Diff for /gforth/Attic/gforth.ds between versions 1.24 and 1.27

version 1.24, 1995/11/15 17:29:07 version 1.27, 1995/12/04 16:38:53
Line 44  Copyright @copyright{} 1995 Free Softwar Line 44  Copyright @copyright{} 1995 Free Softwar
 @center for version 0.1  @center for version 0.1
 @sp 2  @sp 2
 @center Anton Ertl  @center Anton Ertl
   @center Bernd Paysan
 @sp 3  @sp 3
 @center This manual is under construction  @center This manual is under construction
   
Line 2088  machine code), and for defining the the Line 2089  machine code), and for defining the the
 nature of Gforth poses a few problems: First of all. Gforth runs on  nature of Gforth poses a few problems: First of all. Gforth runs on
 several architectures, so it can provide no standard assembler. What's  several architectures, so it can provide no standard assembler. What's
 worse is that the register allocation not only depends on the processor,  worse is that the register allocation not only depends on the processor,
 but also on the gcc version and options used.  but also on the @code{gcc} version and options used.
   
 The words Gforth offers encapsulate some system dependences (e.g., the  The words that Gforth offers encapsulate some system dependences (e.g., the
 header structure), so a system-independent assembler may be used in  header structure), so a system-independent assembler may be used in
 Gforth. If you do not have an assembler, you can compile machine code  Gforth. If you do not have an assembler, you can compile machine code
 directly with @code{,} and @code{c,}.  directly with @code{,} and @code{c,}.
Line 2108  These words are rarely used. Therefore t Line 2109  These words are rarely used. Therefore t
 which is usually not loaded (except @code{flush-icache}, which is always  which is usually not loaded (except @code{flush-icache}, which is always
 present). You can load them with @code{require code.fs}.  present). You can load them with @code{require code.fs}.
   
   In the assembly code you will want to refer to the inner interpreter's
   registers (e.g., the data stack pointer) and you may want to use other
   registers for temporary storage. Unfortunately, the register allocation
   is installation-dependent.
   
   The easiest solution is to use explicit register declarations
   (@pxref{Explicit Reg Vars, , Variables in Specified Registers, gcc.info,
   GNU C Manual}) for all of the inner interpreter's registers: You have to
   compile Gforth with @code{-DFORCE_REG} (configure option
   @code{--enable-force-reg}) and the appropriate declarations must be
   present in the @code{machine.h} file (see @code{mips.h} for an example;
   you can find a full list of all declarable register symbols with
   @code{grep register engine.c}). If you give explicit registers to all
   variables that are declared at the beginning of @code{engine()}, you
   should be able to use the other caller-saved registers for temporary
   storage. Alternatively, you can use the @code{gcc} option
   @code{-ffixed-REG} (@pxref{Code Gen Options, , Options for Code
   Generation Conventions, gcc.info, GNU C Manual}) to reserve a register
   (however, this restriction on register allocation may slow Gforth
   significantly).
   
   If this solution is not viable (e.g., because @code{gcc} does not allow
   you to explicitly declare all the registers you need), you have to find
   out by looking at the code where the inner interpreter's registers
   reside and which registers can be used for temporary storage. You can
   get an assembly listing of the engine's code with @code{make engine.s}.
   
   In any case, it is good practice to abstract your assembly code from the
   actual register allocation. E.g., if the data stack pointer resides in
   register @code{$17}, create an alias for this register called @code{sp},
   and use that in your assembly code.
   
 Another option for implementing normal and defining words efficiently  Another option for implementing normal and defining words efficiently
 is: adding the wanted functionality to the source of Gforth. For normal  is: adding the wanted functionality to the source of Gforth. For normal
 words you just have to edit @file{primitives}, defining words (for fast  words you just have to edit @file{primitives}, defining words (for fast
 defined words) probably require changes in @file{engine.c},  defined words) may require changes in @file{engine.c},
 @file{kernal.fs}, @file{prims2x.fs}, and possibly @file{cross.fs}.  @file{kernal.fs}, @file{prims2x.fs}, and possibly @file{cross.fs}.
   
   
Line 2384  characters is determined by the locale y Line 2417  characters is determined by the locale y
   
 @item division rounding:  @item division rounding:
 installation dependent. @code{s" floored" environment? drop .}. We leave  installation dependent. @code{s" floored" environment? drop .}. We leave
 the choice to gcc (what to use for @code{/}) and to you (whether to use  the choice to @code{gcc} (what to use for @code{/}) and to you (whether to use
 @code{fm/mod}, @code{sm/rem} or simply @code{/}).  @code{fm/mod}, @code{sm/rem} or simply @code{/}).
   
