libjit/doc/libjit.texi

\input texinfo	@c -*-texinfo-*-
@c %** start of header
@setfilename libjit.info
@settitle Just-In-Time Compiler Library
@setchapternewpage off
@c %** end of header

@dircategory Libraries
@direntry
* Libjit: (libjit).                Just-In-Time Compiler Library
@end direntry

@ifinfo
The libjit library assists with the process of building
Just-In-Time compilers for languages, virtual machines,
and emulators.

Copyright @copyright{} 2004 Southern Storm Software, Pty Ltd
@end ifinfo

@titlepage
@sp 10
@center @titlefont{Just-In-Time Compiler Library}

@vskip 0pt plus 1fill
@center{Copyright @copyright{} 2004 Southern Storm Software, Pty Ltd}
@end titlepage

@syncodeindex fn cp
@syncodeindex vr cp
@syncodeindex tp cp

@c -----------------------------------------------------------------------

@node Top, Introduction, , (dir)
@menu
* Introduction::            Introduction and rationale for libjit
* Features::                Features of libjit
* Tutorials::               Tutorials in using libjit
* Initialization::          Initializing the JIT
* Functions::               Building and compiling functions with the JIT
* Types::                   Manipulating system types
* Values::                  Working with temporary values in the JIT
* Instructions::            Working with instructions in the JIT
* Basic Blocks::            Working with basic blocks in the JIT
* Intrinsics::              Intrinsic functions available to libjit users
* Exceptions::              Handling exceptions
* Breakpoint Debugging::    Hooking a breakpoint debugger into libjit
* ELF Binaries::            Manipulating ELF binaries
* Utility Routines::        Miscellaneous utility routines
* Diagnostic Routines::     Diagnostic routines
* C++ Interface::           Using libjit from C++
* Porting::                 Porting libjit to new architectures
* Index::                   Index of concepts and facilities
@end menu

@c -----------------------------------------------------------------------

@node Introduction, Features, Top, Top
@chapter Introduction and rationale for libjit
@cindex Introduction

Just-In-Time compilers are becoming increasingly popular for executing
dynamic languages like Perl and Python and for semi-dynamic languages
like Java and C#.  Studies have shown that JIT techniques can get close to,
and sometimes exceed, the performance of statically-compiled native code.

However, there is a problem with current JIT approaches.  In almost every
case, the JIT is specific to the object model, runtime support library,
garbage collector, or bytecode peculiarities of a particular system.
This inevitably leads to duplication of effort, where all of the good
JIT work that has gone into one virtual machine cannot be reused in another.

JIT's are not only useful for implementing languages.  They can also be used
in other programming fields.  Graphical applications can achieve greater
performance if they can compile a special-purpose rendering routine
on the fly, customized to the rendering task at hand, rather than using
static routines.  Needless to say, such applications have no need for
object models, garbage collectors, or huge runtime class libraries.

Most of the work on a JIT is concerned with arithmetic, numeric type
conversion, memory loads/stores, looping, performing data flow analysis,
assigning registers, and generating the executable machine code.
Only a very small proportion of the work is concerned with language specifics.

The goal of the @code{libjit} project is to provide an extensive set of
routines that takes care of the bulk of the JIT process, without tying the
programmer down with language specifics.  Where we provide support for
common object models, we do so strictly in add-on libraries,
not as part of the core code.

Unlike other systems such as the JVM, .NET, and Parrot, @code{libjit}
is not a virtual machine in its own right.  It is the foundation upon which a
number of different virtual machines, dynamic scripting languages,
or customized rendering routines can be built.

The LLVM project (@uref{http://www.llvm.org/}) has some similar
characteristics to @code{libjit} in that its intermediate format is
generic across front-end languages.  It is written in C++ and provides
a large set of compiler development and optimization components;
much larger than @code{libjit} itself provides.  According to its author,
Chris Lattner, a subset of its capabilities can be used to build JIT's.

Libjit should free developers to think about the design of their front
ends, and not get bogged down in the details of code execution.
Meanwhile, experts in the design and implementation of JIT's can concentrate
on solving code execution problems, instead of front end support issues.

This document describes how to use the library in application programs.
We start with a list of features and some simple tutorials.  Finally,
we provide a complete reference guide for all of the API functions in
@code{libjit}, broken down by function category.

@section Obtaining libjit

The latest version of @code{libjit} can be obtained from Southern
Storm Software, Pty Ltd's Web site:

@quotation
@uref{http://www.southern-storm.com.au/libjit.html}
@end quotation

@section Further reading

While it isn't strictly necessary to know about compiler internals
to use @code{libjit}, you can make more effective use of the library
if you do.  We recommend the "Dragon Book" as an excellent resource
on compiler internals, particularly the sections on code generation
and optimization:

@quotation
Alfred V. Aho, Ravi Sethi, and Jeffrey D. Ullman, "Compilers:
Principles, Techniques, and Tools", Addison-Wesley, 1986.
@end quotation

IBM, Intel, and others have done a lot of research into JIT implementation
techniques over the years.  If you are interested in working on the
internals of @code{libjit}, then you may want to make yourself familiar
with the relevant literature (this is by no means a complete list):

@quotation
IBM's Jikes RVM (Research Virtual Machine), @*
@uref{http://www-124.ibm.com/developerworks/oss/jikesrvm/}.

