File:  [gforth] / gforth / doc / vmgen.texi
Revision 1.10: download - view: text, annotated - select for diffs
Tue Aug 13 19:33:38 2002 UTC (17 years, 3 months ago) by anton
Branches: MAIN
CVS tags: HEAD
documentation changes

    1: \input texinfo    @c -*-texinfo-*-
    2: @comment %**start of header
    3: @setfilename vmgen.info
    4: @include version.texi
    5: @settitle Vmgen (Gforth @value{VERSION})
    6: @c @syncodeindex pg cp
    7: @comment %**end of header
    8: @copying
    9: This manual is for Vmgen
   10: (version @value{VERSION}, @value{UPDATED}),
   11: the virtual machine interpreter generator
   12: 
   13: Copyright @copyright{} 2002 Free Software Foundation, Inc.
   14: 
   15: @quotation
   16: Permission is granted to copy, distribute and/or modify this document
   17: under the terms of the GNU Free Documentation License, Version 1.1 or
   18: any later version published by the Free Software Foundation; with no
   19: Invariant Sections, with the Front-Cover texts being ``A GNU Manual,''
   20: and with the Back-Cover Texts as in (a) below.  A copy of the
   21: license is included in the section entitled ``GNU Free Documentation
   22: License.''
   23: 
   24: (a) The FSF's Back-Cover Text is: ``You have freedom to copy and modify
   25: this GNU Manual, like GNU software.  Copies published by the Free
   26: Software Foundation raise funds for GNU development.''
   27: @end quotation
   28: @end copying
   29: 
   30: @dircategory GNU programming tools
   31: @direntry
   32: * vmgen: (vmgen).               Interpreter generator
   33: @end direntry
   34: 
   35: @titlepage
   36: @title Vmgen
   37: @subtitle for Gforth version @value{VERSION}, @value{UPDATED}
   38: @author M. Anton Ertl (@email{anton@mips.complang.tuwien.ac.at})
   39: @page
   40: @vskip 0pt plus 1filll
   41: @insertcopying
   42: @end titlepage
   43: 
   44: @contents
   45: 
   46: @ifnottex
   47: @node Top, Introduction, (dir), (dir)
   48: @top Vmgen
   49: 
   50: @insertcopying
   51: @end ifnottex
   52: 
   53: @menu
   54: * Introduction::                What can Vmgen do for you?
   55: * Why interpreters?::           Advantages and disadvantages
   56: * Concepts::                    VM interpreter background
   57: * Invoking vmgen::              
   58: * Example::                     
   59: * Input File Format::           
   60: * Using the generated code::    
   61: * Changes::                     from earlier versions
   62: * Contact::                     Bug reporting etc.
   63: * Copying This Manual::         Manual License
   64: * Index::                       
   65: 
   66: @detailmenu
   67:  --- The Detailed Node Listing ---
   68: 
   69: Concepts
   70: 
   71: * Front end and VM interpreter::  Modularizing an interpretive system
   72: * Data handling::               Stacks, registers, immediate arguments
   73: * Dispatch::                    From one VM instruction to the next
   74: 
   75: Example
   76: 
   77: * Example overview::            
   78: * Using profiling to create superinstructions::  
   79: 
   80: Input File Format
   81: 
   82: * Input File Grammar::          
   83: * Simple instructions::         
   84: * Superinstructions::           
   85: 
   86: Simple instructions
   87: 
   88: * C Code Macros::               Macros recognized by Vmgen
   89: * C Code restrictions::         Vmgen makes assumptions about C code
   90: 
   91: Using the generated code
   92: 
   93: * VM engine::                   Executing VM code
   94: * VM instruction table::        
   95: * VM code generation::          Creating VM code (in the front-end)
   96: * Peephole optimization::       Creating VM superinstructions
   97: * VM disassembler::             for debugging the front end
   98: * VM profiler::                 for finding worthwhile superinstructions
   99: 
  100: Copying This Manual
  101: 
  102: * GNU Free Documentation License::  License for copying this manual.
  103: 
  104: @end detailmenu
  105: @end menu
  106: 
  107: @c @ifnottex
  108: This file documents Vmgen (Gforth @value{VERSION}).
  109: 
  110: @c ************************************************************
  111: @node Introduction, Why interpreters?, Top, Top
  112: @chapter Introduction
  113: 
  114: Vmgen is a tool for writing efficient interpreters.  It takes a simple
  115: virtual machine description and generates efficient C code for dealing
  116: with the virtual machine code in various ways (in particular, executing
  117: it).  The run-time efficiency of the resulting interpreters is usually
  118: within a factor of 10 of machine code produced by an optimizing
  119: compiler.
  120: 
  121: The interpreter design strategy supported by vmgen is to divide the
  122: interpreter into two parts:
  123: 
  124: @itemize @bullet
  125: 
  126: @item The @emph{front end} takes the source code of the language to be
  127: implemented, and translates it into virtual machine code.  This is
  128: similar to an ordinary compiler front end; typically an interpreter
  129: front-end performs no optimization, so it is relatively simple to
  130: implement and runs fast.
  131: 
  132: @item The @emph{virtual machine interpreter} executes the virtual
  133: machine code.
  134: 
  135: @end itemize
  136: 
  137: Such a division is usually used in interpreters, for modularity as well
  138: as for efficiency.  The virtual machine code is typically passed between
  139: front end and virtual machine interpreter in memory, like in a
  140: load-and-go compiler; this avoids the complexity and time cost of
  141: writing the code to a file and reading it again.
  142: 
  143: A @emph{virtual machine} (VM) represents the program as a sequence of
  144: @emph{VM instructions}, following each other in memory, similar to real
  145: machine code.  Control flow occurs through VM branch instructions, like
  146: in a real machine.
  147: 
  148: In this setup, vmgen can generate most of the code dealing with virtual
  149: machine instructions from a simple description of the virtual machine
  150: instructions (@pxref...), in particular:
  151: 
  152: @table @emph
  153: 
  154: @item VM instruction execution
  155: 
  156: @item VM code generation
  157: Useful in the front end.
  158: 
  159: @item VM code decompiler
  160: Useful for debugging the front end.
  161: 
  162: @item VM code tracing
  163: Useful for debugging the front end and the VM interpreter.  You will
  164: typically provide other means for debugging the user's programs at the
  165: source level.
  166: 
  167: @item VM code profiling
  168: Useful for optimizing the VM insterpreter with superinstructions
  169: (@pxref...).
  170: 
  171: @end table
  172: 
  173: VMgen supports efficient interpreters though various optimizations, in
  174: particular
  175: 
  176: @itemize
  177: 
  178: @item Threaded code
  179: 
  180: @item Caching the top-of-stack in a register
  181: 
  182: @item Combining VM instructions into superinstructions
  183: 
  184: @item
  185: Replicating VM (super)instructions for better BTB prediction accuracy
  186: (not yet in vmgen-ex, but already in Gforth).
  187: 
  188: @end itemize
  189: 
  190: As a result, vmgen-based interpreters are only about an order of
  191: magintude slower than native code from an optimizing C compiler on small
  192: benchmarks; on large benchmarks, which spend more time in the run-time
  193: system, the slowdown is often less (e.g., the slowdown of a
  194: Vmgen-generated JVM interpreter over the best JVM JIT compiler we
  195: measured is only a factor of 2-3 for large benchmarks; some other JITs
  196: and all other interpreters we looked at were slower than our
  197: interpreter).
