1: \input texinfo @c -*-texinfo-*-
2: @comment %**start of header
3: @setfilename vmgen.info
4: @include version.texi
5: @settitle Vmgen (Gforth @value{VERSION})
6: @c @syncodeindex pg cp
7: @comment %**end of header
8: @copying
9: This manual is for Vmgen
10: (version @value{VERSION}, @value{UPDATED}),
11: the virtual machine interpreter generator
12:
13: Copyright @copyright{} 2002 Free Software Foundation, Inc.
14:
15: @quotation
16: Permission is granted to copy, distribute and/or modify this document
17: under the terms of the GNU Free Documentation License, Version 1.1 or
18: any later version published by the Free Software Foundation; with no
19: Invariant Sections, with the Front-Cover texts being ``A GNU Manual,''
20: and with the Back-Cover Texts as in (a) below. A copy of the
21: license is included in the section entitled ``GNU Free Documentation
22: License.''
23:
24: (a) The FSF's Back-Cover Text is: ``You have freedom to copy and modify
25: this GNU Manual, like GNU software. Copies published by the Free
26: Software Foundation raise funds for GNU development.''
27: @end quotation
28: @end copying
29:
30: @dircategory GNU programming tools
31: @direntry
32: * Vmgen: (vmgen). Interpreter generator
33: @end direntry
34:
35: @titlepage
36: @title Vmgen
37: @subtitle for Gforth version @value{VERSION}, @value{UPDATED}
38: @author M. Anton Ertl (@email{anton@@mips.complang.tuwien.ac.at})
39: @page
40: @vskip 0pt plus 1filll
41: @insertcopying
42: @end titlepage
43:
44: @contents
45:
46: @ifnottex
47: @node Top, Introduction, (dir), (dir)
48: @top Vmgen
49:
50: @insertcopying
51: @end ifnottex
52:
53: @menu
54: * Introduction:: What can Vmgen do for you?
55: * Why interpreters?:: Advantages and disadvantages
56: * Concepts:: VM interpreter background
57: * Invoking Vmgen::
58: * Example::
59: * Input File Format::
60: * Error messages:: reported by Vmgen
61: * Using the generated code::
62: * Hints:: VM archictecture, efficiency
63: * The future::
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::
88: * Register Machines:: How to define register VM instructions
89:
90: Input File Grammar
91:
92: * Eval escapes:: what follows \E
93:
94: Simple instructions
95:
96: * C Code Macros:: Macros recognized by Vmgen
97: * C Code restrictions:: Vmgen makes assumptions about C code
98:
99: Using the generated code
100:
101: * VM engine:: Executing VM code
102: * VM instruction table::
103: * VM code generation:: Creating VM code (in the front-end)
104: * Peephole optimization:: Creating VM superinstructions
105: * VM disassembler:: for debugging the front end
106: * VM profiler:: for finding worthwhile superinstructions
107:
108: Hints
109:
110: * Floating point:: and stacks
111:
112: Copying This Manual
113:
114: * GNU Free Documentation License:: License for copying this manual.
115:
116: @end detailmenu
117: @end menu
118:
119: @c @ifnottex
120: @c This file documents Vmgen (Gforth @value{VERSION}).
121:
122: @c ************************************************************
123: @node Introduction, Why interpreters?, Top, Top
124: @chapter Introduction
125:
126: Vmgen is a tool for writing efficient interpreters. It takes a simple
127: virtual machine description and generates efficient C code for dealing
128: with the virtual machine code in various ways (in particular, executing
129: it). The run-time efficiency of the resulting interpreters is usually
130: within a factor of 10 of machine code produced by an optimizing
131: compiler.
132:
133: The interpreter design strategy supported by Vmgen is to divide the
134: interpreter into two parts:
135:
136: @itemize @bullet
137:
138: @item The @emph{front end} takes the source code of the language to be
139: implemented, and translates it into virtual machine code. This is
140: similar to an ordinary compiler front end; typically an interpreter
141: front-end performs no optimization, so it is relatively simple to
142: implement and runs fast.