Intel's ORP (Open Runtime Platform), @*
@uref{http://orp.sourceforge.net/}.
@end quotation

@c -----------------------------------------------------------------------

@node Features, Tutorials, Introduction, Top
@chapter Features of libjit
@cindex Features

@itemize
@item
The primary interface is in C, for maximal reusability.  Class
interfaces are available for programmers who prefer C++.

@item
Designed for portability to all major 32-bit and 64-bit platforms.

@item
Simple three-address API for library users, but opaque enough that other
representations can be used inside the library in future without
affecting existing users.

@item
Up-front or on-demand compilation of any function.

@item
In-built support to re-compile functions with greater optimization,
automatically redirecting previous callers to the new version.

@item
Fallback interpreter for running code on platforms that don't
have a native code generator yet.  This reduces the need for
programmers to write their own interpreters for such platforms.

@item
Arithmetic, bitwise, conversion, and comparison operators for 8-bit,
16-bit, 32-bit, or 64-bit integer types; and 32-bit, 64-bit, or longer
floating point types.  Includes overflow detecting arithmetic for
integer types.

@item
Large set of mathematical and trigonometric operations
(sqrt, sin, cos, min, abs, etc) for inlining floating-point library functions.

@item
Simplified type layout and exception handling mechanisms, upon which a
variety of different object models can be built.

@item
Support for nested functions, able to access their parent's local variables
(for implementing Pascal-style languages).
@end itemize

@c -----------------------------------------------------------------------

@node Tutorials, Tutorial 1, Features, Top
@chapter Tutorials in using libjit
@cindex Tutorials

In this chapter, we describe how to use @code{libjit} with a number of
short tutorial exercises.  Full source for these tutorials can be found
in the @code{tutorial} directory of the @code{libjit} source tree.

For simplicity, we will ignore errors such as out of memory conditions,
but a real program would be expected to handle such errors.

@menu
* Tutorial 1::              Tutorial 1 - mul_add
* Tutorial 2::              Tutorial 2 - gcd
* Tutorial 3::              Tutorial 3 - compiling on-demand
* Tutorial 4::              Tutorial 4 - mul_add, C++ version
* Tutorial 5::              Tutorial 5 - gcd, with tail calls
* Dynamic Pascal::          Dynamic Pascal - A full JIT example
@end menu

@c -----------------------------------------------------------------------

@node Tutorial 1, Tutorial 2, Tutorials, Tutorials
@section Tutorial 1 - mul_add
@cindex mul_add tutorial

In the first tutorial, we will build and compile the following function
(the source code can be found in @code{tutorial/t1.c}):

@example
int mul_add(int x, int y, int z)
@{
    return x * y + z;
@}
@end example

@noindent
To use the JIT, we first include the @code{<jit/jit.h>} file:

@example
#include <jit/jit.h>
@end example

All of the header files are placed into the @code{jit} sub-directory,
to separate them out from regular system headers.  When @code{libjit}
is installed, you will typically find these headers in
@code{/usr/local/include/jit} or @code{/usr/include/jit}, depending upon
how your system is configured.  You should also link with the
@code{-ljit} option.

@noindent
Every program that uses @code{libjit} needs to call @code{jit_context_create}:

@example
jit_context_t context;
...
context = jit_context_create();
@end example

Almost everything that is done with @code{libjit} is done relative
to a context.  In particular, a context holds all of the functions
that you have built and compiled.

You can have multiple contexts at any one time, but normally you will
only need one.  Multiple contexts may be useful if you wish to
run multiple virtual machines side by side in the same process,
without them interfering with each other.

Whenever we are constructing a function, we need to lock down the
context to prevent multiple threads from using the builder at a time:

@example
jit_context_build_start(context);
@end example

The next step is to construct the function object that will represent
our @code{mul_add} function:

@example
jit_function_t function;
...
function = jit_function_create(context, signature);
@end example

The @code{signature} is a @code{jit_type_t} object that describes the
function's parameters and return value.  This tells @code{libjit} how
to generate the proper calling conventions for the function:

@example
jit_type_t params[3];
jit_type_t signature;
...
params[0] = jit_type_int;
params[1] = jit_type_int;
params[2] = jit_type_int;
signature = jit_type_create_signature
    (jit_abi_cdecl, jit_type_int, params, 3, 1);
@end example

This declares a function that takes three parameters of type
@code{int} and returns a result of type @code{int}.  We've requested
that the function use the @code{cdecl} application binary interface (ABI),
which indicates normal C calling conventions.  @xref{Types}, for
more information on signature types.