  198: 
  199: VMs are usually designed as stack machines (passing data between VM
  200: instructions on a stack), and vmgen supports such designs especially
  201: well; however, you can also use vmgen for implementing a register VM and
  202: still benefit from most of the advantages offered by vmgen.
  203: 
  204: There are many potential uses of the instruction descriptions that are
  205: not implemented at the moment, but we are open for feature requests, and
  206: we will implement new features if someone asks for them; so the feature
  207: list above is not exhaustive.
  208: 
  209: @c *********************************************************************
  210: @node Why interpreters?, Concepts, Introduction, Top
  211: @chapter Why interpreters?
  212: 
  213: Interpreters are a popular language implementation technique because
  214: they combine all three of the following advantages:
  215: 
  216: @itemize
  217: 
  218: @item Ease of implementation
  219: 
  220: @item Portability
  221: 
  222: @item Fast edit-compile-run cycle
  223: 
  224: @end itemize
  225: 
  226: The main disadvantage of interpreters is their run-time speed.  However,
  227: there are huge differences between different interpreters in this area:
  228: the slowdown over optimized C code on programs consisting of simple
  229: operations is typically a factor of 10 for the more efficient
  230: interpreters, and a factor of 1000 for the less efficient ones (the
  231: slowdown for programs executing complex operations is less, because the
  232: time spent in libraries for executing complex operations is the same in
  233: all implementation strategies).
  234: 
  235: Vmgen makes it even easier to implement interpreters.  It also supports
  236: techniques for building efficient interpreters.
  237: 
  238: @c ********************************************************************
  239: @node Concepts, Invoking vmgen, Why interpreters?, Top
  240: @chapter Concepts
  241: 
  242: @menu
  243: * Front end and VM interpreter::  Modularizing an interpretive system
  244: * Data handling::               Stacks, registers, immediate arguments
  245: * Dispatch::                    From one VM instruction to the next
  246: @end menu
  247: 
  248: @c --------------------------------------------------------------------
  249: @node Front end and VM interpreter, Data handling, Concepts, Concepts
  250: @section Front end and VM interpreter
  251: 
  252: @cindex front-end
  253: Interpretive systems are typically divided into a @emph{front end} that
  254: parses the input language and produces an intermediate representation
  255: for the program, and an interpreter that executes the intermediate
  256: representation of the program.
  257: 
  258: @cindex virtual machine
  259: @cindex VM
  260: @cindex instruction, VM
  261: For efficient interpreters the intermediate representation of choice is
  262: virtual machine code (rather than, e.g., an abstract syntax tree).
  263: @emph{Virtual machine} (VM) code consists of VM instructions arranged
  264: sequentially in memory; they are executed in sequence by the VM
  265: interpreter, except for VM branch instructions, which implement control
  266: structures.  The conceptual similarity to real machine code results in
  267: the name @emph{virtual machine}.
  268: 
  269: In this framework, vmgen supports building the VM interpreter and any
  270: other component dealing with VM instructions.  It does not have any
  271: support for the front end, apart from VM code generation support.  The
  272: front end can be implemented with classical compiler front-end
  273: techniques, supported by tools like @command{flex} and @command{bison}.
  274: 
  275: The intermediate representation is usually just internal to the
  276: interpreter, but some systems also support saving it to a file, either
  277: as an image file, or in a full-blown linkable file format (e.g., JVM).
  278: Vmgen currently has no special support for such features, but the
  279: information in the instruction descriptions can be helpful, and we are
  280: open for feature requests and suggestions.
  281: 
  282: @c --------------------------------------------------------------------
  283: @node Data handling, Dispatch, Front end and VM interpreter, Concepts
  284: @section Data handling
  285: 
  286: @cindex stack machine
  287: @cindex register machine
  288: Most VMs use one or more stacks for passing temporary data between VM
  289: instructions.  Another option is to use a register machine architecture
  290: for the virtual machine; however, this option is either slower or
  291: significantly more complex to implement than a stack machine architecture.
  292: 
  293: Vmgen has special support and optimizations for stack VMs, making their
  294: implementation easy and efficient.
  295: 
  296: You can also implement a register VM with vmgen (@pxref{Register
  297: Machines}), and you will still profit from most vmgen features.
  298: 
  299: @cindex stack item size
  300: @cindex size, stack items
  301: Stack items all have the same size, so they typically will be as wide as
  302: an integer, pointer, or floating-point value.  Vmgen supports treating
  303: two consecutive stack items as a single value, but anything larger is
  304: best kept in some other memory area (e.g., the heap), with pointers to
  305: the data on the stack.
  306: 
  307: @cindex instruction stream
  308: @cindex immediate arguments
  309: Another source of data is immediate arguments VM instructions (in the VM
  310: instruction stream).  The VM instruction stream is handled similar to a
  311: stack in vmgen.
  312: 
  313: @cindex garbage collection
  314: @cindex reference counting
  315: Vmgen has no built-in support for nor restrictions against @emph{garbage
  316: collection}.  If you need garbage collection, you need to provide it in
  317: your run-time libraries.  Using @emph{reference counting} is probably
  318: harder, but might be possible (contact us if you are interested).
  319: @c reference counting might be possible by including counting code in 
  320: @c the conversion macros.
  321: 
  322: @c --------------------------------------------------------------------
  323: @node Dispatch,  , Data handling, Concepts
  324: @section Dispatch
  325: 
  326: Understanding this section is probably not necessary for using vmgen,
  327: but it may help.  You may want to skip it now, and read it if you find statements about dispatch methods confusing.
  328: 
  329: After executing one VM instruction, the VM interpreter has to dispatch
  330: the next VM instruction (vmgen calls the dispatch routine @samp{NEXT}).
  331: Vmgen supports two methods of dispatch:
  332: 
  333: @table
  334: 
  335: @item switch dispatch
  336: In this method the VM interpreter contains a giant @code{switch}
  337: statement, with one @code{case} for each VM instruction.  The VM
  338: instructions are represented by integers (e.g., produced by an
  339: @code{enum}) in the VM code, and dipatch occurs by loading the next
  340: integer from the VM code, @code{switch}ing on it, and continuing at the
  341: appropriate @code{case}; after executing the VM instruction, jump back
  342: to the dispatch code.
  343: 
  344: @item threaded code
  345: This method represents a VM instruction in the VM code by the address of
  346: the start of the machine code fragment for executing the VM instruction.
  347: Dispatch consists of loading this address, jumping to it, and
  348: incrementing the VM instruction pointer.  Typically the threaded-code
  349: dispatch code is appended directly to the code for executing the VM
  350: instruction.  Threaded code cannot be implemented in ANSI C, but it can
  351: be implemented using GNU C's labels-as-values extension (@pxref{labels
  352: as values}).
