Difference between revisions of "ZSeries/Part 2"
(more details) |
(→pp.pas) |
||
Line 21: | Line 21: | ||
==pp.pas== | ==pp.pas== | ||
:'''''main program''', includes '''fpcdefs.inc''', calls units '''cmem, profile, catch, globals, compiler''''' | :'''''main program''', includes '''fpcdefs.inc''', calls units '''cmem, profile, catch, globals, compiler''''' | ||
− | '''pp.pas''' is the main program of the compiler. We're going to edit this file to create an IBM-370 cross-compiler. First, we'll decide what is the switch for this, as well as the names of the components. For the compiler switch or target environment, we'll use I370. Within the code, the cpu name is '''cpu_i370''', floating-point processor is '''fpu_i370''', and the operating system target is '''osvs1'''. | + | '''pp.pas''' is the main program of the compiler. We're going to edit this file to create an IBM-370 cross-compiler. First, we'll decide what is the switch for this, as well as the names of the components. For the compiler switch or target environment, we'll use I370. Within the code, the cpu name is '''cpu_i370''', floating-point processor is '''fpu_i370''', and the operating system target is <s>'''osvs1'''</s> '''linux'''. (I have access to Linux on a zSystem so I'm going to use that instead.) |
So we'll start by adding "i370" to the source comments and wherever else needed to indicate that, by inserting the middle line in the comment block (about line 35): | So we'll start by adding "i370" to the source comments and wherever else needed to indicate that, by inserting the middle line in the comment block (about line 35): |
Revision as of 17:04, 18 October 2012
Because the number of places I have to look is lots of them, the list of units I go through is broken up into more than one page. I'll try to keep page lengths to a reasonable factor, if a page seems to be too large, I will split it and start a new section.
To Begin
This compiler is huge. It's hundreds of source files, and is going to be an enormous task. Where do you start? Well, you start with the main program of the command-line compiler, and you look at it. That file is pp.pas. Then you collect all of the units it uses, and you look at each one's source file, then you repeat this recursively, finding every place where there is a reference to a particular machine type or operating system and you add the particular cpu or operating system (or both) to it.
Note that line numbers indicated in any source file are from the version 2.6.0 compiler sources and as such, as lines are added, other line numbers where things were found and changed will increase. So line numbers will be referenced in a file from top to bottom so the references should match. Also, so as not to brand this as "windows centric" since the hope is to build a cross-compiler for I370 that could run on either Windows or Linux, when file names are specified, directory separators will use /.
Note that from this point on, all editing occurs in our "sandbox" directory separate from the original compiler.
How units and include files are listed
Since this series of articles is intended to be a tutorial on how to add a new architecture to the compiler, files that call other files or units are listed at the top of the section dealing with that file. Where a unit is called by another, that caller is listed. Same for files that are included by other files. Any units the unit or file uses, are also listed. As new units which are referenced are found, they are added to the list of all the units that need to be checked and inspected, and potentially edited, as well as new units that have to be created.
This also helps since the list of modules is broken over several pages, so you can see where a module cross-references another one even if the references for both are not on the same page.
The format of the first line is generally like this (items not used will not appear on the line):
- Called from: [file names that call this file as a unit, printed in bold] (or) main program, included by [files that call this file using include], includes [files that this file calls as include files], requires external procedures [non-Pascal procedures this unit uses], calls units [units that this unit or file uses] or "no called units noted" if it does not use any units
If the file is not normally in the Compiler directory that is noted, or if it is in a subdirectory. As items are noted in the source code of that file or unit, they will be mentioned.
The Free Pascal Compiler
pp.pas
- main program, includes fpcdefs.inc, calls units cmem, profile, catch, globals, compiler
pp.pas is the main program of the compiler. We're going to edit this file to create an IBM-370 cross-compiler. First, we'll decide what is the switch for this, as well as the names of the components. For the compiler switch or target environment, we'll use I370. Within the code, the cpu name is cpu_i370, floating-point processor is fpu_i370, and the operating system target is osvs1 linux. (I have access to Linux on a zSystem so I'm going to use that instead.)