Now that we have a function object, we need to construct the instructions
in its body.  First, we obtain references to each of the function's
parameter values:

@example
jit_value_t x, y, z;
...
x = jit_value_get_param(function, 0);
y = jit_value_get_param(function, 1);
z = jit_value_get_param(function, 2);
@end example

Values are one of the two cornerstones of the @code{libjit} process.
Values represent parameters, local variables, and intermediate
temporary results.  Once we have the parameters, we compute
the result of @code{x * y + z} as follows:

@example
jit_value_t temp1, temp2;
...
temp1 = jit_insn_mul(function, x, y);
temp2 = jit_insn_add(function, temp1, z);
@end example

This demonstrates the other cornerstone of the @code{libjit} process:
instructions.  Each of these instructions takes two values as arguments
and returns a new temporary value with the result.

Students of compiler design will notice that the above statements look
very suspiciously like the "three address statements" that are described
in compiler textbooks.  And that is indeed what they are internally within
@code{libjit}.

If you don't know what three address statements are, then don't worry.
The library hides most of the details from you.  All you need to do is
break your code up into simple operation steps (addition, multiplication,
negation, copy, etc).  Then perform the steps one at a time, using
the temporary values in subsequent steps.  @xref{Instructions}, for
a complete list of all instructions that are supported by @code{libjit}.

Now that we have computed the desired result, we return it to the caller
using @code{jit_insn_return}:

@example
jit_insn_return(function, temp2);
@end example

We have completed the process of building the function body.  Now we
compile it into its executable form:

@example
jit_function_compile(function);
jit_context_build_end(context);
@end example

As a side-effect, this will discard all of the memory associated with
the values and instructions that we constructed while building the
function.  They are no longer required, because we now have the
executable form that we require.

We also unlock the context, because it is now safe for other threads
to access the function building process.

Up until this point, we haven't executed the @code{mul_add} function.
All we have done is build and compile it, ready for execution.  To execute it,
we call @code{jit_function_apply}:

@example
jit_int arg1, arg2, arg3;
void *args[3];
jit_int result;
...
arg1 = 3;
arg2 = 5;
arg3 = 2;
args[0] = &arg1;
args[1] = &arg2;
args[2] = &arg3;
jit_function_apply(function, args, &result);
printf("mul_add(3, 5, 2) = %d\n", (int)result);
@end example

We pass an array of pointers to @code{jit_function_apply}, each one
pointing to the corresponding argument value.  This gives us a very
general purpose mechanism for calling any function that may be
built and compiled using @code{libjit}.  If all went well, the
program should print the following:

@example
mul_add(3, 5, 2) = 17
@end example

You will notice that we used @code{jit_int} as the type of the arguments,
not @code{int}.  The @code{jit_int} type is guaranteed to be 32 bits
in size on all platforms, whereas @code{int} varies in size from platform
to platform.  Since we wanted our function to work the same everywhere,
we used a type with a predictable size.

If you really wanted the system @code{int} type, you would use
@code{jit_type_sys_int} instead of @code{jit_type_int} when you
created the function's signature.  The @code{jit_type_sys_int} type
is guaranteed to match the local system's @code{int} precision.

@noindent
Finally, we clean up the context and all of the memory that was used:

@example
jit_context_destroy(context);
@end example

@c -----------------------------------------------------------------------

@node Tutorial 2, Tutorial 3, Tutorial 1, Tutorials
@section Tutorial 2 - gcd
@cindex gcd tutorial

In this second tutorial, we implement the subtracting Euclidean
Greatest Common Divisor (GCD) algorithm over positive integers.
This tutorial demonstrates how to handle conditional branching
and function calls.  In C, the code for the @code{gcd} function
is as follows:

@example
unsigned int gcd(unsigned int x, unsigned int y)
@{
    if(x == y)
    @{
        return x;
    @}
    else if(x < y)
    @{
        return gcd(x, y - x);
    @}
    else
    @{
        return gcd(x - y, y);
    @}
@}
@end example

The source code for this tutorial can be found in @code{tutorial/t2.c}.
Many of the details are similar to the previous tutorial.  We omit
those details here and concentrate on how to build the function body.
@xref{Tutorial 1}, for more information.

@noindent
We start by checking the condition @code{x == y}:

@example
jit_value_t x, y, temp1;
...
x = jit_value_get_param(function, 0);
y = jit_value_get_param(function, 1);
temp1 = jit_insn_eq(function, x, y);
@end example

This is very similar to our previous tutorial, except that we are using
the @code{eq} operator this time.  If the condition is not true, we
want to skip the @code{return} statement.  We achieve this with the
@code{jit_insn_branch_if_not} instruction:

@example
jit_label_t label1 = jit_label_undefined;
...
jit_insn_branch_if_not(function, temp1, &label1);
@end example

The label must be initialized to @code{jit_label_undefined}.  It will be
updated by @code{jit_insn_branch_if_not} to refer to a future position in
the code that we haven't seen yet.