  353: 
  354: @end table
  355: 
  356: @c *************************************************************
  357: @node Invoking vmgen, Example, Concepts, Top
  358: @chapter Invoking vmgen
  359: 
  360: The usual way to invoke vmgen is as follows:
  361: 
  362: @example
  363: vmgen @var{infile}
  364: @end example
  365: 
  366: Here @var{infile} is the VM instruction description file, which usually
  367: ends in @file{.vmg}.  The output filenames are made by taking the
  368: basename of @file{infile} (i.e., the output files will be created in the
  369: current working directory) and replacing @file{.vmg} with @file{-vm.i},
  370: @file{-disasm.i}, @file{-gen.i}, @file{-labels.i}, @file{-profile.i},
  371: and @file{-peephole.i}.  E.g., @command{bison hack/foo.vmg} will create
  372: @file{foo-vm.i} etc.
  373: 
  374: The command-line options supported by vmgen are
  375: 
  376: @table @option
  377: 
  378: @cindex -h, command-line option
  379: @cindex --help, command-line option
  380: @item --help
  381: @itemx -h
  382: Print a message about the command-line options
  383: 
  384: @cindex -v, command-line option
  385: @cindex --version, command-line option
  386: @item --version
  387: @itemx -v
  388: Print version and exit
  389: @end table
  390: 
  391: @c env vars GFORTHDIR GFORTHDATADIR
  392: 
  393: @c ****************************************************************
  394: @node Example, Input File Format, Invoking vmgen, Top
  395: @chapter Example
  396: 
  397: @menu
  398: * Example overview::            
  399: * Using profiling to create superinstructions::  
  400: @end menu
  401: 
  402: @c --------------------------------------------------------------------
  403: @node Example overview, Using profiling to create superinstructions, Example, Example
  404: @section Example overview
  405: 
  406: There are two versions of the same example for using vmgen:
  407: @file{vmgen-ex} and @file{vmgen-ex2} (you can also see Gforth as
  408: example, but it uses additional (undocumented) features, and also
  409: differs in some other respects).  The example implements @emph{mini}, a
  410: tiny Modula-2-like language with a small JavaVM-like virtual machine.
  411: The difference between the examples is that @file{vmgen-ex} uses many
  412: casts, and @file{vmgen-ex2} tries to avoids most casts and uses unions
  413: instead.
  414: 
  415: The files provided with each example are:
  416: 
  417: @example
  418: Makefile
  419: README
  420: disasm.c           wrapper file
  421: engine.c           wrapper file
  422: peephole.c         wrapper file
  423: profile.c          wrapper file
  424: mini-inst.vmg      simple VM instructions
  425: mini-super.vmg     superinstructions (empty at first)
  426: mini.h             common declarations
  427: mini.l             scanner
  428: mini.y             front end (parser, VM code generator)
  429: support.c          main() and other support functions
  430: fib.mini           example mini program
  431: simple.mini        example mini program
  432: test.mini          example mini program (tests everything)
  433: test.out           test.mini output
  434: stat.awk           script for aggregating profile information
  435: peephole-blacklist list of instructions not allowed in superinstructions
  436: seq2rule.awk       script for creating superinstructions
  437: @end example
  438: 
  439: For your own interpreter, you would typically copy the following files
  440: and change little, if anything:
  441: 
  442: @example
  443: disasm.c           wrapper file
  444: engine.c           wrapper file
  445: peephole.c         wrapper file
  446: profile.c          wrapper file
  447: stat.awk           script for aggregating profile information
  448: seq2rule.awk       script for creating superinstructions
  449: @end example
  450: 
  451: You would typically change much in or replace the following files:
  452: 
  453: @example
  454: Makefile
  455: mini-inst.vmg      simple VM instructions
  456: mini.h             common declarations
  457: mini.l             scanner
  458: mini.y             front end (parser, VM code generator)
  459: support.c          main() and other support functions
  460: peephole-blacklist list of instructions not allowed in superinstructions
  461: @end example
  462: 
  463: You can build the example by @code{cd}ing into the example's directory,
  464: and then typing @samp{make}; you can check that it works with @samp{make
  465: check}.  You can run run mini programs like this:
  466: 
  467: @example
  468: ./mini fib.mini
  469: @end example
  470: 
  471: To learn about the options, type @samp{./mini -h}.
  472: 
  473: @c --------------------------------------------------------------------
  474: @node Using profiling to create superinstructions,  , Example overview, Example
  475: @section Using profiling to create superinstructions
  476: 
  477: I have not added rules for this in the @file{Makefile} (there are many
  478: options for selecting superinstructions, and I did not want to hardcode
  479: one into the @file{Makefile}), but there are some supporting scripts, and
  480: here's an example:
  481: 
  482: Suppose you want to use @file{fib.mini} and @file{test.mini} as training
  483: programs, you get the profiles like this:
  484: 
  485: @example
  486: make fib.prof test.prof #takes a few seconds
  487: @end example
  488: 
  489: You can aggregate these profiles with @file{stat.awk}:
  490: 
  491: @example
  492: awk -f stat.awk fib.prof test.prof
  493: @end example
  494: 
  495: The result contains lines like:
  496: 
  497: @example
  498:       2      16        36910041 loadlocal lit
  499: @end example
  500: 
  501: This means that the sequence @code{loadlocal lit} statically occurs a
  502: total of 16 times in 2 profiles, with a dynamic execution count of
  503: 36910041.
  504: 
  505: The numbers can be used in various ways to select superinstructions.
  506: E.g., if you just want to select all sequences with a dynamic
  507: execution count exceeding 10000, you would use the following pipeline:
  508: 
  509: @example
  510: awk -f stat.awk fib.prof test.prof|
  511: awk '$3>=10000'|                #select sequences
  512: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
  513: awk -f seq2rule.awk|      #transform sequences into superinstruction rules
  514: sort -k 3 >mini-super.vmg       #sort sequences
  515: @end example
  516: 
  517: The file @file{peephole-blacklist} contains all instructions that
  518: directly access a stack or stack pointer (for mini: @code{call},
  519: @code{return}); the sort step is necessary to ensure that prefixes
  520: preceed larger superinstructions.
  521: 
  522: Now you can create a version of mini with superinstructions by just
  523: saying @samp{make}
  524: 
  525: 
  526: @c ***************************************************************
  527: @node Input File Format, Using the generated code, Example, Top
  528: @chapter Input File Format
  529: 
  530: Vmgen takes as input a file containing specifications of virtual machine
  531: instructions.  This file usually has a name ending in @file{.vmg}.
  532: 
  533: Most examples are taken from the example in @file{vmgen-ex}.
  534: 
  535: @menu
  536: * Input File Grammar::          
  537: * Simple instructions::         
  538: * Superinstructions::           
  539: @end menu
  540: 
  541: @c --------------------------------------------------------------------
  542: @node Input File Grammar, Simple instructions, Input File Format, Input File Format
  543: @section Input File Grammar
  544: 
  545: The grammar is in EBNF format, with @code{@var{a}|@var{b}} meaning
  546: ``@var{a} or @var{b}'', @code{@{@var{c}@}} meaning 0 or more repetitions
  547: of @var{c} and @code{[@var{d}]} meaning 0 or 1 repetitions of @var{d}.
  548: 
  549: Vmgen input is not free-format, so you have to take care where you put
  550: spaces and especially newlines; it's not as bad as makefiles, though:
  551: any sequence of spaces and tabs is equivalent to a single space.