So we'll start by adding "i370" to the source comments and wherever else needed to indicate that, by inserting the middle line in the comment block (about line 35):
VIS generate a compiler for the VIS I370 generate a compiler for the IBM 370/390/zSeries mainframes DEBUG version with debug code is generated
This program includes fpcdefs.inc so we'll check that later. We have to add the indication to only select one target compiler, so we'll select for I370 (about line 142, where we'll add everything starting at the sixth line where we check to see if i370 is defined):
{$ifdef support_mmx} {$ifndef i386} {$fatal I386 switch must be on for MMX support} {$endif i386} {$endif support_mmx} {$ifdef i370} {$ifdef CPUDEFINED} {$fatal ONLY one of the switches for the CPU type must be defined} {$endif CPUDEFINED} {$define CPUDEFINED} {$endif i370}
The rest of the main program seems to be okay, but we will have to go through and look at all the units that this program uses, which, depending on which options have been set, are or can be: cmem, profile, catch, globals,
and compiler
. Then follow those and see where they lead, and "lather, rinse, repeat." The rest of this file seems okay, so we'll save it. (If we don't change a file, we don't re-save it so as not to change the date and time of the last edit.)
fpcdefs.inc
- Called as include file from pp.pas, called as include file from units catch, compiler, verbose (and virtually all other units), no called units noted
fpcdefs.inc provides various definitions regarding what processor we're compiling for to most units and many other files. We need to define the processor, so we'll borrow the generic one, and add or remove items as we need them. So we'll start with the block beginning with the line {$ifdef generic_cpu}
through the line {$endif generic_cpu}
. From later work, I discover I'll have to mark the target machine as big endian, so I'll include that. (Note that the definitions will include things as I discover them, so this may include things I haven't explained.)
So around line 151, between the lines
{$endif mips} {$IFDEF MACOS}
We'll update this to the following:
{$endif mips} {$ifdef i370} {$define cpu32bit} {$define cpu32bitaddr} {$define cpu32bitalu} {$define cpuflags} {$define cpuextended} {$define ENDIAN_BIG} {$endif i370} {$IFDEF MACOS}
Otherwise, now, fpcdefs.inc looks okay, so I'll save it.
Unit cmem
- Called from: pp.pas, file located in directory rtl/inc outside of the compiler directory, requires external procedures malloc, free, realloc, calloc, no called units noted
This is used as a bridge to the C memory management library and its functions malloc, free, realloc, calloc
. In some architectures it uses a specific library. I will probably borrow these from an existing C library or simulate them. Or this module may be rewritten. For now, there is nothing I need to do here.
Unit profile
- Called from: pp.pas, file located in directory rtl/go32v2 outside of the compiler directory, no units noted
This is only used for profiling under the Go32 system, so this module is not relevant to our cross-compiler.
Unit catch
- Called from: pp.pas, includes fpcdefs.inc, no called units noted
catch.pas deals with handling control-c and segfaults. We might have to throw some flags to enable this or to handle it since the compiler doesn't run locally on the mainframe, it runs on the user's PC. Right now, we're just trying to get to the point of being able to create code for the zSeries. So in the absence of anything requiring we deal with this, we'll leave it alone for now.
Unit globals
- Called from: pp.pas; options.pas , includes fpcdefs.inc, calls unit comphook
globals.pas, starting at about line 372, has some definitions for specific target processors. So, again, I'll borrow from {$ifdef GENERIC_CPU}
and change as necessary. At line 424, we have
{$endif mips}
so we'll add the 370 after this, and put in the following:
{$endif mips} {$ifdef i370} cputype : cpu_i370; optimizecputype : cpu_i370; fputype : fpu_i370; {$endif i370}
Note, these values will be defined elsewhere. Note that this unit references a few other units depending on the OS and machine, but the only one relevant to us will be comphook
so we'll put that on the "stack" of units that have to be inspected and possibly edited.