If the condition is true, then execution falls through to the next
instruction where we return @code{x} to the caller:

@example
jit_insn_return(function, x);
@end example

If the condition was not true, then we branched to @code{label1} above.
We fix the location of the label using @code{jit_insn_label}:

@example
jit_insn_label(function, &label1);
@end example

@noindent
We use similar code to check the condition @code{x < y}, and branch
to @code{label2} if it is not true:

@example
jit_value_t temp2;
jit_label_t label2 = jit_label_undefined;
...
temp2 = jit_insn_lt(function, x, y);
jit_insn_branch_if_not(function, temp2, &label2);
@end example

At this point, we need to call the @code{gcd} function with the
arguments @code{x} and @code{y - x}.  The code for this is
fairly straight-forward.  The @code{jit_insn_call} instruction calls
the function listed in its third argument.  In this case, we are calling
ourselves recursively:

@example
jit_value_t temp_args[2];
jit_value_t temp3;
...
temp_args[0] = x;
temp_args[1] = jit_insn_sub(function, y, x);
temp3 = jit_insn_call
    (function, "gcd", function, 0, temp_args, 2, 0);
jit_insn_return(function, temp3);
@end example

The string @code{"gcd"} in the second argument is for diagnostic purposes
only.  It can be helpful when debugging, but the @code{libjit} library
otherwise makes no use of it.  You can set it to NULL if you wish.

In general, @code{libjit} does not maintain mappings from names to
@code{jit_function_t} objects.  It is assumed that the front end will
take care of that, using whatever naming scheme is appropriate to
its needs.

@noindent
The final part of the @code{gcd} function is similar to the previous one:

@example
jit_value_t temp4;
...
jit_insn_label(function, &label2);
temp_args[0] = jit_insn_sub(function, x, y);
temp_args[1] = y;
temp4 = jit_insn_call
    (function, "gcd", function, 0, temp_args, 2, 0);
jit_insn_return(function, temp4);
@end example

@noindent
We can now compile the function and execute it in the usual manner.

@c -----------------------------------------------------------------------

@node Tutorial 3, Tutorial 4, Tutorial 2, Tutorials
@section Tutorial 3 - compiling on-demand
@cindex On-demand compilation tutorial

In the previous tutorials, we compiled everything that we needed
at startup time, and then entered the execution phase.  The real power
of a JIT becomes apparent when you use it to compile functions
only as they are called.  You can thus avoid compiling functions
that are never called in a given program run, saving memory and
startup time.

We demonstrate how to do on-demand compilation by rewriting Tutorial 1.
The source code for the modified version is in @code{tutorial/t3.c}.

When the @code{mul_add} function is created, we don't create its function
body or call @code{jit_function_compile}.  We instead provide a
C function called @code{compile_mul_add} that performs on-demand
compilation:

@example
jit_function_t function;
...
function = jit_function_create(context, signature);
jit_function_set_on_demand_compiler(function, compile_mul_add);
@end example

We can now call this function with @code{jit_function_apply}, and the
system will automatically call @code{compile_mul_add} for us if the
function hasn't been built yet.  The contents of @code{compile_mul_add}
are fairly obvious:

@example
int compile_mul_add(jit_function_t function)
@{
    jit_value_t x, y, z;
    jit_value_t temp1, temp2;

    x = jit_value_get_param(function, 0);
    y = jit_value_get_param(function, 1);
    z = jit_value_get_param(function, 2);

    temp1 = jit_insn_mul(function, x, y);
    temp2 = jit_insn_add(function, temp1, z);

    jit_insn_return(function, temp2);
    return 1;
@}
@end example

When the on-demand compiler returns, @code{libjit} will call
@code{jit_function_compile} and then jump to the newly compiled code.
Upon the second and subsequent calls to the function, @code{libjit}
will bypass the on-demand compiler and call the compiled code directly.
Note that in case of on-demand compilation @code{libjit} automatically
locks and unlocks the corresponding context with
@code{jit_context_build_start} and @code{jit_context_build_end} calls.

Sometimes you may wish to force a commonly used function to
be recompiled, so that you can apply additional optimization.
To do this, you must set the "recompilable" flag just after the
function is first created:

@example
jit_function_t function;
...
function = jit_function_create(context, signature);
jit_function_set_recompilable(function);
jit_function_set_on_demand_compiler(function, compile_mul_add);
@end example

Once the function is compiled (either on-demand or up-front) its
intermediate representation built by @code{libjit} is discarded.
To force the function to be recompiled you need to build it again
and call @code{jit_function_compile} after that.  As always when
the function is built and compiled manually it is necessary
to take care of context locking:

@example
jit_context_build_start(context);
jit_function_get_on_demand_compiler(function)(function);
jit_function_compile(function);
jit_context_build_end(context);
@end example

After this, any existing references to the function will be redirected
to the new version.  However, if some thread is currently executing the
previous version, then it will keep doing so until the previous version
exits.  Only after that will subsequent calls go to the new version.

In this tutorial, we use the same on-demand compiler when we
recompile @code{mul_add}.  In a real program, you would probably call
@code{jit_function_set_on_demand_compiler} to set a new on-demand
compiler that performs greater levels of optimization.

If you no longer intend to recompile the function, you should call
@code{jit_function_clear_recompilable} so that @code{libjit} can
manage the function more efficiently from then on.

