Annotation of gforth/doc/vmgen.texi, revision 1.29

1.10      anton       1: \input texinfo    @c -*-texinfo-*-
                      2: @comment %**start of header
                      3: @setfilename vmgen.info
1.1       anton       4: @include version.texi
1.10      anton       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: 
1.27      anton      13: Copyright @copyright{} 2002,2003,2005 Free Software Foundation, Inc.
1.10      anton      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: 
1.23      anton      30: @dircategory Software development
1.10      anton      31: @direntry
1.24      anton      32: * Vmgen: (vmgen).               Virtual machine interpreter generator
1.10      anton      33: @end direntry
                     34: 
                     35: @titlepage
                     36: @title Vmgen
                     37: @subtitle for Gforth version @value{VERSION}, @value{UPDATED}
1.11      anton      38: @author M. Anton Ertl (@email{anton@@mips.complang.tuwien.ac.at})
1.10      anton      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
1.11      anton      57: * Invoking Vmgen::              
1.10      anton      58: * Example::                     
                     59: * Input File Format::           
1.13      anton      60: * Error messages::              reported by Vmgen
1.10      anton      61: * Using the generated code::    
1.13      anton      62: * Hints::                       VM archictecture, efficiency
                     63: * The future::                  
1.10      anton      64: * Changes::                     from earlier versions
                     65: * Contact::                     Bug reporting etc.
                     66: * Copying This Manual::         Manual License
                     67: * Index::                       
                     68: 
                     69: @detailmenu
                     70:  --- The Detailed Node Listing ---
                     71: 
                     72: Concepts
                     73: 
                     74: * Front end and VM interpreter::  Modularizing an interpretive system
                     75: * Data handling::               Stacks, registers, immediate arguments
                     76: * Dispatch::                    From one VM instruction to the next
                     77: 
                     78: Example
                     79: 
                     80: * Example overview::            
                     81: * Using profiling to create superinstructions::  
                     82: 
                     83: Input File Format
                     84: 
                     85: * Input File Grammar::          
                     86: * Simple instructions::         
                     87: * Superinstructions::           
1.18      anton      88: * Store Optimization::          
1.11      anton      89: * Register Machines::           How to define register VM instructions
1.10      anton      90: 
1.17      anton      91: Input File Grammar
                     92: 
                     93: * Eval escapes::                what follows \E
                     94: 
1.10      anton      95: Simple instructions
                     96: 
1.29    ! anton      97: * Explicit stack access::       If the C code accesses a stack pointer
1.10      anton      98: * C Code Macros::               Macros recognized by Vmgen
                     99: * C Code restrictions::         Vmgen makes assumptions about C code
1.22      anton     100: * Stack growth direction::      is configurable per stack
1.10      anton     101: 
                    102: Using the generated code
                    103: 
                    104: * VM engine::                   Executing VM code
                    105: * VM instruction table::        
                    106: * VM code generation::          Creating VM code (in the front-end)
                    107: * Peephole optimization::       Creating VM superinstructions
                    108: * VM disassembler::             for debugging the front end
                    109: * VM profiler::                 for finding worthwhile superinstructions
                    110: 
1.13      anton     111: Hints
                    112: 
                    113: * Floating point::              and stacks
                    114: 
1.10      anton     115: Copying This Manual
                    116: 
                    117: * GNU Free Documentation License::  License for copying this manual.
                    118: 
                    119: @end detailmenu
                    120: @end menu
1.1       anton     121: 
                    122: @c @ifnottex
1.11      anton     123: @c This file documents Vmgen (Gforth @value{VERSION}).
1.1       anton     124: 
1.10      anton     125: @c ************************************************************
                    126: @node Introduction, Why interpreters?, Top, Top
1.2       anton     127: @chapter Introduction
1.1       anton     128: 
                    129: Vmgen is a tool for writing efficient interpreters.  It takes a simple
                    130: virtual machine description and generates efficient C code for dealing
                    131: with the virtual machine code in various ways (in particular, executing
                    132: it).  The run-time efficiency of the resulting interpreters is usually
                    133: within a factor of 10 of machine code produced by an optimizing
                    134: compiler.
                    135: 
1.11      anton     136: The interpreter design strategy supported by Vmgen is to divide the
1.1       anton     137: interpreter into two parts:
                    138: 
                    139: @itemize @bullet
                    140: 
                    141: @item The @emph{front end} takes the source code of the language to be
                    142: implemented, and translates it into virtual machine code.  This is
                    143: similar to an ordinary compiler front end; typically an interpreter
                    144: front-end performs no optimization, so it is relatively simple to
                    145: implement and runs fast.
                    146: 
                    147: @item The @emph{virtual machine interpreter} executes the virtual
                    148: machine code.
                    149: 
                    150: @end itemize
                    151: 
                    152: Such a division is usually used in interpreters, for modularity as well
1.6       anton     153: as for efficiency.  The virtual machine code is typically passed between
                    154: front end and virtual machine interpreter in memory, like in a
1.1       anton     155: load-and-go compiler; this avoids the complexity and time cost of
                    156: writing the code to a file and reading it again.
                    157: 
                    158: A @emph{virtual machine} (VM) represents the program as a sequence of
                    159: @emph{VM instructions}, following each other in memory, similar to real
                    160: machine code.  Control flow occurs through VM branch instructions, like
                    161: in a real machine.
                    162: 
1.12      anton     163: @cindex functionality features overview
1.11      anton     164: In this setup, Vmgen can generate most of the code dealing with virtual
1.1       anton     165: machine instructions from a simple description of the virtual machine
1.11      anton     166: instructions (@pxref{Input File Format}), in particular:
1.1       anton     167: 
1.13      anton     168: @table @strong
1.1       anton     169: 
                    170: @item VM instruction execution
                    171: 
                    172: @item VM code generation
                    173: Useful in the front end.
                    174: 
                    175: @item VM code decompiler
                    176: Useful for debugging the front end.
                    177: 
                    178: @item VM code tracing
                    179: Useful for debugging the front end and the VM interpreter.  You will
                    180: typically provide other means for debugging the user's programs at the
                    181: source level.
                    182: 
                    183: @item VM code profiling
1.12      anton     184: Useful for optimizing the VM interpreter with superinstructions
1.11      anton     185: (@pxref{VM profiler}).
1.1       anton     186: 
                    187: @end table
                    188: 
1.13      anton     189: To create parts of the interpretive system that do not deal with VM
                    190: instructions, you have to use other tools (e.g., @command{bison}) and/or
                    191: hand-code them.
                    192: 
1.12      anton     193: @cindex efficiency features overview
1.11      anton     194: @noindent
                    195: Vmgen supports efficient interpreters though various optimizations, in
1.1       anton     196: particular
                    197: 
1.11      anton     198: @itemize @bullet
1.1       anton     199: 
                    200: @item Threaded code
                    201: 
                    202: @item Caching the top-of-stack in a register
                    203: 
                    204: @item Combining VM instructions into superinstructions
                    205: 
                    206: @item
                    207: Replicating VM (super)instructions for better BTB prediction accuracy
                    208: (not yet in vmgen-ex, but already in Gforth).
                    209: 
                    210: @end itemize
                    211: 
1.12      anton     212: @cindex speed for JVM
1.11      anton     213: As a result, Vmgen-based interpreters are only about an order of
                    214: magnitude slower than native code from an optimizing C compiler on small
1.1       anton     215: benchmarks; on large benchmarks, which spend more time in the run-time
1.2       anton     216: system, the slowdown is often less (e.g., the slowdown of a
                    217: Vmgen-generated JVM interpreter over the best JVM JIT compiler we
                    218: measured is only a factor of 2-3 for large benchmarks; some other JITs
                    219: and all other interpreters we looked at were slower than our
                    220: interpreter).
1.1       anton     221: 
                    222: VMs are usually designed as stack machines (passing data between VM
1.11      anton     223: instructions on a stack), and Vmgen supports such designs especially
1.12      anton     224: well; however, you can also use Vmgen for implementing a register VM
                    225: (@pxref{Register Machines}) and still benefit from most of the advantages
                    226: offered by Vmgen.
1.1       anton     227: 
1.2       anton     228: There are many potential uses of the instruction descriptions that are
                    229: not implemented at the moment, but we are open for feature requests, and
1.13      anton     230: we will consider new features if someone asks for them; so the feature
1.2       anton     231: list above is not exhaustive.
1.1       anton     232: 
1.2       anton     233: @c *********************************************************************
1.10      anton     234: @node Why interpreters?, Concepts, Introduction, Top
1.2       anton     235: @chapter Why interpreters?
1.12      anton     236: @cindex interpreters, advantages
                    237: @cindex advantages of interpreters
                    238: @cindex advantages of vmgen
1.2       anton     239: 
                    240: Interpreters are a popular language implementation technique because
                    241: they combine all three of the following advantages:
                    242: 
1.11      anton     243: @itemize @bullet
1.2       anton     244: 
                    245: @item Ease of implementation
                    246: 
                    247: @item Portability
                    248: 
                    249: @item Fast edit-compile-run cycle
                    250: 
                    251: @end itemize
                    252: 
1.12      anton     253: Vmgen makes it even easier to implement interpreters.
                    254: 
                    255: @cindex speed of interpreters
1.2       anton     256: The main disadvantage of interpreters is their run-time speed.  However,
                    257: there are huge differences between different interpreters in this area:
                    258: the slowdown over optimized C code on programs consisting of simple
                    259: operations is typically a factor of 10 for the more efficient
                    260: interpreters, and a factor of 1000 for the less efficient ones (the
                    261: slowdown for programs executing complex operations is less, because the
                    262: time spent in libraries for executing complex operations is the same in
                    263: all implementation strategies).
                    264: 
1.12      anton     265: Vmgen supports techniques for building efficient interpreters.
1.2       anton     266: 
                    267: @c ********************************************************************
1.11      anton     268: @node Concepts, Invoking Vmgen, Why interpreters?, Top
1.2       anton     269: @chapter Concepts
                    270: 
1.10      anton     271: @menu
                    272: * Front end and VM interpreter::  Modularizing an interpretive system
                    273: * Data handling::               Stacks, registers, immediate arguments
                    274: * Dispatch::                    From one VM instruction to the next
                    275: @end menu
                    276: 
1.2       anton     277: @c --------------------------------------------------------------------
1.10      anton     278: @node Front end and VM interpreter, Data handling, Concepts, Concepts
                    279: @section Front end and VM interpreter
1.12      anton     280: @cindex modularization of interpreters
1.2       anton     281: 
                    282: @cindex front-end
                    283: Interpretive systems are typically divided into a @emph{front end} that
                    284: parses the input language and produces an intermediate representation
                    285: for the program, and an interpreter that executes the intermediate
                    286: representation of the program.
                    287: 
                    288: @cindex virtual machine
                    289: @cindex VM
1.12      anton     290: @cindex VM instruction
1.2       anton     291: @cindex instruction, VM
1.12      anton     292: @cindex VM branch instruction
                    293: @cindex branch instruction, VM
                    294: @cindex VM register
                    295: @cindex register, VM
                    296: @cindex opcode, VM instruction
                    297: @cindex immediate argument, VM instruction
1.2       anton     298: For efficient interpreters the intermediate representation of choice is
                    299: virtual machine code (rather than, e.g., an abstract syntax tree).
                    300: @emph{Virtual machine} (VM) code consists of VM instructions arranged
                    301: sequentially in memory; they are executed in sequence by the VM
1.12      anton     302: interpreter, but VM branch instructions can change the control flow and
                    303: are used for implementing control structures.  The conceptual similarity
                    304: to real machine code results in the name @emph{virtual machine}.
                    305: Various terms similar to terms for real machines are used; e.g., there
                    306: are @emph{VM registers} (like the instruction pointer and stack
                    307: pointer(s)), and the VM instruction consists of an @emph{opcode} and
                    308: @emph{immediate arguments}.
1.2       anton     309: 
1.11      anton     310: In this framework, Vmgen supports building the VM interpreter and any
1.2       anton     311: other component dealing with VM instructions.  It does not have any
                    312: support for the front end, apart from VM code generation support.  The
                    313: front end can be implemented with classical compiler front-end
1.3       anton     314: techniques, supported by tools like @command{flex} and @command{bison}.