  552: 
  553: @example
  554: description: {instruction|comment|eval-escape}
  555: 
  556: instruction: simple-inst|superinst
  557: 
  558: simple-inst: ident " (" stack-effect " )" newline c-code newline newline
  559: 
  560: stack-effect: {ident} " --" {ident}
  561: 
  562: super-inst: ident " =" ident {ident}  
  563: 
  564: comment:      "\ "  text newline
  565: 
  566: eval-escape:  "\e " text newline
  567: @end example
  568: @c \+ \- \g \f \c
  569: 
  570: Note that the @code{\}s in this grammar are meant literally, not as
  571: C-style encodings for non-printable characters.
  572: 
  573: The C code in @code{simple-inst} must not contain empty lines (because
  574: vmgen would mistake that as the end of the simple-inst.  The text in
  575: @code{comment} and @code{eval-escape} must not contain a newline.
  576: @code{Ident} must conform to the usual conventions of C identifiers
  577: (otherwise the C compiler would choke on the vmgen output).
  578: 
  579: Vmgen understands a few extensions beyond the grammar given here, but
  580: these extensions are only useful for building Gforth.  You can find a
  581: description of the format used for Gforth in @file{prim}.
  582: 
  583: @subsection Eval escapes
  584: @c woanders?
  585: The text in @code{eval-escape} is Forth code that is evaluated when
  586: vmgen reads the line.  If you do not know (and do not want to learn)
  587: Forth, you can build the text according to the following grammar; these
  588: rules are normally all Forth you need for using vmgen:
  589: 
  590: @example
  591: text: stack-decl|type-prefix-decl|stack-prefix-decl
  592: 
  593: stack-decl: "stack " ident ident ident
  594: type-prefix-decl: 
  595:     's" ' string '" ' ("single"|"double") ident "type-prefix" ident
  596: stack-prefix-decl:  ident "stack-prefix" string
  597: @end example
  598: 
  599: Note that the syntax of this code is not checked thoroughly (there are
  600: many other Forth program fragments that could be written there).
  601: 
  602: If you know Forth, the stack effects of the non-standard words involved
  603: are:
  604: 
  605: @example
  606: stack        ( "name" "pointer" "type" -- )
  607:              ( name execution: -- stack )
  608: type-prefix  ( addr u xt1 xt2 n stack "prefix" -- )
  609: single       ( -- xt1 xt2 n )
  610: double       ( -- xt1 xt2 n )
  611: stack-prefix ( stack "prefix" -- )
  612: @end example
  613: 
  614: 
  615: @c --------------------------------------------------------------------
  616: @node Simple instructions, Superinstructions, Input File Grammar, Input File Format
  617: @section Simple instructions
  618: 
  619: We will use the following simple VM instruction description as example:
  620: 
  621: @example
  622: sub ( i1 i2 -- i )
  623: i = i1-i2;
  624: @end example
  625: 
  626: The first line specifies the name of the VM instruction (@code{sub}) and
  627: its stack effect (@code{i1 i2 -- i}).  The rest of the description is
  628: just plain C code.
  629: 
  630: @cindex stack effect
  631: The stack effect specifies that @code{sub} pulls two integers from the
  632: data stack and puts them in the C variables @code{i1} and @code{i2} (with
  633: the rightmost item (@code{i2}) taken from the top of stack) and later
  634: pushes one integer (@code{i)) on the data stack (the rightmost item is
  635: on the top afterwards).
  636: 
  637: How do we know the type and stack of the stack items?  Vmgen uses
  638: prefixes, similar to Fortran; in contrast to Fortran, you have to
  639: define the prefix first:
  640: 
  641: @example
  642: \E s" Cell"   single data-stack type-prefix i
  643: @end example
  644: 
  645: This defines the prefix @code{i} to refer to the type @code{Cell}
  646: (defined as @code{long} in @file{mini.h}) and, by default, to the
  647: @code{data-stack}.  It also specifies that this type takes one stack
  648: item (@code{single}).  The type prefix is part of the variable name.
  649: 
  650: Before we can use @code{data-stack} in this way, we have to define it:
  651: 
  652: @example
  653: \E stack data-stack sp Cell
  654: @end example
  655: @c !! use something other than Cell
  656: 
  657: This line defines the stack @code{data-stack}, which uses the stack
  658: pointer @code{sp}, and each item has the basic type @code{Cell}; other
  659: types have to fit into one or two @code{Cell}s (depending on whether the
  660: type is @code{single} or @code{double} wide), and are converted from and
  661: to Cells on accessing the @code{data-stack) with conversion macros
  662: (@pxref{Conversion macros}).  Stacks grow towards lower addresses in
  663: vmgen-erated interpreters.
  664: 
  665: We can override the default stack of a stack item by using a stack
  666: prefix.  E.g., consider the following instruction:
  667: 
  668: @example
  669: lit ( #i -- i )
  670: @end example
  671: 
  672: The VM instruction @code{lit} takes the item @code{i} from the
  673: instruction stream (indicated by the prefix @code{#}), and pushes it on
  674: the (default) data stack.  The stack prefix is not part of the variable
  675: name.  Stack prefixes are defined like this:
  676: 
  677: @example
  678: \E inst-stream stack-prefix #
  679: @end example
  680: 
  681: This definition defines that the stack prefix @code{#} specifies the
  682: ``stack'' @code{inst-stream}.  Since the instruction stream behaves a
  683: little differently than an ordinary stack, it is predefined, and you do
  684: not need to define it.
  685: 
  686: The instruction stream contains instructions and their immediate
  687: arguments, so specifying that an argument comes from the instruction
  688: stream indicates an immediate argument.  Of course, instruction stream
  689: arguments can only appear to the left of @code{--} in the stack effect.
  690: If there are multiple instruction stream arguments, the leftmost is the
  691: first one (just as the intuition suggests).
  692: 
  693: @menu
  694: * C Code Macros::               Macros recognized by Vmgen
  695: * C Code restrictions::         Vmgen makes assumptions about C code
  696: @end menu
  697: 
  698: @c --------------------------------------------------------------------
  699: @node C Code Macros, C Code restrictions, Simple instructions, Simple instructions
  700: @subsection C Code Macros
  701: 
  702: Vmgen recognizes the following strings in the C code part of simple
  703: instructions:
  704: 
  705: @table @samp
  706: 
  707: @item SET_IP
  708: As far as vmgen is concerned, a VM instruction containing this ends a VM
  709: basic block (used in profiling to delimit profiled sequences).  On the C
  710: level, this also sets the instruction pointer.
  711: 
  712: @item SUPER_END
  713: This ends a basic block (for profiling), without a SET_IP.
  714: 
  715: @item TAIL;
  716: Vmgen replaces @samp{TAIL;} with code for ending a VM instruction and
  717: dispatching the next VM instruction.  This happens automatically when
  718: control reaches the end of the C code.  If you want to have this in the
  719: middle of the C code, you need to use @samp{TAIL;}.  A typical example
  720: is a conditional VM branch:
  721: 
  722: @example
  723: if (branch_condition) {
  724:   SET_IP(target); TAIL;
  725: }
  726: /* implicit tail follows here */
  727: @end example
  728: 
  729: In this example, @samp{TAIL;} is not strictly necessary, because there
  730: is another one implicitly after the if-statement, but using it improves
  731: branch prediction accuracy slightly and allows other optimizations.