The exact conditions under which a function should be recompiled
are not specified by @code{libjit}.  It may be because the function
has been called several times and has reached some threshold.
Or it may be because some other function that it calls has become a
candidate for inlining.  It is up to the front end to decide when
recompilation is warranted, usually based on language-specific
heuristics.

@c -----------------------------------------------------------------------

@node Tutorial 4, Tutorial 5, Tutorial 3, Tutorials
@section Tutorial 4 - mul_add, C++ version
@cindex mul_add C++ tutorial

While @code{libjit} can be easily accessed from C++ programs using
the C API's, you may instead wish to use an API that better reflects
the C++ programming paradigm.  We demonstrate how to do this by rewriting
Tutorial 3 using the @code{libjitplus} library.

@noindent
To use the @code{libjitplus} library, we first include
the @code{<jit/jit-plus.h>} file:

@example
#include <jit/jit-plus.h>
@end example

This file incorporates all of the definitions from @code{<jit/jit.h>},
so you have full access to the underlying C API if you need it.

This time, instead of building the @code{mul_add} function with
@code{jit_function_create} and friends, we define a class to represent it:

@example
class mul_add_function : public jit_function
@{
public:
    mul_add_function(jit_context& context) : jit_function(context)
    @{
        create();
        set_recompilable();
    @}

    virtual void build();

protected:
    virtual jit_type_t create_signature();
@};
@end example

Where we used @code{jit_function_t} and @code{jit_context_t} before,
we now use the C++ @code{jit_function} and @code{jit_context} classes.

In our constructor, we attach ourselves to the context and then call
the @code{create()} method.  This is in turn will call our overridden
virtual method @code{create_signature()} to obtain the signature:

@example
jit_type_t mul_add_function::create_signature()
@{
    // Return type, followed by three parameters,
    // terminated with "end_params".
    return signature_helper
        (jit_type_int, jit_type_int, jit_type_int,
         jit_type_int, end_params);
@}
@end example

The @code{signature_helper()} method is provided for your convenience,
to help with building function signatures.  You can create your own
signature manually using @code{jit_type_create_signature} if you wish.

The final thing we do in the constructor is call @code{set_recompilable()}
to mark the @code{mul_add} function as recompilable, just as we did in
Tutorial 3.

The C++ library will create the function as compilable on-demand for
us, so we don't have to do that explicitly.  But we do have to override
the virtual @code{build()} method to build the function's body on-demand:

@example
void mul_add_function::build()
@{
    jit_value x = get_param(0);
    jit_value y = get_param(1);
    jit_value z = get_param(2);

    insn_return(x * y + z);
@}
@end example

This is similar to the first version that we wrote in Tutorial 1.
Instructions are created with @code{insn_*} methods that correspond
to their @code{jit_insn_*} counterparts in the C library.

One of the nice things about the C++ API compared to the C API is that we
can use overloaded operators to manipulate @code{jit_value} objects.
This can simplify the function build process considerably when we
have lots of expressions to compile.  We could have used @code{insn_mul}
and @code{insn_add} instead in this example and the result would have
been the same.

Now that we have our @code{mul_add_function} class, we can create
an instance of the function and apply it as follows:

@example
jit_context context;
mul_add_function mul_add(context);

jit_int arg1 = 3;
jit_int arg2 = 5;
jit_int arg3 = 2;
jit_int args[3];
args[0] = &arg1;
args[1] = &arg2;
args[2] = &arg3;

mul_add.apply(args, &result);
@end example

@noindent
@xref{C++ Interface}, for more information on the @code{libjitplus}
library.

@c -----------------------------------------------------------------------

@node Tutorial 5, Dynamic Pascal, Tutorial 4, Tutorials
@section Tutorial 5 - gcd, with tail calls
@cindex gcd with tail calls

Astute readers would have noticed that Tutorial 2 included two instances
of "tail calls".  That is, calls to the same function that are immediately
followed by a @code{return} instruction.

Libjit can optimize tail calls if you provide the @code{JIT_CALL_TAIL}
flag to @code{jit_insn_call}.  Previously, we used the following code
to call @code{gcd} recursively:

@example
temp3 = jit_insn_call
    (function, "gcd", function, 0, temp_args, 2, 0);
jit_insn_return(function, temp3);
@end example

@noindent
In Tutorial 5, this is modified to the following:

@example
jit_insn_call(function, "gcd", function, 0, temp_args, 2, JIT_CALL_TAIL);
@end example

There is no need for the @code{jit_insn_return}, because the call
will never return to that point in the code.  Behind the scenes,
@code{libjit} will convert the call into a jump back to the head
of the function.

Tail calls can only be used in certain circumstances.  The source
and destination of the call must have the same function signatures.
None of the parameters should point to local variables in the current
stack frame.  And tail calls cannot be used from any source function
that uses @code{try} or @code{alloca} statements.

Because it can be difficult for @code{libjit} to determine when these
conditions have been met, it relies upon the caller to supply the
@code{JIT_CALL_TAIL} flag when it is appropriate to use a tail call.

