Skip to content
This repository
tree: e0f82981e7
Fetching contributors…

Octocat-spinner-32-eaf2f5

Cannot retrieve contributors at this time

file 293 lines (232 sloc) 13.654 kb
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292
/*
 * Copyright (c) 1998-1999 Stephen Williams (steve@icarus.com)
 *
 * This source code is free software; you can redistribute it
 * and/or modify it in source code form under the terms of the GNU
 * General Public License as published by the Free Software
 * Foundation; either version 2 of the License, or (at your option)
 * any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 */


Note that the netlist.h header contains detailed descriptions of how
things work. This is just an overview.

NETLIST FORMAT

The output from the parse and elaboration steps is a "netlist" rooted
in a Design object. Parsing translates the design described in the
initial source file into a temporary symbolic "pform". Elaboration
then expands the design, resolving references and expanding
hierarchies, to produce a flattened netlist. This is the form that
optimizers and code generators use.

The design optimization processes all manipulate the netlist,
translating it to a (hopefully) better netlist after each step. The
complete netlist is then passed to the code generator, the emit
function, where the final code (in the target format) is produced.

STRUCTURAL ITEMS: NetNode and NetNet

Components and wires, memories and registers all at their base are
either NetNode objects or NetNet objects. Even these classes are
derived from the NetObj class.

All NetNode and NetNet objects have a name and some number of
pins. The name usually comes from the Verilog source that represents
that object, although objects that are artifacts of elaboration will
have a generated (and probably unreadable) name. The pins are the
ports into the device. NetNode objects have a pin for each pin of the
component it represents, and NetNet objects have a pin for each signal
in the vector.

Node and net pins can be connected together via the connect
function. Connections are transitive (A==B and B==c means A==C) so
connections accumulate on a link as items are connected to it. The
destructors for nets and nodes automatically arrange for pins to be
disconnected when the item is deleted, so that the netlist can be
changed during processing.

STRUCTURAL LINKS

The NetNode and NetNet classes contain arrays of Link objects, one
object per pin. Each pin is a single bit. The Link objects link to all
the NetNode and NetNet objects' links that are connected together in
the design, and to a Nexus object. This way, code that examines a node
of the design can discover what is connected to each pin.

The connected set of links also has common properties that are stored
or access from the Nexus object. All the Links that are connected
together are also connected to a single Nexus object. This object is
useful for accessing the properties and values that come from the
connected set of links. The Nexus object is also handy for iterating
over the connected set of Links.

See the Link class definition in netlist.h for a description of the link
methods, and the Nexus class for nexus global methods.

Currently, a link has 3 possible direction properties:

PASSIVE -- These pins are sampled by the object that holds the
pin based on some external event. These are used,
for example, by NetESignal objects that read a
point for a procedural expression.

INPUT -- These pins potentially react to the setting of its
input.

OUTPUT -- These pins potentially drive the node. (They may be
three-state.)


BEHAVIORAL ITEMS: NetProcTop, NetProc and derived classes

Behavioral items are not in general linked to the netlist. Instead,
they represent elaborated behavioral statements. The type of the object
implies what the behavior of the statement does. For example, a
NetCondit object represents an ``if'' statement, and carries a
condition expression and up to two alternative sub-statements.

At the root of a process is a NetProcTop object. This class carries a
type flag (initial or always) and a single NetProc object. The
contained statement may, depending on the derived class, refer to
other statements, compound statements, so on. But at the root of the
tree is the NetProcTop object. The Design class keeps a list of the
elaborated NetProcTop objects. That list represents the list of
processes in the design.

INTERACTION OF BEHAVIORAL AND STRUCTURAL: NetAssign_

The behavioral statements in a Verilog design effect the structural
aspects through assignments to registers. Registers are structural
items represented by the NetNet class, linked to the assignment
statement through pins. This implies that the l-value of an assignment
is structural. It also implies that the statement itself is
structural, and indeed it is derived from NetNode.

The NetAssign_ class is also derived from the NetProc class because
what it does is brought on by executing the process. By multiple
inheritance we have therefore that the assignment is both a NetNode
and a NetProc. The NetAssign_ node has pins that represent the l-value
of the statement, and carries behavioral expressions that represent
the r-value of the assignment.

MEMORIES

The netlist form includes the NetMemory type to hold the content of a
memory. Instances of this type represent the declaration of a memory,
and occur once for each memory. References to the memory are managed
by the NetEMemory and NetAssignMem_ classes.

An instance of the NetEMemory class is created whenever a procedural
expression references a memory element. The operand is the index to
use to address (and read) the memory.

An instance of the NetAssignMem_ class is created when there is a
procedural assignment to the memory. The NetAssignMem_ object
represents the l-value reference (a write) to the memory. As with the
NetEMemory class, this is a procedural reference only.

When a memory reference appears in structural context (i.e. continuous
assignments) elaboration creates a NetRamDq. This is a LPM_RAM_DQ
device. Elaboration leaves the write control and data input pins
unconnected for now, because memories cannot appear is l-values of
continuous assignments. However, the synthesis functor may connect
signals to the write control lines to get a fully operational RAM.

By the time elaboration completes, there may be many NetAssignMem_,
NetEMemory and NetRamDq objects referencing the same NetMemory
object. Each represents a port into the memory. It is up to the
synthesis steps (and the target code) to figure out what to do with
these ports.

