1, r[0]-1, n[1]+1, r[1]-1, n[2]+1, r[2]-1, n[3]+1, r[3]-1),
w[0]),
w[1]),
w[2]),
w[3]);
break;
default:
abort();
}
return clamp(-0.99999f, 0.99999f, value);
}