143:
144: @item The @emph{virtual machine interpreter} executes the virtual
145: machine code.
146:
147: @end itemize
148:
149: Such a division is usually used in interpreters, for modularity as well
150: as for efficiency. The virtual machine code is typically passed between
151: front end and virtual machine interpreter in memory, like in a
152: load-and-go compiler; this avoids the complexity and time cost of
153: writing the code to a file and reading it again.
154:
155: A @emph{virtual machine} (VM) represents the program as a sequence of
156: @emph{VM instructions}, following each other in memory, similar to real
157: machine code. Control flow occurs through VM branch instructions, like
158: in a real machine.
159:
160: @cindex functionality features overview
161: In this setup, Vmgen can generate most of the code dealing with virtual
162: machine instructions from a simple description of the virtual machine
163: instructions (@pxref{Input File Format}), in particular:
164:
165: @table @strong
166:
167: @item VM instruction execution
168:
169: @item VM code generation
170: Useful in the front end.
171:
172: @item VM code decompiler
173: Useful for debugging the front end.
174:
175: @item VM code tracing
176: Useful for debugging the front end and the VM interpreter. You will
177: typically provide other means for debugging the user's programs at the
178: source level.
179:
180: @item VM code profiling
181: Useful for optimizing the VM interpreter with superinstructions
182: (@pxref{VM profiler}).
183:
184: @end table
185:
186: To create parts of the interpretive system that do not deal with VM
187: instructions, you have to use other tools (e.g., @command{bison}) and/or
188: hand-code them.
189:
190: @cindex efficiency features overview
191: @noindent
192: Vmgen supports efficient interpreters though various optimizations, in
193: particular
194:
195: @itemize @bullet
196:
197: @item Threaded code
198:
199: @item Caching the top-of-stack in a register
200:
201: @item Combining VM instructions into superinstructions
202:
203: @item
204: Replicating VM (super)instructions for better BTB prediction accuracy
205: (not yet in vmgen-ex, but already in Gforth).
206:
207: @end itemize
208:
209: @cindex speed for JVM
210: As a result, Vmgen-based interpreters are only about an order of
211: magnitude slower than native code from an optimizing C compiler on small
212: benchmarks; on large benchmarks, which spend more time in the run-time
213: system, the slowdown is often less (e.g., the slowdown of a
214: Vmgen-generated JVM interpreter over the best JVM JIT compiler we
215: measured is only a factor of 2-3 for large benchmarks; some other JITs
216: and all other interpreters we looked at were slower than our
217: interpreter).
218:
219: VMs are usually designed as stack machines (passing data between VM
220: instructions on a stack), and Vmgen supports such designs especially
221: well; however, you can also use Vmgen for implementing a register VM
222: (@pxref{Register Machines}) and still benefit from most of the advantages
223: offered by Vmgen.
224:
225: There are many potential uses of the instruction descriptions that are
226: not implemented at the moment, but we are open for feature requests, and
227: we will consider new features if someone asks for them; so the feature
228: list above is not exhaustive.
229:
230: @c *********************************************************************
231: @node Why interpreters?, Concepts, Introduction, Top
232: @chapter Why interpreters?
233: @cindex interpreters, advantages
234: @cindex advantages of interpreters
235: @cindex advantages of vmgen
236:
237: Interpreters are a popular language implementation technique because
238: they combine all three of the following advantages:
239:
240: @itemize @bullet
241:
242: @item Ease of implementation
243:
244: @item Portability
245:
246: @item Fast edit-compile-run cycle
247:
248: @end itemize
249:
250: Vmgen makes it even easier to implement interpreters.
251:
252: @cindex speed of interpreters
253: The main disadvantage of interpreters is their run-time speed. However,
254: there are huge differences between different interpreters in this area:
255: the slowdown over optimized C code on programs consisting of simple
256: operations is typically a factor of 10 for the more efficient
257: interpreters, and a factor of 1000 for the less efficient ones (the
258: slowdown for programs executing complex operations is less, because the
259: time spent in libraries for executing complex operations is the same in
260: all implementation strategies).
261:
262: Vmgen supports techniques for building efficient interpreters.
263:
264: @c ********************************************************************
265: @node Concepts, Invoking Vmgen, Why interpreters?, Top
266: @chapter Concepts
267:
268: @menu
269: * Front end and VM interpreter:: Modularizing an interpretive system
270: * Data handling:: Stacks, registers, immediate arguments
271: * Dispatch:: From one VM instruction to the next
272: @end menu
273:
274: @c --------------------------------------------------------------------
275: @node Front end and VM interpreter, Data handling, Concepts, Concepts
276: @section Front end and VM interpreter
277: @cindex modularization of interpreters
278:
279: @cindex front-end
280: Interpretive systems are typically divided into a @emph{front end} that
281: parses the input language and produces an intermediate representation
282: for the program, and an interpreter that executes the intermediate
283: representation of the program.
284:
285: @cindex virtual machine
286: @cindex VM
287: @cindex VM instruction
288: @cindex instruction, VM
289: @cindex VM branch instruction
290: @cindex branch instruction, VM
291: @cindex VM register
292: @cindex register, VM
293: @cindex opcode, VM instruction
294: @cindex immediate argument, VM instruction
295: For efficient interpreters the intermediate representation of choice is
296: virtual machine code (rather than, e.g., an abstract syntax tree).
297: @emph{Virtual machine} (VM) code consists of VM instructions arranged
298: sequentially in memory; they are executed in sequence by the VM
299: interpreter, but VM branch instructions can change the control flow and
300: are used for implementing control structures. The conceptual similarity
301: to real machine code results in the name @emph{virtual machine}.
302: Various terms similar to terms for real machines are used; e.g., there
303: are @emph{VM registers} (like the instruction pointer and stack
304: pointer(s)), and the VM instruction consists of an @emph{opcode} and
305: @emph{immediate arguments}.
306:
307: In this framework, Vmgen supports building the VM interpreter and any
308: other component dealing with VM instructions. It does not have any
309: support for the front end, apart from VM code generation support. The
310: front end can be implemented with classical compiler front-end
311: techniques, supported by tools like @command{flex} and @command{bison}.
312:
313: The intermediate representation is usually just internal to the
314: interpreter, but some systems also support saving it to a file, either
315: as an image file, or in a full-blown linkable file format (e.g., JVM).
316: Vmgen currently has no special support for such features, but the
317: information in the instruction descriptions can be helpful, and we are
318: open to feature requests and suggestions.
319:
320: @c --------------------------------------------------------------------
321: @node Data handling, Dispatch, Front end and VM interpreter, Concepts
322: @section Data handling
323:
324: @cindex stack machine
325: @cindex register machine
326: Most VMs use one or more stacks for passing temporary data between VM
327: instructions. Another option is to use a register machine architecture
328: for the virtual machine; we believe that using a stack architecture is
329: usually both simpler and faster.
330:
331: however, this option is slower or
332: significantly more complex to implement than a stack machine architecture.
333:
334: Vmgen has special support and optimizations for stack VMs, making their
335: implementation easy and efficient.
336:
337: You can also implement a register VM with Vmgen (@pxref{Register
338: Machines}), and you will still profit from most Vmgen features.
339:
340: @cindex stack item size
341: @cindex size, stack items
342: Stack items all have the same size, so they typically will be as wide as
343: an integer, pointer, or floating-point value. Vmgen supports treating
344: two consecutive stack items as a single value, but anything larger is
345: best kept in some other memory area (e.g., the heap), with pointers to
346: the data on the stack.
347:
348: @cindex instruction stream
349: @cindex immediate arguments
350: Another source of data is immediate arguments VM instructions (in the VM
351: instruction stream). The VM instruction stream is handled similar to a
352: stack in Vmgen.
353:
354: @cindex garbage collection
355: @cindex reference counting
356: Vmgen has no built-in support for, nor restrictions against
357: @emph{garbage collection}. If you need garbage collection, you need to
358: provide it in your run-time libraries. Using @emph{reference counting}
359: is probably harder, but might be possible (contact us if you are
360: interested).
361: @c reference counting might be possible by including counting code in
362: @c the conversion macros.
363:
364: @c --------------------------------------------------------------------
365: @node Dispatch, , Data handling, Concepts
366: @section Dispatch
367: @cindex Dispatch of VM instructions
368: @cindex main interpreter loop
369:
370: Understanding this section is probably not necessary for using Vmgen,
371: but it may help. You may want to skip it now, and read it if you find statements about dispatch methods confusing.
372:
373: After executing one VM instruction, the VM interpreter has to dispatch
374: the next VM instruction (Vmgen calls the dispatch routine @samp{NEXT}).
375: Vmgen supports two methods of dispatch:
376:
377: @table @strong
378:
379: @item switch dispatch
380: @cindex switch dispatch
381: In this method the VM interpreter contains a giant @code{switch}
382: statement, with one @code{case} for each VM instruction. The VM
383: instruction opcodes are represented by integers (e.g., produced by an
384: @code{enum}) in the VM code, and dispatch occurs by loading the next
385: opcode, @code{switch}ing on it, and continuing at the appropriate
386: @code{case}; after executing the VM instruction, the VM interpreter
387: jumps back to the dispatch code.
388:
389: @item threaded code
390: @cindex threaded code
391: This method represents a VM instruction opcode by the address of the
392: start of the machine code fragment for executing the VM instruction.
393: Dispatch consists of loading this address, jumping to it, and
394: incrementing the VM instruction pointer. Typically the threaded-code
395: dispatch code is appended directly to the code for executing the VM
396: instruction. Threaded code cannot be implemented in ANSI C, but it can
397: be implemented using GNU C's labels-as-values extension (@pxref{Labels
398: as Values, , Labels as Values, gcc.info, GNU C Manual}).
399:
400: @c call threading
401: @end table
402:
403: Threaded code can be twice as fast as switch dispatch, depending on the
404: interpreter, the benchmark, and the machine.
405:
406: @c *************************************************************
407: @node Invoking Vmgen, Example, Concepts, Top
408: @chapter Invoking Vmgen
409: @cindex Invoking Vmgen
410:
411: The usual way to invoke Vmgen is as follows:
412:
413: @example
414: vmgen @var{inputfile}
415: @end example
416:
417: Here @var{inputfile} is the VM instruction description file, which
418: usually ends in @file{.vmg}. The output filenames are made by taking
419: the basename of @file{inputfile} (i.e., the output files will be created
420: in the current working directory) and replacing @file{.vmg} with
421: @file{-vm.i}, @file{-disasm.i}, @file{-gen.i}, @file{-labels.i},
422: @file{-profile.i}, and @file{-peephole.i}. E.g., @command{vmgen
423: hack/foo.vmg} will create @file{foo-vm.i}, @file{foo-disasm.i},
424: @file{foo-gen.i}, @file{foo-labels.i}, @file{foo-profile.i} and
425: @file{foo-peephole.i}.
426:
427: The command-line options supported by Vmgen are
428:
429: @table @option
430:
431: @cindex -h, command-line option
432: @cindex --help, command-line option
433: @item --help
434: @itemx -h
435: Print a message about the command-line options
436:
437: @cindex -v, command-line option
438: @cindex --version, command-line option
439: @item --version
440: @itemx -v
441: Print version and exit
442: @end table
443:
444: @c env vars GFORTHDIR GFORTHDATADIR
445:
446: @c ****************************************************************
447: @node Example, Input File Format, Invoking Vmgen, Top
448: @chapter Example
449: @cindex example of a Vmgen-based interpreter
450:
451: @menu
452: * Example overview::
453: * Using profiling to create superinstructions::
454: @end menu
455:
456: @c --------------------------------------------------------------------
457: @node Example overview, Using profiling to create superinstructions, Example, Example
458: @section Example overview
459: @cindex example overview
460: @cindex @file{vmgen-ex}
461: @cindex @file{vmgen-ex2}
462:
463: There are two versions of the same example for using Vmgen:
464: @file{vmgen-ex} and @file{vmgen-ex2} (you can also see Gforth as
465: example, but it uses additional (undocumented) features, and also
466: differs in some other respects). The example implements @emph{mini}, a
467: tiny Modula-2-like language with a small JavaVM-like virtual machine.
468:
469: The difference between the examples is that @file{vmgen-ex} uses many
470: casts, and @file{vmgen-ex2} tries to avoids most casts and uses unions
471: instead. In the rest of this manual we usually mention just files in
472: @file{vmgen-ex}; if you want to use unions, use the equivalent file in
473: @file{vmgen-ex2}.
474: @cindex unions example
475: @cindex casts example
476:
477: The files provided with each example are:
478: @cindex example files
479:
480: @example
481: Makefile
482: README
483: disasm.c wrapper file
484: engine.c wrapper file
485: peephole.c wrapper file
486: profile.c wrapper file
487: mini-inst.vmg simple VM instructions
488: mini-super.vmg superinstructions (empty at first)
489: mini.h common declarations
490: mini.l scanner
491: mini.y front end (parser, VM code generator)
492: support.c main() and other support functions
493: fib.mini example mini program
494: simple.mini example mini program
495: test.mini example mini program (tests everything)
496: test.out test.mini output
497: stat.awk script for aggregating profile information
498: peephole-blacklist list of instructions not allowed in superinstructions
499: seq2rule.awk script for creating superinstructions
500: @end example
501:
502: For your own interpreter, you would typically copy the following files
503: and change little, if anything:
504: @cindex wrapper files
505:
506: @example
507: disasm.c wrapper file
508: engine.c wrapper file
509: peephole.c wrapper file
510: profile.c wrapper file
511: stat.awk script for aggregating profile information
512: seq2rule.awk script for creating superinstructions
513: @end example
514:
515: @noindent
516: You would typically change much in or replace the following files:
517:
518: @example
519: Makefile
520: mini-inst.vmg simple VM instructions
521: mini.h common declarations
522: mini.l scanner
523: mini.y front end (parser, VM code generator)
524: support.c main() and other support functions
525: peephole-blacklist list of instructions not allowed in superinstructions
526: @end example
527:
528: You can build the example by @code{cd}ing into the example's directory,
529: and then typing @code{make}; you can check that it works with @code{make
530: check}. You can run run mini programs like this:
531:
532: @example
533: ./mini fib.mini
534: @end example
535:
536: To learn about the options, type @code{./mini -h}.
537:
538: @c --------------------------------------------------------------------
539: @node Using profiling to create superinstructions, , Example overview, Example
540: @section Using profiling to create superinstructions
541: @cindex profiling example
542: @cindex superinstructions example
543:
544: I have not added rules for this in the @file{Makefile} (there are many
545: options for selecting superinstructions, and I did not want to hardcode
546: one into the @file{Makefile}), but there are some supporting scripts, and
547: here's an example:
548:
549: Suppose you want to use @file{fib.mini} and @file{test.mini} as training
550: programs, you get the profiles like this:
551:
552: @example
553: make fib.prof test.prof #takes a few seconds
554: @end example
555:
556: You can aggregate these profiles with @file{stat.awk}:
557:
558: @example
559: awk -f stat.awk fib.prof test.prof
560: @end example
561:
562: The result contains lines like:
563:
564: @example
565: 2 16 36910041 loadlocal lit
566: @end example
567:
568: This means that the sequence @code{loadlocal lit} statically occurs a
569: total of 16 times in 2 profiles, with a dynamic execution count of
570: 36910041.
571:
572: The numbers can be used in various ways to select superinstructions.
