summaryrefslogtreecommitdiff
path: root/Parser/acceler.c
blob: 6ed12c3074235bb6d298cad5a1b7b536aab177b1 (plain)
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

/* Parser accelerator module */

/* The parser as originally conceived had disappointing performance.
   This module does some precomputation that speeds up the selection
   of a DFA based upon a token, turning a search through an array
   into a simple indexing operation.  The parser now cannot work
   without the accelerators installed.  Note that the accelerators
   are installed dynamically when the parser is initialized, they
   are not part of the static data structure written on graminit.[ch]
   by the parser generator. */

#include "pgenheaders.h"
#include "grammar.h"
#include "node.h"
#include "token.h"
#include "parser.h"

/* Forward references */
static void fixdfa(grammar *, dfa *);
static void fixstate(grammar *, state *);

void
PyGrammar_AddAccelerators(grammar *g)
{
	dfa *d;
	int i;
	d = g->g_dfa;
	for (i = g->g_ndfas; --i >= 0; d++)
		fixdfa(g, d);
	g->g_accel = 1;
}

void
PyGrammar_RemoveAccelerators(grammar *g)
{
	dfa *d;
	int i;
	g->g_accel = 0;
	d = g->g_dfa;
	for (i = g->g_ndfas; --i >= 0; d++) {
		state *s;
		int j;
		s = d->d_state;
		for (j = 0; j < d->d_nstates; j++, s++) {
			if (s->s_accel)
				PyMem_DEL(s->s_accel);
			s->s_accel = NULL;
		}
	}
}

static void
fixdfa(grammar *g, dfa *d)
{
	state *s;
	int j;
	s = d->d_state;
	for (j = 0; j < d->d_nstates; j++, s++)
		fixstate(g, s);
}

static void
fixstate(grammar *g, state *s)
{
	arc *a;
	int k;
	int *accel;
	int nl = g->g_ll.ll_nlabels;
	s->s_accept = 0;
	accel = PyMem_NEW(int, nl);
	for (k = 0; k < nl; k++)
		accel[k] = -1;
	a = s->s_arc;
	for (k = s->s_narcs; --k >= 0; a++) {
		int lbl = a->a_lbl;
		label *l = &g->g_ll.ll_label[lbl];
		int type = l->lb_type;
		if (a->a_arrow >= (1 << 7)) {
			printf("XXX too many states!\n");
			continue;
		}
		if (ISNONTERMINAL(type)) {
			dfa *d1 = PyGrammar_FindDFA(g, type);
			int ibit;
			if (type - NT_OFFSET >= (1 << 7)) {
				printf("XXX too high nonterminal number!\n");
				continue;
			}
			for (ibit = 0; ibit < g->g_ll.ll_nlabels; ibit++) {
				if (testbit(d1->d_first, ibit)) {
#ifdef applec
#define MPW_881_BUG			/* Undefine if bug below is fixed */
#endif
#ifdef MPW_881_BUG
					/* In 881 mode MPW 3.1 has a code
					   generation bug which seems to
					   set the upper bits; fix this by
					   explicitly masking them off */
					int temp;
#endif
					if (accel[ibit] != -1)
						printf("XXX ambiguity!\n");
#ifdef MPW_881_BUG
					temp = 0xFFFF &
						(a->a_arrow | (1 << 7) |
						 ((type - NT_OFFSET) << 8));
					accel[ibit] = temp;
#else
					accel[ibit] = a->a_arrow | (1 << 7) |
						((type - NT_OFFSET) << 8);
#endif
				}
			}
		}
		else if (lbl == EMPTY)
			s->s_accept = 1;
		else if (lbl >= 0 && lbl < nl)
			accel[lbl] = a->a_arrow;
	}
	while (nl > 0 && accel[nl-1] == -1)
		nl--;
	for (k = 0; k < nl && accel[k] == -1;)
		k++;
	if (k < nl) {
		int i;
		s->s_accel = PyMem_NEW(int, nl-k);
		if (s->s_accel == NULL) {
			fprintf(stderr, "no mem to add parser accelerators\n");
			exit(1);
		}
		s->s_lower = k;
		s->s_upper = nl;
		for (i = 0; k < nl; i++, k++)
			s->s_accel[i] = accel[k];
	}
	PyMem_DEL(accel);
}