1.2       anton     315: 
                    316: The intermediate representation is usually just internal to the
                    317: interpreter, but some systems also support saving it to a file, either
                    318: as an image file, or in a full-blown linkable file format (e.g., JVM).
                    319: Vmgen currently has no special support for such features, but the
                    320: information in the instruction descriptions can be helpful, and we are
1.13      anton     321: open to feature requests and suggestions.
1.3       anton     322: 
1.10      anton     323: @c --------------------------------------------------------------------
                    324: @node Data handling, Dispatch, Front end and VM interpreter, Concepts
1.3       anton     325: @section Data handling
                    326: 
                    327: @cindex stack machine
                    328: @cindex register machine
                    329: Most VMs use one or more stacks for passing temporary data between VM
                    330: instructions.  Another option is to use a register machine architecture
1.13      anton     331: for the virtual machine; we believe that using a stack architecture is
                    332: usually both simpler and faster.
                    333: 
                    334: however, this option is slower or
1.3       anton     335: significantly more complex to implement than a stack machine architecture.
                    336: 
                    337: Vmgen has special support and optimizations for stack VMs, making their
                    338: implementation easy and efficient.
                    339: 
1.11      anton     340: You can also implement a register VM with Vmgen (@pxref{Register
                    341: Machines}), and you will still profit from most Vmgen features.
1.3       anton     342: 
                    343: @cindex stack item size
                    344: @cindex size, stack items
                    345: Stack items all have the same size, so they typically will be as wide as
                    346: an integer, pointer, or floating-point value.  Vmgen supports treating
                    347: two consecutive stack items as a single value, but anything larger is
                    348: best kept in some other memory area (e.g., the heap), with pointers to
                    349: the data on the stack.
                    350: 
                    351: @cindex instruction stream
                    352: @cindex immediate arguments
                    353: Another source of data is immediate arguments VM instructions (in the VM
                    354: instruction stream).  The VM instruction stream is handled similar to a
1.11      anton     355: stack in Vmgen.
1.3       anton     356: 
                    357: @cindex garbage collection
                    358: @cindex reference counting
1.12      anton     359: Vmgen has no built-in support for, nor restrictions against
                    360: @emph{garbage collection}.  If you need garbage collection, you need to
                    361: provide it in your run-time libraries.  Using @emph{reference counting}
                    362: is probably harder, but might be possible (contact us if you are
                    363: interested).
1.3       anton     364: @c reference counting might be possible by including counting code in 
                    365: @c the conversion macros.
                    366: 
1.10      anton     367: @c --------------------------------------------------------------------
                    368: @node Dispatch,  , Data handling, Concepts
1.6       anton     369: @section Dispatch
1.12      anton     370: @cindex Dispatch of VM instructions
                    371: @cindex main interpreter loop
1.6       anton     372: 
1.11      anton     373: Understanding this section is probably not necessary for using Vmgen,
1.6       anton     374: but it may help.  You may want to skip it now, and read it if you find statements about dispatch methods confusing.
                    375: 
                    376: After executing one VM instruction, the VM interpreter has to dispatch
1.11      anton     377: the next VM instruction (Vmgen calls the dispatch routine @samp{NEXT}).
1.6       anton     378: Vmgen supports two methods of dispatch:
                    379: 
1.13      anton     380: @table @strong
1.6       anton     381: 
                    382: @item switch dispatch
1.12      anton     383: @cindex switch dispatch
1.6       anton     384: In this method the VM interpreter contains a giant @code{switch}
                    385: statement, with one @code{case} for each VM instruction.  The VM
1.12      anton     386: instruction opcodes are represented by integers (e.g., produced by an
                    387: @code{enum}) in the VM code, and dispatch occurs by loading the next
                    388: opcode, @code{switch}ing on it, and continuing at the appropriate
                    389: @code{case}; after executing the VM instruction, the VM interpreter
                    390: jumps back to the dispatch code.
1.6       anton     391: 
                    392: @item threaded code
1.12      anton     393: @cindex threaded code
                    394: This method represents a VM instruction opcode by the address of the
                    395: start of the machine code fragment for executing the VM instruction.
1.6       anton     396: Dispatch consists of loading this address, jumping to it, and
                    397: incrementing the VM instruction pointer.  Typically the threaded-code
                    398: dispatch code is appended directly to the code for executing the VM
                    399: instruction.  Threaded code cannot be implemented in ANSI C, but it can
1.11      anton     400: be implemented using GNU C's labels-as-values extension (@pxref{Labels
                    401: as Values, , Labels as Values, gcc.info, GNU C Manual}).
1.6       anton     402: 
1.13      anton     403: @c call threading
1.6       anton     404: @end table
                    405: 
1.12      anton     406: Threaded code can be twice as fast as switch dispatch, depending on the
                    407: interpreter, the benchmark, and the machine.
                    408: 
1.3       anton     409: @c *************************************************************
1.11      anton     410: @node Invoking Vmgen, Example, Concepts, Top
                    411: @chapter Invoking Vmgen
1.12      anton     412: @cindex Invoking Vmgen
1.3       anton     413: 
1.11      anton     414: The usual way to invoke Vmgen is as follows:
1.3       anton     415: 
                    416: @example
1.13      anton     417: vmgen @var{inputfile}
1.3       anton     418: @end example
                    419: 
1.13      anton     420: Here @var{inputfile} is the VM instruction description file, which
                    421: usually ends in @file{.vmg}.  The output filenames are made by taking
                    422: the basename of @file{inputfile} (i.e., the output files will be created
                    423: in the current working directory) and replacing @file{.vmg} with
                    424: @file{-vm.i}, @file{-disasm.i}, @file{-gen.i}, @file{-labels.i},
                    425: @file{-profile.i}, and @file{-peephole.i}.  E.g., @command{vmgen
                    426: hack/foo.vmg} will create @file{foo-vm.i}, @file{foo-disasm.i},
                    427: @file{foo-gen.i}, @file{foo-labels.i}, @file{foo-profile.i} and
                    428: @file{foo-peephole.i}.
1.3       anton     429: 
1.11      anton     430: The command-line options supported by Vmgen are
1.3       anton     431: 
                    432: @table @option
                    433: 
                    434: @cindex -h, command-line option
                    435: @cindex --help, command-line option
                    436: @item --help
                    437: @itemx -h
                    438: Print a message about the command-line options
                    439: 
                    440: @cindex -v, command-line option
                    441: @cindex --version, command-line option
                    442: @item --version
                    443: @itemx -v
                    444: Print version and exit
                    445: @end table
                    446: 
                    447: @c env vars GFORTHDIR GFORTHDATADIR
                    448: 
1.5       anton     449: @c ****************************************************************
1.11      anton     450: @node Example, Input File Format, Invoking Vmgen, Top
1.5       anton     451: @chapter Example
1.12      anton     452: @cindex example of a Vmgen-based interpreter
1.5       anton     453: 
1.10      anton     454: @menu
                    455: * Example overview::            
                    456: * Using profiling to create superinstructions::  
                    457: @end menu
                    458: 
                    459: @c --------------------------------------------------------------------
                    460: @node Example overview, Using profiling to create superinstructions, Example, Example
1.5       anton     461: @section Example overview
1.12      anton     462: @cindex example overview
                    463: @cindex @file{vmgen-ex}
                    464: @cindex @file{vmgen-ex2}
1.5       anton     465: 
1.11      anton     466: There are two versions of the same example for using Vmgen:
1.5       anton     467: @file{vmgen-ex} and @file{vmgen-ex2} (you can also see Gforth as
                    468: example, but it uses additional (undocumented) features, and also
                    469: differs in some other respects).  The example implements @emph{mini}, a
                    470: tiny Modula-2-like language with a small JavaVM-like virtual machine.
1.12      anton     471: 
1.5       anton     472: The difference between the examples is that @file{vmgen-ex} uses many
                    473: casts, and @file{vmgen-ex2} tries to avoids most casts and uses unions
1.12      anton     474: instead.  In the rest of this manual we usually mention just files in
                    475: @file{vmgen-ex}; if you want to use unions, use the equivalent file in
                    476: @file{vmgen-ex2}.
                    477: @cindex unions example
                    478: @cindex casts example
1.5       anton     479: 
                    480: The files provided with each example are:
1.12      anton     481: @cindex example files
1.5       anton     482: 
                    483: @example
                    484: Makefile
                    485: README
                    486: disasm.c           wrapper file
                    487: engine.c           wrapper file
                    488: peephole.c         wrapper file
                    489: profile.c          wrapper file
                    490: mini-inst.vmg      simple VM instructions
                    491: mini-super.vmg     superinstructions (empty at first)
                    492: mini.h             common declarations
                    493: mini.l             scanner
                    494: mini.y             front end (parser, VM code generator)
                    495: support.c          main() and other support functions
                    496: fib.mini           example mini program
                    497: simple.mini        example mini program
                    498: test.mini          example mini program (tests everything)
                    499: test.out           test.mini output
                    500: stat.awk           script for aggregating profile information
                    501: peephole-blacklist list of instructions not allowed in superinstructions
                    502: seq2rule.awk       script for creating superinstructions
                    503: @end example
                    504: 
                    505: For your own interpreter, you would typically copy the following files
                    506: and change little, if anything:
1.12      anton     507: @cindex wrapper files
1.5       anton     508: 
                    509: @example
                    510: disasm.c           wrapper file
                    511: engine.c           wrapper file
                    512: peephole.c         wrapper file
                    513: profile.c          wrapper file
                    514: stat.awk           script for aggregating profile information
                    515: seq2rule.awk       script for creating superinstructions
                    516: @end example
                    517: 
1.11      anton     518: @noindent
1.5       anton     519: You would typically change much in or replace the following files:
                    520: 
                    521: @example
                    522: Makefile
                    523: mini-inst.vmg      simple VM instructions
                    524: mini.h             common declarations
                    525: mini.l             scanner
                    526: mini.y             front end (parser, VM code generator)
                    527: support.c          main() and other support functions
                    528: peephole-blacklist list of instructions not allowed in superinstructions
                    529: @end example
                    530: 
                    531: You can build the example by @code{cd}ing into the example's directory,
1.12      anton     532: and then typing @code{make}; you can check that it works with @code{make
1.5       anton     533: check}.  You can run run mini programs like this:
                    534: 
                    535: @example
                    536: ./mini fib.mini
                    537: @end example
                    538: 
1.12      anton     539: To learn about the options, type @code{./mini -h}.
1.5       anton     540: 
1.10      anton     541: @c --------------------------------------------------------------------
                    542: @node Using profiling to create superinstructions,  , Example overview, Example
1.5       anton     543: @section Using profiling to create superinstructions
1.12      anton     544: @cindex profiling example
                    545: @cindex superinstructions example
1.5       anton     546: 
                    547: I have not added rules for this in the @file{Makefile} (there are many
                    548: options for selecting superinstructions, and I did not want to hardcode
                    549: one into the @file{Makefile}), but there are some supporting scripts, and
                    550: here's an example:
                    551: 
                    552: Suppose you want to use @file{fib.mini} and @file{test.mini} as training
                    553: programs, you get the profiles like this:
                    554: 
                    555: @example
                    556: make fib.prof test.prof #takes a few seconds
                    557: @end example
                    558: 
                    559: You can aggregate these profiles with @file{stat.awk}:
                    560: 
                    561: @example
                    562: awk -f stat.awk fib.prof test.prof
                    563: @end example
                    564: 
                    565: The result contains lines like:
                    566: 
                    567: @example
                    568:       2      16        36910041 loadlocal lit
                    569: @end example
                    570: 
                    571: This means that the sequence @code{loadlocal lit} statically occurs a
                    572: total of 16 times in 2 profiles, with a dynamic execution count of
                    573: 36910041.
                    574: 
                    575: The numbers can be used in various ways to select superinstructions.
                    576: E.g., if you just want to select all sequences with a dynamic
                    577: execution count exceeding 10000, you would use the following pipeline:
                    578: 
                    579: @example
                    580: awk -f stat.awk fib.prof test.prof|
                    581: awk '$3>=10000'|                #select sequences
                    582: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
1.12      anton     583: awk -f seq2rule.awk|  #transform sequences into superinstruction rules
1.5       anton     584: sort -k 3 >mini-super.vmg       #sort sequences
                    585: @end example
                    586: 
                    587: The file @file{peephole-blacklist} contains all instructions that
                    588: directly access a stack or stack pointer (for mini: @code{call},
                    589: @code{return}); the sort step is necessary to ensure that prefixes
1.13      anton     590: precede larger superinstructions.