 @item values of @code{STATE} when true:  @item values of @code{STATE} when true:
Line 2495  The next invocation of a parsing word re Line 2528  The next invocation of a parsing word re
 Compiles a recursive call to the defining word not to the defined word.  Compiles a recursive call to the defining word not to the defined word.
   
 @item argument input source different than current input source for @code{RESTORE-INPUT}:  @item argument input source different than current input source for @code{RESTORE-INPUT}:
 !!???If the argument input source is a valid input source then it gets  @code{-12 THROW}. Note that, once an input file is closed (e.g., because
 restored. Otherwise causes @code{-12 THROW}, which, unless caught, issues  the end of the file was reached), its source-id may be
 the message "argument type mismatch" and aborts.  reused. Therefore, restoring an input source specification referencing a
   closed file may lead to unpredictable results instead of a @code{-12
   THROW}.
   
   In the future, Gforth may be able to retore input source specifications
   from other than the current input soruce.
   
 @item data space containing definitions gets de-allocated:  @item data space containing definitions gets de-allocated:
 Deallocation with @code{allot} is not checked. This typically resuls in  Deallocation with @code{allot} is not checked. This typically resuls in
Line 2569  Not checked. As usual, you can expect me Line 2607  Not checked. As usual, you can expect me
 None.  None.
   
 @item operator's terminal facilities available:  @item operator's terminal facilities available:
 !!??  After processing the command line, Gforth goes into interactive mode,
   and you can give commands to Gforth interactively. The actual facilities
   available depend on how you invoke Gforth.
   
 @item program data space available:  @item program data space available:
 @code{sp@ here - .} gives the space remaining for dictionary and data  @code{sp@ here - .} gives the space remaining for dictionary and data
 stack together.  stack together.
   
 @item return stack space available:  @item return stack space available:
 !!??  By default 16 KBytes. The default can be overridden with the @code{-r}
   switch (@pxref{Invocation}) when Gforth starts up.
   
 @item stack space available:  @item stack space available:
 @code{sp@ here - .} gives the space remaining for dictionary and data  @code{sp@ here - .} gives the space remaining for dictionary and data
Line 2897  System dependent; @code{REPRESENT} is im Line 2938  System dependent; @code{REPRESENT} is im
 function @code{ecvt()} and inherits its behaviour in this respect.  function @code{ecvt()} and inherits its behaviour in this respect.
   
 @item rounding or truncation of floating-point numbers:  @item rounding or truncation of floating-point numbers:
 What's the question?!!  System dependent; the rounding behaviour is inherited from the hosting C
   compiler. IEEE-FP-based (i.e., most) systems by default round to
   nearest, and break ties by rounding to even (i.e., such that the last
   bit of the mantissa is 0).
   
 @item size of floating-point stack:  @item size of floating-point stack:
 @code{s" FLOATING-STACK" environment? drop .}. Can be changed at startup  @code{s" FLOATING-STACK" environment? drop .}. Can be changed at startup
Line 3608  Sieve benchmark on a 486DX2/66 than Gfor Line 3652  Sieve benchmark on a 486DX2/66 than Gfor
   
 However, this potential advantage of assembly language implementations  However, this potential advantage of assembly language implementations
 is not necessarily realized in complete Forth systems: We compared  is not necessarily realized in complete Forth systems: We compared
 Gforth (compiled with @code{gcc-2.6.3} and @code{-DFORCE_REG}) with  Gforth (direct threaded, compiled with @code{gcc-2.6.3} and
 Win32Forth 1.2093 and LMI's NT Forth (Beta, May 1994), two systems  @code{-DFORCE_REG}) with Win32Forth 1.2093, LMI's NT Forth (Beta, May
 written in assembly, and with two systems written in C: PFE-0.9.11  1994) and Eforth (with and without peephole (aka pinhole) optimization
 (compiled with @code{gcc-2.6.3} with the default configuration for  of the threaded code); all these systems were written in assembly
 Linux: @code{-O2 -fomit-frame-pointer -DUSE_REGS}) and ThisForth Beta  language. We also compared Gforth with two systems written in C:
 (compiled with gcc-2.6.3 -O3 -fomit-frame-pointer). We benchmarked  PFE-0.9.11 (compiled with @code{gcc-2.6.3} with the default
 Gforth, PFE and ThisForth on a 486DX2/66 under Linux. Kenneth O'Heskin  configuration for Linux: @code{-O2 -fomit-frame-pointer -DUSE_REGS}) and
 kindly provided the results for Win32Forth and NT Forth on a 486DX2/66  ThisForth Beta (compiled with gcc-2.6.3 -O3 -fomit-frame-pointer;
 with similar memory performance under Windows NT.  ThisForth employs peephole optimization of the threaded code). We
   benchmarked Gforth, PFE and ThisForth on a 486DX2/66 under
   Linux. Kenneth O'Heskin kindly provided the results for Win32Forth and
   NT Forth on a 486DX2/66 with similar memory performance under Windows
   NT. Marcel Hendrix ported Eforth to Linux, then extended it to run the
   benchmarks, added the peephole optimizer, ran the benchmarks and
   reported the results.
     