Unit compiler
- Called from: pp.pas, includes fpcdefs.inc, calls units fksysutils, sysutils, math, verbose, comphook, systems, cutils, cfileutl, cclasses, globals, options, fmodule, parser, symtable, assemble, link, dbgbase, import, export, tokens, pass_1, wpobase, wpo, cpupara, cpupi, cgcpu, cpunode, cputarg, i_i370, globtype
Starting at about line 24, this unit lists the units it uses. Some units depend on other flags, but the units it does always require are: verbose, comphook, systems, cutils, cfileutl, cclasses, globals, options, fmodule, parser, symtable, assemble, link, dbgbase, import, export, tokens, pass_1, wpobase, wpo, cpupara, cpupi, cgcpu, cpunode,
and cputarg
. These are also added to the "stack" for checking on references.
At this point I do not know whether or not I need to set the flag USE_FAKE_SYSUTILS
, if I do, I only need to add one unit, fksysutils
, but if I do need the "real" sysutils unit, then it will automatically include the units sysutils, math
. So I'll presume all three have to be looked at. They also go on the stack for checking. Doing a directory scan, apparently every architecture does use the sysutils
unit, and none use fksysutils
, so I'll presume that to be the case.
This unit also defines the machine that is the target for this compiler, so we have to add a unit which will define our target architecture. There are units for every machine, so we have to insert one. At the code around line 116:
{$ifdef nativent} ,i_nativent {$endif nativent} ,globtype;
we change to:
{$ifdef nativent} ,i_nativent {$endif nativent} {$ifdef i370} ,i_i370 {$endif i370} ,globtype;
This means we've now added a new source file which will eventually have to be created, i_i370.pas for the new unit i_i370
, which almost certainly will be in the I370 subdirectory. Also unit globtype
which is at the end of the list of units used. The rest looks okay so for the moment we're done with this unit.
Unit fksysutils
This is only used if the particular architecture sets the flag that says it does not use the sysutils
and math
units. Given that all of them do use them, I will presume for now I do not need to create this file. Presume every unit that includes sysutils includes this file if necessary.
Unit sysutils
- Called from: compiler.pas; verbose.pas; cclasses.pas; parser.pas; assemble.pas; rtl/objpas/math.pas (outside Compiler directory) located in rtl/i370 (outside Compiler directory), includes fpcdefs.inc, no called units noted
This file is a local file for each architecture, so it will be in the rtl subdirectory in its i370 subdirectory. I'll take a look at a couple of architectures to see what's expected to be included. It is used for platform dependent calls, is used by the listed units, and if there are no platform dependent calls, it will simply be a stub unit. For the moment, I'll just create a stub file and come back later if I need to.
{
This file is part of the Free Pascal run time library. Copyright (c) 2012 by Viridian Development Corporation Sysutils unit for IBM 370/390/zSystem See the file COPYING.FPC, included in this distribution, for details about the copyright. 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. **********************************************************************} unit sysutils; interface implementation end.
Unit math
- Called from: compiler.pas, located in rtl/objpas outside the compiler directory, includes fpcdefs.inc, calls unit sysutils
This might need to be changed later. The IBM 370 series uses a different floating point system and its numeric limits are not the same as IEEE. For the moment I'll leave this alone and come back later. Note that when it is compiled, a copy of the PPU file needs to be in rtl/i370 (outside of the Compiler directory).
Unit verbose
- Called from: compiler.pas; options.pas; parser.pas, includes msgtxt.inc, msgidx.inc, uses units sysutils, fksysutl, cutils, globtype, finput, cmsgs
verbose.pas handles generating long ("verbose") error messages. It uses unit sysutils
unless USE_FAKE_SYSUTILS
is defined, then it uses unit fksysutl
instead. If EXTERN_MSG
is not defined, it includes msgtxt.inc. Always includes msgidx.inc. Since this is in the compiler, not the user's code I can ignore it for now.
msgtxt.inc
- Called as include file from verbose.pas, no called units noted
Causes the entire 253 different error messages to be included in the verbose
unit as a block of strings. Since it's in the compiler, not the user's program, it's nothing I have to worry about right now.
msgidx.inc
- Called as include file from verbose.pas, no called units noted
This is a list of constants that translate the names of error messages into reference numbers. It is intended to be included into the verbose
unit. Again, I can ignore this for now.
Unit comphook
- Called from: compiler.pas; parser.pas, includes fpcdefs.inc, uses units sysutils, fksysutl, globtype, finput;
comphook.pas only uses fksysutl if sysutils isn't present. Is a compiler hook to external programs, right now I don't need to do anything with it.