1.5       anton     591: 
                    592: Now you can create a version of mini with superinstructions by just
                    593: saying @samp{make}
                    594: 
1.10      anton     595: 
1.3       anton     596: @c ***************************************************************
1.13      anton     597: @node Input File Format, Error messages, Example, Top
1.3       anton     598: @chapter Input File Format
1.12      anton     599: @cindex input file format
                    600: @cindex format, input file
1.3       anton     601: 
                    602: Vmgen takes as input a file containing specifications of virtual machine
                    603: instructions.  This file usually has a name ending in @file{.vmg}.
                    604: 
1.5       anton     605: Most examples are taken from the example in @file{vmgen-ex}.
1.3       anton     606: 
1.10      anton     607: @menu
                    608: * Input File Grammar::          
                    609: * Simple instructions::         
                    610: * Superinstructions::           
1.18      anton     611: * Store Optimization::          
1.11      anton     612: * Register Machines::           How to define register VM instructions
1.10      anton     613: @end menu
                    614: 
                    615: @c --------------------------------------------------------------------
                    616: @node Input File Grammar, Simple instructions, Input File Format, Input File Format
1.3       anton     617: @section Input File Grammar
1.12      anton     618: @cindex grammar, input file
                    619: @cindex input file grammar
1.3       anton     620: 
                    621: The grammar is in EBNF format, with @code{@var{a}|@var{b}} meaning
                    622: ``@var{a} or @var{b}'', @code{@{@var{c}@}} meaning 0 or more repetitions
                    623: of @var{c} and @code{[@var{d}]} meaning 0 or 1 repetitions of @var{d}.
                    624: 
1.12      anton     625: @cindex free-format, not
1.15      anton     626: @cindex newlines, significance in syntax
1.3       anton     627: Vmgen input is not free-format, so you have to take care where you put
1.15      anton     628: newlines (and, in a few cases, white space).
1.3       anton     629: 
                    630: @example
1.15      anton     631: description: @{instruction|comment|eval-escape|c-escape@}
1.3       anton     632: 
                    633: instruction: simple-inst|superinst
                    634: 
1.15      anton     635: simple-inst: ident '(' stack-effect ')' newline c-code newline newline
1.3       anton     636: 
1.15      anton     637: stack-effect: @{ident@} '--' @{ident@}
1.3       anton     638: 
1.15      anton     639: super-inst: ident '=' ident @{ident@}  
1.3       anton     640: 
1.12      anton     641: comment:      '\ '  text newline
1.3       anton     642: 
1.13      anton     643: eval-escape:  '\E ' text newline
1.15      anton     644: 
                    645: c-escape:     '\C ' text newline
1.3       anton     646: @end example
                    647: @c \+ \- \g \f \c
                    648: 
                    649: Note that the @code{\}s in this grammar are meant literally, not as
1.5       anton     650: C-style encodings for non-printable characters.
1.3       anton     651: 
1.15      anton     652: There are two ways to delimit the C code in @code{simple-inst}:
                    653: 
                    654: @itemize @bullet
                    655: 
                    656: @item
                    657: If you start it with a @samp{@{} at the start of a line (i.e., not even
                    658: white space before it), you have to end it with a @samp{@}} at the start
                    659: of a line (followed by a newline).  In this case you may have empty
                    660: lines within the C code (typically used between variable definitions and
                    661: statements).
                    662: 
                    663: @item
                    664: You do not start it with @samp{@{}.  Then the C code ends at the first
                    665: empty line, so you cannot have empty lines within this code.
                    666: 
                    667: @end itemize
                    668: 
                    669: The text in @code{comment}, @code{eval-escape} and @code{c-escape} must
                    670: not contain a newline.  @code{Ident} must conform to the usual
                    671: conventions of C identifiers (otherwise the C compiler would choke on
                    672: the Vmgen output), except that idents in @code{stack-effect} may have a
                    673: stack prefix (for stack prefix syntax, @pxref{Eval escapes}).
                    674: 
                    675: @cindex C escape
                    676: @cindex @code{\C}
                    677: @cindex conditional compilation of Vmgen output
                    678: The @code{c-escape} passes the text through to each output file (without
                    679: the @samp{\C}).  This is useful mainly for conditional compilation
                    680: (i.e., you write @samp{\C #if ...} etc.).
                    681: 
                    682: @cindex sync lines
                    683: @cindex @code{#line}
                    684: In addition to the syntax given in the grammer, Vmgen also processes
                    685: sync lines (lines starting with @samp{#line}), as produced by @samp{m4
                    686: -s} (@pxref{Invoking m4, , Invoking m4, m4.info, GNU m4}) and similar
                    687: tools.  This allows associating C compiler error messages with the
                    688: original source of the C code.
1.3       anton     689: 
                    690: Vmgen understands a few extensions beyond the grammar given here, but
                    691: these extensions are only useful for building Gforth.  You can find a
                    692: description of the format used for Gforth in @file{prim}.
                    693: 
1.17      anton     694: @menu
                    695: * Eval escapes::                what follows \E
                    696: @end menu
                    697: 
                    698: @node Eval escapes,  , Input File Grammar, Input File Grammar
1.10      anton     699: @subsection Eval escapes
1.12      anton     700: @cindex escape to Forth
                    701: @cindex eval escape
1.15      anton     702: @cindex @code{\E}
1.13      anton     703: 
1.3       anton     704: @c woanders?
                    705: The text in @code{eval-escape} is Forth code that is evaluated when
1.13      anton     706: Vmgen reads the line.  You will normally use this feature to define
                    707: stacks and types.
                    708: 
                    709: If you do not know (and do not want to learn) Forth, you can build the
                    710: text according to the following grammar; these rules are normally all
                    711: Forth you need for using Vmgen:
1.3       anton     712: 
                    713: @example
1.18      anton     714: text: stack-decl|type-prefix-decl|stack-prefix-decl|set-flag
1.3       anton     715: 
1.12      anton     716: stack-decl: 'stack ' ident ident ident
1.3       anton     717: type-prefix-decl: 
1.12      anton     718:     's" ' string '" ' ('single'|'double') ident 'type-prefix' ident
                    719: stack-prefix-decl:  ident 'stack-prefix' string
1.22      anton     720: set-flag: ('store-optimization'|'include-skipped-insts') ('on'|'off')
1.3       anton     721: @end example
                    722: 
                    723: Note that the syntax of this code is not checked thoroughly (there are
1.13      anton     724: many other Forth program fragments that could be written in an
                    725: eval-escape).
1.3       anton     726: 
1.14      anton     727: A stack prefix can contain letters, digits, or @samp{:}, and may start
                    728: with an @samp{#}; e.g., in Gforth the return stack has the stack prefix
                    729: @samp{R:}.  This restriction is not checked during the stack prefix
                    730: definition, but it is enforced by the parsing rules for stack items
                    731: later.
                    732: 
1.3       anton     733: If you know Forth, the stack effects of the non-standard words involved
                    734: are:
1.12      anton     735: @findex stack
                    736: @findex type-prefix
                    737: @findex single
                    738: @findex double
                    739: @findex stack-prefix
1.18      anton     740: @findex store-optimization
1.3       anton     741: @example
1.22      anton     742: stack                 ( "name" "pointer" "type" -- )
                    743:                       ( name execution: -- stack )
                    744: type-prefix           ( addr u item-size stack "prefix" -- )
                    745: single                ( -- item-size )
                    746: double                ( -- item-size )
                    747: stack-prefix          ( stack "prefix" -- )
                    748: store-optimization    ( -- addr )
                    749: include-skipped-insts ( -- addr )
1.3       anton     750: @end example
                    751: 
1.14      anton     752: An @var{item-size} takes three cells on the stack.
1.5       anton     753: 
1.10      anton     754: @c --------------------------------------------------------------------
                    755: @node Simple instructions, Superinstructions, Input File Grammar, Input File Format
1.3       anton     756: @section Simple instructions
1.12      anton     757: @cindex simple VM instruction
                    758: @cindex instruction, simple VM
1.3       anton     759: 
                    760: We will use the following simple VM instruction description as example:
                    761: 
                    762: @example
                    763: sub ( i1 i2 -- i )
                    764: i = i1-i2;
                    765: @end example
                    766: 
                    767: The first line specifies the name of the VM instruction (@code{sub}) and
                    768: its stack effect (@code{i1 i2 -- i}).  The rest of the description is
                    769: just plain C code.
                    770: 
                    771: @cindex stack effect
1.12      anton     772: @cindex effect, stack
1.3       anton     773: The stack effect specifies that @code{sub} pulls two integers from the
1.12      anton     774: data stack and puts them in the C variables @code{i1} and @code{i2}
                    775: (with the rightmost item (@code{i2}) taken from the top of stack;
                    776: intuition: if you push @code{i1}, then @code{i2} on the stack, the
                    777: resulting stack picture is @code{i1 i2}) and later pushes one integer
                    778: (@code{i}) on the data stack (the rightmost item is on the top
                    779: afterwards).
                    780: 
                    781: @cindex prefix, type
                    782: @cindex type prefix
                    783: @cindex default stack of a type prefix
1.3       anton     784: How do we know the type and stack of the stack items?  Vmgen uses
                    785: prefixes, similar to Fortran; in contrast to Fortran, you have to
                    786: define the prefix first:
                    787: 
                    788: @example
                    789: \E s" Cell"   single data-stack type-prefix i
                    790: @end example
                    791: 
                    792: This defines the prefix @code{i} to refer to the type @code{Cell}
                    793: (defined as @code{long} in @file{mini.h}) and, by default, to the
                    794: @code{data-stack}.  It also specifies that this type takes one stack
                    795: item (@code{single}).  The type prefix is part of the variable name.
                    796: 
1.12      anton     797: @cindex stack definition
                    798: @cindex defining a stack
1.3       anton     799: Before we can use @code{data-stack} in this way, we have to define it:
                    800: 
                    801: @example
                    802: \E stack data-stack sp Cell
                    803: @end example
                    804: @c !! use something other than Cell
                    805: 
1.12      anton     806: @cindex stack basic type
                    807: @cindex basic type of a stack
                    808: @cindex type of a stack, basic
1.3       anton     809: This line defines the stack @code{data-stack}, which uses the stack
                    810: pointer @code{sp}, and each item has the basic type @code{Cell}; other
                    811: types have to fit into one or two @code{Cell}s (depending on whether the
1.12      anton     812: type is @code{single} or @code{double} wide), and are cast from and to
                    813: Cells on accessing the @code{data-stack} with type cast macros
1.22      anton     814: (@pxref{VM engine}).  By default, stacks grow towards lower addresses in
                    815: Vmgen-erated interpreters (@pxref{Stack growth direction}).
1.3       anton     816: 
1.12      anton     817: @cindex stack prefix
                    818: @cindex prefix, stack
1.3       anton     819: We can override the default stack of a stack item by using a stack
                    820: prefix.  E.g., consider the following instruction:
                    821: 
                    822: @example
                    823: lit ( #i -- i )
                    824: @end example
                    825: 
                    826: The VM instruction @code{lit} takes the item @code{i} from the
1.5       anton     827: instruction stream (indicated by the prefix @code{#}), and pushes it on
1.3       anton     828: the (default) data stack.  The stack prefix is not part of the variable
                    829: name.  Stack prefixes are defined like this:
                    830: 
                    831: @example
                    832: \E inst-stream stack-prefix #
1.29    ! anton     833: \E data-stack  stack-prefix S:
1.3       anton     834: @end example
                    835: 
1.5       anton     836: This definition defines that the stack prefix @code{#} specifies the
1.3       anton     837: ``stack'' @code{inst-stream}.  Since the instruction stream behaves a
                    838: little differently than an ordinary stack, it is predefined, and you do
                    839: not need to define it.
                    840: 
1.12      anton     841: @cindex instruction stream
1.3       anton     842: The instruction stream contains instructions and their immediate
                    843: arguments, so specifying that an argument comes from the instruction
                    844: stream indicates an immediate argument.  Of course, instruction stream
                    845: arguments can only appear to the left of @code{--} in the stack effect.
