paweld/fpc
1
0
Fork 0
mirror of https://gitlab.com/freepascal.org/fpc/source.git synced 2025-10-31 01:11:29 +01:00
Code Issues Packages Projects Releases Wiki Activity

* some spelling errors corrected

Browse Source
This commit is contained in:
Jonas Maebe 1998-09-17 21:51:22 +00:00
parent 1e3a448793
commit c4dd8d472a
1 changed files with 30 additions and 31 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

61
docs/internal.tex
View File

@ -54,7 +54,7 @@ building.
This document describes the compiler as it is/functions at the time of This document describes the compiler as it is/functions at the time of
writing. Since the compiler is under continuous development, some of the writing. Since the compiler is under continuous development, some of the
things described here may be outdated. In case of doubt, consult the things described here may be outdated. In case of doubt, consult the
\file{README} files, distributed with the compiler. \file{README} files distributed with the compiler.
The \file{README} files are, in case of conflict with this manual, The \file{README} files are, in case of conflict with this manual,
authoritative. authoritative.
@ -69,9 +69,9 @@ spelling mistakes.
% Getting more information. % Getting more information.
\section{Getting more information.} \section{Getting more information.}
The ultimative source for informations about compiler internals is The ultimate source for information about compiler internals is
the compiler source though it isn't very well documented. If you the compiler source, though it isn't very well documented. If you
need more infomrations you should join the developers mailing need more infomration you should join the developers mailing
list or you can contact the developers. list or you can contact the developers.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@ -89,18 +89,18 @@ list or you can contact the developers.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\chapter{The symbol tables} \chapter{The symbol tables}
The symbol table is used to store informations about all The symbol table is used to store information about all
symbols, declarations and definitions in a program. symbols, declarations and definitions in a program.
In an abstract view, a symbol table is a data base with a string field In an abstract view, a symbol table is a data base with a string field
as index. \fpc implements the symbol table mainly as a binary tree, as index. \fpc implements the symbol table mainly as a binary tree, but
for big symbol tables some hash technics are used. The implementation for big symbol tables some hash technics are used. The implementation
can be found in symtable.pas, object tsymtable. can be found in symtable.pas, object tsymtable.
The symbol table module can't be associated with a stage of the compiler, The symbol table module can't be associated with a stage of the compiler,
each stage does accesses to the symbol table. each stage accesses it.
The scanner uses a symbol table to handle preprocessor symbols, the The scanner uses a symbol table to handle preprocessor symbols, the
parser inserts declaration and the code generator uses the collected parser inserts declaration and the code generator uses the collected
informations about symbols and types to generate the code. information about symbols and types to generate the code.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Definitions % Definitions
@ -111,10 +111,10 @@ They are used to describe types, for example the type of a variable
symbol is given by a definition and the result type symbol is given by a definition and the result type
of a expression is given as a definition. of a expression is given as a definition.
They have nothing to do with the definition of a procedure. They have nothing to do with the definition of a procedure.
Definitions are implemented as a object (in file \file{symtable.pas}, Definitions are implemented as an object (in file \file{symtable.pas},
\var{tdef} and it's descendents). There are a lot of different \var{tdef} and its descendents). There are a lot of different
definitions: for example to describe definitions, for example to describe
ordinal types, arrays, pointers, procedures ordinal types, arrays, pointers, procedures, ...
To make it more clear let's have a look at the fields of tdef: To make it more clear let's have a look at the fields of tdef:
@ -177,26 +177,25 @@ The register allocation is very hairy, so it gets
an own chapter in this manual. Please be careful when changing things an own chapter in this manual. Please be careful when changing things
regarding the register allocation and test such changes intensive. regarding the register allocation and test such changes intensive.
Future versions will may be implement another kind of register allocation Future versions will may implement another kind of register allocation
to make this part of the compiler more robust, see to make this part of the compiler more robust, see
\ref{se:future_plans}. But the current \ref{se:future_plans}. But the current
system is less or more working and changing it would be a lot of system is less or more working and changing it would be a lot of
work, so we have to live with it. work, so we have to live with it.
The current register allocation mechanism was implemented 5 years The current register allocation mechanism was implemented 5 years