  732: 
  733: @item SUPER_CONTINUE
  734: This indicates that the implicit tail at the end of the VM instruction
  735: dispatches the sequentially next VM instruction even if there is a
  736: @code{SET_IP} in the VM instruction.  This enables an optimization that
  737: is not yet implemented in the vmgen-ex code (but in Gforth).  The
  738: typical application is in conditional VM branches:
  739: 
  740: @example
  741: if (branch_condition) {
  742:   SET_IP(target); TAIL; /* now this TAIL is necessary */
  743: }
  744: SUPER_CONTINUE;
  745: @end example
  746: 
  747: @end table
  748: 
  749: Note that vmgen is not smart about C-level tokenization, comments,
  750: strings, or conditional compilation, so it will interpret even a
  751: commented-out SUPER_END as ending a basic block (or, e.g.,
  752: @samp{RETAIL;} as @samp{TAIL;}).  Conversely, vmgen requires the literal
  753: presence of these strings; vmgen will not see them if they are hiding in
  754: a C preprocessor macro.
  755: 
  756: 
  757: @c --------------------------------------------------------------------
  758: @node C Code restrictions,  , C Code Macros, Simple instructions
  759: @subsection C Code restrictions
  760: 
  761: Vmgen generates code and performs some optimizations under the
  762: assumption that the user-supplied C code does not access the stack
  763: pointers or stack items, and that accesses to the instruction pointer
  764: only occur through special macros.  In general you should heed these
  765: restrictions.  However, if you need to break these restrictions, read
  766: the following.
  767: 
  768: Accessing a stack or stack pointer directly can be a problem for several
  769: reasons: 
  770: 
  771: @itemize
  772: 
  773: @item
  774: You may cache the top-of-stack item in a local variable (that is
  775: allocated to a register).  This is the most frequent source of trouble.
  776: You can deal with it either by not using top-of-stack caching (slowdown
  777: factor 1-1.4, depending on machine), or by inserting flushing code
  778: (e.g., @samp{IF_spTOS(sp[...] = spTOS);}) at the start and reloading
  779: code (e.g., @samp{IF_spTOS(spTOS = sp[0])}) at the end of problematic C
  780: code.  Vmgen inserts a stack pointer update before the start of the
  781: user-supplied C code, so the flushing code has to use an index that
  782: corrects for that.  In the future, this flushing may be done
  783: automatically by mentioning a special string in the C code.
  784: @c sometimes flushing and/or reloading unnecessary
  785: 
  786: @item
  787: The vmgen-erated code loads the stack items from stack-pointer-indexed
  788: memory into variables before the user-supplied C code, and stores them
  789: from variables to stack-pointer-indexed memory afterwards.  If you do
  790: any writes to the stack through its stack pointer in your C code, it
  791: will not affact the variables, and your write may be overwritten by the
  792: stores after the C code.  Similarly, a read from a stack using a stack
  793: pointer will not reflect computations of stack items in the same VM
  794: instruction.
  795: 
  796: @item
  797: Superinstructions keep stack items in variables across the whole
  798: superinstruction.  So you should not include VM instructions, that
  799: access a stack or stack pointer, as components of superinstructions.
  800: 
  801: @end itemize
  802: 
  803: You should access the instruction pointer only through its special
  804: macros (@samp{IP}, @samp{SET_IP}, @samp{IPTOS}); this ensure that these
  805: macros can be implemented in several ways for best performance.
  806: @samp{IP} points to the next instruction, and @samp{IPTOS} is its
  807: contents.
  808: 
  809: 
  810: @c --------------------------------------------------------------------
  811: @node Superinstructions,  , Simple instructions, Input File Format
  812: @section Superinstructions
  813: 
  814: Note: don't invest too much work in (static) superinstructions; a future
  815: version of vmgen will support dynamic superinstructions (see Ian
  816: Piumarta and Fabio Riccardi, @cite{Optimizing Direct Threaded Code by
  817: Selective Inlining}, PLDI'98), and static superinstructions have much
  818: less benefit in that context.
  819: 
  820: Here is an example of a superinstruction definition:
  821: 
  822: @example
  823: lit_sub = lit sub
  824: @end example
  825: 
  826: @code{lit_sub} is the name of the superinstruction, and @code{lit} and
  827: @code{sub} are its components.  This superinstruction performs the same
  828: action as the sequence @code{lit} and @code{sub}.  It is generated
  829: automatically by the VM code generation functions whenever that sequence
  830: occurs, so you only need to add this definition if you want to use this
  831: superinstruction (and even that can be partially automatized,
  832: @pxref{...}).
  833: 
  834: Vmgen requires that the component instructions are simple instructions
  835: defined before superinstructions using the components.  Currently, vmgen
  836: also requires that all the subsequences at the start of a
  837: superinstruction (prefixes) must be defined as superinstruction before
  838: the superinstruction.  I.e., if you want to define a superinstruction
  839: 
  840: @example
  841: sumof5 = add add add add
  842: @end example
  843: 
  844: you first have to define
  845: 
  846: @example
  847: add ( n1 n2 -- n )
  848: n = n1+n2;
  849: 
  850: sumof3 = add add
  851: sumof4 = add add add
  852: @end example
  853: 
  854: Here, @code{sumof4} is the longest prefix of @code{sumof5}, and @code{sumof3}
  855: is the longest prefix of @code{sumof4}.
  856: 
  857: Note that vmgen assumes that only the code it generates accesses stack
  858: pointers, the instruction pointer, and various stack items, and it
  859: performs optimizations based on this assumption.  Therefore, VM
  860: instructions that change the instruction pointer should only be used as
  861: last component; a VM instruction that accesses a stack pointer should
  862: not be used as component at all.  Vmgen does not check these
  863: restrictions, they just result in bugs in your interpreter.
  864: 
  865: @c ********************************************************************
  866: @node Using the generated code, Changes, Input File Format, Top
  867: @chapter Using the generated code
  868: 
  869: The easiest way to create a working VM interpreter with vmgen is
  870: probably to start with one of the examples, and modify it for your
  871: purposes.  This chapter is just the reference manual for the macros
  872: etc. used by the generated code, the other context expected by the
  873: generated code, and what you can do with the various generated files.
  874: 
  875: @menu
  876: * VM engine::                   Executing VM code
  877: * VM instruction table::        
  878: * VM code generation::          Creating VM code (in the front-end)
  879: * Peephole optimization::       Creating VM superinstructions
  880: * VM disassembler::             for debugging the front end
  881: * VM profiler::                 for finding worthwhile superinstructions
  882: @end menu
  883: 
  884: @c --------------------------------------------------------------------
  885: @node VM engine, VM instruction table, Using the generated code, Using the generated code
  886: @section VM engine
  887: 
  888: The VM engine is the VM interpreter that executes the VM code.  It is
  889: essential for an interpretive system.
  890: 
  891: Vmgen supports two methods of VM instruction dispatch: @emph{threaded
  892: code} (fast, but gcc-specific), and @emph{switch dispatch} (slow, but
  893: portable across C compilers); you can use conditional compilation
  894: (@samp{defined(__GNUC__)}) to choose between these methods, and our
  895: example does so.
  896: 
  897: For both methods, the VM engine is contained in a C-level function.
  898: Vmgen generates most of the contents of the function for you
  899: (@file{@var{name}-vm.i}), but you have to define this function, and
  900: macros and variables used in the engine, and initialize the variables.
  901: In our example the engine function also includes
  902: @file{@var{name}-labels.i} (@pxref{VM instruction table}).
