Gpc November 2004

gpc@gnu.de

18 participants
22 discussions

Inefficiency of `mod'
by Emil Jerabek 25 Oct '10

25 Oct '10

Hi, during some debugging I noticed that GPC generates very hairy code for the `mod' operator with signed arguments. A closer look revealed that the code contains a lot of branching on conditions which can be determined on compile time, and most surprisingly, the value of X mod Y is recomputed _twice_, if X is negative. (I've attached a commented assembler dump.) This may be partially a back-end problem (CSE?), but I feel there's something wrong in the relevant code in gpc-common.c (build_pascal_binary_op) as well. `exp1 mod exp2' gets translated into something like exp2 <= 0 ? error : exp1 >= 0 ? exp1 mod exp2 : ((-exp1) mod exp2) == 0 ? 0 : exp2 - (-exp1) mod exp2 This arranges that both arguments supplied to the "internal mod" are non-negative, but this information is not passed on -- the arguments are still declared as signed types, which apparently makes the back-end to produce the mess it does. I've done a little test: the following replacement for `mod', which differs from the built-in one essentially just by the type-casts to `Cardinal', inline operator xmod (X, Y: Integer) = Z: Integer; begin if Y <= 0 then RuntimeError (714); if X >= 0 then Z := Cardinal (X) mod Y else begin Z := Cardinal (-X) mod Y; if Z <> 0 then Z := Y - Z end end; indeed generates a better code, it measures about 15% faster for positive X and more than twice faster for negative X. YMMV. If this is possible in user code, it should work in the compiler internals as well :) The result type of the whole `mod' expression looks dubious too. IIUIC, `exp1 mod exp2' inherits the type of exp1. Now, the result of `mod' is guaranteed to be nonnegative and to fit in the type of exp2 (or more precisely, it is in 0 .. High (type of (exp2)) - 1), whereas it needn't have anything to do with the type of exp1: consider var X, Z: Integer; Y: LongInt; begin X := -8; Y := High (LongInt); Z := X mod Y { -> overflow } end Or am I missing something? Emil (BTW, where is fjf434c.pas?)

10 16

-mlong-double-128
by Adriaan van Os 13 May '05

13 May '05

The following produces an internal compiler error: [G4:~/gnu/testgpc/adriaan] adriaan% cat compare.pas program Compare; procedure R ( var r1, r2: real); begin if r1 <> r2 then r1:= r2 - 1.0 end; begin end. [G4:~/gnu/testgpc/adriaan] adriaan% gpc -o compare compare.pas -mlong-double-128 compare.pas: In procedure `R': compare.pas:6: error: unrecognizable insn: (insn 26 34 27 2 0x0 (set (reg:SI 126) (subreg:SI (reg:TF 33 f1) 0)) -1 (nil) (nil)) compare.pas:6: internal compiler error: in extract_insn, at recog.c:2175 It may be, of course, that the Darwin target doesn't properly support the -mlong-double-128 PowerPC processor option. If it's a gcc back-end bug, I can report it to Bugzilla. However, the "real" type used is 64-bit on the Darwin target, with or without -mlong-double-128 ... The option only changes the size of the "longreal" and "extended" types to 128-bit. Regards, Adriaan van Os [G4:~/gnu/testgpc/adriaan] adriaan% gpc -v Reading specs from /Developer/Pascal/gpc33d7/lib/gcc-lib/powerpc-apple-darwin/3.3/specs Configured with: ../gpc-3.3d7/configure --enable-languages=pascal,c --enable-threads=posix --prefix=/Developer/Pascal/gpc33d7 --target=powerpc-apple-darwin Thread model: posix gpc version 20030507, based on gcc-3.3