@c -----------------------------------------------------------------------

@node Dynamic Pascal, Initialization, Tutorial 5, Tutorials
@section Dynamic Pascal - A full JIT example
@cindex Dynamic Pascal

This @code{libjit/dpas} directory contains an implementation of
"Dynamic Pascal", or "dpas" as we like to call it.  It is provided
as an example of using @code{libjit} in a real working environment.
We also use it to write test programs that exercise the JIT's capabilities.

Other Pascal implementations compile the source to executable form,
which is then run separately.  Dynamic Pascal loads the source code
at runtime, dynamically JIT'ing the program as it goes.  It thus has
a lot in common with scripting languages like Perl and Python.

If you are writing a bytecode-based virtual machine, you would use
a similar approach to Dynamic Pascal.  The key difference is that
you would build the JIT data structures after loading the bytecode
rather than after parsing the source code.

To run a Dynamic Pascal program, use @code{dpas name.pas}.  You may also
need to pass the @code{-I} option to specify the location of the system
library if you have used an @code{import} clause in your program.  e.g.
@code{dpas -I$HOME/libjit/dpas/library name.pas}.

@noindent
This Pascal grammar is based on the EBNF description at the following URL:

@uref{http://www.cs.qub.ac.uk/~S.Fitzpatrick/Teaching/Pascal/EBNF.html}

@noindent
There are a few differences to "Standard Pascal":

@enumerate
@item
Identifiers are case-insensitive, but case-preserving.

@item
Program headings are normally @code{program Name (Input, Output);}.  This can
be abbreviated to @code{program Name;} as the program modifiers are ignored.

@item
Some GNU Pascal operators like @code{xor}, @code{shl}, @code{@@}, etc
have been added.

@item
The integer type names (@code{Integer}, @code{Cardinal}, @code{LongInt}, etc)
follow those used in GNU Pascal also.  The @code{Integer} type is always
32-bits in size, while @code{LongInt} is always 64-bits in size.

@item
The types @code{SysInt}, @code{SysCard}, @code{SysLong}, @code{SysLongCard},
@code{SysLongestInt}, and @code{SysLongestCard} are guaranteed to be the
same size as the underlying C system's @code{int}, @code{unsigned int},
@code{long}, @code{unsigned long}, @code{long long}, and
@code{unsigned long long} types.

@item
The type @code{Address} is logically equivalent to C's @code{void *}.
Any pointer or array can be implicitly cast to @code{Address}.  An explicit
cast is required to cast back to a typed pointer (you cannot cast back
to an array).

@item
The @code{String} type is declared as @code{^Char}.  Single-dimensional
arrays of @code{Char} can be implicitly cast to any @code{String}
destination.  Strings are not bounds-checked, so be careful.  Arrays
are bounds-checked.

@item
Pointers can be used as arrays.  e.g. @code{p[n]} will access the n'th
item of an unbounded array located at @code{p}.  Use with care.

@item
We don't support @code{file of} types.  Data can be written to stdout
using @code{Write} and @code{WriteLn}, but that is the extent of
the I/O facilities.

@item
The declaration @code{import Name1, Name2, ...;} can be used at the head of a
program to declare additional files to include.  e.g. @code{import stdio} will
import the contents of @code{stdio.pas}.  We don't support units.

@item
The idiom @code{; ..} can be used at the end of a formal parameter list to
declare that the procedure or function takes a variable number of arguments.
The builtin function @code{va_arg(Type)} is used to extract the arguments.

@item
The directive @code{import("Library")} can be used to declare that a function
or procedure was imported from an external C library.  For example, the
following imports the C @code{puts} and @code{printf} functions:

@example
function puts (str : String) : SysInt; import ("libc")
function printf (format : String; ..) : SysInt; import ("libc")
@end example

Functions that are imported in this manner have case-sensitive names.
i.e. using @code{Printf} above will fail.

@item
The @code{throw} keyword can be used to throw an exception.  The argument
must be a pointer.  The @code{try}, @code{catch}, and @code{finally}
keywords are used to manage such exceptions further up the stack.  e.g.

@example
try
    ...
catch Name : Type
    ...
finally
    ...
end
@end example

The @code{catch} block will be invoked with the exception pointer that was
supplied to @code{throw}, after casting it to @code{Type} (which must
be a pointer type).  Specifying @code{throw} on its own without an argument
will rethrow the current exception pointer, and can only be used inside a
@code{catch} block.

Dynamic Pascal does not actually check the type of the thrown pointer.
If you have multiple kinds of exceptions, then you must store some kind
of type indicator in the block that is thrown and then inspect @code{^Name}
to see what the indicator says.

@item
The @code{exit} keyword can be used to break out of a loop.

@item
Function calls can be used as procedure calls.  The return value is ignored.

@item
Hexadecimal constants can be expressed as @code{XXH}.  The first digit
must be between 0 and 9, but the remaining digits can be any hex digit.

@item
Ternary conditionals can be expressed as @code{(if e1 then e2 else e3)}.
The brackets are required.  This is equivalent to C's @code{e1 ? e2 : e3}.