                    846: If there are multiple instruction stream arguments, the leftmost is the
                    847: first one (just as the intuition suggests).
                    848: 
1.10      anton     849: @menu
1.29    ! anton     850: * Explicit stack access::       If the C code accesses a stack pointer
1.10      anton     851: * C Code Macros::               Macros recognized by Vmgen
                    852: * C Code restrictions::         Vmgen makes assumptions about C code
1.22      anton     853: * Stack growth direction::      is configurable per stack
1.10      anton     854: @end menu
                    855: 
                    856: @c --------------------------------------------------------------------
1.29    ! anton     857: @node  Explicit stack access, C Code Macros, Simple instructions, Simple instructions
        !           858: @subsection Explicit stack access
        !           859: @cindex stack access, explicit
        !           860: @cindex Stack pointer access
        !           861: @cindex explicit stack access
        !           862: 
        !           863: Not all stack effects can be specified using the stack effect
        !           864: specifications above.  For VM instructions that have other stack
        !           865: effects, you can specify them explicitly by accessing the stack
        !           866: pointer in the C code; however, you have to notify Vmgen of such
        !           867: explicit stack accesses, otherwise Vmgens optimizations could conflict
        !           868: with your explicit stack accesses.
        !           869: 
        !           870: You notify Vmgen by putting @code{...} with the appropriate stack
        !           871: prefix into the stack comment.  Then the VM instruction will first
        !           872: take the other stack items specified in the stack effect into C
        !           873: variables, then make sure that all other stack items for that stack
        !           874: are in memory, and that the stack pointer for the stack points to the
        !           875: top-of-stack (by default, unless you change the stack access
        !           876: transformation: @pxref{Stack growth direction}).
        !           877: 
        !           878: The general rule is: If you mention a stack pointer in the C code of a
        !           879: VM instruction, you should put a @code{...} for that stack in the stack
        !           880: effect.
        !           881: 
        !           882: Consider this example:
        !           883: 
        !           884: @example
        !           885: return ( #iadjust S:... target afp i1 -- i2 )
        !           886: SET_IP(target);
        !           887: sp = (Cell *)(((char *)sp)+iadjust);
        !           888: fp = afp;
        !           889: i2=i1;
        !           890: @end example
        !           891: 
        !           892: First the variables @code{target afp i1} are popped off the stack,
        !           893: then the stack pointer @code{sp} is set correctly for the new stack
        !           894: depth, then the C code changes the stack depth and does other things,
        !           895: and finally @code{i2} is pushed on the stack with the new depth.
        !           896: 
        !           897: The position of the @code{...} within the stack effect does not
        !           898: matter.  You can use several @code{...}s, for different stacks, and
        !           899: also several for the same stack (that has no additional effect).  If
        !           900: you use @code{...} without a stack prefix, this specifies all the
        !           901: stacks except the instruction stream.
        !           902: 
        !           903: You cannot use @code{...} for the instruction stream, but that is not
        !           904: necessary: At the start of the C code, @code{IP} points to the start
        !           905: of the next VM instruction (i.e., right beyond the end of the current
        !           906: VM instruction), and you can change the instruction pointer with
        !           907: @code{SET_IP} (@pxref{VM engine}).
        !           908: 
        !           909: 
        !           910: @c --------------------------------------------------------------------
        !           911: @node C Code Macros, C Code restrictions, Explicit stack access, Simple instructions
1.10      anton     912: @subsection C Code Macros
1.12      anton     913: @cindex macros recognized by Vmgen
                    914: @cindex basic block, VM level
1.5       anton     915: 
                    916: Vmgen recognizes the following strings in the C code part of simple
                    917: instructions:
                    918: 
1.12      anton     919: @table @code
1.5       anton     920: 
                    921: @item SET_IP
1.12      anton     922: @findex SET_IP
1.11      anton     923: As far as Vmgen is concerned, a VM instruction containing this ends a VM
1.5       anton     924: basic block (used in profiling to delimit profiled sequences).  On the C
                    925: level, this also sets the instruction pointer.
                    926: 
                    927: @item SUPER_END
1.12      anton     928: @findex SUPER_END
                    929: This ends a basic block (for profiling), even if the instruction
                    930: contains no @code{SET_IP}.
1.5       anton     931: 
1.13      anton     932: @item INST_TAIL;
                    933: @findex INST_TAIL;
                    934: Vmgen replaces @samp{INST_TAIL;} with code for ending a VM instruction and
                    935: dispatching the next VM instruction.  Even without a @samp{INST_TAIL;} this
1.12      anton     936: happens automatically when control reaches the end of the C code.  If
                    937: you want to have this in the middle of the C code, you need to use
1.13      anton     938: @samp{INST_TAIL;}.  A typical example is a conditional VM branch:
1.5       anton     939: 
                    940: @example
1.11      anton     941: if (branch_condition) @{
1.13      anton     942:   SET_IP(target); INST_TAIL;
1.11      anton     943: @}
1.5       anton     944: /* implicit tail follows here */
                    945: @end example
                    946: 
1.13      anton     947: In this example, @samp{INST_TAIL;} is not strictly necessary, because there
1.5       anton     948: is another one implicitly after the if-statement, but using it improves
                    949: branch prediction accuracy slightly and allows other optimizations.
                    950: 
                    951: @item SUPER_CONTINUE
1.12      anton     952: @findex SUPER_CONTINUE
1.5       anton     953: This indicates that the implicit tail at the end of the VM instruction
                    954: dispatches the sequentially next VM instruction even if there is a
                    955: @code{SET_IP} in the VM instruction.  This enables an optimization that
                    956: is not yet implemented in the vmgen-ex code (but in Gforth).  The
                    957: typical application is in conditional VM branches:
                    958: 
                    959: @example
1.11      anton     960: if (branch_condition) @{
1.13      anton     961:   SET_IP(target); INST_TAIL; /* now this INST_TAIL is necessary */
1.11      anton     962: @}
1.5       anton     963: SUPER_CONTINUE;
                    964: @end example
                    965: 
1.27      anton     966: @item VM_JUMP
                    967: @findex VM_JUMP
                    968: @code{VM_JUMP(target)} is equivalent to @code{goto *(target)}, but
                    969: allows Vmgen to do dynamic superinstructions and replication.  You
1.28      anton     970: still need to say @code{SUPER_END}.  Also, the goto only happens at
                    971: the end (wherever the VM_JUMP is).  Essentially, this just suppresses
                    972: much of the ordinary dispatch mechanism.
1.27      anton     973: 
1.5       anton     974: @end table
                    975: 
1.11      anton     976: Note that Vmgen is not smart about C-level tokenization, comments,
1.5       anton     977: strings, or conditional compilation, so it will interpret even a
                    978: commented-out SUPER_END as ending a basic block (or, e.g.,
1.13      anton     979: @samp{RESET_IP;} as @samp{SET_IP;}).  Conversely, Vmgen requires the literal
1.11      anton     980: presence of these strings; Vmgen will not see them if they are hiding in
1.5       anton     981: a C preprocessor macro.
                    982: 
                    983: 
1.10      anton     984: @c --------------------------------------------------------------------
1.22      anton     985: @node C Code restrictions, Stack growth direction, C Code Macros, Simple instructions
1.10      anton     986: @subsection C Code restrictions
1.12      anton     987: @cindex C code restrictions
                    988: @cindex restrictions on C code
                    989: @cindex assumptions about C code
                    990: 
                    991: @cindex accessing stack (pointer)
                    992: @cindex stack pointer, access
                    993: @cindex instruction pointer, access
1.5       anton     994: Vmgen generates code and performs some optimizations under the
                    995: assumption that the user-supplied C code does not access the stack
                    996: pointers or stack items, and that accesses to the instruction pointer
                    997: only occur through special macros.  In general you should heed these
                    998: restrictions.  However, if you need to break these restrictions, read
                    999: the following.
                   1000: 
                   1001: Accessing a stack or stack pointer directly can be a problem for several
                   1002: reasons: 
1.12      anton    1003: @cindex stack caching, restriction on C code
                   1004: @cindex superinstructions, restrictions on components
1.5       anton    1005: 
1.11      anton    1006: @itemize @bullet
1.5       anton    1007: 
                   1008: @item
1.12      anton    1009: Vmgen optionally supports caching the top-of-stack item in a local
                   1010: variable (that is allocated to a register).  This is the most frequent
                   1011: source of trouble.  You can deal with it either by not using
                   1012: top-of-stack caching (slowdown factor 1-1.4, depending on machine), or
                   1013: by inserting flushing code (e.g., @samp{IF_spTOS(sp[...] = spTOS);}) at
                   1014: the start and reloading code (e.g., @samp{IF_spTOS(spTOS = sp[0])}) at
                   1015: the end of problematic C code.  Vmgen inserts a stack pointer update
                   1016: before the start of the user-supplied C code, so the flushing code has
                   1017: to use an index that corrects for that.  In the future, this flushing
                   1018: may be done automatically by mentioning a special string in the C code.
1.5       anton    1019: @c sometimes flushing and/or reloading unnecessary
                   1020: 
                   1021: @item
1.11      anton    1022: The Vmgen-erated code loads the stack items from stack-pointer-indexed
1.5       anton    1023: memory into variables before the user-supplied C code, and stores them
                   1024: from variables to stack-pointer-indexed memory afterwards.  If you do
                   1025: any writes to the stack through its stack pointer in your C code, it
1.13      anton    1026: will not affect the variables, and your write may be overwritten by the
1.5       anton    1027: stores after the C code.  Similarly, a read from a stack using a stack
                   1028: pointer will not reflect computations of stack items in the same VM
                   1029: instruction.
                   1030: 
                   1031: @item
                   1032: Superinstructions keep stack items in variables across the whole
                   1033: superinstruction.  So you should not include VM instructions, that
1.12      anton    1034: access a stack or stack pointer, as components of superinstructions
                   1035: (@pxref{VM profiler}).
1.5       anton    1036: 
                   1037: @end itemize
                   1038: 
                   1039: You should access the instruction pointer only through its special
                   1040: macros (@samp{IP}, @samp{SET_IP}, @samp{IPTOS}); this ensure that these
                   1041: macros can be implemented in several ways for best performance.
                   1042: @samp{IP} points to the next instruction, and @samp{IPTOS} is its
                   1043: contents.
                   1044: 
1.22      anton    1045: @c --------------------------------------------------------------------
                   1046: @node Stack growth direction,  , C Code restrictions, Simple instructions
                   1047: @subsection Stack growth direction
                   1048: @cindex stack growth direction
                   1049: 
                   1050: @cindex @code{stack-access-transform}
                   1051: By default, the stacks grow towards lower addresses.  You can change
                   1052: this for a stack by setting the @code{stack-access-transform} field of
                   1053: the stack to an xt @code{( itemnum -- index )} that performs the
                   1054: appropriate index transformation.
                   1055: 
                   1056: E.g., if you want to let @code{data-stack} grow towards higher
                   1057: addresses, with the stack pointer always pointing just beyond the
                   1058: top-of-stack, use this right after defining @code{data-stack}:
                   1059: 
                   1060: @example
                   1061: \E : sp-access-transform ( itemnum -- index ) negate 1- ;
                   1062: \E ' sp-access-transform ' data-stack >body stack-access-transform !
                   1063: @end example
                   1064: 
                   1065: This means that @code{sp-access-transform} will be used to generate
                   1066: indexes for accessing @code{data-stack}.  The definition of
                   1067: @code{sp-access-transform} above transforms n into -n-1, e.g, 1 into -2.
                   1068: This will access the 0th data-stack element (top-of-stack) at sp[-1],
                   1069: the 1st at sp[-2], etc., which is the typical way upward-growing
                   1070: stacks are used.  If you need a different transform and do not know
                   1071: enough Forth to program it, let me know.
1.5       anton    1072: 
1.10      anton    1073: @c --------------------------------------------------------------------
1.18      anton    1074: @node Superinstructions, Store Optimization, Simple instructions, Input File Format
1.3       anton    1075: @section Superinstructions
1.12      anton    1076: @cindex superinstructions, defining
                   1077: @cindex defining superinstructions
1.5       anton    1078: 
1.8       anton    1079: Note: don't invest too much work in (static) superinstructions; a future
1.11      anton    1080: version of Vmgen will support dynamic superinstructions (see Ian
1.8       anton    1081: Piumarta and Fabio Riccardi, @cite{Optimizing Direct Threaded Code by
                   1082: Selective Inlining}, PLDI'98), and static superinstructions have much
1.12      anton    1083: less benefit in that context (preliminary results indicate only a factor
                   1084: 1.1 speedup).