 We used four small benchmarks: the ubiquitous Sieve; bubble-sorting and  We used four small benchmarks: the ubiquitous Sieve; bubble-sorting and
 matrix multiplication come from the Stanford integer benchmarks and have  matrix multiplication come from the Stanford integer benchmarks and have
Line 3628  other words, it shows the speedup factor Line 3678  other words, it shows the speedup factor
 other systems).  other systems).
   
 @example  @example
 relative             Win32-        NT               This-  relative        Win32-    NT       eforth       This-
   time     Gforth     Forth     Forth       PFE     Forth  time      Gforth Forth Forth eforth  +opt   PFE Forth
 sieve        1.00      1.30      1.07      1.67      2.98  sieve       1.00  1.39  1.14   1.39  0.85  1.78  3.18
 bubble       1.00      1.30      1.40      1.66  bubble      1.00  1.33  1.43   1.51  0.89  1.70
 matmul       1.00      1.40      1.29      2.24  matmul      1.00  1.43  1.31   1.42  1.12  2.28
 fib          1.00      1.44      1.26      1.82      2.82  fib         1.00  1.55  1.36   1.24  1.15  1.97  3.04
 @end example  @end example
   
 You may find the good performance of Gforth compared with the systems  You may find the good performance of Gforth compared with the systems
Line 3645  method for relocating the Forth image: l Line 3695  method for relocating the Forth image: l
 the actual addresses at run time, resulting in two address computations  the actual addresses at run time, resulting in two address computations
 per NEXT (@pxref{System Architecture}).  per NEXT (@pxref{System Architecture}).
   
   Only Eforth with the peephole optimizer performs comparable to
   Gforth. The speedups achieved with peephole optimization of threaded
   code are quite remarkable. Adding a peephole optimizer to Gforth should
   cause similar speedups.
   
 The speedup of Gforth over PFE and ThisForth can be easily explained  The speedup of Gforth over PFE and ThisForth can be easily explained
 with the self-imposed restriction to standard C (although the measured  with the self-imposed restriction to standard C (although the measured
 implementation of PFE uses a GNU C extension: global register  implementation of PFE uses a GNU C extension: global register
Line 3659  machine registers by itself and would no Line 3714  machine registers by itself and would no
 register declarations, giving a 1.3 times slower engine (on a 486DX2/66  register declarations, giving a 1.3 times slower engine (on a 486DX2/66
 running the Sieve) than the one measured above.  running the Sieve) than the one measured above.
   
 The numbers in this section have also been published in the paper  In @cite{Translating Forth to Efficient C} by M. Anton Ertl and Martin
 @cite{Translating Forth to Efficient C} by M. Anton Ertl and Martin  Maierhofer (presented at EuroForth '95), an indirect threaded version of
 Maierhofer, presented at EuroForth '95. It is available at  Gforth is compared with Win32Forth, NT Forth, PFE, and ThisForth; that
   version of Gforth is 2\%@minus{}8\% slower on a 486 than the version
   used here. The paper available at
 @*@file{http://www.complang.tuwien.ac.at/papers/ertl&maierhofer95.ps.gz};  @*@file{http://www.complang.tuwien.ac.at/papers/ertl&maierhofer95.ps.gz};
 it also contains numbers for some native code systems. You can find  it also contains numbers for some native code systems. You can find
 numbers for Gforth on various machines in @file{Benchres}.  numbers for Gforth on various machines in @file{Benchres}.
Line 3701  VolksForth descends from F83. It was wri Line 3758  VolksForth descends from F83. It was wri
 Pennemann, Georg Rehfeld and Dietrich Weineck for the C64 (called  Pennemann, Georg Rehfeld and Dietrich Weineck for the C64 (called
 UltraForth there) in the mid-80s and ported to the Atari ST in 1986.  UltraForth there) in the mid-80s and ported to the Atari ST in 1986.
   
 Laxen and Perry wrote F83 as a model implementation of the  Hennry Laxen and Mike Perry wrote F83 as a model implementation of the
 Forth-83 standard. !! Pedigree? When?  Forth-83 standard. !! Pedigree? When?
   
 A team led by Bill Ragsdale implemented fig-Forth on many processors in  A team led by Bill Ragsdale implemented fig-Forth on many processors in

Removed from v.1.24  
changed lines
  Added in v.1.27


FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>