summaryrefslogtreecommitdiffstatshomepage
path: root/src/lib/netlist/solver/nld_ms_sor.h
blob: b7012f5532d820d6e0346bdef616c59a18d7d23a (plain) (blame)
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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
// license:GPL-2.0+
// copyright-holders:Couriersud
/*
 * nld_ms_sor.h
 *
 * Generic successive over relaxation solver.
 *
 * Fow w==1 we will do the classic Gauss-Seidel approach
 *
 */

#ifndef NLD_MS_SOR_H_
#define NLD_MS_SOR_H_

#include <algorithm>

#include "solver/nld_ms_direct.h"
#include "solver/nld_solver.h"

namespace netlist
{
	namespace devices
	{
template <unsigned m_N, unsigned storage_N>
class matrix_solver_SOR_t: public matrix_solver_direct_t<m_N, storage_N>
{
public:

	matrix_solver_SOR_t(netlist_t &anetlist, const pstring &name, const solver_parameters_t *params, int size)
		: matrix_solver_direct_t<m_N, storage_N>(anetlist, name, matrix_solver_t::ASCENDING, params, size)
		, m_lp_fact(*this, "m_lp_fact", 0)
		{
		}

	virtual ~matrix_solver_SOR_t() {}

	virtual void vsetup(analog_net_t::list_t &nets) override;
	virtual int vsolve_non_dynamic(const bool newton_raphson) override;

private:
	state_var<nl_double> m_lp_fact;
};

// ----------------------------------------------------------------------------------------
// matrix_solver - Gauss - Seidel
// ----------------------------------------------------------------------------------------


template <unsigned m_N, unsigned storage_N>
void matrix_solver_SOR_t<m_N, storage_N>::vsetup(analog_net_t::list_t &nets)
{
	matrix_solver_direct_t<m_N, storage_N>::vsetup(nets);
}

template <unsigned m_N, unsigned storage_N>
int matrix_solver_SOR_t<m_N, storage_N>::vsolve_non_dynamic(const bool newton_raphson)
{
	const unsigned iN = this->N();
	bool resched = false;
	int  resched_cnt = 0;

	/* ideally, we could get an estimate for the spectral radius of
	 * Inv(D - L) * U
	 *
	 * and estimate using
	 *
	 * omega = 2.0 / (1.0 + std::sqrt(1-rho))
	 */

	const nl_double ws = this->m_params.m_sor;

	nl_double w[storage_N];
	nl_double one_m_w[storage_N];
	nl_double RHS[storage_N];
	nl_double new_V[storage_N];

	for (unsigned k = 0; k < iN; k++)
	{
		nl_double gtot_t = 0.0;
		nl_double gabs_t = 0.0;
		nl_double RHS_t = 0.0;

		const unsigned term_count = this->m_terms[k]->count();
		const nl_double * const RESTRICT gt = this->m_terms[k]->gt();
		const nl_double * const RESTRICT go = this->m_terms[k]->go();
		const nl_double * const RESTRICT Idr = this->m_terms[k]->Idr();
		const nl_double * const *other_cur_analog = this->m_terms[k]->other_curanalog();

		new_V[k] = this->m_nets[k]->m_cur_Analog;

		for (unsigned i = 0; i < term_count; i++)
		{
			gtot_t = gtot_t + gt[i];
			RHS_t = RHS_t + Idr[i];
		}

		for (unsigned i = this->m_terms[k]->m_railstart; i < term_count; i++)
			RHS_t = RHS_t  + go[i] * *other_cur_analog[i];

		RHS[k] = RHS_t;

		if (USE_GABS)
		{
			for (unsigned i = 0; i < term_count; i++)
				gabs_t = gabs_t + std::abs(go[i]);

			gabs_t *= NL_FCONST(0.5); // derived by try and error
			if (gabs_t <= gtot_t)
			{
				w[k] = ws / gtot_t;
				one_m_w[k] = NL_FCONST(1.0) - ws;
			}
			else
			{
				w[k] = NL_FCONST(1.0) / (gtot_t + gabs_t);
				one_m_w[k] = NL_FCONST(1.0) - NL_FCONST(1.0) * gtot_t / (gtot_t + gabs_t);
			}
		}
		else
		{
			w[k] = ws / gtot_t;
			one_m_w[k] = NL_FCONST(1.0) - ws;
		}
	}

	const nl_double accuracy = this->m_params.m_accuracy;

	/* uncommenting the line below will force dynamic updates every X iterations
	 * althought the system has not converged yet. This is a proof of concept,
	 *
	 */
	const bool interleaved_dynamic_updates = false;
	//const bool interleaved_dynamic_updates = newton_raphson;

	do {
		resched = false;
		nl_double err = 0;
		for (unsigned k = 0; k < iN; k++)
		{
			const int * RESTRICT net_other = this->m_terms[k]->net_other();
			const unsigned railstart = this->m_terms[k]->m_railstart;
			const nl_double * RESTRICT go = this->m_terms[k]->go();

			nl_double Idrive = 0.0;
			for (unsigned i = 0; i < railstart; i++)
				Idrive = Idrive + go[i] * new_V[net_other[i]];

			const nl_double new_val = new_V[k] * one_m_w[k] + (Idrive + RHS[k]) * w[k];

			err = std::max(std::abs(new_val - new_V[k]), err);
			new_V[k] = new_val;
		}

		if (err > accuracy)
			resched = true;

		resched_cnt++;
	//} while (resched && (resched_cnt < this->m_params.m_gs_loops));
	} while (resched && ((!interleaved_dynamic_updates && resched_cnt < this->m_params.m_gs_loops) || (interleaved_dynamic_updates && resched_cnt < 5 )));

	this->m_iterative_total += resched_cnt;

	if (resched && !interleaved_dynamic_updates)
	{
		// Fallback to direct solver ...
		this->m_iterative_fail++;
		return matrix_solver_direct_t<m_N, storage_N>::vsolve_non_dynamic(newton_raphson);
	}

	this->m_stat_calculations++;

	if (interleaved_dynamic_updates)
	{
		for (unsigned k = 0; k < iN; k++)
			this->m_nets[k]->m_cur_Analog += 1.0 * (new_V[k] - this->m_nets[k]->m_cur_Analog);
	}
	else
	{
		for (unsigned k = 0; k < iN; k++)
			this->m_nets[k]->m_cur_Analog = new_V[k];
	}

	return resched_cnt;
}

	} //namespace devices
} // namespace netlist

#endif /* NLD_MS_SOR_H_ */