6 15

cast increases required alignment of target type
by Adriaan van Os 15 Feb '05

15 Feb '05

Consider the followig test program: program TestCast; type BytePtr = ^Byte; IntPtr = ^Integer; var theBytePtr : BytePtr; theIntPtr: IntPtr; begin theBytePtr := nil; theIntPtr:= nil; theIntPtr:= IntPtr( theBytePtr); {testcast.pas:11: warning: cast increases required alignment of target type} BytePtr( theIntPtr) := theBytePtr; theBytePtr:= BytePtr( theIntPtr); IntPtr( theBytePtr):= theIntPtr; {testcast.pas:14: warning: cast increases required alignment of target type} end. 1. Compiling with gpc produces two warnings as indicated in the comment. The compiler warns if the "casting" pointer type (outside the parenthesis) points to a larger type than the "casted" pointer type (inside the parenthesis). Gale Paper brought to my attention that the same holds on the left side of an assignment and I am inclined to regard this as a bug - the compiler should check here for a smaller type, rather than a larger type. 2. the -Wcast-align and -Wno-cast-align command-line switches don't seem to work with gpc. Regards, Adriaan van Os

2 5

gpc-20041017 released
by Frank Heckenbach 15 Feb '05

15 Feb '05

I have uploaded a new beta version of GPC to <http://www.g-n-u.de/gpc/>! Because our main server is still out of order, please note: - The new URL, as stated above. www.gnu-pascal.de still doesn't work. - There are just the "bare bones" files (full and minimal source distribution, binaries for Linux/IA32 with and without gcc). No diffs, no separate test files, no HTML pages etc. - The main directory in the full source distribution is probably different! Be careful with automatic build scripts. - The releases are signed with a new OpenPGP key. The fingerprint is given in my signature in this mail and my other mails to the gpc(a)gnu.de list for some weeks. When the main server is back again, I'll do the usual release procedures there. One major change is that the types `Integer', `Word', `Cardinal' are now equivalent to `PtrInt', `PtrWord', `PtrCard', no more (necessarily) to C's `int' and `unsigned int' types. New types `CInteger', `CWord', `CCardinal' are provided for the latter purpose. This affects C interfaces in particular. A "porting guide" is given below. If you maintain any such interfaces, be sure to read it carefully. Another major improvement is in the area of initializers, thanks to Waldek Hebisch. GPC now allows initialization of variant records (fjf259.pas, peter6.pas) and packed arrays (emil5.pas) and supports Extended Pascal's structured initializers (fjf964*.pas, fjf967*.pas, fjf968*.pas) as well as record, array and set values (constdef.pas, fjf966*.pas, fjf971*.pas). Unfortunately, we don't have very many test programs for these new features yet. If you can send us some (according to the guidelines described in the manual), this will be very welcome. Other new features: * `Discard' * new make variable `GPC_PAGESIZE' to set the page size when building the manual (PDF, PostScript, DVI) * `qualified' and import lists are no more allowed after `uses' (only after `import', as EP demands) (@) * the `GMP' unit doesn't support gmp-2.x anymore (if you used it, just upgrade to a newer GMP version) (@) * conflicts between object fields/methods and ancestor type names are detected as required by OOE (fjf945*.pas) (@) (O) * repeated function headings (in `forward' declarations and interfaces) are checked stricter: if one has a result variable, so must the other (according to the OOE draft) (@) (O) Bugfixes: * records with no fields are invalid according to EP (fjf972.pas) * `Index' and `Pos' cannot be used in constant expressions (fjf970*.pas) * the `__FILE__' and `__BASE_FILE__' macros should return full paths * `Sqr' sometimes evaluates its argument twice (fjf963.pas) * memory leak in routines with a local variable of file type that are left via `Exit' or `Return' (fjf962.pas) * using a Boolean function as a condition in a `repeat' loop doesn't work correctly (artur1*.pas) * overstrict type-checking in comparisons involving procedural types (fjf960*.pas) * `Read', `Write' etc. evaluate its arguments in the wrong order (az43*.pas) * `Read' etc. evaluate references to string parameters twice (fjf958.pas) * on targets with 16 bit `Integer' type, the `MicroSecond' field of `TimeStamp' causes a compilation error * character arrays indexed by non-integer ordinal types treated as strings cause internal