@item
Assigning to a function result will immediately return.  i.e. it is
similar to @code{return value;} in C.  It isn't necessary to arrange for
execution to flow through to the end of the function as in regular Pascal.

@item
The term @code{sizeof(Type)} can be used to get the size of a type.

@item
Procedure and function headings can appear in a record type to declare a
field with a @code{pointer to procedure/function} type.
@end enumerate

@c -----------------------------------------------------------------------

@node Initialization, Functions, Dynamic Pascal, Top
@chapter Initializing the JIT
@cindex Initialization
@cindex Contexts

@include libjitext-init.texi
@include libjitext-context.texi

@c -----------------------------------------------------------------------

@node Functions, Types, Initialization, Top
@chapter Building and compiling functions with the JIT
@cindex Building functions
@cindex Compiling functions

@include libjitext-function.texi

@c -----------------------------------------------------------------------

@node Types, Values, Functions, Top
@chapter Manipulating system types
@cindex Manipulating system types

@include libjitext-type.texi

@c -----------------------------------------------------------------------

@node Values, Instructions, Types, Top
@chapter Working with temporary values in the JIT
@cindex Working with values

@include libjitext-value.texi

@c -----------------------------------------------------------------------

@node Instructions, Basic Blocks, Values, Top
@chapter Working with instructions in the JIT
@cindex Working with instructions

@include libjitext-insn.texi

@c -----------------------------------------------------------------------

@node Basic Blocks, Intrinsics, Instructions, Top
@chapter Working with basic blocks in the JIT
@cindex Working with basic blocks

@include libjitext-block.texi

@c -----------------------------------------------------------------------

@node Intrinsics, Exceptions, Basic Blocks, Top
@chapter Intrinsic functions available to libjit users
@cindex Intrinsics

@include libjitext-intrinsic.texi

@c -----------------------------------------------------------------------

@node Exceptions, Breakpoint Debugging, Intrinsics, Top
@chapter Handling exceptions
@cindex Handling exceptions

@include libjitext-except.texi

@c -----------------------------------------------------------------------

@node Breakpoint Debugging, ELF Binaries, Exceptions, Top
@chapter Hooking a breakpoint debugger into libjit
@cindex Breakpoint debugging

@include libjitext-debugger.texi

@c -----------------------------------------------------------------------

@node ELF Binaries, Utility Routines, Breakpoint Debugging, Top
@chapter Manipulating ELF binaries
@cindex ELF binaries

@include libjitext-elf-read.texi

@c -----------------------------------------------------------------------

@node Utility Routines, Diagnostic Routines, ELF Binaries, Top
@chapter Miscellaneous utility routines
@cindex Utility routines
@cindex jit-util.h

The @code{libjit} library provides a number of utility routines
that it itself uses internally, but which may also be useful to front ends.

@include libjitext-alloc.texi
@include libjitext-memory.texi
@include libjitext-string.texi
@include libjitext-meta.texi
@include libjitext-apply.texi
@include libjitext-walk.texi
@include libjitext-dynlib.texi
@include libjitext-cpp-mangle.texi

@c -----------------------------------------------------------------------

@node Diagnostic Routines, C++ Interface, Utility Routines, Top
@chapter Diagnostic routines
@cindex Diagnostic routines

@include libjitext-dump.texi

@c -----------------------------------------------------------------------

@node C++ Interface, C++ Contexts, Diagnostic Routines, Top
@chapter Using libjit from C++
@cindex Using libjit from C++

This chapter describes the classes and methods that are available
in the @code{libjitplus} library.  To use this library, you must
include the header @code{<jit/jit-plus.h>} and link with the
@code{-ljitplus} and @code{-ljit} options.

@menu
* C++ Contexts::            Contexts in C++
* C++ Values::              Values in C++
* C++ Functions::           Functions in C++
@end menu

@c -----------------------------------------------------------------------

@node C++ Contexts, C++ Values, C++ Interface, C++ Interface
@chapter Contexts in C++
@cindex C++ contexts

@include libjitext-plus-context.texi

@c -----------------------------------------------------------------------

@node C++ Values, C++ Functions, C++ Contexts, C++ Interface
@chapter Values in C++
@cindex C++ values

@include libjitext-plus-value.texi

@c -----------------------------------------------------------------------

@node C++ Functions, Porting, C++ Values, C++ Interface
@chapter Functions in C++
@cindex C++ functions

@include libjitext-plus-function.texi

@c -----------------------------------------------------------------------

@node Porting, Porting Apply, C++ Functions, Top
@chapter Porting libjit to new architectures
@cindex Porting libjit

This chapter describes what needs to be done to port @code{libjit}
to a new CPU architecture.  It is assumed that the reader is familiar
with compiler implementation techniques and the particulars of their
target CPU's instruction set.

We will use @code{ARCH} to represent the name of the architecture
in the sections that follow.  It is usually the name of the CPU in
lower case (e.g. @code{x86}, @code{arm}, @code{ppc}, etc).  By
convention, all back end functions should be prefixed with @code{_jit},
because they are not part of the public API.

