Last active
August 29, 2015 13:57
-
-
Save alexshen/9781052 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
#include <xmmintrin.h> | |
#include <chrono> | |
#include <iostream> | |
using namespace std; | |
using namespace chrono; | |
#ifdef _MSC_VER | |
__declspec(align(16)) struct vector4 | |
#else | |
struct alignas(16) vector4 | |
#endif | |
{ | |
float v[4]; | |
}; | |
template<int N> | |
struct dot_unroller; | |
template<> | |
struct dot_unroller<1> | |
{ | |
static float eval(float const* v1, float const* v2) | |
{ | |
return *v1 * *v2; | |
} | |
}; | |
template<int N> | |
struct dot_unroller | |
{ | |
static float eval(float const* v1, float const* v2) | |
{ | |
return dot_unroller<1>::eval(v1, v2) + dot_unroller<N - 1>::eval(v1 + 1, v2 + 1); | |
} | |
}; | |
float dot_std(vector4 const& v1, vector4 const& v2) | |
{ | |
return dot_unroller<4>::eval(v1.v, v2.v); | |
} | |
float dot_unroll(vector4 const& v1, vector4 const& v2) | |
{ | |
return v1.v[0] * v2.v[0] + | |
v1.v[1] * v2.v[1] + | |
v1.v[2] * v2.v[2] + | |
v1.v[3] * v2.v[3]; | |
} | |
float dot_sse(vector4 const& v1, vector4 const& v2) | |
{ | |
#ifdef _MSC_VER | |
__declspec(align(16)) float r[4]; | |
#else | |
alignas(16) float r[4]; | |
#endif | |
__m128 a, b; | |
a = _mm_load_ps(v1.v); | |
b = _mm_load_ps(v2.v); | |
_mm_store_ps(r, _mm_mul_ps(a, b)); | |
return r[0] + r[1] + r[2] + r[3]; | |
} | |
struct timer | |
{ | |
char const* name; | |
high_resolution_clock::time_point start; | |
timer(char const* p) | |
: name{ p } | |
, start{ high_resolution_clock::now() } | |
{ | |
} | |
~timer() | |
{ | |
auto duration = high_resolution_clock::now() - start; | |
cout << name << ": " << duration_cast<nanoseconds>(duration).count() << endl; | |
} | |
}; | |
template<typename F> | |
void time_it(char const* name, F f, int count) | |
{ | |
float res; | |
{ | |
timer t{ name }; | |
vector4 a = { 1, 2, 3, 4 }; | |
for (int i = 0; i < count; ++i) | |
{ | |
res = f(a, a); | |
} | |
} | |
cout << res << endl; | |
} | |
int main() | |
{ | |
int const count = 1000000; | |
for (int i = 0; i < 3; ++i) | |
{ | |
time_it("dot_std", dot_std, count); | |
time_it("dot_unroll", dot_unroll, count); | |
time_it("dot_sse", dot_sse, count); | |
} | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment