gforth: gforth/engine/main.c

updated copyright year

Added fft bench

Generate image for Android Gforth

Added gforth_free_stacks() and gforth_free()

Nested signal handler in gforth_go

Another small refactoring

Use gforth_execute to show bootmessage

Refactoring for library use: completed

Refactoring almost finished - lacks a small test

Refactoring mostly completed

Refactored engine entries

Removed engine stack arguments - now global states

Refactoring for libgforth

Refactoring main to make Gforth library actually useful

Makefile changes to compile libraries

use PER_THREAD

More consistent use of %p, sigaltstack is back

Don't use alternate signal stack (collides with multithreading)

loop to install signal handler and setjmp for Android

Looping setjmp, use %p where appropriate

throw_jmp_buf only through a handler in the per-thread space

Better signal debug info

converted gforth_pointers into a function - necessary, since all pointers are now thread-specific

Fixed last bug in threading

thread-safe cstr (requires explicit mfree\!)

Started work on multithreading

Fixed register allocation problem with gcc 4.6 for x86 (all memory words now as subroutines)

updated copyright years

Debugging stack check is in again

Segfault when GFORTH_DEBUGGING is defined

updated copyright years

Make dynamic superinstructions work for MIPS by defining a
architecture-specific test macro CHECK_PRIM() that return whether a memory
region might contain non-relocatable code.  On MIPS that check is implementad
as a simple comparison against a blacklist of problematic opcodes.

updated image file magic number

Removed special casing of DOES> for relocater

Better compilation of code words (no need to use LIT-EXECUTE for them)

added ;ABI-CODE (not yet documented, not interactively usable yet)
deleted does-handler and doesjump stuff (unused)
moved abi-call to a new group
Extended AMD64 calling convention documentation

prepare for removing DOESJUMP: assert(0) when it occurs

support for assembler definitions with host platform's ABI calling conventions
('ABI-CODE')

Moved setlocale out of io.c

C-based EC improved a lot

updated copyright years

Fixed --clear-dictionary bug
Fixed marker bug (Aleksej)

added option --code-block-size=SIZE

--debug now reports unused code at end of code blocks

Bugfix: now makes sure that superinstruction is not divided between
  code blocks, to avoid having a transition alone with a NEXT.
Dictionary allocation now does not add the image preamble size
Minor debugging output changes

disabled setting precision to 53 bits (glibc's libm relies on 64 bits)

set 387 precision control to 53-bit mantissae to get rid of FLAW in paranoia.fs

more concrete report of performance problems

Makefile bugfix
fixed some portability issues for MIPS
fixed some portability issues for old platforms

Added cstr and tilde_cstr to gforth_pointers

Windows version of libcc interface (first attempt)

workarounds for MacOS X and /lib64 portability issues (/lib64 incomplete)

fixed BSD portability issue (reported by Aleksej Saushev)

fixed some warnings that were reported on sparc-solaris

fix if libltdl is not present

Now builds with llvm-gcc (but libcc.fs does not work)

eliminated warnings
added Benchmark results

The Makefile now uses an EMACS variable
Eliminated most compilation warnings

updated copyright years
updated copyright-blacklist (added libltdl)
updated distributed files (don't distribute files without distribution terms)
added copyright to preforth.in and build-ec.in

Removed remainders of FFI in C code

removed usage of libtool-2.2 ltdl functions

added CLEAR-LIB ADD-LIB (libcc.fs), documented them,
  and used them in libffi.fs and fflib.fs
OPEN-LIB no longer guesses library extensions (lt_dladvise_ext())

Bugfix: make and make dist don't need pre-installed Gforth
FORCE_REG_UNNECESSARY now defined automatically and used better
documentation bugfix

CPPFLAGS are now used by the Makefiles
bugfix in libcc.fs
OPEN-LIB now can open libraries without extension and the library's
   symbols become global
libltdl from libtool-2.2 or higher required (will be included soon)

removed --[no-]check-alignment (no implementation)
  on Alpha use <http://www.complang.tuwien.ac.at/anton/uace.c> instead

added --[no-]check-alignment (currently no effect)

--diag now does not complain about reg. alloc. on PPC (FORCE_REG_UNNECESSARY)

Fixed build problem
Some changes for NXT

Building on i386 works again

enable dynamic code generation for (debugging) gforth on AMD64 by accessing
  the global variable through a local pointer.  Mixed results up to now, may
  be better with explicit register allocation.

Now FORCE_REG is tried and used by default
Removed --enable-force-reg
Removed suggestions to use --enable-force-* from --diag option

refined static super vs. TPA check (TPA ok if state equivalence disabled).

reenabled static superinstructions
disable tpa-automaton if static superinstrutions are used

updated copyright notices for GPL v3

updated copyright years

libcc.fs now uses libtool
OPEN-LIB and LIB-SYM now call libltdl (libtool) functions
  probably needs some additional adjustment in configure.in
added LIB-ERROR

added -DBURG_FORMAT compile-time option for generating burg grammars etc.
cleaned up debugp() definition
documentation bugfix (ELSE)

added --print-sequences

Constants compile to literals

bugfix (reserve stack space for FTOS even if USE_NO_FTOS is defined)
  A better way to fix this would be to use an appropriate cache-fast.vmg

generalized code padding and added 386 and AMD64 padding

native branch target alignment for Alpha.  Speedup:
   sieve bubble matrix  fib
    1.309 1.160  1.409 1.834 without code alignment
    0.906 1.040  1.242 1.040 with code alignment

fixed minor bug in -falign-* configure tests
cleaned up dependencies on .h files in engine/Makefile.in
made Alpha port compile with gcc-2.95 (disabled longlong.h division)
optimized Alpha port:
  use old division code
  native code after a jump is now aligned.  Speed effect:
   sieve bubble matrix  fib
    1.950 1.605  1.259 1.653 b1 gcc-3.3.5
    1.453 1.932  1.230 1.458 b1 gcc-3.3.5 with alignment
  aligning branch targets would probably also help

Backed out bugfix for old Cygwin
Bug fix + correct cygwin fix in history.fs

Fixes for cygwin

Use MAP_NORESERVE on OSs that support it to overcommit dict and stack memory
Added --vm-commit option to turn off MAP_NORESERVE

Gforth says "hello"

Configure work to make Gforth on NXT compile without much effort

Some stuff to get closer to run Gforth on NXT

Further work on NXT Gforth
Fixed problem with useless kernel error messages
TAGS generation works again

C-based Gforth EC starts to work

Another step in direction to compile for NXT

Some more work to get a C-based Gforth EC work with autoconf

minor changes

improved stack boundary memory protection

lists and C prelude stuff in libcc.fs
minor changes

updated copyright years

added --ignore-async-signals
fixed typo

Save callback global variables

reduced CODE_BLOCK_SIZE to 512k (for release)

Minor fix

Environment OS from $host_os
R8C changes
Better output when there are no performance problems

Added gforth_ prefix to some functions

Warnings squashing cont.

Reduced warnings

Modified some exported symbols to start with gforth_

updated the copyright year on many files
added FSF copyright header to complex.fs fft.fs regexp-test.fs regexp.fs
added fsl-util.fs to update-copyright-blacklist

documented 2 bugs
minor bugfixes and cleanups

added options for disabling tree-parsing automaton features
added tree parsing automaton statistics

added TPA state equivalence, but disabled static superinstructions for now

finished simple tree-parsing automaton

initial work for tree-parsing automaton for stack caching

bugfix
--debug now also outputs some cache state information

Checkin of somewhat working ffilib bindings (not completely tested)

added STACK_CACHE_DEFAULT_FAST configuration variable
fixed some bugs
added some primitive/state combinations
changes to support up to 8 stack cache registers
Worked around an unresolved bug in kernel/toolsext.fs

First working version with default state with 2 regs
added configuration variable STACK_CACHE_DEFAULT_FAST

slight tweak in dict_alloc_read()

bugfix: only one allocation when the read-mmap fails (e.g., in Windows).

added configure option --disable-gcc-pr15242-workaround
bugfixes for the gcc pr15242 workaround

further work on gcc PR 15242 workaround (now works on AMD64 gforth-fast)
made treatment of Priminfo length and restlen more regular

preparation for gcc PR 15242 workaround
minor changes

Added --diag switch, diagnostics output.
Fixed "broken link" bug when referring to current file

Added --diag switch

updated copyright years for files changed in 2004

moved imm_ops field of costs up a bit (better factoring and readability)

added imm_ops operand to costs structure and generate it

fixed "too many crashes" in httpd.fs
fixed reentrance problem with callbacks

fixed bug in cache1.vmg
prims2x.fs: better comments for prim*.i
main.c: better --debug output (added primitive names)

fixed buggy commit (uncovered by SUSE 9.0 gcc)

minor changes

fixed gforth-native bug (branch target resolution)
fixed "make dist" bug (arch/misc/optcmove.fs now included)
Documentation changes

minor cleanups

changed compile_call2() interface

changed compile_prim1arg() interface

minor fix

minor fixes

bugfixes and cleanup in reloation bitset handling
more primtests

Threw out some unused code
recorded a bug

activated gforth-native again
some gforth-native bugfixes and changes

made classical 1-register stack caching work again (and probably
  general stack caching with non-empty canonical state).  There was a
  subtle bug involving ?DUP-?BRANCH.
made this the default for gforth-fast and turned off anything beyond
deleted the now-unused rewrite_bb().

Now the _fast.i files are included for gforth-fast and gforth-native

stack caching works now (at least for make check)
  the main change is to the optimize_... stuff in main.c
  also restricted static optimization area to basic blocks
minor bugfixes

put stack caching stuff in prims2x.fs (cvs log prims2y.fs for
	fine-grained changes)
set up prim etc. to work with the changes prims2x.fs (disabled USE_TOS for now)

now an is-branch? field is generated engine/costs.i
fixed some typos in README
changed default threading scheme for generic machine

fixed various minor issues:
  workaround for HPUX make and our dependence cycles
  suppress warnings for unused variables in superinstructions
  added some superinstructions
  useful code_size metrics for primitives with funny K and J labels
  default metric for --no-dynamic is now cost_nexts

Defined run-time routines (docol etc.) as primitives, eliminated some
  special-casing and "+DOESJUMP+1" etc. in engine.c and main.c
Defined the enum type PrimNum and replaced many shorts with PrimNums
  (gives nicer gdb output)
Workaround for the absence of the FPE_... macros in FreeBSD/Alpha 4_STABLE

Callback stuff tested

made superinstructions compatible with conditional compilation of primitives
  (in Gforth; don't use conditional compilation in vmgen).

decompile_prim can now deal with static superinstructions

fixed Windows PATHSEP bug
Makefile now tolerates ";" as PATHSEP
eliminatd some warnings

Fixed alloc-callback

Inclusion of ffcall stuff
ftp://ftp.santafe.edu/pub/gnu/ffcall-1.8.tar.gz
High level stuff missing

changed relocation validity check (now searches for next K label, then
  checks if J label is between I and this K label).

better checking of primitives for dynamic code generation (works with gcc-3.3)

gforth-prof is built by default again
default CFLAGS is now "-O2" instead of "-g -O2"
added some static superinstructions
fixed computation of engine/costs.i
minor changes

minor bugfix

added --ss-greedy (greedy superinstruction selection)
renamed --print-codesize into --print-metrics (and added stuff to print)

added various flags for controlling static superinstruction selection

dynamic superinstructions enabled again, further cleanups

refactored static superinstruction stuff

implemented new approach for static superinstructions (cleanup needed)
386 disasm bugfix (enter)

preparing for new approach to static superinstructions
fixed bug in 386 disassembler (fucompp)
minor cleanup

added --print-codesize option

updated copyright years

Gforth now works properly when its stdout is to a broken pipe (but not stderr)
special SIGPIPE throw code (-2049)
minor bugfix ("uncaught exception" is now printed on stderr)

portability changes
bugfixes in prims2x.fs, kernel/int.fs

eliminated some (hopefully outdated) hppa special treatments
rewrote hppa cacheflush
prims2x can now process CRLF inputs (but the output is partly unixified)
prims2x can now process several sync lines in sequence
minor fixes

fixed some portability bugs and other minor bugs

minor changes (64-bit-cleanups)

minor changes

undid changes to copyright notices

Updated copyright notices
Added stack effects to kernel/input.fs

flush_to_here now not called on superblock ends

Icache-flushing is now not performed on generating a NEXT, just on the
   end of a definition through finish-code.
Eliminated MIPS linker testing (not necessary with hybrid D/ITC).
--no-dynamic is default on systems that don't define FLUSH_ICACHE.

moved DOES-HANDLER! from prim to kernel/int.fs
eliminated MAKE_DOES_HANDLER

deleted unused prims (HASHKEY) and COMPILE-PRIM

changes to eliminate gcc warnings

undid the unnecessary/broken stuff Bernd has done
moved a change from machpc.fs to machpc.fs.in
removed machpc.fs (generated file)

Fixed look problem
Added benchmarking of all engines

Alpha bug workaround (disabled ALIGN_CODE)
fixed bugs in F>D and D>F

restored the old engine/prim_lab.i format (for vmgen compatibility)
introduced engine/prim_grp.i for the new use

First inclusion of group-based primitive tokens (no reordering happend yet)

increased image version

replaced some automake macros with autoconf macros; related changes in
  autoconf variables (e.g., @VERSION@ -> @PACKAGE_VERSION@)
gforth-prof and gforth-native are now not built by default
modernized "Automatic Remaking" section of Makefile.in
Updated bug reporting instructions
Eliminated use of old bug reporting email address

Gforth now supports large files (>2GB) on small machines (32-bits/cell).
forth.h now asks for all kinds of POSIX, X/Open, and GNU support.
rearranged include files such that forth.h precedes the system files.

some configure.in cleanups.  In particular, I replaced all echos with
  AC_MSG_... macros.
The gcc version (for disabling dynamic code) is now checked in configure (not
  when compiling main.c)
updated elisp-comp install-sh missing mkinstalldirs from autoconf-2.54

"compile," now uses compile-prim1 instead of compile-prim
disablesd compile-prim

moved global/static vars from engine.c into main.c (to avoid duplicating them)

"make check" now checks all engines and some variations
bugfix in forget-dyncode

bugfixes in DECOMPILE-PRIM (no crash on non-address,
   find @ instead of >code-address etc.)

implemented DECOMPILE-PRIM and use it in (LOOK); SIMPLE-SEE works again
minor bugfix in SIMPLE-SEE
HEAD? now delivers 1 on likely-false positives

dynamically generated code is now (sort of) released on executing a MARKER

code_area now allocated on demand (unlimited size)
some refactoring in dynamic code generation

now engine/prim_num.i is generated, and used in main.c (for main-native.o)

bugfix ("gforth" crashed, "./gforth" didn't)

Added check for directory separation

new engine gforth-native (works on 386 arch only for now).
  appropriate changes in control-flow instructions in prim
  new primitives SET-NEXT-CODE and CALL2 (not necessary for the other engines)
  new primitives COMPILE-PRIM1 and FINISH-CODE
  prims2x.fs now produces IMMARG(...) macros for initializing immediate args
  prims2x.fs: changes in some of the output-c-tail words (goes with the
     changes in the control-flow words).
  appropriate changes in engine.c
  engine.c: rewrite of check_prims, support for gforth-native (NO_IP)
  threaded.c: support for NO_IP
  various kernel files: started to eliminate return stack manipulations for
    embedding data (e.g. string literals); incomplete.
dynamic superinstructions now use LABEL2 instead of IS_NEXT_JUMP
FORCE_REG has no effect if DOUBLY_INDIRECT (gcc-2.95.1 crashes otherwise;
   it's unclear which change provoked this).

No engine2 and other dynamic stuff is compiled for gforth-itc, -iditc, -prof

Gforth now survives destruction of the C stack pointer (it now uses
  sigaltstack() to set up signal handling on another stack).

Fixes for GCC 3.2

Disabled dynamic native code generation for gcc >=3.1
added flag --dynamic

Bugfixes: compile_prim now flushes the I-cache; fixed test case

changed version
disabled dynamic code generation for now (problems with gcc-3.2).
documentation changes

gforth now stores a local ip to saved_ip instead of using a global ip
  associated changes in register allocation, threading, documentation
testdist now also tests vmgen and vmgen-ex

added regression test for signal on broken execute (not working on Linux-PPC)
new error message and code for ticking compile-only words
bug workaround for gforth binary (dynamic superinsts produced wrong code for
    ?dup-0=-branch)

cleaned up #defines of DIRECT_THREADED and INDIRECT_THREADED
introduced engine gforth-itc (indirect threading, no replication)
fixed bug in indirect threading

support for --no-super (no dynamic superinstructions) and
            --no-dynamic (use only static primitives)

lit@ and lit+ are now defined as superinstructions
compile lit @ and lit + instead of lit@ and lit+
extended prims2x to support superinstructions with non-C-names
  (syntax: forth-name /c-name = ...)
support profiling of interpreters with superinstructions
  (with simple instructions in the output).
profile output with prefixes only (enable by editing profile.c).
optional reporting of static superinstruction lengths (compared to
  dynamic superinstructions); enable by compiling with -DPRINT_SUPER_LENGTHS

eliminated searching in compile_prim (faster startup and compilation)

dynamic superinstructions for the code in the image
SEE works again for CODE words

hybrid direct/indirect threading (replaces direct threading)

bugfixes

compile_prim now converts "xts" into "threaded-code" for DOUBLY_INDIRECT,
     and reports non-primitive non-xt input

Differentiate between threaded code and xts in gforth-ditc and in gforthmi

Thrown out static vocabulary stack
Changed cross to make mixed threading workable

preparations in loader for hybrid threading

default dictionary size 1M cells
machine-specific stuff for dynamic superinstructions is now in machine.h
support for dynamic superinstructions for Alpha

gforth no longer uses TOS caching (for improved error detection, suggested by
                                   Dennis Donnelly).
The default FP stack size is 16K (for improved error detection).
minor changes.

added dynamic superinstructions (currently for direct threading on 386,
                                 without checking for other platforms)

determine primitive relocatability for dynamic superinstructions
documentation changes (from bug reports by Robert Epprecht and others)

Non-relocatable images with fixed base work again.
Base address of $100 and 0 are now special cases that indicate
a relocatable image.

Image starts at $100 for cross-compilation now. A NULL-reference
is really a reference to address 0 and NIL is 0.
In the kernel image-header to get the headers address.
Image loading is only changed minimal: The stuff with fixed
addressed images is thrown out, because every image is relocatable.
No header change. Old images should work.

Made some Gforth-EC targets work again
Started grouping (no renumbering yet, groups are just dummies)

continued block profiling

good start at profiling for peephole optimization
backtrace now also works for calls done with CALL

rewrote large parts of prims2x.fs to become more flexible (not restricted to
  2 stacks, factored out common code for the stacks, etc.).
Changes in other files to go with the prims2x.fs changes

changed FSF address in copyright messages

updated copyright dates in many files (not in ec-related files)

bugfix (determination of whether rp is in range after the setjmp in go_forth.

The relocator now checks for unimplemented primitives used in the image.

Made Gforth compile on DOS. Cleaned up some minor parts in make binonlydist
Also renamed

unbuffered.fs -> unbuffer.fs
exceptions.fs -> except.fs
backtrace.fs -> backtrac.fs

in the CVS archive (do that in your gforth/CVS/Entries before checking out!!!)

If you want to make my life more easy, please refrain from using >8.3 file
names for vital parts.

Added some files to .cvsignore
Fixed problem when relocate() tries to relocate beyond the last image entry

Dictionary allocation should now work again on OSs that don't treat
  mmap beyond EOF nicely.

Gforth can now use mmap to read the image
small changes to reduce the warnings

Got SHARC port to comile again (untested yet)
Added GCC patch for SHARC port

Made block 0 the first block and provided OFFSET for backward compatibility
(store 1 there).
Fixed bug for negative buffers.
Added a type in write-file for OS-less ports.
Some corrections to make the SHARC port compile again.
HAS_DEBUG is now a central flag which replaces perror and fprintf if not set.

Made USE_TOS work with 386 on gcc 2.95

added engine option --appl-image and gforthmi option --application
minor objects.fs changes

fixed bug (path handling when opening image).

fixed a Cell==int assumption

Just some nice casts.

gforth (non-fast) now uses a plain threading scheme, making error
  positions more accurate
minor changes

On platforms that support the SA_SIGINFO flag (Linux is not quite
  there yet), there are now new signal handlers that perform more
  precise error reporting (in particular for SIGFPE and SIGSEGV).
Minor fixes.

moved signal handling into a new file signals.c
minor changes to window size handling

Minor fixes for SHARC
KEY returns now EOF (-1) when at end of file (non-standard?)

factored out print_sizes

bugfix

Improved dump information for magic
Added code in cross.fs to support new magic
Fixed makefile problem with cp (cp -p to preserve date)

there is now a debugging version of the engine that maintains ip and
rp in global variables (to allow backtrace on signals). The debugging
engine is called gforth and the original engine is called gforth-fast.

fixed --debug bug (printed non-printing characters (magic[7])).
fixed buglet in configure.in (now heeds LIBS even for hppa).
set version to snapshot

Install process fixes for DOS and OS/2
EC primitive count fixed
EC relocate problem fixed

Changed format to Gforth2x (0.4 or greater)

Changes to make gforth run on DOS and Win32 (I hope it's not broken by the
latest Makefile.in changes)
Some minor tweaks

Added some documentation (files stup, blocks stub, Mini-OOF implementation)
Added Benchres for my machine
made DOS and Win32 compile and run
New gforthmi.bat script for DOS - needs a temporary file for the commands
instead of the -e option.
Added select.o again for DOS (DJGPP's select is broken wrt timing)
Improved select.c
Bug with DOS: engine-ditc doesn't compile with optimization on. Maybe I need
to get a new GCC version for DOS?

updated dates in copyright messages
inserted copyright messages in most files that did not have them
removed outdated files engine/32bit.h engine/strsig.c

Added system-specific options for C-part of primitives

Bug fix: print cause for problems in command line
Cleaned up Makefiles
Adjusted configures for DOS and OS/2 (not tested yet)
New feature: link image into executable (for non-OS systems)

Default path now has "." in front
Worked a bit on the documentation
fixed handling of "." in open-path-file (now also works with "." from the path)

various changes in configure.in and the Makefile.in's:
   symlinks kernel.fi and arch/machine eliminated
   utilize some automake stuff (e.g., @VERSION@ and @lispdir@)
   eliminated engine/version.h
   eliminated replace.a; gforth[-ditc] is now built by engine/Makefile
   getopt_long replacement is now donw through LIBOBJS, like everything else
automake-induced changes in acconfig.h; added acinclude.m4 and AUTHORS.
aclocal.m4 is now generated (it's still in CVS because there's not yet a
	make rule for it).
made gforthmi more conformant with GNU standards (--version etc.).
prim: fixed portability bug (RTLD_GLOBAL does not exist everywhere)
arch/power/machine.h: refined cache flush code
doc/gforth.ds now uses @VERSION{} (but we need to generate version.texi)
	documented new buffering behaviour

emacs update bug fix in gforth.el
bug fix in struct.fs, compat/struct.fs
main.c: stdout is now unbuffered, if it is a tty

Changed offset for relocation

Dos Changes:
!!Moved:
engine/32bit.h moved to arch/generic/machine.h
cache.c moved to arch/hppa/
cleanalign.c moved to engine/
!!New Files:
envos.* *.bsh
Instead of linking arch/xy/machine.h an custom include path is used
engine.a engine.o main.o makes with engine/Makefile now
history.fs and startup.fs works for dos and unix now
environment variable osclass is set in envos.fs, wich is os-specific
any comments?
supressed some warnings in main.c

Added documentation for structures and object.fs
Changed representation of structures from "size align" to "align size",
   and renamed 1 cells: to cell% etc.
added %size and %alignment
fixed search bug
added command-line option --die-on-signal

now uses mmap for the stacks and malloc for dict on MIPS with direct threading

Some bug fixes.

jwilke's changes:
Moved many files to other directories
renamed many files
other changes unknown to me.

	Diffs between Use Text Field v0-7-0 v0-6-2 v0-6-1 v0-6-0 v0-5-0 v0-4-0 MAIN HEAD and Use Text Field v0-7-0 v0-6-2 v0-6-1 v0-6-0 v0-5-0 v0-4-0 MAIN HEAD
	Generate output suitable for use with a patch program
	Type of Diff should be a Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

---

View only Branch: Show all branches MAIN

---

Sort log by: Not sorted Commit date Revision

---

CVS Admin Powered by ViewCVS 1.0-dev

ViewCVS and CVS Help