CSB/Semirings.h at 6f7e8a146594375ee022142fc52aa694e31e9398 · PASSIONLab/CSB · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130

#ifndef _SEMIRINGS_H_
#define _SEMIRINGS_H_

#include <utility>
#include <climits>
#include <cmath>
#include <tr1/array>
#include <memory>
#include "promote.h"

template <typename T>
struct inf_plus{
  T operator()(const T& a, const T& b) const {
	T inf = std::numeric_limits<T>::max();
    	if (a == inf || b == inf){
      		return inf;
    	}
    	return a + b;
  }
};

// (+,*) on scalars
template <class T1, class T2>
struct PTSR
{
	typedef typename promote_trait<T1,T2>::T_promote T_promote;

	static T_promote add(const T1 & arg1, const T2 & arg2)
	{
		return (static_cast<T_promote>(arg1) +
			static_cast<T_promote>(arg2) );
	}
	static T_promote multiply(const T1 & arg1, const T2 & arg2)
	{
		return (static_cast<T_promote>(arg1) *
			static_cast<T_promote>(arg2) );
	}
	// y += ax overload with a=1
	static void axpy(const T2 & x, T_promote & y)
	{
		y += x;
	}

	static void axpy(T1 a, const T2 & x, T_promote & y)
	{
		y += a*x;
	}
};


template<int Begin, int End, int Step>
struct UnrollerL {
    template<typename Lambda>
    static void step(Lambda& func) {
        func(Begin);
        UnrollerL<Begin+Step, End, Step>::step(func);
    }
};

template<int End, int Step>
struct UnrollerL<End, End, Step> {
    template<typename Lambda>
    [[maybe_unused]] static void step([[maybe_unused]] Lambda& func) {
		// base case is when Begin=End; do nothing
    }
};


// (+,*) on std:array's
template<class T1, class T2, unsigned D>
struct PTSRArray
{
	typedef typename promote_trait<T1,T2>::T_promote T_promote;

	// y <- a*x + y overload with a=1
	static void axpy(const array<T2, D> & b, array<T_promote, D> & c)
	{
		// const T2 * __restrict barr =  std::assume_aligned<ALIGN>(b.data());
		// T_promote * __restrict carr = std::assume_aligned<ALIGN>(c.data());
		const T2 * __restrict barr =  b.data();
		T_promote * __restrict carr = c.data();

		#pragma simd
		for(unsigned int i=0; i<D; ++i)
		{
			carr[i] +=  barr[i];
		}
		// auto multadd = [&] (int i) { c[i] +=  b[i]; };
		// UnrollerL<0, D, 1>::step ( multadd );
	}

	// Todo: Do partial unrolling; this code will bloat for D > 32
	static void axpy(T1 a, const array<T2,D> & b, array<T_promote,D> & c)
	{
		// const T2 * __restrict barr =  std::assume_aligned<ALIGN>(b.data());
		// T_promote * __restrict carr = std::assume_aligned<ALIGN>(c.data());
		const T2 * __restrict barr =  b.data();
		T_promote * __restrict carr = c.data();

		#pragma simd
		for(unsigned int i=0; i<D; ++i)
		{
			carr[i] +=  a* barr[i];
		}
		//auto multadd = [&] (int i) { carr[i] +=  a* barr[i]; };
		//UnrollerL<0, D, 1>::step ( multadd );
	}
};

// (min,+) on scalars
template <class T1, class T2>
struct MPSR
{
	typedef typename promote_trait<T1,T2>::T_promote T_promote;

	static T_promote add(const T1 & arg1, const T2 & arg2)
	{
		return std::min<T_promote>
		(static_cast<T_promote>(arg1), static_cast<T_promote>(arg2));
	}
	static T_promote multiply(const T1 & arg1, const T2 & arg2)
	{
		return inf_plus< T_promote >
		(static_cast<T_promote>(arg1), static_cast<T_promote>(arg2));
	}
};


#endif