1.8       anton    1085: 
1.5       anton    1086: Here is an example of a superinstruction definition:
                   1087: 
                   1088: @example
                   1089: lit_sub = lit sub
                   1090: @end example
                   1091: 
                   1092: @code{lit_sub} is the name of the superinstruction, and @code{lit} and
                   1093: @code{sub} are its components.  This superinstruction performs the same
                   1094: action as the sequence @code{lit} and @code{sub}.  It is generated
                   1095: automatically by the VM code generation functions whenever that sequence
1.11      anton    1096: occurs, so if you want to use this superinstruction, you just need to
                   1097: add this definition (and even that can be partially automatized,
                   1098: @pxref{VM profiler}).
1.5       anton    1099: 
1.12      anton    1100: @cindex prefixes of superinstructions
1.5       anton    1101: Vmgen requires that the component instructions are simple instructions
1.11      anton    1102: defined before superinstructions using the components.  Currently, Vmgen
1.5       anton    1103: also requires that all the subsequences at the start of a
                   1104: superinstruction (prefixes) must be defined as superinstruction before
                   1105: the superinstruction.  I.e., if you want to define a superinstruction
                   1106: 
                   1107: @example
1.12      anton    1108: foo4 = load add sub mul
1.5       anton    1109: @end example
                   1110: 
1.12      anton    1111: you first have to define @code{load}, @code{add}, @code{sub} and
                   1112: @code{mul}, plus
1.5       anton    1113: 
                   1114: @example
1.12      anton    1115: foo2 = load add
                   1116: foo3 = load add sub
1.5       anton    1117: @end example
                   1118: 
                   1119: Here, @code{sumof4} is the longest prefix of @code{sumof5}, and @code{sumof3}
                   1120: is the longest prefix of @code{sumof4}.
                   1121: 
1.11      anton    1122: Note that Vmgen assumes that only the code it generates accesses stack
1.5       anton    1123: pointers, the instruction pointer, and various stack items, and it
                   1124: performs optimizations based on this assumption.  Therefore, VM
1.12      anton    1125: instructions where your C code changes the instruction pointer should
                   1126: only be used as last component; a VM instruction where your C code
                   1127: accesses a stack pointer should not be used as component at all.  Vmgen
                   1128: does not check these restrictions, they just result in bugs in your
                   1129: interpreter.
1.5       anton    1130: 
1.22      anton    1131: @cindex include-skipped-insts
                   1132: The Vmgen flag @code{include-skipped-insts} influences superinstruction
                   1133: code generation.  Currently there is no support in the peephole
                   1134: optimizer for both variations, so leave this flag alone for now.
                   1135: 
1.12      anton    1136: @c -------------------------------------------------------------------
1.18      anton    1137: @node  Store Optimization, Register Machines, Superinstructions, Input File Format
                   1138: @section Store Optimization
                   1139: @cindex store optimization
                   1140: @cindex optimization, stack stores
                   1141: @cindex stack stores, optimization
                   1142: @cindex eliminating stack stores
                   1143: 
                   1144: This minor optimization (0.6\%--0.8\% reduction in executed instructions
                   1145: for Gforth) puts additional requirements on the instruction descriptions
                   1146: and is therefore disabled by default.
                   1147: 
                   1148: What does it do?  Consider an instruction like
                   1149: 
                   1150: @example
                   1151: dup ( n -- n n )
                   1152: @end example
                   1153: 
                   1154: For simplicity, also assume that we are not caching the top-of-stack in
                   1155: a register.  Now, the C code for dup first loads @code{n} from the
                   1156: stack, and then stores it twice to the stack, one time to the address
                   1157: where it came from; that time is unnecessary, but gcc does not optimize
                   1158: it away, so vmgen can do it instead (if you turn on the store
                   1159: optimization).
                   1160: 
                   1161: Vmgen uses the stack item's name to determine if the stack item contains
                   1162: the same value as it did at the start.  Therefore, if you use the store
                   1163: optimization, you have to ensure that stack items that have the same
                   1164: name on input and output also have the same value, and are not changed
                   1165: in the C code you supply.  I.e., the following code could fail if you
                   1166: turn on the store optimization:
                   1167: 
                   1168: @example
                   1169: add1 ( n -- n )
                   1170: n++;
                   1171: @end example
                   1172: 
                   1173: Instead, you have to use different names, i.e.:
                   1174: 
                   1175: @example
1.27      anton    1176: add1 ( n1 -- n2 )
1.18      anton    1177: n2=n1+1;
                   1178: @end example
                   1179: 
1.22      anton    1180: Similarly, the store optimization assumes that the stack pointer is only
                   1181: changed by Vmgen-erated code.  If your C code changes the stack pointer,
                   1182: use different names in input and output stack items to avoid a (probably
                   1183: wrong) store optimization, or turn the store optimization off for this
                   1184: VM instruction.
                   1185: 
1.18      anton    1186: To turn on the store optimization, write
                   1187: 
                   1188: @example
                   1189: \E store-optimization on
                   1190: @end example
                   1191: 
                   1192: at the start of the file.  You can turn this optimization on or off
                   1193: between any two VM instruction descriptions.  For turning it off again,
                   1194: you can use
                   1195: 
                   1196: @example
                   1197: \E store-optimization off
                   1198: @end example
                   1199: 
                   1200: @c -------------------------------------------------------------------
                   1201: @node Register Machines,  , Store Optimization, Input File Format
1.11      anton    1202: @section Register Machines
1.12      anton    1203: @cindex Register VM
                   1204: @cindex Superinstructions for register VMs
                   1205: @cindex tracing of register VMs
1.11      anton    1206: 
                   1207: If you want to implement a register VM rather than a stack VM with
                   1208: Vmgen, there are two ways to do it: Directly and through
                   1209: superinstructions.
                   1210: 
                   1211: If you use the direct way, you define instructions that take the
                   1212: register numbers as immediate arguments, like this:
                   1213: 
                   1214: @example
                   1215: add3 ( #src1 #src2 #dest -- )
                   1216: reg[dest] = reg[src1]+reg[src2];
                   1217: @end example
                   1218: 
1.12      anton    1219: A disadvantage of this method is that during tracing you only see the
                   1220: register numbers, but not the register contents.  Actually, with an
                   1221: appropriate definition of @code{printarg_src} (@pxref{VM engine}), you
                   1222: can print the values of the source registers on entry, but you cannot
                   1223: print the value of the destination register on exit.
                   1224: 
1.11      anton    1225: If you use superinstructions to define a register VM, you define simple
                   1226: instructions that use a stack, and then define superinstructions that
                   1227: have no overall stack effect, like this:
                   1228: 
                   1229: @example
                   1230: loadreg ( #src -- n )
                   1231: n = reg[src];
                   1232: 
                   1233: storereg ( n #dest -- )
                   1234: reg[dest] = n;
                   1235: 
                   1236: adds ( n1 n2 -- n )
                   1237: n = n1+n2;
                   1238: 
                   1239: add3 = loadreg loadreg adds storereg
                   1240: @end example
                   1241: 
                   1242: An advantage of this method is that you see the values and not just the
1.12      anton    1243: register numbers in tracing.  A disadvantage of this method is that
1.11      anton    1244: currently you cannot generate superinstructions directly, but only
                   1245: through generating a sequence of simple instructions (we might change
                   1246: this in the future if there is demand).
                   1247: 
                   1248: Could the register VM support be improved, apart from the issues
                   1249: mentioned above?  It is hard to see how to do it in a general way,
                   1250: because there are a number of different designs that different people
                   1251: mean when they use the term @emph{register machine} in connection with
                   1252: VM interpreters.  However, if you have ideas or requests in that
                   1253: direction, please let me know (@pxref{Contact}).
                   1254: 
1.5       anton    1255: @c ********************************************************************
1.13      anton    1256: @node Error messages, Using the generated code, Input File Format, Top
                   1257: @chapter Error messages
                   1258: @cindex error messages
                   1259: 
                   1260: These error messages are created by Vmgen:
                   1261: 
                   1262: @table @code
                   1263: 
                   1264: @cindex @code{# can only be on the input side} error
                   1265: @item # can only be on the input side
                   1266: You have used an instruction-stream prefix (usually @samp{#}) after the
                   1267: @samp{--} (the output side); you can only use it before (the input
                   1268: side).
                   1269: 
                   1270: @cindex @code{prefix for this combination must be defined earlier} error
1.20      anton    1271: @item the prefix for this superinstruction must be defined earlier
1.13      anton    1272: You have defined a superinstruction (e.g. @code{abc = a b c}) without
                   1273: defining its direct prefix (e.g., @code{ab = a b}),
                   1274: @xref{Superinstructions}.
                   1275: 
                   1276: @cindex @code{sync line syntax} error
                   1277: @item sync line syntax
                   1278: If you are using a preprocessor (e.g., @command{m4}) to generate Vmgen
                   1279: input code, you may want to create @code{#line} directives (aka sync
                   1280: lines).  This error indicates that such a line is not in th syntax
1.16      anton    1281: expected by Vmgen (this should not happen; please report the offending
                   1282: line in a bug report).
1.13      anton    1283: 
                   1284: @cindex @code{syntax error, wrong char} error
1.22      anton    1285: @item syntax error, wrong char
1.16      anton    1286: A syntax error.  If you do not see right away where the error is, it may
                   1287: be helpful to check the following: Did you put an empty line in a VM
                   1288: instruction where the C code is not delimited by braces (then the empty
                   1289: line ends the VM instruction)?  If you used brace-delimited C code, did
                   1290: you put the delimiting braces (and only those) at the start of the line,
                   1291: without preceding white space?  Did you forget a delimiting brace?
1.13      anton    1292: 
                   1293: @cindex @code{too many stacks} error
                   1294: @item too many stacks
1.16      anton    1295: Vmgen currently supports 3 stacks (plus the instruction stream); if you
                   1296: need more, let us know.
1.13      anton    1297: 
                   1298: @cindex @code{unknown prefix} error
                   1299: @item unknown prefix
                   1300: The stack item does not match any defined type prefix (after stripping
                   1301: away any stack prefix).  You should either declare the type prefix you
                   1302: want for that stack item, or use a different type prefix
                   1303: 
1.22      anton    1304: @cindex @code{unknown primitive} error
1.13      anton    1305: @item unknown primitive
                   1306: You have used the name of a simple VM instruction in a superinstruction
                   1307: definition without defining the simple VM instruction first.
                   1308: 
                   1309: @end table
                   1310: 
                   1311: In addition, the C compiler can produce errors due to code produced by
                   1312: Vmgen; e.g., you need to define type cast functions.
                   1313: 
                   1314: @c ********************************************************************
                   1315: @node Using the generated code, Hints, Error messages, Top
1.5       anton    1316: @chapter Using the generated code
1.12      anton    1317: @cindex generated code, usage
                   1318: @cindex Using vmgen-erated code
1.5       anton    1319: 
1.11      anton    1320: The easiest way to create a working VM interpreter with Vmgen is
1.12      anton    1321: probably to start with @file{vmgen-ex}, and modify it for your purposes.
1.13      anton    1322: This chapter explains what the various wrapper and generated files do.
                   1323: It also contains reference-manual style descriptions of the macros,
                   1324: variables etc. used by the generated code, and you can skip that on
                   1325: first reading.
1.5       anton    1326: 
1.10      anton    1327: @menu
                   1328: * VM engine::                   Executing VM code
                   1329: * VM instruction table::        
                   1330: * VM code generation::          Creating VM code (in the front-end)
                   1331: * Peephole optimization::       Creating VM superinstructions
                   1332: * VM disassembler::             for debugging the front end
                   1333: * VM profiler::                 for finding worthwhile superinstructions
                   1334: @end menu
1.6       anton    1335: 
1.10      anton    1336: @c --------------------------------------------------------------------
                   1337: @node VM engine, VM instruction table, Using the generated code, Using the generated code
1.5       anton    1338: @section VM engine
1.12      anton    1339: @cindex VM instruction execution
                   1340: @cindex engine
                   1341: @cindex executing VM code
                   1342: @cindex @file{engine.c}
                   1343: @cindex @file{-vm.i} output file
1.5       anton    1344: 
                   1345: The VM engine is the VM interpreter that executes the VM code.  It is
                   1346: essential for an interpretive system.