  903: 
  904: The following macros and variables are used in @file{@var{name}-vm.i}:
  905: 
  906: @table @code
  907: 
  908: @item LABEL(@var{inst_name})
  909: This is used just before each VM instruction to provide a jump or
  910: @code{switch} label (the @samp{:} is provided by vmgen).  For switch
  911: dispatch this should expand to @samp{case @var{label}}; for
  912: threaded-code dispatch this should just expand to @samp{case
  913: @var{label}}.  In either case @var{label} is usually the @var{inst_name}
  914: with some prefix or suffix to avoid naming conflicts.
  915: 
  916: @item LABEL2(@var{inst_name})
  917: This will be used for dynamic superinstructions; at the moment, this
  918: should expand to nothing.
  919: 
  920: @item NAME(@var{inst_name_string})
  921: Called on entering a VM instruction with a string containing the name of
  922: the VM instruction as parameter.  In normal execution this should be a
  923: noop, but for tracing this usually prints the name, and possibly other
  924: information (several VM registers in our example).
  925: 
  926: @item DEF_CA
  927: Usually empty.  Called just inside a new scope at the start of a VM
  928: instruction.  Can be used to define variables that should be visible
  929: during every VM instruction.  If you define this macro as non-empty, you
  930: have to provide the finishing @samp{;} in the macro.
  931: 
  932: @item NEXT_P0 NEXT_P1 NEXT_P2
  933: The three parts of instruction dispatch.  They can be defined in
  934: different ways for best performance on various processors (see
  935: @file{engine.c} in the example or @file{engine/threaded.h} in Gforth).
  936: @samp{NEXT_P0} is invoked right at the start of the VM isntruction (but
  937: after @samp{DEF_CA}), @samp{NEXT_P1} right after the user-supplied C
  938: code, and @samp{NEXT_P2} at the end.  The actual jump has to be
  939: performed by @samp{NEXT_P2}.
  940: 
  941: The simplest variant is if @samp{NEXT_P2} does everything and the other
  942: macros do nothing.  Then also related macros like @samp{IP},
  943: @samp{SET_IP}, @samp{IP}, @samp{INC_IP} and @samp{IPTOS} are very
  944: straightforward to define.  For switch dispatch this code consists just
  945: of a jump to the dispatch code (@samp{goto next_inst;} in our example;
  946: for direct threaded code it consists of something like
  947: @samp{({cfa=*ip++; goto *cfa;})}.
  948: 
  949: Pulling code (usually the @samp{cfa=*ip;}) up into @samp{NEXT_P1}
  950: usually does not cause problems, but pulling things up into
  951: @samp{NEXT_P0} usually requires changing the other macros (and, at least
  952: for Gforth on Alpha, it does not buy much, because the compiler often
  953: manages to schedule the relevant stuff up by itself).  An even more
  954: extreme variant is to pull code up even further, into, e.g., NEXT_P1 of
  955: the previous VM instruction (prefetching, useful on PowerPCs).
  956: 
  957: @item INC_IP(@var{n})
  958: This increments @code{IP} by @var{n}.
  959: 
  960: @item SET_IP(@var{target})
  961: This sets @code{IP} to @var{target}.
  962: 
  963: @item vm_@var{A}2@var{B}(a,b)
  964: Type casting macro that assigns @samp{a} (of type @var{A}) to @samp{b}
  965: (of type @var{B}).  This is mainly used for getting stack items into
  966: variables and back.  So you need to define macros for every combination
  967: of stack basic type (@code{Cell} in our example) and type-prefix types
  968: used with that stack (in both directions).  For the type-prefix type,
  969: you use the type-prefix (not the C type string) as type name (e.g.,
  970: @samp{vm_Cell2i}, not @samp{vm_Cell2Cell}).  In addition, you have to
  971: define a vm_@var{X}2@var{X} macro for the stack basic type (used in
  972: superinstructions).
  973: 
  974: The stack basic type for the predefined @samp{inst-stream} is
  975: @samp{Cell}.  If you want a stack with the same item size, making its
  976: basic type @samp{Cell} usually reduces the number of macros you have to
  977: define.
  978: 
  979: Here our examples differ a lot: @file{vmgen-ex} uses casts in these
  980: macros, whereas @file{vmgen-ex2} uses union-field selection (or
  981: assignment to union fields).
  982: 
  983: @item vm_two@var{A}2@var{B}(a1,a2,b)
  984: @item vm_@var{B}2two@var{A}(b,a1,a2)
  985: Conversions between two stack items (@code{a1}, @code{a2}) and a
  986: variable @code{b} of a type that takes two stack items.  This does not
  987: occur in our small examples, but you can look at Gforth for examples.
  988: 
  989: @item @var{stackpointer}
  990: For each stack used, the stackpointer name given in the stack
  991: declaration is used.  For a regular stack this must be an l-expression;
  992: typically it is a variable declared as a pointer to the stack's basic
  993: type.  For @samp{inst-stream}, the name is @samp{IP}, and it can be a
  994: plain r-value; typically it is a macro that abstracts away the
  995: differences between the various implementations of NEXT_P*.
  996: 
  997: @item @var{stackpointer}TOS
  998: The top-of-stack for the stack pointed to by @var{stackpointer}.  If you
  999: are using top-of-stack caching for that stack, this should be defined as
 1000: variable; if you are not using top-of-stack caching for that stack, this
 1001: should be a macro expanding to @samp{@var{stackpointer}[0]}.  The stack
 1002: pointer for the predefined @samp{inst-stream} is called @samp{IP}, so
 1003: the top-of-stack is called @samp{IPTOS}.
 1004: 
 1005: @item IF_@var{stackpointer}TOS(@var{expr})
 1006: Macro for executing @var{expr}, if top-of-stack caching is used for the
 1007: @var{stackpointer} stack.  I.e., this should do @var{expr} if there is
 1008: top-of-stack caching for @var{stackpointer}; otherwise it should do
 1009: nothing.
 1010: 
 1011: @item SUPER_END
 1012: This is used by the VM profiler (@pxref{VM profiler}); it should not do
 1013: anything in normal operation, and call @code{vm_count_block(IP)} for
 1014: profiling.
 1015: 
 1016: @item SUPER_CONTINUE
 1017: This is just a hint to vmgen and does nothing at the C level.
 1018: 
 1019: @item VM_DEBUG
 1020: If this is defined, the tracing code will be compiled in (slower
 1021: interpretation, but better debugging).  Our example compiles two
 1022: versions of the engine, a fast-running one that cannot trace, and one
 1023: with potential tracing and profiling.
 1024: 
 1025: @item vm_debug
 1026: Needed only if @samp{VM_DEBUG} is defined.  If this variable contains
 1027: true, the VM instructions produce trace output.  It can be turned on or
 1028: off at any time.
 1029: 
 1030: @item vm_out
 1031: Needed only if @samp{VM_DEBUG} is defined.  Specifies the file on which
 1032: to print the trace output (type @samp{FILE *}).
 1033: 
 1034: @item printarg_@var{type}(@var{value})
 1035: Needed only if @samp{VM_DEBUG} is defined.  Macro or function for
 1036: printing @var{value} in a way appropriate for the @var{type}.  This is
 1037: used for printing the values of stack items during tracing.  @var{Type}
 1038: is normally the type prefix specified in a @code{type-prefix} definition
 1039: (e.g., @samp{printarg_i}); in superinstructions it is currently the
 1040: basic type of the stack.
