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
|
/**CFile****************************************************************
FileName [fpgaCore.c]
PackageName [MVSIS 1.3: Multi-valued logic synthesis system.]
Synopsis [Technology mapping for variable-size-LUT FPGAs.]
Author [MVSIS Group]
Affiliation [UC Berkeley]
Date [Ver. 2.0. Started - August 18, 2004.]
Revision [$Id: fpgaCore.c,v 1.7 2004/10/01 23:41:04 satrajit Exp $]
***********************************************************************/
#include "fpgaInt.h"
////////////////////////////////////////////////////////////////////////
/// DECLARATIONS ///
////////////////////////////////////////////////////////////////////////
static int Fpga_MappingPostProcess( Fpga_Man_t * p );
////////////////////////////////////////////////////////////////////////
/// FUNCTION DEFITIONS ///
////////////////////////////////////////////////////////////////////////
/**Function*************************************************************
Synopsis [Performs technology mapping for the given object graph.]
Description [The object graph is stored in the mapping manager.
First, all the AND-nodes, which fanout into the POs, are collected
in the DFS fashion. Next, three steps are performed: the k-feasible
cuts are computed for each node, the truth tables are computed for
each cut, and the delay-optimal matches are assigned for each node.]
SideEffects []
SeeAlso []
***********************************************************************/
int Fpga_Mapping( Fpga_Man_t * p )
{
int clk, clkTotal = clock();
// collect the nodes reachable from POs in the DFS order (including the choices)
p->vAnds = Fpga_MappingDfs( p, 1 );
Fpga_ManReportChoices( p ); // recomputes levels
Fpga_MappingSetChoiceLevels( p );
// compute the cuts of nodes in the DFS order
clk = clock();
Fpga_MappingCuts( p );
p->timeCuts = clock() - clk;
// match the truth tables to the supergates
clk = clock();
if ( !Fpga_MappingMatches( p, 1 ) )
return 0;
p->timeMatch = clock() - clk;
// perform area recovery
if ( p->fAreaRecovery )
{
clk = clock();
if ( !Fpga_MappingPostProcess( p ) )
return 0;
p->timeRecover = clock() - clk;
}
PRT( "Total mapping time", clock() - clkTotal );
// print the AI-graph used for mapping
//Fpga_ManShow( p, "test" );
if ( p->fVerbose )
Fpga_MappingPrintOutputArrivals( p );
return 1;
}
/**Function*************************************************************
Synopsis [Postprocesses the mapped network for area recovery.]
Description [This procedure assumes that the mapping is assigned.
It iterates the loop, in which the required times are computed and
the mapping is updated. It is conceptually similar to the paper:
V. Manohararajah, S. D. Brown, Z. G. Vranesic, Heuristics for area
minimization in LUT-based FGPA technology mapping. Proc. IWLS '04.]
SideEffects []
SeeAlso []
***********************************************************************/
int Fpga_MappingPostProcess( Fpga_Man_t * p )
{
float aAreaTotalPrev, aAreaTotalCur, aAreaTotalCur2;
int Iter, clk;
// compute area, set references, and collect nodes used in the mapping
aAreaTotalCur = Fpga_MappingSetRefsAndArea( p );
if ( p->fVerbose )
{
printf( "Iteration %dD : Area = %11.1f ", 0, aAreaTotalCur );
PRT( "Time", p->timeMatch );
}
Iter = 1;
do {
clk = clock();
// save the previous area flow
aAreaTotalPrev = aAreaTotalCur;
// compute the required times and the fanouts
Fpga_TimeComputeRequiredGlobal( p );
// remap topologically
Fpga_MappingMatches( p, 0 );
// get the resulting area
// aAreaTotalCur = Fpga_MappingSetRefsAndArea( p );
aAreaTotalCur = Fpga_MappingAreaTrav( p );
// note that here we do not update the reference counter
// for some reason, this works better on benchmarks
if ( p->fVerbose )
{
printf( "Iteration %dF : Area = %11.1f ", Iter++, aAreaTotalCur );
PRT( "Time", clock() - clk );
}
// quit if this iteration reduced area flow by less than 1%
} while ( aAreaTotalPrev > 1.02 * aAreaTotalCur );
// update reference counters
aAreaTotalCur2 = Fpga_MappingSetRefsAndArea( p );
assert( aAreaTotalCur == aAreaTotalCur2 );
do {
clk = clock();
// save the previous area flow
aAreaTotalPrev = aAreaTotalCur;
// compute the required times and the fanouts
Fpga_TimeComputeRequiredGlobal( p );
// remap topologically
Fpga_MappingMatchesArea( p );
// get the resulting area
aAreaTotalCur = Fpga_MappingSetRefsAndArea( p );
if ( p->fVerbose )
{
printf( "Iteration %dA : Area = %11.1f ", Iter++, aAreaTotalCur );
PRT( "Time", clock() - clk );
}
// quit if this iteration reduced area flow by less than 1%
} while ( aAreaTotalPrev > 1.02 * aAreaTotalCur );
p->fAreaGlo = aAreaTotalCur;
return 1;
}
|