                   1347: 
1.6       anton    1348: Vmgen supports two methods of VM instruction dispatch: @emph{threaded
                   1349: code} (fast, but gcc-specific), and @emph{switch dispatch} (slow, but
                   1350: portable across C compilers); you can use conditional compilation
                   1351: (@samp{defined(__GNUC__)}) to choose between these methods, and our
                   1352: example does so.
                   1353: 
                   1354: For both methods, the VM engine is contained in a C-level function.
                   1355: Vmgen generates most of the contents of the function for you
                   1356: (@file{@var{name}-vm.i}), but you have to define this function, and
                   1357: macros and variables used in the engine, and initialize the variables.
                   1358: In our example the engine function also includes
                   1359: @file{@var{name}-labels.i} (@pxref{VM instruction table}).
                   1360: 
1.12      anton    1361: @cindex tracing VM code
1.13      anton    1362: @cindex superinstructions and tracing
1.12      anton    1363: In addition to executing the code, the VM engine can optionally also
                   1364: print out a trace of the executed instructions, their arguments and
                   1365: results.  For superinstructions it prints the trace as if only component
                   1366: instructions were executed; this allows to introduce new
                   1367: superinstructions while keeping the traces comparable to old ones
                   1368: (important for regression tests).
                   1369: 
                   1370: It costs significant performance to check in each instruction whether to
                   1371: print tracing code, so we recommend producing two copies of the engine:
                   1372: one for fast execution, and one for tracing.  See the rules for
                   1373: @file{engine.o} and @file{engine-debug.o} in @file{vmgen-ex/Makefile}
                   1374: for an example.
                   1375: 
1.6       anton    1376: The following macros and variables are used in @file{@var{name}-vm.i}:
1.5       anton    1377: 
                   1378: @table @code
                   1379: 
1.12      anton    1380: @findex LABEL
1.5       anton    1381: @item LABEL(@var{inst_name})
                   1382: This is used just before each VM instruction to provide a jump or
1.11      anton    1383: @code{switch} label (the @samp{:} is provided by Vmgen).  For switch
1.13      anton    1384: dispatch this should expand to @samp{case @var{label}:}; for
                   1385: threaded-code dispatch this should just expand to @samp{@var{label}:}.
1.12      anton    1386: In either case @var{label} is usually the @var{inst_name} with some
                   1387: prefix or suffix to avoid naming conflicts.
1.5       anton    1388: 
1.12      anton    1389: @findex LABEL2
1.9       anton    1390: @item LABEL2(@var{inst_name})
                   1391: This will be used for dynamic superinstructions; at the moment, this
                   1392: should expand to nothing.
                   1393: 
1.12      anton    1394: @findex NAME
1.5       anton    1395: @item NAME(@var{inst_name_string})
                   1396: Called on entering a VM instruction with a string containing the name of
1.13      anton    1397: the VM instruction as parameter.  In normal execution this should be
                   1398: expand to nothing, but for tracing this usually prints the name, and
                   1399: possibly other information (several VM registers in our example).
1.5       anton    1400: 
1.12      anton    1401: @findex DEF_CA
1.5       anton    1402: @item DEF_CA
                   1403: Usually empty.  Called just inside a new scope at the start of a VM
                   1404: instruction.  Can be used to define variables that should be visible
                   1405: during every VM instruction.  If you define this macro as non-empty, you
                   1406: have to provide the finishing @samp{;} in the macro.
                   1407: 
1.12      anton    1408: @findex NEXT_P0
                   1409: @findex NEXT_P1
                   1410: @findex NEXT_P2
1.5       anton    1411: @item NEXT_P0 NEXT_P1 NEXT_P2
                   1412: The three parts of instruction dispatch.  They can be defined in
                   1413: different ways for best performance on various processors (see
                   1414: @file{engine.c} in the example or @file{engine/threaded.h} in Gforth).
1.12      anton    1415: @samp{NEXT_P0} is invoked right at the start of the VM instruction (but
1.5       anton    1416: after @samp{DEF_CA}), @samp{NEXT_P1} right after the user-supplied C
                   1417: code, and @samp{NEXT_P2} at the end.  The actual jump has to be
1.13      anton    1418: performed by @samp{NEXT_P2} (if you would do it earlier, important parts
                   1419: of the VM instruction would not be executed).
1.5       anton    1420: 
                   1421: The simplest variant is if @samp{NEXT_P2} does everything and the other
                   1422: macros do nothing.  Then also related macros like @samp{IP},
                   1423: @samp{SET_IP}, @samp{IP}, @samp{INC_IP} and @samp{IPTOS} are very
                   1424: straightforward to define.  For switch dispatch this code consists just
1.12      anton    1425: of a jump to the dispatch code (@samp{goto next_inst;} in our example);
1.5       anton    1426: for direct threaded code it consists of something like
1.11      anton    1427: @samp{(@{cfa=*ip++; goto *cfa;@})}.
1.5       anton    1428: 
1.12      anton    1429: Pulling code (usually the @samp{cfa=*ip++;}) up into @samp{NEXT_P1}
1.5       anton    1430: usually does not cause problems, but pulling things up into
                   1431: @samp{NEXT_P0} usually requires changing the other macros (and, at least
                   1432: for Gforth on Alpha, it does not buy much, because the compiler often
                   1433: manages to schedule the relevant stuff up by itself).  An even more
                   1434: extreme variant is to pull code up even further, into, e.g., NEXT_P1 of
                   1435: the previous VM instruction (prefetching, useful on PowerPCs).
                   1436: 
1.12      anton    1437: @findex INC_IP
1.5       anton    1438: @item INC_IP(@var{n})
1.8       anton    1439: This increments @code{IP} by @var{n}.
                   1440: 
1.12      anton    1441: @findex SET_IP
1.8       anton    1442: @item SET_IP(@var{target})
                   1443: This sets @code{IP} to @var{target}.
1.5       anton    1444: 
1.12      anton    1445: @cindex type cast macro
                   1446: @findex vm_@var{A}2@var{B}
1.5       anton    1447: @item vm_@var{A}2@var{B}(a,b)
                   1448: Type casting macro that assigns @samp{a} (of type @var{A}) to @samp{b}
                   1449: (of type @var{B}).  This is mainly used for getting stack items into
                   1450: variables and back.  So you need to define macros for every combination
                   1451: of stack basic type (@code{Cell} in our example) and type-prefix types
                   1452: used with that stack (in both directions).  For the type-prefix type,
                   1453: you use the type-prefix (not the C type string) as type name (e.g.,
                   1454: @samp{vm_Cell2i}, not @samp{vm_Cell2Cell}).  In addition, you have to
1.12      anton    1455: define a vm_@var{X}2@var{X} macro for the stack's basic type @var{X}
                   1456: (used in superinstructions).
1.5       anton    1457: 
1.12      anton    1458: @cindex instruction stream, basic type
1.5       anton    1459: The stack basic type for the predefined @samp{inst-stream} is
                   1460: @samp{Cell}.  If you want a stack with the same item size, making its
                   1461: basic type @samp{Cell} usually reduces the number of macros you have to
                   1462: define.
                   1463: 
1.12      anton    1464: @cindex unions in type cast macros
                   1465: @cindex casts in type cast macros
                   1466: @cindex type casting between floats and integers
1.5       anton    1467: Here our examples differ a lot: @file{vmgen-ex} uses casts in these
                   1468: macros, whereas @file{vmgen-ex2} uses union-field selection (or
1.12      anton    1469: assignment to union fields).  Note that casting floats into integers and
                   1470: vice versa changes the bit pattern (and you do not want that).  In this
                   1471: case your options are to use a (temporary) union, or to take the address
                   1472: of the value, cast the pointer, and dereference that (not always
                   1473: possible, and sometimes expensive).
1.5       anton    1474: 
1.12      anton    1475: @findex vm_two@var{A}2@var{B}
                   1476: @findex vm_@var{B}2two@var{A}
1.5       anton    1477: @item vm_two@var{A}2@var{B}(a1,a2,b)
                   1478: @item vm_@var{B}2two@var{A}(b,a1,a2)
1.12      anton    1479: Type casting between two stack items (@code{a1}, @code{a2}) and a
1.5       anton    1480: variable @code{b} of a type that takes two stack items.  This does not
1.12      anton    1481: occur in our small examples, but you can look at Gforth for examples
                   1482: (see @code{vm_twoCell2d} in @file{engine/forth.h}).
1.5       anton    1483: 
1.12      anton    1484: @cindex stack pointer definition
                   1485: @cindex instruction pointer definition
1.5       anton    1486: @item @var{stackpointer}
                   1487: For each stack used, the stackpointer name given in the stack
                   1488: declaration is used.  For a regular stack this must be an l-expression;
                   1489: typically it is a variable declared as a pointer to the stack's basic
                   1490: type.  For @samp{inst-stream}, the name is @samp{IP}, and it can be a
                   1491: plain r-value; typically it is a macro that abstracts away the
1.12      anton    1492: differences between the various implementations of @code{NEXT_P*}.
1.22      anton    1493: 
                   1494: @cindex IMM_ARG
                   1495: @findex IMM_ARG
                   1496: @item IMM_ARG(access,value)
                   1497: Define this to expland to ``(access)''.  This is just a placeholder for
                   1498: future extensions.
1.5       anton    1499: 
1.12      anton    1500: @cindex top of stack caching
                   1501: @cindex stack caching
                   1502: @cindex TOS
                   1503: @findex IPTOS
1.5       anton    1504: @item @var{stackpointer}TOS
                   1505: The top-of-stack for the stack pointed to by @var{stackpointer}.  If you
                   1506: are using top-of-stack caching for that stack, this should be defined as
                   1507: variable; if you are not using top-of-stack caching for that stack, this
                   1508: should be a macro expanding to @samp{@var{stackpointer}[0]}.  The stack
                   1509: pointer for the predefined @samp{inst-stream} is called @samp{IP}, so
                   1510: the top-of-stack is called @samp{IPTOS}.
                   1511: 
1.12      anton    1512: @findex IF_@var{stackpointer}TOS
1.5       anton    1513: @item IF_@var{stackpointer}TOS(@var{expr})
                   1514: Macro for executing @var{expr}, if top-of-stack caching is used for the
                   1515: @var{stackpointer} stack.  I.e., this should do @var{expr} if there is
                   1516: top-of-stack caching for @var{stackpointer}; otherwise it should do
                   1517: nothing.
                   1518: 
1.12      anton    1519: @findex SUPER_END
1.8       anton    1520: @item SUPER_END
                   1521: This is used by the VM profiler (@pxref{VM profiler}); it should not do
                   1522: anything in normal operation, and call @code{vm_count_block(IP)} for
                   1523: profiling.
                   1524: 
1.12      anton    1525: @findex SUPER_CONTINUE
1.8       anton    1526: @item SUPER_CONTINUE
1.11      anton    1527: This is just a hint to Vmgen and does nothing at the C level.
1.8       anton    1528: 
1.26      anton    1529: @findex MAYBE_UNUSED
                   1530: @item MAYBE_UNUSED
                   1531: This should be defined as @code{__attribute__((unused))} for gcc-2.7 and
                   1532: higher.  It suppresses the warnings about unused variables in the code
                   1533: for superinstructions.  You need to define this only if you are using
                   1534: superinstructions.
                   1535: 
1.12      anton    1536: @findex VM_DEBUG
1.5       anton    1537: @item VM_DEBUG
                   1538: If this is defined, the tracing code will be compiled in (slower
                   1539: interpretation, but better debugging).  Our example compiles two
                   1540: versions of the engine, a fast-running one that cannot trace, and one
                   1541: with potential tracing and profiling.
                   1542: 
1.12      anton    1543: @findex vm_debug
1.5       anton    1544: @item vm_debug
                   1545: Needed only if @samp{VM_DEBUG} is defined.  If this variable contains
                   1546: true, the VM instructions produce trace output.  It can be turned on or
                   1547: off at any time.