compiler errors (fjf957*.pas) * `-W[no-]cast-align' does not work (fjf956*.pas) * in `FormatTime' `%Z' and `%z' unless provided by system library ignore DST * powerpc: `--strength-reduce' doesn't work with `for' loops (was kludged before, fixed in gcc-3.3.3) * AIX: `ReturnAddress' doesn't work after use of dynmamic variables (backend bug, fixed in gcc-3.3.3) * functions returning sets are called twice if range-checking is on (inga1*.pas) PORTING GUIDE FOR THE NEW C COMPATIBLE INTEGER TYPES ==================================================== In my experience, the transition goes as follows: 1. If you want your code to keep working with older GPC versions after this change, add (near the top) the following part. You can omit the types you will not need in the following. {$if __GPC_RELEASE__ < 20041017} type CInteger = Integer; CCardinal = Cardinal; CWord = Word; {$endif} 2. In all C-visible declarations, replace `Integer' with `CInteger' and `Cardinal'/`Word' with `CCardinal'/`CWord'. C-visible declarations are usually `external' declarations (routines and/or variables) in Pascal which are implemented in C, but they can also be Pascal-implemented declarations (usually with a `name' attribute) which are declared `extern' in C code. Do not forget structures that are used within such declarations, e.g. parameters of type [pointer to] a record, or procedural parameters, etc. The same goes for other languages such as C++. 3. Fix any resulting incompatibilites. (This is the nasty part.) Value parameters are usually no problem. Anything else might be a problem, e.g., reference parameters, variables, structure fields, procedural parameters' parameters (including value parameters). As long as they are used only in the implementation of a unit/module which exports a Pascal interface, necessary changes can be confined to and handled within this interfaec. Otherwise, you have the choice whether to change the interface, which will require further changes in code using the interface, or to add wrappers to keep the interface. Example: unit MyCInterface; interface uses GPC; type t = record x, y, z: Integer end; procedure a (n: Integer); external name 'a'; procedure b (var n: Integer); external name 'b'; procedure c (var r: t); external name 'c'; function d (var n: Integer): TString; implementation function cd (var n: Integer): CString; external name 'd'; function d (var n: Integer): TString; begin d := CString2String (cd (n)) end; end. Changing a to procedure a (n: CInteger); external name 'a'; is usually harmless. Changing b to procedure b (var n: CInteger); external name 'b'; is problematic, since all callers would need to change their types (which might cause further ramifications in indirectly dependent code). So it might be in order to add a wrapper, such as: procedure b (var n: Integer); [implementation] procedure cb (var n: CInteger); external name 'b'; procedure b (var n: Integer); var cn: CInteger; begin cn := n; { omit if n is used only as an "output" parameter } cb (n); n := cn end; Procedure c uses a record containing `Integer' fields. Here, it's a hard choice between adding a wrapper (might be expensive, as the record gets bigger, and if more routines are affected) or changing the record to: type t = record x, y, z: CInteger end; The choice might depend on how many other routines use this structure and how many of them are written in Pascal or C, and how much existing code already uses the structure and what it does with it. Function d already has a wrapper, so it's usually best to also take care of the new change there: function d (var n: Integer): TString; var cn: CInteger; begin cn := n; { omit if n is used only as an "output" parameter } d := CString2String (cd (n)) n := cn end; Finally, note that `Integer' might be bigger than `CInteger', especially on newer platforms (that's the reason for the whole change in the first place). So if you want to have a bigger range available, you might want to consider changing from `CInteger' etc. (`[unsigned] int' in C) to `MedInt' etc. (`[unsigned] long' in C). This, however, requires (possibly substantial) changes in the C code, and might be out of your control (when interfacing to 3rd party libraries). Frank -- Frank Heckenbach, frank(a)g-n-u.de, http://fjf.gnu.de/, 7977168E GPC To-Do list, latest features, fixed bugs: http://www.gnu-pascal.de/todo.html NEW! GPC download signing key: ACB3 79B2 7EB2 B7A7 EFDE D101 CD02 4C9D 0FE0 E5E8