EXPRESSIONS

Expressions are represented as a tree of NetExpr nodes. The NetExpr
base class contains the core methods that represent an expression
node, including virtual methods to help with dealing with nested
complexities of expressions.

Expressions (as expressed in the source and p-form) may also be
elaborated structurally, where it makes sense. For example, assignment
l-value expressions are represented as connections to pins. Also,
continuous assignment module items are elaborated as gates instead of
as a procedural expression. Event expressions are also elaborated
structurally as events are like devices that trigger behavioral
statements.

However, typical expressions the behavioral description are
represented as a tree of NetExpr nodes. The derived class of the node
encodes what kind of operator the node represents.

EXPRESSION BIT WIDTH

The expression (represented by the NetExpr class) has a bit width that
it either explicitly specified, or implied by context or contents.
When each node of the expression is first constructed during
elaboration, it is given, by type and parameters, an idea what its
width should be. It certain cases, this is definitive, for example
with signals. In others, it is ambiguous, as with unsized constants.

As the expression is built up by elaboration, operators that combine
expressions impose bit widths of the environment or expose the bit
widths of the sub expressions. For example, the bitwise AND (&)
operator has a bit size implied by its operands, whereas the
comparison (==) operator has a bit size of 1. The building up of the
elaborated expression checks and adjusts the bit widths as the
expression is built up, until finally the context of the expression
takes the final bit width and makes any final adjustments.

The NetExpr::expr_width() method returns the calculated (or guessed)
expression width. This method will return 0 until the width is set by
calculation or context. If this method returns false, then it is up to
the context that wants the width to set one. The elaboration phase
will call the NetExpr::set_width method on an expression as soon as it
gets to a point where it believes that it knows what the width should
be.

The NetExpr::set_width(unsigned) virtual method is used by the context
of an expression node to note to the expression that the width is
determined and please adapt. If the expression cannot reasonably
adapt, it will return false. Otherwise, it will adjust bit widths and
return true.

XXXX I do not yet properly deal with cases where elaboration knows for
XXXX certain that the bit width does not matter. In this case, I
XXXX really should tell the expression node about it so that it can
XXXX pick a practical (and optimal) width.

INTERACTION OF EXPRESSIONS AND STRUCTURE: NetESignal

The NetAssign_ class described above is the means for processes to
manipulate the net, but values are read from the net by NetESignal
objects. These objects are class NetExpr because they can appear in
expressions (and have width). They are not NetNode object, but hold
pointers to a NetNet object, which is used to retrieve values with the
expression is evaluated.


HIERARCHY IN NETLISTS

The obvious hierarchical structure of Verilog is the module. The
Verilog program may contain any number of instantiations of modules in
order to form an hierarchical design. However, the elaboration of the
design into a netlist erases module boundaries. Modules are expanded
each place they are used, with the hierarchical instance name used to
name the components of the module instance. However, the fact that a
wire or register is a module port is lost.

The advantage of this behavior is first the simplification of the
netlist structure itself. Backends that process netlists only need to
cope with a list of nets, a list of nodes and a list of
processes. This eases the task of the backend code generators.

Another advantage of this flattening of the netlist is that optimizers
can operate globally, with optimizations freely crossing module
boundaries. This makes coding of netlist transform functions such as
constant propagation more effective and easier to write.


SCOPE REPRESENTATION IN NETLISTS

In spite of the literal flattening of the design, scope information is
preserved in the netlist, with the NetScope class. The Design class
keeps a single pointer to the root scope of the design. This is the
scope of the root module. Scopes that are then created within that
(or any nested) module are placed as children of the root scope, and
those children can have further children, and so on.

Each scope in the tree carries its own name, and its relationship to
its parent and children. This makes it possible to walk the tree of
scopes. In practice, the walking of the scopes is handled by recursive
methods.

Each scope also carries the parameters that are applicable to the
scope itself. The parameter expression (possibly evaluated) can be
located by name, given the scope object itself. The scan of the pform
to generate scopes also places the parameters that are declared in the
scope. Overrides are managed during the scan, and once the scan is
complete, defparam overrides are applied.


TASKS IN NETLISTS

The flattening of the design does not include tasks and named
begin-end blocks. Tasks are behavioral hierarchy (whereas modules are
structural) so do not easily succumb to the flattening process. In
particular, it is logically impossible to flatten tasks that
recurse. (The elaboration process does reserve the right to flatten
some task calls. C++ programmers recognize this as inlining a task.)


TIME SCALE IN NETLISTS

The Design class and the NetScope classes carry time scale and
resolution information of the elaborated design. There is a global
resolution, and there are scope specific units and resolutions. Units
and resolutions are specified as signed integers, and interpreted as
the power of 10 of the value. For example, a resolution "-9" means
that "1" is 1ns (1e-9). The notation supports units from -128 to +127.
It is up to the back-ends to interpret "-4" as "100us".

Delays are expressed in the netlist by integers. The units of these
delays are always given in the units of the design precision. This
allows everything to work with integers, and generally places the
burden of scaling delays into elaboration. This is, after all, a
common task. The Design::get_precision() method gets the global design
precision.

Each NetScope also carries its local time_units and time_precision
values. These are filled in during scope elaboration and are used in
subsequent elaboration phases to arrange for scaling of delays. This
information can also be used by the code generator to scale times back
to the units of the scope, if that is desired.
Something went wrong with that request. Please try again.