                   1548: 
1.12      anton    1549: @findex vm_out
1.5       anton    1550: @item vm_out
                   1551: Needed only if @samp{VM_DEBUG} is defined.  Specifies the file on which
                   1552: to print the trace output (type @samp{FILE *}).
                   1553: 
1.12      anton    1554: @findex printarg_@var{type}
1.5       anton    1555: @item printarg_@var{type}(@var{value})
                   1556: Needed only if @samp{VM_DEBUG} is defined.  Macro or function for
                   1557: printing @var{value} in a way appropriate for the @var{type}.  This is
                   1558: used for printing the values of stack items during tracing.  @var{Type}
                   1559: is normally the type prefix specified in a @code{type-prefix} definition
                   1560: (e.g., @samp{printarg_i}); in superinstructions it is currently the
                   1561: basic type of the stack.
                   1562: 
                   1563: @end table
                   1564: 
1.6       anton    1565: 
1.10      anton    1566: @c --------------------------------------------------------------------
                   1567: @node VM instruction table, VM code generation, VM engine, Using the generated code
                   1568: @section VM instruction table
1.12      anton    1569: @cindex instruction table
                   1570: @cindex opcode definition
                   1571: @cindex labels for threaded code
                   1572: @cindex @code{vm_prim}, definition
                   1573: @cindex @file{-labels.i} output file
1.6       anton    1574: 
                   1575: For threaded code we also need to produce a table containing the labels
                   1576: of all VM instructions.  This is needed for VM code generation
                   1577: (@pxref{VM code generation}), and it has to be done in the engine
                   1578: function, because the labels are not visible outside.  It then has to be
                   1579: passed outside the function (and assigned to @samp{vm_prim}), to be used
                   1580: by the VM code generation functions.
                   1581: 
                   1582: This means that the engine function has to be called first to produce
                   1583: the VM instruction table, and later, after generating VM code, it has to
                   1584: be called again to execute the generated VM code (yes, this is ugly).
                   1585: In our example program, these two modes of calling the engine function
                   1586: are differentiated by the value of the parameter ip0 (if it equals 0,
                   1587: then the table is passed out, otherwise the VM code is executed); in our
                   1588: example, we pass the table out by assigning it to @samp{vm_prim} and
                   1589: returning from @samp{engine}.
                   1590: 
1.12      anton    1591: In our example (@file{vmgen-ex/engine.c}), we also build such a table for
                   1592: switch dispatch; this is mainly done for uniformity.
1.6       anton    1593: 
                   1594: For switch dispatch, we also need to define the VM instruction opcodes
                   1595: used as case labels in an @code{enum}.
                   1596: 
                   1597: For both purposes (VM instruction table, and enum), the file
1.11      anton    1598: @file{@var{name}-labels.i} is generated by Vmgen.  You have to define
1.6       anton    1599: the following macro used in this file:
1.5       anton    1600: 
1.12      anton    1601: @table @code
1.5       anton    1602: 
1.12      anton    1603: @findex INST_ADDR
1.5       anton    1604: @item INST_ADDR(@var{inst_name})
                   1605: For switch dispatch, this is just the name of the switch label (the same
1.6       anton    1606: name as used in @samp{LABEL(@var{inst_name})}), for both uses of
                   1607: @file{@var{name}-labels.i}.  For threaded-code dispatch, this is the
                   1608: address of the label defined in @samp{LABEL(@var{inst_name})}); the
1.11      anton    1609: address is taken with @samp{&&} (@pxref{Labels as Values, , Labels as
                   1610: Values, gcc.info, GNU C Manual}).
1.5       anton    1611: 
                   1612: @end table
                   1613: 
                   1614: 
1.10      anton    1615: @c --------------------------------------------------------------------
                   1616: @node VM code generation, Peephole optimization, VM instruction table, Using the generated code
1.6       anton    1617: @section VM code generation
1.12      anton    1618: @cindex VM code generation
                   1619: @cindex code generation, VM
                   1620: @cindex @file{-gen.i} output file
1.6       anton    1621: 
                   1622: Vmgen generates VM code generation functions in @file{@var{name}-gen.i}
                   1623: that the front end can call to generate VM code.  This is essential for
                   1624: an interpretive system.
                   1625: 
1.12      anton    1626: @findex gen_@var{inst}
1.11      anton    1627: For a VM instruction @samp{x ( #a b #c -- d )}, Vmgen generates a
1.6       anton    1628: function with the prototype
                   1629: 
                   1630: @example
                   1631: void gen_x(Inst **ctp, a_type a, c_type c)
                   1632: @end example
                   1633: 
                   1634: The @code{ctp} argument points to a pointer to the next instruction.
                   1635: @code{*ctp} is increased by the generation functions; i.e., you should
                   1636: allocate memory for the code to be generated beforehand, and start with
                   1637: *ctp set at the start of this memory area.  Before running out of
                   1638: memory, allocate a new area, and generate a VM-level jump to the new
1.12      anton    1639: area (this overflow handling is not implemented in our examples).
1.6       anton    1640: 
1.12      anton    1641: @cindex immediate arguments, VM code generation
1.6       anton    1642: The other arguments correspond to the immediate arguments of the VM
                   1643: instruction (with their appropriate types as defined in the
                   1644: @code{type_prefix} declaration.
                   1645: 
                   1646: The following types, variables, and functions are used in
                   1647: @file{@var{name}-gen.i}:
                   1648: 
1.12      anton    1649: @table @code
1.6       anton    1650: 
1.12      anton    1651: @findex Inst
1.6       anton    1652: @item Inst
                   1653: The type of the VM instruction; if you use threaded code, this is
                   1654: @code{void *}; for switch dispatch this is an integer type.
                   1655: 
1.12      anton    1656: @cindex @code{vm_prim}, use
1.6       anton    1657: @item vm_prim
                   1658: The VM instruction table (type: @code{Inst *}, @pxref{VM instruction table}).
                   1659: 
1.12      anton    1660: @findex gen_inst
1.6       anton    1661: @item gen_inst(Inst **ctp, Inst i)
                   1662: This function compiles the instruction @code{i}.  Take a look at it in
                   1663: @file{vmgen-ex/peephole.c}.  It is trivial when you don't want to use
                   1664: superinstructions (just the last two lines of the example function), and
                   1665: slightly more complicated in the example due to its ability to use
                   1666: superinstructions (@pxref{Peephole optimization}).
                   1667: 
1.12      anton    1668: @findex genarg_@var{type_prefix}
1.6       anton    1669: @item genarg_@var{type_prefix}(Inst **ctp, @var{type} @var{type_prefix})
                   1670: This compiles an immediate argument of @var{type} (as defined in a
                   1671: @code{type-prefix} definition).  These functions are trivial to define
                   1672: (see @file{vmgen-ex/support.c}).  You need one of these functions for
                   1673: every type that you use as immediate argument.
                   1674: 
                   1675: @end table
                   1676: 
1.12      anton    1677: @findex BB_BOUNDARY
1.6       anton    1678: In addition to using these functions to generate code, you should call
                   1679: @code{BB_BOUNDARY} at every basic block entry point if you ever want to
                   1680: use superinstructions (or if you want to use the profiling supported by
1.12      anton    1681: Vmgen; but this support is also useful mainly for selecting
                   1682: superinstructions).  If you use @code{BB_BOUNDARY}, you should also
                   1683: define it (take a look at its definition in @file{vmgen-ex/mini.y}).
1.6       anton    1684: 
                   1685: You do not need to call @code{BB_BOUNDARY} after branches, because you
                   1686: will not define superinstructions that contain branches in the middle
                   1687: (and if you did, and it would work, there would be no reason to end the
                   1688: superinstruction at the branch), and because the branches announce
                   1689: themselves to the profiler.
                   1690: 
                   1691: 
1.10      anton    1692: @c --------------------------------------------------------------------
                   1693: @node Peephole optimization, VM disassembler, VM code generation, Using the generated code
1.6       anton    1694: @section Peephole optimization
1.12      anton    1695: @cindex peephole optimization
                   1696: @cindex superinstructions, generating
                   1697: @cindex @file{peephole.c}
                   1698: @cindex @file{-peephole.i} output file
1.6       anton    1699: 
                   1700: You need peephole optimization only if you want to use
                   1701: superinstructions.  But having the code for it does not hurt much if you
                   1702: do not use superinstructions.
                   1703: 
                   1704: A simple greedy peephole optimization algorithm is used for
                   1705: superinstruction selection: every time @code{gen_inst} compiles a VM
1.12      anton    1706: instruction, it checks if it can combine it with the last VM instruction
1.6       anton    1707: (which may also be a superinstruction resulting from a previous peephole
                   1708: optimization); if so, it changes the last instruction to the combined
                   1709: instruction instead of laying down @code{i} at the current @samp{*ctp}.
                   1710: 
                   1711: The code for peephole optimization is in @file{vmgen-ex/peephole.c}.
                   1712: You can use this file almost verbatim.  Vmgen generates
1.26      anton    1713: @file{@var{file}-peephole.i} which contains data for the peephole
1.6       anton    1714: optimizer.
                   1715: 
1.12      anton    1716: @findex init_peeptable
1.6       anton    1717: You have to call @samp{init_peeptable()} after initializing
                   1718: @samp{vm_prim}, and before compiling any VM code to initialize data
                   1719: structures for peephole optimization.  After that, compiling with the VM
                   1720: code generation functions will automatically combine VM instructions
                   1721: into superinstructions.  Since you do not want to combine instructions
                   1722: across VM branch targets (otherwise there will not be a proper VM
                   1723: instruction to branch to), you have to call @code{BB_BOUNDARY}
                   1724: (@pxref{VM code generation}) at branch targets.
                   1725: 
                   1726: 
1.10      anton    1727: @c --------------------------------------------------------------------
                   1728: @node VM disassembler, VM profiler, Peephole optimization, Using the generated code
1.6       anton    1729: @section VM disassembler
1.12      anton    1730: @cindex VM disassembler
                   1731: @cindex disassembler, VM code
                   1732: @cindex @file{disasm.c}
                   1733: @cindex @file{-disasm.i} output file
1.6       anton    1734: 
                   1735: A VM code disassembler is optional for an interpretive system, but
                   1736: highly recommended during its development and maintenance, because it is
                   1737: very useful for detecting bugs in the front end (and for distinguishing
                   1738: them from VM interpreter bugs).
                   1739: 
                   1740: Vmgen supports VM code disassembling by generating
                   1741: @file{@var{file}-disasm.i}.  This code has to be wrapped into a
1.12      anton    1742: function, as is done in @file{vmgen-ex/disasm.c}.  You can use this file
1.6       anton    1743: almost verbatim.  In addition to @samp{vm_@var{A}2@var{B}(a,b)},
                   1744: @samp{vm_out}, @samp{printarg_@var{type}(@var{value})}, which are
                   1745: explained above, the following macros and variables are used in
                   1746: @file{@var{file}-disasm.i} (and you have to define them):
                   1747: 
1.12      anton    1748: @table @code
1.6       anton    1749: 
                   1750: @item ip
                   1751: This variable points to the opcode of the current VM instruction.
                   1752: 
1.12      anton    1753: @cindex @code{IP}, @code{IPTOS} in disassmbler
1.6       anton    1754: @item IP IPTOS
                   1755: @samp{IPTOS} is the first argument of the current VM instruction, and
                   1756: @samp{IP} points to it; this is just as in the engine, but here
                   1757: @samp{ip} points to the opcode of the VM instruction (in contrast to the
                   1758: engine, where @samp{ip} points to the next cell, or even one further).
                   1759: 
1.12      anton    1760: @findex VM_IS_INST
1.6       anton    1761: @item VM_IS_INST(Inst i, int n)
                   1762: Tests if the opcode @samp{i} is the same as the @samp{n}th entry in the
                   1763: VM instruction table.
                   1764: 
                   1765: @end table
                   1766: 
                   1767: 
1.10      anton    1768: @c --------------------------------------------------------------------
                   1769: @node VM profiler,  , VM disassembler, Using the generated code
1.7       anton    1770: @section VM profiler
1.12      anton    1771: @cindex VM profiler
                   1772: @cindex profiling for selecting superinstructions
                   1773: @cindex superinstructions and profiling
                   1774: @cindex @file{profile.c}
                   1775: @cindex @file{-profile.i} output file
1.7       anton    1776: 
                   1777: The VM profiler is designed for getting execution and occurence counts
                   1778: for VM instruction sequences, and these counts can then be used for
                   1779: selecting sequences as superinstructions.  The VM profiler is probably
1.8       anton    1780: not useful as profiling tool for the interpretive system.  I.e., the VM
1.7       anton    1781: profiler is useful for the developers, but not the users of the
1.8       anton    1782: interpretive system.