@menu
* Porting Apply::           Porting the function apply facility
* Instruction Generation::  Creating the instruction generation macros
* Architecture Rules::      Writing the architecture definition rules
* Register Allocation::     Allocating registers in the back end
@end menu

@c -----------------------------------------------------------------------

@node Porting Apply, Instruction Generation, Porting, Porting
@section Porting the function apply facility
@cindex Porting apply

The first step in porting @code{libjit} to a new architecture is to port
the @code{jit_apply} facility.  This provides support for calling
arbitrary C functions from your application or from JIT'ed code.
If you are familiar with @code{libffi} or @code{ffcall}, then
@code{jit_apply} provides a similar facility.

Even if you don't intend to write a native code generator, you will
probably still need to port @code{jit_apply} to each new architecture.

The @code{libjit} library makes use of gcc's @code{__builtin_apply}
facility to do most of the hard work of function application.
This gcc facility takes three arguments: a pointer to the function
to invoke, a structure containing register arguments, and a size
value that indicates the number of bytes to push onto the stack
for the call.

Unfortunately, the register argument structure is very system dependent.
There is no standard format for it, but it usually looks something
like this:

@table @code
@item stack_args
Pointer to an array of argument values to push onto the stack.

@item struct_ptr
Pointer to the buffer to receive a @code{struct} return value.
The @code{struct_ptr} field is only present if the architecture
passes @code{struct} pointers in a special register.

@item word_reg[0..N]
Values for the word registers.  Platforms that pass values in
registers will populate these fields.  Not present if the architecture
does not use word registers for function calls.

@item float_reg[0..N]
Values for the floating-point registers.  Not present if the architecture
does not use floating-point registers for function calls.
@end table

It is possible to automatically detect the particulars of this structure
by making test function calls and inspecting where the arguments end up
in the structure.  The @code{gen-apply} program in @code{libjit/tools}
takes care of this.  It outputs the @code{jit-apply-rules.h} file,
which tells @code{jit_apply} how to operate.

The @code{gen-apply} program will normally "just work", but it is possible
that some architectures will be stranger than usual.  You will need to modify
@code{gen-apply} to detect this additional strangeness, and perhaps
also modify @code{libjit/jit/jit-apply.c}.

If you aren't using gcc to compile @code{libjit}, then things may
not be quite this easy.  You may have to write some inline assembly
code to emulate @code{__builtin_apply}.  See the file
@code{jit-apply-x86.h} for an example of how to do this.
Be sure to add an @code{#include} line to @code{jit-apply-func.h}
once you do this.

The other half of @code{jit_apply} is closure and redirector support.
Closures are used to wrap up interpreted functions so that they can be
called as regular C functions.  Redirectors are used to help compile a
JIT'ed function on-demand, and then redirect control to it.

Unfortunately, you will have to write some assembly code to support
closures and redirectors.  The builtin gcc facilities are not complete
enough to handle the task.  See @code{jit-apply-x86.c} and
@code{jit-apply-arm.c} for some examples from existing architectures.
You may be able to get some ideas from the @code{libffi} and
@code{ffcall} libraries as to what you need to do on your architecture.

@c -----------------------------------------------------------------------

@node Instruction Generation, Architecture Rules, Porting Apply, Porting
@section Creating the instruction generation macros
@cindex Instruction generation macros

You will need a large number of macros and support functions to
generate the raw instructions for your chosen CPU.  These macros are
fairly generic and are not necessarily specific to @code{libjit}.
There may already be a suitable set of macros for your CPU in
some other Free Software project.

Typically, the macros are placed into a file called @code{jit-gen-ARCH.h}
in the @code{libjit/jit} directory.  If some of the macros are complicated,
you can place helper functions into the file @code{jit-gen-ARCH.c}.
Remember to add both @code{jit-gen-ARCH.h} and @code{jit-gen-ARCH.c}
to @code{Makefile.am} in @code{libjit/jit}.

Existing examples that you can look at for ideas are @code{jit-gen-x86.h}
and @code{jit-gen-arm.h}.  The macros in these existing files assume that
instructions can be output to a buffer in a linear fashion, and that each
instruction is relatively independent of the next.

This independence principle may not be true of all CPU's.  For example,
the @code{ia64} packs up to three instructions into a single "bundle"
for parallel execution.  We recommend that the macros should appear to
use linear output, but call helper functions to pack bundles after the fact.
This will make it easier to write the architecture definition rules.
A similar approach could be used for performing instruction scheduling
on platforms that require it.

@c -----------------------------------------------------------------------

@node Architecture Rules, Register Allocation, Instruction Generation, Porting
@section Writing the architecture definition rules
@cindex Architecture definition rules

@include libjitext-rules-interp.texi

@c -----------------------------------------------------------------------

@node Register Allocation, Index, Architecture Rules, Porting
@section Allocating registers in the back end
@cindex Register allocation

@include libjitext-reg-alloc.texi

@c -----------------------------------------------------------------------

@page

@node Index, , Register Allocation, Top
@unnumbered Index of concepts and facilities

@printindex cp

@contents
@bye