9 23

not 0
by Adriaan van Os 08 Feb '05

08 Feb '05

GPC accepts: var i: integer value not 0; but it doesn't accept: var w: word value not 0; not.pas:4: error: constant out of range not.pas:4: error: initial value is of wrong type so, the workaround is: var w: word value not word( 0); At first sight, this looks a bit clumsy (but at second sight it may not). It gives rise to general questions about the type of constants in Pascal, e.g. * is there a difference between $FF.. and $0FF.. ? * is $FF.. signed or not ? Any comments ? Regards, Adriaan van Os

5 18

GCC integration?
by E. Weddington 22 Dec '04

22 Dec '04

Hello, I couldn't find the answer anywhere on the website, FAQ, or mailing list archives: Why isn't GNU Pascal integrated with the rest of GCC? and the related question: What would it take to integrate GNU Pascal with the rest of GCC? Thanks for your time. Eric

11 30

GPC on AIX - memory limit?
by Peter Keenan 29 Nov '04

29 Nov '04

Hi, I've built gpc 20041017, based on gcc-3.3.2 in my home directory on an AIX machine which has the GCC 3.3.2 binaries installed in usr/local/bin. The testsuite programs all run fine and I am able to compile and run my own programs. I am using a program converted from Delphi, where I increased the stack and heap limits to allow the program use more memory. The program ran successfully several times on a PC, using more than 1GB of memory. However on AIX the program seem to be hitting a ceiling as they crash out with SIGILL, this happened with several datasets. The crash occurs at a New() statement asking for an additional 1600 bytes of memory and seems to occur when about 256MB of memory has been used. I tried the memavail function and it shows around 256 MB available at the start of the program, however memavail also sometimes causes a segmentation fault so I cannot use it to plot the memory use of the program . Using ulimit -a core file size (blocks, -c) unlimited data seg size (kbytes, -d) unlimited file size (blocks, -f) unlimited max memory size (kbytes, -m) unlimited open files (-n) unlimited pipe size (512 bytes, -p) 64 stack size (kbytes, -s) unlimited cpu time (seconds, -t) unlimited max user processes (-u) 10000 virtual memory (kbytes, -v) unlimited so no obviou problem there., If I use ulimit -d to reduce the data size below 256MB then that affects the program, but increasing it to a fixed value above 256MB does not seem to stop it crashing. I am inexperienced with the UNIX type environment and would appreciate any suggestions for changes needed at the GPC end or at the AIX and to get around this problem. Peter.

2 2

-Wno-underscore
by Adriaan van Os 28 Nov '04

28 Nov '04

[G4:~/testgpcadriaan] adriaan% cat NoUnderscore.p program NoUnderscore; procedure _Run; begin writeln('_Run') end; begin _Run end. [G4:~/testgpcadriaan] adriaan% gpc NoUnderscore.p NoUnderscore.p:4: warning: identifiers should not start with an underscore [G4:~/testgpcadriaan] adriaan% gpc NoUnderscore.p -Wno-underscore [G4:~/testgpcadriaan] adriaan% gpc NoUnderscore.p --gnu-pascal -Wno-underscore NoUnderscore.p:4: warning: identifiers should not start with an underscore So, it seems that -Wno-underscore is broken when used with --gnu-pascal. Regards, Adriaan van Os