 1041: 
 1042: @end table
 1043: 
 1044: 
 1045: @c --------------------------------------------------------------------
 1046: @node VM instruction table, VM code generation, VM engine, Using the generated code
 1047: @section VM instruction table
 1048: 
 1049: For threaded code we also need to produce a table containing the labels
 1050: of all VM instructions.  This is needed for VM code generation
 1051: (@pxref{VM code generation}), and it has to be done in the engine
 1052: function, because the labels are not visible outside.  It then has to be
 1053: passed outside the function (and assigned to @samp{vm_prim}), to be used
 1054: by the VM code generation functions.
 1055: 
 1056: This means that the engine function has to be called first to produce
 1057: the VM instruction table, and later, after generating VM code, it has to
 1058: be called again to execute the generated VM code (yes, this is ugly).
 1059: In our example program, these two modes of calling the engine function
 1060: are differentiated by the value of the parameter ip0 (if it equals 0,
 1061: then the table is passed out, otherwise the VM code is executed); in our
 1062: example, we pass the table out by assigning it to @samp{vm_prim} and
 1063: returning from @samp{engine}.
 1064: 
 1065: In our example, we also build such a table for switch dispatch; this is
 1066: mainly done for uniformity.
 1067: 
 1068: For switch dispatch, we also need to define the VM instruction opcodes
 1069: used as case labels in an @code{enum}.
 1070: 
 1071: For both purposes (VM instruction table, and enum), the file
 1072: @file{@var{name}-labels.i} is generated by vmgen.  You have to define
 1073: the following macro used in this file:
 1074: 
 1075: @table @samp
 1076: 
 1077: @item INST_ADDR(@var{inst_name})
 1078: For switch dispatch, this is just the name of the switch label (the same
 1079: name as used in @samp{LABEL(@var{inst_name})}), for both uses of
 1080: @file{@var{name}-labels.i}.  For threaded-code dispatch, this is the
 1081: address of the label defined in @samp{LABEL(@var{inst_name})}); the
 1082: address is taken with @samp{&&} (@pxref{labels-as-values}).
 1083: 
 1084: @end table
 1085: 
 1086: 
 1087: @c --------------------------------------------------------------------
 1088: @node VM code generation, Peephole optimization, VM instruction table, Using the generated code
 1089: @section VM code generation
 1090: 
 1091: Vmgen generates VM code generation functions in @file{@var{name}-gen.i}
 1092: that the front end can call to generate VM code.  This is essential for
 1093: an interpretive system.
 1094: 
 1095: For a VM instruction @samp{x ( #a b #c -- d )}, vmgen generates a
 1096: function with the prototype
 1097: 
 1098: @example
 1099: void gen_x(Inst **ctp, a_type a, c_type c)
 1100: @end example
 1101: 
 1102: The @code{ctp} argument points to a pointer to the next instruction.
 1103: @code{*ctp} is increased by the generation functions; i.e., you should
 1104: allocate memory for the code to be generated beforehand, and start with
 1105: *ctp set at the start of this memory area.  Before running out of
 1106: memory, allocate a new area, and generate a VM-level jump to the new
 1107: area (this is not implemented in our examples).
 1108: 
 1109: The other arguments correspond to the immediate arguments of the VM
 1110: instruction (with their appropriate types as defined in the
 1111: @code{type_prefix} declaration.
 1112: 
 1113: The following types, variables, and functions are used in
 1114: @file{@var{name}-gen.i}:
 1115: 
 1116: @table @samp
 1117: 
 1118: @item Inst
 1119: The type of the VM instruction; if you use threaded code, this is
 1120: @code{void *}; for switch dispatch this is an integer type.
 1121: 
 1122: @item vm_prim
 1123: The VM instruction table (type: @code{Inst *}, @pxref{VM instruction table}).
 1124: 
 1125: @item gen_inst(Inst **ctp, Inst i)
 1126: This function compiles the instruction @code{i}.  Take a look at it in
 1127: @file{vmgen-ex/peephole.c}.  It is trivial when you don't want to use
 1128: superinstructions (just the last two lines of the example function), and
 1129: slightly more complicated in the example due to its ability to use
 1130: superinstructions (@pxref{Peephole optimization}).
 1131: 
 1132: @item genarg_@var{type_prefix}(Inst **ctp, @var{type} @var{type_prefix})
 1133: This compiles an immediate argument of @var{type} (as defined in a
 1134: @code{type-prefix} definition).  These functions are trivial to define
 1135: (see @file{vmgen-ex/support.c}).  You need one of these functions for
 1136: every type that you use as immediate argument.
 1137: 
 1138: @end table
 1139: 
 1140: In addition to using these functions to generate code, you should call
 1141: @code{BB_BOUNDARY} at every basic block entry point if you ever want to
 1142: use superinstructions (or if you want to use the profiling supported by
 1143: vmgen; however, this is mainly useful for selecting superinstructions).
 1144: If you use @code{BB_BOUNDARY}, you should also define it (take a look at
 1145: its definition in @file{vmgen-ex/mini.y}).
 1146: 
 1147: You do not need to call @code{BB_BOUNDARY} after branches, because you
 1148: will not define superinstructions that contain branches in the middle
 1149: (and if you did, and it would work, there would be no reason to end the
 1150: superinstruction at the branch), and because the branches announce
 1151: themselves to the profiler.
 1152: 
 1153: 
 1154: @c --------------------------------------------------------------------
 1155: @node Peephole optimization, VM disassembler, VM code generation, Using the generated code
 1156: @section Peephole optimization
 1157: 
 1158: You need peephole optimization only if you want to use
 1159: superinstructions.  But having the code for it does not hurt much if you
 1160: do not use superinstructions.
 1161: 
 1162: A simple greedy peephole optimization algorithm is used for
 1163: superinstruction selection: every time @code{gen_inst} compiles a VM
 1164: instruction, it looks if it can combine it with the last VM instruction
 1165: (which may also be a superinstruction resulting from a previous peephole
 1166: optimization); if so, it changes the last instruction to the combined
 1167: instruction instead of laying down @code{i} at the current @samp{*ctp}.
 1168: 
 1169: The code for peephole optimization is in @file{vmgen-ex/peephole.c}.
 1170: You can use this file almost verbatim.  Vmgen generates
 1171: @file{@var{file}-peephole.i} which contains data for the peephoile
 1172: optimizer.
 1173: 
 1174: You have to call @samp{init_peeptable()} after initializing
 1175: @samp{vm_prim}, and before compiling any VM code to initialize data
 1176: structures for peephole optimization.  After that, compiling with the VM
 1177: code generation functions will automatically combine VM instructions
 1178: into superinstructions.  Since you do not want to combine instructions
 1179: across VM branch targets (otherwise there will not be a proper VM
 1180: instruction to branch to), you have to call @code{BB_BOUNDARY}
 1181: (@pxref{VM code generation}) at branch targets.
 1182: 
 1183: 
 1184: @c --------------------------------------------------------------------
 1185: @node VM disassembler, VM profiler, Peephole optimization, Using the generated code
 1186: @section VM disassembler
 1187: 
 1188: A VM code disassembler is optional for an interpretive system, but
 1189: highly recommended during its development and maintenance, because it is
 1190: very useful for detecting bugs in the front end (and for distinguishing
 1191: them from VM interpreter bugs).