573: E.g., if you just want to select all sequences with a dynamic
574: execution count exceeding 10000, you would use the following pipeline:
575:
576: @example
577: awk -f stat.awk fib.prof test.prof|
578: awk '$3>=10000'| #select sequences
579: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
580: awk -f seq2rule.awk| #transform sequences into superinstruction rules
581: sort -k 3 >mini-super.vmg #sort sequences
582: @end example
583:
584: The file @file{peephole-blacklist} contains all instructions that
585: directly access a stack or stack pointer (for mini: @code{call},
586: @code{return}); the sort step is necessary to ensure that prefixes
587: precede larger superinstructions.
588:
589: Now you can create a version of mini with superinstructions by just
590: saying @samp{make}
591:
592:
593: @c ***************************************************************
594: @node Input File Format, Error messages, Example, Top
595: @chapter Input File Format
596: @cindex input file format
597: @cindex format, input file
598:
599: Vmgen takes as input a file containing specifications of virtual machine
600: instructions. This file usually has a name ending in @file{.vmg}.
601:
602: Most examples are taken from the example in @file{vmgen-ex}.
603:
604: @menu
605: * Input File Grammar::
606: * Simple instructions::
607: * Superinstructions::
608: * Register Machines:: How to define register VM instructions
609: @end menu
610:
611: @c --------------------------------------------------------------------
612: @node Input File Grammar, Simple instructions, Input File Format, Input File Format
613: @section Input File Grammar
614: @cindex grammar, input file
615: @cindex input file grammar
616:
617: The grammar is in EBNF format, with @code{@var{a}|@var{b}} meaning
618: ``@var{a} or @var{b}'', @code{@{@var{c}@}} meaning 0 or more repetitions
619: of @var{c} and @code{[@var{d}]} meaning 0 or 1 repetitions of @var{d}.
620:
621: @cindex free-format, not
622: @cindex newlines, significance in syntax
623: Vmgen input is not free-format, so you have to take care where you put
624: newlines (and, in a few cases, white space).
625:
626: @example
627: description: @{instruction|comment|eval-escape|c-escape@}
628:
629: instruction: simple-inst|superinst
630:
631: simple-inst: ident '(' stack-effect ')' newline c-code newline newline
632:
633: stack-effect: @{ident@} '--' @{ident@}
634:
635: super-inst: ident '=' ident @{ident@}
636:
637: comment: '\ ' text newline
638:
639: eval-escape: '\E ' text newline
640:
641: c-escape: '\C ' text newline
642: @end example
643: @c \+ \- \g \f \c
644:
645: Note that the @code{\}s in this grammar are meant literally, not as
646: C-style encodings for non-printable characters.
647:
648: There are two ways to delimit the C code in @code{simple-inst}:
649:
650: @itemize @bullet
651:
652: @item
653: If you start it with a @samp{@{} at the start of a line (i.e., not even
654: white space before it), you have to end it with a @samp{@}} at the start
655: of a line (followed by a newline). In this case you may have empty
656: lines within the C code (typically used between variable definitions and
657: statements).
658:
659: @item
660: You do not start it with @samp{@{}. Then the C code ends at the first
661: empty line, so you cannot have empty lines within this code.
662:
663: @end itemize
664:
665: The text in @code{comment}, @code{eval-escape} and @code{c-escape} must
666: not contain a newline. @code{Ident} must conform to the usual
667: conventions of C identifiers (otherwise the C compiler would choke on
668: the Vmgen output), except that idents in @code{stack-effect} may have a
669: stack prefix (for stack prefix syntax, @pxref{Eval escapes}).
670:
671: @cindex C escape
672: @cindex @code{\C}
673: @cindex conditional compilation of Vmgen output
674: The @code{c-escape} passes the text through to each output file (without
675: the @samp{\C}). This is useful mainly for conditional compilation
676: (i.e., you write @samp{\C #if ...} etc.).
677:
678: @cindex sync lines
679: @cindex @code{#line}
680: In addition to the syntax given in the grammer, Vmgen also processes
681: sync lines (lines starting with @samp{#line}), as produced by @samp{m4
682: -s} (@pxref{Invoking m4, , Invoking m4, m4.info, GNU m4}) and similar
683: tools. This allows associating C compiler error messages with the
684: original source of the C code.