1.7       anton    1783: 
1.8       anton    1784: The output of the profiler is: for each basic block (executed at least
                   1785: once), it produces the dynamic execution count of that basic block and
                   1786: all its subsequences; e.g.,
1.7       anton    1787: 
1.8       anton    1788: @example
                   1789:        9227465  lit storelocal 
                   1790:        9227465  storelocal branch 
                   1791:        9227465  lit storelocal branch 
                   1792: @end example
1.7       anton    1793: 
1.8       anton    1794: I.e., a basic block consisting of @samp{lit storelocal branch} is
                   1795: executed 9227465 times.
1.6       anton    1796: 
1.12      anton    1797: @cindex @file{stat.awk}
                   1798: @cindex @file{seq2rule.awk}
1.8       anton    1799: This output can be combined in various ways.  E.g.,
1.12      anton    1800: @file{vmgen-ex/stat.awk} adds up the occurences of a given sequence wrt
1.8       anton    1801: dynamic execution, static occurence, and per-program occurence.  E.g.,
1.3       anton    1802: 
1.8       anton    1803: @example
                   1804:       2      16        36910041 loadlocal lit 
                   1805: @end example
1.2       anton    1806: 
1.12      anton    1807: @noindent
1.8       anton    1808: indicates that the sequence @samp{loadlocal lit} occurs in 2 programs,
                   1809: in 16 places, and has been executed 36910041 times.  Now you can select
                   1810: superinstructions in any way you like (note that compile time and space
                   1811: typically limit the number of superinstructions to 100--1000).  After
                   1812: you have done that, @file{vmgen/seq2rule.awk} turns lines of the form
1.11      anton    1813: above into rules for inclusion in a Vmgen input file.  Note that this
1.8       anton    1814: script does not ensure that all prefixes are defined, so you have to do
                   1815: that in other ways.  So, an overall script for turning profiles into
                   1816: superinstructions can look like this:
1.2       anton    1817: 
1.8       anton    1818: @example
                   1819: awk -f stat.awk fib.prof test.prof|
                   1820: awk '$3>=10000'|                #select sequences
                   1821: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
                   1822: awk -f seq2rule.awk|            #turn into superinstructions
                   1823: sort -k 3 >mini-super.vmg       #sort sequences
                   1824: @end example
1.2       anton    1825: 
1.8       anton    1826: Here the dynamic count is used for selecting sequences (preliminary
                   1827: results indicate that the static count gives better results, though);
1.12      anton    1828: the third line eliminates sequences containing instructions that must not
1.8       anton    1829: occur in a superinstruction, because they access a stack directly.  The
                   1830: dynamic count selection ensures that all subsequences (including
                   1831: prefixes) of longer sequences occur (because subsequences have at least
                   1832: the same count as the longer sequences); the sort in the last line
                   1833: ensures that longer superinstructions occur after their prefixes.
                   1834: 
1.12      anton    1835: But before using this, you have to have the profiler.  Vmgen supports its
1.8       anton    1836: creation by generating @file{@var{file}-profile.i}; you also need the
                   1837: wrapper file @file{vmgen-ex/profile.c} that you can use almost verbatim.
                   1838: 
1.12      anton    1839: @cindex @code{SUPER_END} in profiling
                   1840: @cindex @code{BB_BOUNDARY} in profiling
1.8       anton    1841: The profiler works by recording the targets of all VM control flow
                   1842: changes (through @code{SUPER_END} during execution, and through
                   1843: @code{BB_BOUNDARY} in the front end), and counting (through
                   1844: @code{SUPER_END}) how often they were targeted.  After the program run,
                   1845: the numbers are corrected such that each VM basic block has the correct
1.12      anton    1846: count (entering a block without executing a branch does not increase the
                   1847: count, and the correction fixes that), then the subsequences of all
                   1848: basic blocks are printed.  To get all this, you just have to define
                   1849: @code{SUPER_END} (and @code{BB_BOUNDARY}) appropriately, and call
                   1850: @code{vm_print_profile(FILE *file)} when you want to output the profile
                   1851: on @code{file}.
1.8       anton    1852: 
1.12      anton    1853: @cindex @code{VM_IS_INST} in profiling
                   1854: The @file{@var{file}-profile.i} is similar to the disassembler file, and
1.8       anton    1855: it uses variables and functions defined in @file{vmgen-ex/profile.c},
                   1856: plus @code{VM_IS_INST} already defined for the VM disassembler
                   1857: (@pxref{VM disassembler}).
                   1858: 
1.13      anton    1859: @c **********************************************************
                   1860: @node Hints, The future, Using the generated code, Top
                   1861: @chapter Hints
                   1862: @cindex hints
                   1863: 
                   1864: @menu
                   1865: * Floating point::              and stacks
                   1866: @end menu
                   1867: 
                   1868: @c --------------------------------------------------------------------
                   1869: @node Floating point,  , Hints, Hints
                   1870: @section Floating point
                   1871: 
                   1872: How should you deal with floating point values?  Should you use the same
                   1873: stack as for integers/pointers, or a different one?  This section
                   1874: discusses this issue with a view on execution speed.
                   1875: 
                   1876: The simpler approach is to use a separate floating-point stack.  This
                   1877: allows you to choose FP value size without considering the size of the
                   1878: integers/pointers, and you avoid a number of performance problems.  The
                   1879: main downside is that this needs an FP stack pointer (and that may not
                   1880: fit in the register file on the 386 arhitecture, costing some
                   1881: performance, but comparatively little if you take the other option into
                   1882: account).  If you use a separate FP stack (with stack pointer @code{fp}),
                   1883: using an fpTOS is helpful on most machines, but some spill the fpTOS
                   1884: register into memory, and fpTOS should not be used there.
                   1885: 
                   1886: The other approach is to share one stack (pointed to by, say, @code{sp})
                   1887: between integer/pointer and floating-point values.  This is ok if you do
                   1888: not use @code{spTOS}.  If you do use @code{spTOS}, the compiler has to
                   1889: decide whether to put that variable into an integer or a floating point
                   1890: register, and the other type of operation becomes quite expensive on
                   1891: most machines (because moving values between integer and FP registers is
                   1892: quite expensive).  If a value of one type has to be synthesized out of
                   1893: two values of the other type (@code{double} types), things are even more
                   1894: interesting.
                   1895: 
                   1896: One way around this problem would be to not use the @code{spTOS}
                   1897: supported by Vmgen, but to use explicit top-of-stack variables (one for
                   1898: integers, one for FP values), and having a kind of accumulator+stack
                   1899: architecture (e.g., Ocaml bytecode uses this approach); however, this is
                   1900: a major change, and it's ramifications are not completely clear.
1.10      anton    1901: 
                   1902: @c **********************************************************
1.13      anton    1903: @node The future, Changes, Hints, Top
                   1904: @chapter The future
                   1905: @cindex future ideas
                   1906: 
1.21      anton    1907: We have a number of ideas for future versions of Vmgen.  However, there
1.13      anton    1908: are so many possible things to do that we would like some feedback from
                   1909: you.  What are you doing with Vmgen, what features are you missing, and
                   1910: why?
                   1911: 
                   1912: One idea we are thinking about is to generate just one @file{.c} file
                   1913: instead of letting you copy and adapt all the wrapper files (you would
                   1914: still have to define stuff like the type-specific macros, and stack
                   1915: pointers etc. somewhere).  The advantage would be that, if we change the
                   1916: wrapper files between versions, you would not need to integrate your
                   1917: changes and our changes to them; Vmgen would also be easier to use for
                   1918: beginners.  The main disadvantage of that is that it would reduce the
                   1919: flexibility of Vmgen a little (well, those who like flexibility could
                   1920: still patch the resulting @file{.c} file, like they are now doing for
                   1921: the wrapper files).  In any case, if you are doing things to the wrapper
                   1922: files that would cause problems in a generated-@file{.c}-file approach,
                   1923: please let us know.
                   1924: 
                   1925: @c **********************************************************
                   1926: @node Changes, Contact, The future, Top
1.8       anton    1927: @chapter Changes
1.12      anton    1928: @cindex Changes from old versions
1.8       anton    1929: 
1.19      anton    1930: User-visible changes between 0.5.9-20020822 and 0.5.9-20020901:
                   1931: 
                   1932: The store optimization is now disabled by default, but can be enabled by
                   1933: the user (@pxref{Store Optimization}).  Documentation for this
                   1934: optimization is also new.
                   1935: 
                   1936: User-visible changes between 0.5.9-20010501 and 0.5.9-20020822:
1.17      anton    1937: 
                   1938: There is now a manual (in info, HTML, Postscript, or plain text format).
                   1939: 
                   1940: There is the vmgen-ex2 variant of the vmgen-ex example; the new
                   1941: variant uses a union type instead of lots of casting.
                   1942: 
                   1943: Both variants of the example can now be compiled with an ANSI C compiler
                   1944: (using switch dispatch and losing quite a bit of performance); tested
                   1945: with @command{lcc}.
                   1946: 
1.11      anton    1947: Users of the gforth-0.5.9-20010501 version of Vmgen need to change
1.8       anton    1948: several things in their source code to use the current version.  I
                   1949: recommend keeping the gforth-0.5.9-20010501 version until you have
                   1950: completed the change (note that you can have several versions of Gforth
                   1951: installed at the same time).  I hope to avoid such incompatible changes
                   1952: in the future.
1.2       anton    1953: 
1.8       anton    1954: The required changes are:
                   1955: 
                   1956: @table @code
1.13      anton    1957: 
                   1958: @cindex @code{TAIL;}, changes
                   1959: @item TAIL;
                   1960: has been renamed into @code{INST_TAIL;} (less chance of an accidental
                   1961: match).
1.2       anton    1962: 
1.12      anton    1963: @cindex @code{vm_@var{A}2@var{B}}, changes
1.8       anton    1964: @item vm_@var{A}2@var{B}
                   1965: now takes two arguments.
                   1966: 
1.12      anton    1967: @cindex @code{vm_two@var{A}2@var{B}}, changes
1.8       anton    1968: @item vm_two@var{A}2@var{B}(b,a1,a2);
                   1969: changed to vm_two@var{A}2@var{B}(a1,a2,b) (note the absence of the @samp{;}).
                   1970: 
                   1971: @end table
1.2       anton    1972: 
1.8       anton    1973: Also some new macros have to be defined, e.g., @code{INST_ADDR}, and
                   1974: @code{LABEL}; some macros have to be defined in new contexts, e.g.,
                   1975: @code{VM_IS_INST} is now also needed in the disassembler.
1.4       anton    1976: 
1.12      anton    1977: @c *********************************************************
1.10      anton    1978: @node Contact, Copying This Manual, Changes, Top
1.8       anton    1979: @chapter Contact
1.17      anton    1980: 
                   1981: To report a bug, use
                   1982: @url{https://savannah.gnu.org/bugs/?func=addbug&group_id=2672}.
                   1983: 
                   1984: For discussion on Vmgen (e.g., how to use it), use the mailing list
                   1985: @email{bug-vmgen@@mail.freesoftware.fsf.org} (use
                   1986: @url{http://mail.gnu.org/mailman/listinfo/help-vmgen} to subscribe).
                   1987: 
                   1988: You can find vmgen information at
                   1989: @url{http://www.complang.tuwien.ac.at/anton/vmgen/}.
1.4       anton    1990: 
1.12      anton    1991: @c ***********************************************************
1.10      anton    1992: @node Copying This Manual, Index, Contact, Top
                   1993: @appendix Copying This Manual
                   1994: 
                   1995: @menu
                   1996: * GNU Free Documentation License::  License for copying this manual.
                   1997: @end menu
                   1998: 
                   1999: @include fdl.texi
                   2000: 
                   2001: 
                   2002: @node Index,  , Copying This Manual, Top
                   2003: @unnumbered Index
                   2004: 
                   2005: @printindex cp
                   2006: 
                   2007: @bye

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