Unit systems
- Called from: compiler.pas; cfileutl.pas; options.pas; parser.pas; assemble.pas, includes fpcdefs.inc; systems.inc, calls unit cutils
Relates to information about the target systems supported. Lines 303-305 need to be altered to add the 370 (TsystemCpu will be adjusted to compensate):
cpu2str : array[TSystemCpu] of string[10] = (,'i386','m68k','alpha','powerpc','sparc','vm','ia64','x86_64', 'mips','arm', 'powerpc64', 'avr', 'mipsel','i370');
At around line 851 we have to indicate the target, replacing the one line
{$endif mips}
to
{$endif mips} {$ifdef i370} default_target(system_i370_mainframe); {$endif i370}
systems.inc
- Called as include file from: systems.pas, no called units noted
This adds the cpu types available. About line 50, change
cpu_mipsel { 13 }
to
cpu_mipsel, { 13 } cpu_i370 { 14 }
About line 67, change
,asmmode_avr_gas
to
,asmmode_avr_gas ,asmmode_i370
About line 151, change
system_powerpc_wii { 70 }
to
system_powerpc_wii, { 70 } system_i370_mainframe { 71 }
About line 185, change
,as_i386_nlmcoff
to
,as_i386_nlmcoff ,as_i370
Otherwise seems okay for now.
Unit cutils
- Called from: compiler.pas; verbose.pas; systems.pas; cfileutl.pas; cclasses.pas; parser.pas; symtable.pas, includes fpcdefs.inc, no called units noted
This provides a number of utility functions. It also requires the target be defined as big endian, although that may only be important if the compiler is running on the target rather than on the cross-compiling host.
Unit cfileutl
- Called from: compiler.pas; options.pas, includes fpcdefs.inc, uses units SysUtils; fksysutl; GlobType; CUtils; CClasses; Systems
Contains file handling code. May fix later.
Unit cclasses
- Called from: compiler.pas; cfileutl.pas; options,pas; parser.pas; symtable.pas, includes fpcdefs.inc, calls units SysUtils; fksysutl; globtype; CUtils; CStreams;
Creates some basic classes for the compiler
Unit globals
- Called from: compiler.pas; parser.pas; assemble.pas, includes fpcdefs.inc, no called units noted
globals.pas definitely provides direct processor-dependent information. We'll copy the default {$ifdef GENERIC_CPU}
and change it for the i370. At about line 424 substitute
{$endif mips}
with
{$endif mips} {$ifdef i370} cputype : cpu_i370; optimizecputype : cpu_i370; fputype : fpu_i370; {$endif i370}
No other processor-dependent code needs to be fixed for the moment.
Unit options
- Called from: compiler.pas, includes fpcdefs.inc, calls units CClasses; cfileutl; globtype; globals; verbose; systems; cpuinfo; comprsrc
Command-line options. If I need to include them for the IBM 370 they'll go here. Right now I don't see any I'll need.
Unit fmodule
- Called from: compiler.pas; parser.pas, includes fpcdefs.inc, no called units noted
Loading and searching modules. If the compiler ever becomes self-hosting on the zSystem this will probably need to be changed.
Unit parser
- Called from: compiler.pas, includes fpcdefs.inc, calls units sysutils; fksysutl; cutils; cclasses; globtype; version; tokens; systems; globals; verbose; switches; symbase; symtable; symdef; symsym; finput; fmodule; fppu; aasmbase; aasmtai; aasmdata; cgbase; script; gendef; comphook; scanner; scandir; pbase; ptype; psystem; pmodules; psub; ncgrtti; htypechk; cresstr; cpuinfo; procinfo
Does source code parsing.
Unit symtable
- Called from: compiler.pas; parser.pas; symtable.pas, includes fpcdefs.inc, no called units noted
Handles symbol tables
Unit assemble
- Called from: compiler.pas, includes fpcdefs.inc, calls units SysUtils; systems; globtype; globals; aasmbase; aasmtai; aasmdata; ogbase; finput
Handles assembler write and assembler calls. Does include some architecture-specific items.
To make this page a reasonable size, this list continues in part 3.