685:
686: Vmgen understands a few extensions beyond the grammar given here, but
687: these extensions are only useful for building Gforth. You can find a
688: description of the format used for Gforth in @file{prim}.
689:
690: @menu
691: * Eval escapes:: what follows \E
692: @end menu
693:
694: @node Eval escapes, , Input File Grammar, Input File Grammar
695: @subsection Eval escapes
696: @cindex escape to Forth
697: @cindex eval escape
698: @cindex @code{\E}
699:
700: @c woanders?
701: The text in @code{eval-escape} is Forth code that is evaluated when
702: Vmgen reads the line. You will normally use this feature to define
703: stacks and types.
704:
705: If you do not know (and do not want to learn) Forth, you can build the
706: text according to the following grammar; these rules are normally all
707: Forth you need for using Vmgen:
708:
709: @example
710: text: stack-decl|type-prefix-decl|stack-prefix-decl
711:
712: stack-decl: 'stack ' ident ident ident
713: type-prefix-decl:
714: 's" ' string '" ' ('single'|'double') ident 'type-prefix' ident
715: stack-prefix-decl: ident 'stack-prefix' string
716: @end example
717:
718: Note that the syntax of this code is not checked thoroughly (there are
719: many other Forth program fragments that could be written in an
720: eval-escape).
721:
722: A stack prefix can contain letters, digits, or @samp{:}, and may start
723: with an @samp{#}; e.g., in Gforth the return stack has the stack prefix
724: @samp{R:}. This restriction is not checked during the stack prefix
725: definition, but it is enforced by the parsing rules for stack items
726: later.
727:
728: If you know Forth, the stack effects of the non-standard words involved
729: are:
730: @findex stack
731: @findex type-prefix
732: @findex single
733: @findex double
734: @findex stack-prefix
735: @example
736: stack ( "name" "pointer" "type" -- )
737: ( name execution: -- stack )
738: type-prefix ( addr u item-size stack "prefix" -- )
739: single ( -- item-size )
740: double ( -- item-size )
741: stack-prefix ( stack "prefix" -- )
742: @end example
743:
744: An @var{item-size} takes three cells on the stack.
745:
746: @c --------------------------------------------------------------------
747: @node Simple instructions, Superinstructions, Input File Grammar, Input File Format
748: @section Simple instructions
749: @cindex simple VM instruction
750: @cindex instruction, simple VM
751:
752: We will use the following simple VM instruction description as example:
753:
754: @example
755: sub ( i1 i2 -- i )
756: i = i1-i2;
757: @end example
758:
759: The first line specifies the name of the VM instruction (@code{sub}) and
760: its stack effect (@code{i1 i2 -- i}). The rest of the description is
761: just plain C code.
762:
763: @cindex stack effect
764: @cindex effect, stack
765: The stack effect specifies that @code{sub} pulls two integers from the
766: data stack and puts them in the C variables @code{i1} and @code{i2}
767: (with the rightmost item (@code{i2}) taken from the top of stack;
768: intuition: if you push @code{i1}, then @code{i2} on the stack, the
769: resulting stack picture is @code{i1 i2}) and later pushes one integer
770: (@code{i}) on the data stack (the rightmost item is on the top
771: afterwards).
772:
773: @cindex prefix, type
774: @cindex type prefix
775: @cindex default stack of a type prefix
776: How do we know the type and stack of the stack items? Vmgen uses
777: prefixes, similar to Fortran; in contrast to Fortran, you have to
778: define the prefix first:
779:
780: @example
781: \E s" Cell" single data-stack type-prefix i
782: @end example
783:
784: This defines the prefix @code{i} to refer to the type @code{Cell}
785: (defined as @code{long} in @file{mini.h}) and, by default, to the
786: @code{data-stack}. It also specifies that this type takes one stack
787: item (@code{single}). The type prefix is part of the variable name.