 1192: 
 1193: Vmgen supports VM code disassembling by generating
 1194: @file{@var{file}-disasm.i}.  This code has to be wrapped into a
 1195: function, as is done in @file{vmgen-ex/disasm.i}.  You can use this file
 1196: almost verbatim.  In addition to @samp{vm_@var{A}2@var{B}(a,b)},
 1197: @samp{vm_out}, @samp{printarg_@var{type}(@var{value})}, which are
 1198: explained above, the following macros and variables are used in
 1199: @file{@var{file}-disasm.i} (and you have to define them):
 1200: 
 1201: @table @samp
 1202: 
 1203: @item ip
 1204: This variable points to the opcode of the current VM instruction.
 1205: 
 1206: @item IP IPTOS
 1207: @samp{IPTOS} is the first argument of the current VM instruction, and
 1208: @samp{IP} points to it; this is just as in the engine, but here
 1209: @samp{ip} points to the opcode of the VM instruction (in contrast to the
 1210: engine, where @samp{ip} points to the next cell, or even one further).
 1211: 
 1212: @item VM_IS_INST(Inst i, int n)
 1213: Tests if the opcode @samp{i} is the same as the @samp{n}th entry in the
 1214: VM instruction table.
 1215: 
 1216: @end table
 1217: 
 1218: 
 1219: @c --------------------------------------------------------------------
 1220: @node VM profiler,  , VM disassembler, Using the generated code
 1221: @section VM profiler
 1222: 
 1223: The VM profiler is designed for getting execution and occurence counts
 1224: for VM instruction sequences, and these counts can then be used for
 1225: selecting sequences as superinstructions.  The VM profiler is probably
 1226: not useful as profiling tool for the interpretive system.  I.e., the VM
 1227: profiler is useful for the developers, but not the users of the
 1228: interpretive system.
 1229: 
 1230: The output of the profiler is: for each basic block (executed at least
 1231: once), it produces the dynamic execution count of that basic block and
 1232: all its subsequences; e.g.,
 1233: 
 1234: @example
 1235:        9227465  lit storelocal 
 1236:        9227465  storelocal branch 
 1237:        9227465  lit storelocal branch 
 1238: @end example
 1239: 
 1240: I.e., a basic block consisting of @samp{lit storelocal branch} is
 1241: executed 9227465 times.
 1242: 
 1243: This output can be combined in various ways.  E.g.,
 1244: @file{vmgen/stat.awk} adds up the occurences of a given sequence wrt
 1245: dynamic execution, static occurence, and per-program occurence.  E.g.,
 1246: 
 1247: @example
 1248:       2      16        36910041 loadlocal lit 
 1249: @end example
 1250: 
 1251: indicates that the sequence @samp{loadlocal lit} occurs in 2 programs,
 1252: in 16 places, and has been executed 36910041 times.  Now you can select
 1253: superinstructions in any way you like (note that compile time and space
 1254: typically limit the number of superinstructions to 100--1000).  After
 1255: you have done that, @file{vmgen/seq2rule.awk} turns lines of the form
 1256: above into rules for inclusion in a vmgen input file.  Note that this
 1257: script does not ensure that all prefixes are defined, so you have to do
 1258: that in other ways.  So, an overall script for turning profiles into
 1259: superinstructions can look like this:
 1260: 
 1261: @example
 1262: awk -f stat.awk fib.prof test.prof|
 1263: awk '$3>=10000'|                #select sequences
 1264: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
 1265: awk -f seq2rule.awk|            #turn into superinstructions
 1266: sort -k 3 >mini-super.vmg       #sort sequences
 1267: @end example
 1268: 
 1269: Here the dynamic count is used for selecting sequences (preliminary
 1270: results indicate that the static count gives better results, though);
 1271: the third line eliminats sequences containing instructions that must not
 1272: occur in a superinstruction, because they access a stack directly.  The
 1273: dynamic count selection ensures that all subsequences (including
 1274: prefixes) of longer sequences occur (because subsequences have at least
 1275: the same count as the longer sequences); the sort in the last line
 1276: ensures that longer superinstructions occur after their prefixes.
 1277: 
 1278: But before using it, you have to have the profiler.  Vmgen supports its
 1279: creation by generating @file{@var{file}-profile.i}; you also need the
 1280: wrapper file @file{vmgen-ex/profile.c} that you can use almost verbatim.
 1281: 
 1282: The profiler works by recording the targets of all VM control flow
 1283: changes (through @code{SUPER_END} during execution, and through
 1284: @code{BB_BOUNDARY} in the front end), and counting (through
 1285: @code{SUPER_END}) how often they were targeted.  After the program run,
 1286: the numbers are corrected such that each VM basic block has the correct
 1287: count (originally entering a block without executing a branch does not
 1288: increase the count), then the subsequences of all basic blocks are
 1289: printed.  To get all this, you just have to define @code{SUPER_END} (and
 1290: @code{BB_BOUNDARY}) appropriately, and call @code{vm_print_profile(FILE
 1291: *file)} when you want to output the profile on @code{file}.
 1292: 
 1293: The @file{@var{file}-profile.i} is simular to the disassembler file, and
 1294: it uses variables and functions defined in @file{vmgen-ex/profile.c},
 1295: plus @code{VM_IS_INST} already defined for the VM disassembler
 1296: (@pxref{VM disassembler}).
 1297: 
 1298: 
 1299: @c **********************************************************
 1300: @node Changes, Contact, Using the generated code, Top
 1301: @chapter Changes
 1302: 
 1303: Users of the gforth-0.5.9-20010501 version of vmgen need to change
 1304: several things in their source code to use the current version.  I
 1305: recommend keeping the gforth-0.5.9-20010501 version until you have
 1306: completed the change (note that you can have several versions of Gforth
 1307: installed at the same time).  I hope to avoid such incompatible changes
 1308: in the future.
 1309: 
 1310: The required changes are:
 1311: 
 1312: @table @code
 1313: 
 1314: @item vm_@var{A}2@var{B}
 1315: now takes two arguments.
 1316: 
 1317: @item vm_two@var{A}2@var{B}(b,a1,a2);
 1318: changed to vm_two@var{A}2@var{B}(a1,a2,b) (note the absence of the @samp{;}).
 1319: 
 1320: @end table
 1321: 
 1322: Also some new macros have to be defined, e.g., @code{INST_ADDR}, and
 1323: @code{LABEL}; some macros have to be defined in new contexts, e.g.,
 1324: @code{VM_IS_INST} is now also needed in the disassembler.
 1325: 
 1326: @node Contact, Copying This Manual, Changes, Top
 1327: @chapter Contact
 1328: 
 1329: @node Copying This Manual, Index, Contact, Top
 1330: @appendix Copying This Manual
 1331: 
 1332: @menu
 1333: * GNU Free Documentation License::  License for copying this manual.
 1334: @end menu
 1335: 
 1336: @include fdl.texi
 1337: 
 1338: 
 1339: @node Index,  , Copying This Manual, Top
 1340: @unnumbered Index
 1341: 
 1342: @printindex cp
 1343: 
 1344: @bye

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