ago and I didn't think, that the compiler would become ago and I didn't think that the compiler would become
so popular, so not much time was spent in the design so popular, so not much time was spent in the design of it.
of the register allocation.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Basics % Basics
\section{Basics} \section{Basics}
The register allocation is done in the first and second pass of The register allocation is done in the first and the second pass of
the compiler. the compiler.
The first pass of a node has to calculate how much registers The first pass of a node has to calculate how much registers
are necessary to generate code for the node, it has are necessary to generate code for the node, but it also has
also to take care of child nodes i.e. how much registers to take care of child nodes i.e. how much registers
they need. they need.
The register allocation is done via \var{getregister\*} The register allocation is done via \var{getregister\*}
@ -296,8 +295,8 @@ procedure secondderef(var p : ptree);
// the contents of the registers isn't destroyed // the contents of the registers isn't destroyed
del_reference(p^.left^.location.reference); del_reference(p^.left^.location.reference);
// now should be at least one register free, so we // now there should be at least one register free, so
// can allocate one for the base of the result // we can allocate one for the base of the result
hr:=getregister32; hr:=getregister32;
// generate dereferencing instruction // generate dereferencing instruction
@ -316,18 +315,18 @@ procedure secondderef(var p : ptree);
% Binary nodes % Binary nodes
\section{Binary nodes} \section{Binary nodes}
The whole thing becomes a little bit more hairy, if you have to The whole thing becomes a little bit more hairy if you have to
generate code for a binary+ node (a node with two or more generate code for a binary+ node (a node with two or more
childs). If a node calls second pass for a child node, childs). If a node calls second pass for a child node,
it has to ensure that enough registers are free it has to ensure that enough registers are free
to evalute the child node (\var{usableregs>=childnode\^.registers32}). to evaluate the child node (\var{usableregs>=childnode\^.registers32}).
If this condition isn't true, the current node has If this condition isn't met, the current node has
to store and restore all registers which the node owns to to store and restore all registers which the node owns to
release registers. This should be done using the release registers. This should be done using the
procedures \var{maybe\_push} and \var{restore}. If still procedures \var{maybe\_push} and \var{restore}. If still
\var{usableregs<childnode\^.registers32}, the child nodes have to solve \var{usableregs<childnode\^.registers32}, the child nodes have to solve
the problem. The point is: if \var{usableregs<childnode\^.registers32}, the problem. The point is: if \var{usableregs<childnode\^.registers32},
the current node have to release all registers which it owns the current node has to release all registers which it owns
before the second pass is called. An example for generating before the second pass is called. An example for generating
code of a binary node is \var{cg386add.secondadd}. code of a binary node is \var{cg386add.secondadd}.
@ -335,10 +334,10 @@ code of a binary node is \var{cg386add.secondadd}.
% FPU registers % FPU registers
\section{FPU registers} \section{FPU registers}
The number of required FPU registers must be also calculated with The number of required FPU registers also has to be calculated, but
one difference: you needn't to save registers, if too few registers there's one difference: you don't have to save registers. If not
are free, just an error message is generated, the user enough FPU registers are free, an error message is generated, as the user
have to take care of too few FPU registers, this is a consequence has to take care of this situation since this is a consequence
of the stack structure of the FPU. of the stack structure of the FPU.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@ -433,7 +432,7 @@ The target processor must be set always and it can be:
% Leave Out specific Parts % Leave Out specific Parts
\section{Leave Out specific Parts} \section{Leave Out specific Parts}
Leaving of parts of the compiler is useful, if you want to create Leaving out parts of the compiler can be useful if you want to create
a compiler which should also run on systems with less memory a compiler which should also run on systems with less memory
requirements (for example a real mode version compiled with Turbo Pascal). requirements (for example a real mode version compiled with Turbo Pascal).
@ -446,7 +445,7 @@ requirements (for example a real mode version compiled with Turbo Pascal).
The following defines apply only to the i386 version of the compiler. The following defines apply only to the i386 version of the compiler.
\begin{description} \begin{description}
\item[\var{NoAg386Int}] No Intel styled assembler (for the MASM/TASM) writer \item[\var{NoAg386Int}] No Intel styled assembler (for MASM/TASM) writer
\item[\var{NoAg386Nsm}] No NASM assembler writer \item[\var{NoAg386Nsm}] No NASM assembler writer
\item[\var{NoAg386Att}] No AT\&T assembler (for the GNU AS) writer \item[\var{NoAg386Att}] No AT\&T assembler (for the GNU AS) writer
\item[\var{NoRA386Int}] No Intel assembler parser \item[\var{NoRA386Int}] No Intel assembler parser