4 6

Re: Building GPC with GCC 3.4.2 (in Fedora Core 3)
by Wood David 24 Nov '04

24 Nov '04

A big thanks Waldek. I seem to have a gpc built with gcc 3.4.2. Just a few minor hiccups which don't appear to have been too critical: 1) The man pages didn't build because help2man wasn't found. I'm not sure where this normally comes from, but the man pages have usually been built ok. 2) When I did 'cd gcc' and 'make pascal.install-with-gcc' , the gpc -v command gave me the correct build version, but a gpc compilation resulted in the error 'installation problem, cannot exec gpc1: no such file or directory'. Repeating the installation just using 'make install' solved the problem. 3) At the end of the install, I get the error /bin/sh: -ranlib: command not found. I presume the leading '-' is messing this up. I found ranlib in /usr/bin. Apologies for the corporate nonsense that follows. I'll try to get our company policy changed but it's likely to be a David and Goliath battle! Regards David. > -----Original Message----- > From: gpc-owner(a)gnu.de [mailto:gpc-owner@gnu.de]On Behalf Of Waldek > Hebisch > Sent: Wednesday, November 24, 2004 2:08 PM > To: gpc(a)gnu.de > Subject: Re: Building GPC with GCC 3.4.2 (in Fedora Core 3) > > > Wood David wrote: > > I'm trying to compile gpc-20040516 with gcc-core-3.4.2 in > the hope it will > > sort out a compatibility problem when linked with other > c-code we've built. > > I found Waldek's patches to gpc proper and gcc 3.4.20041015 > (I've also tried > > the 3.4.0 patch) and applied these using $ patch -p1 < > XX.diff in the > > respective source directories. However, the make crashes, > firstly with > > > > ../../gcc-3.4.2/gcc/p/mk_lang_opt: Permission denied. > > > > In normal build `mk_lang_opt' should not run (it is needed only after > a modification). The problem is due to limitation of `patch' (inabilty > to set file mode or modification time) and microsecond timestamps > in new kernels. Before the build you should touch `lang.opt' and > `handle-opts.c' to mark them as up to date. > > > So I do a chmod +x on this file and type make again. The > next worry comes > > when it says I should have 16 shift/reduce conflicts and 30 > reduce/reduce > > conflicts but I get 16 and 295! > > > > That is harmless. > > > The final nail in the coffin appears when p/handle-opts.c > produces a string > > of errors because the OPT_* components are undeclared. > Just before this, > > lang.c reports a number of warnings about function > declarations not being > > prototypes and no previous prototype for 'pascal_post-options' or > > 'pascal_handle_option'. > > > > Trying to run `mk_lang_opt' had wiped-out pre-built `lang.opt' (you > get empty file instead). `lang.opt' is needed to define > `OPT_*' constants. > Making `mk_lang_opt' executable _before_ the build should also correct > the problem. > > -- > Waldek Hebisch > hebisch(a)math.uni.wroc.pl > The Information contained in this E-Mail and any subsequent correspondence is private and is intended solely for the intended recipient(s). For those other than the recipient any disclosure, copying, distribution, or any action taken or omitted to be taken in reliance on such information is prohibited and may be unlawful. Emails and other electronic communication with QinetiQ may be monitored. Calls to QinetiQ may be recorded for quality control, regulatory and monitoring purposes.

1 0

Building GPC with GCC 3.4.2 (in Fedora Core 3)
by Wood David 24 Nov '04

24 Nov '04

I'm trying to compile gpc-20040516 with gcc-core-3.4.2 in the hope it will sort out a compatibility problem when linked with other c-code we've built. I found Waldek's patches to gpc proper and gcc 3.4.20041015 (I've also tried the 3.4.0 patch) and applied these using $ patch -p1 < XX.diff in the respective source directories. However, the make crashes, firstly with ../../gcc-3.4.2/gcc/p/mk_lang_opt: Permission denied. So I do a chmod +x on this file and type make again. The next worry comes when it says I should have 16 shift/reduce conflicts and 30 reduce/reduce conflicts but I get 16 and 295! The final nail in the coffin appears when p/handle-opts.c produces a string of errors because the OPT_* components are undeclared. Just before this, lang.c reports a number of warnings about function declarations not being prototypes and no previous prototype for 'pascal_post-options' or 'pascal_handle_option'. Am I doing something silly? I've been building gpc with gcc-3.3.2 without any difficulty. Regards David Wood QinetiQ Farnborough The Information contained in this E-Mail and any subsequent correspondence is private and is intended solely for the intended recipient(s). For those other than the recipient any disclosure, copying, distribution, or any action taken or omitted to be taken in reliance on such information is prohibited and may be unlawful. Emails and other electronic communication with QinetiQ may be monitored. Calls to QinetiQ may be recorded for quality control, regulatory and monitoring purposes.

3 2

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

2000

1999

1998

1997

1996

1995

Gpc November 2004