788:
789: @cindex stack definition
790: @cindex defining a stack
791: Before we can use @code{data-stack} in this way, we have to define it:
792:
793: @example
794: \E stack data-stack sp Cell
795: @end example
796: @c !! use something other than Cell
797:
798: @cindex stack basic type
799: @cindex basic type of a stack
800: @cindex type of a stack, basic
801: @cindex stack growth direction
802: This line defines the stack @code{data-stack}, which uses the stack
803: pointer @code{sp}, and each item has the basic type @code{Cell}; other
804: types have to fit into one or two @code{Cell}s (depending on whether the
805: type is @code{single} or @code{double} wide), and are cast from and to
806: Cells on accessing the @code{data-stack} with type cast macros
807: (@pxref{VM engine}). Stacks grow towards lower addresses in
808: Vmgen-erated interpreters.
809:
810: @cindex stack prefix
811: @cindex prefix, stack
812: We can override the default stack of a stack item by using a stack
813: prefix. E.g., consider the following instruction:
814:
815: @example
816: lit ( #i -- i )
817: @end example
818:
819: The VM instruction @code{lit} takes the item @code{i} from the
820: instruction stream (indicated by the prefix @code{#}), and pushes it on
821: the (default) data stack. The stack prefix is not part of the variable
822: name. Stack prefixes are defined like this:
823:
824: @example
825: \E inst-stream stack-prefix #
826: @end example
827:
828: This definition defines that the stack prefix @code{#} specifies the
829: ``stack'' @code{inst-stream}. Since the instruction stream behaves a
830: little differently than an ordinary stack, it is predefined, and you do
831: not need to define it.
832:
833: @cindex instruction stream
834: The instruction stream contains instructions and their immediate
835: arguments, so specifying that an argument comes from the instruction
836: stream indicates an immediate argument. Of course, instruction stream
837: arguments can only appear to the left of @code{--} in the stack effect.
838: If there are multiple instruction stream arguments, the leftmost is the
839: first one (just as the intuition suggests).
840:
841: @menu
842: * C Code Macros:: Macros recognized by Vmgen
843: * C Code restrictions:: Vmgen makes assumptions about C code
844: @end menu
845:
846: @c --------------------------------------------------------------------
847: @node C Code Macros, C Code restrictions, Simple instructions, Simple instructions
848: @subsection C Code Macros
849: @cindex macros recognized by Vmgen
850: @cindex basic block, VM level
851:
852: Vmgen recognizes the following strings in the C code part of simple
853: instructions:
854:
855: @table @code
856:
857: @item SET_IP
858: @findex SET_IP
859: As far as Vmgen is concerned, a VM instruction containing this ends a VM
860: basic block (used in profiling to delimit profiled sequences). On the C
861: level, this also sets the instruction pointer.
862:
863: @item SUPER_END
864: @findex SUPER_END
865: This ends a basic block (for profiling), even if the instruction
866: contains no @code{SET_IP}.
867:
868: @item INST_TAIL;
869: @findex INST_TAIL;
870: Vmgen replaces @samp{INST_TAIL;} with code for ending a VM instruction and
871: dispatching the next VM instruction. Even without a @samp{INST_TAIL;} this
872: happens automatically when control reaches the end of the C code. If
873: you want to have this in the middle of the C code, you need to use
874: @samp{INST_TAIL;}. A typical example is a conditional VM branch:
875:
876: @example
877: if (branch_condition) @{
878: SET_IP(target); INST_TAIL;
879: @}
880: /* implicit tail follows here */
881: @end example
882:
883: In this example, @samp{INST_TAIL;} is not strictly necessary, because there
884: is another one implicitly after the if-statement, but using it improves
885: branch prediction accuracy slightly and allows other optimizations.
886:
887: @item SUPER_CONTINUE
888: @findex SUPER_CONTINUE
889: This indicates that the implicit tail at the end of the VM instruction
890: dispatches the sequentially next VM instruction even if there is a
891: @code{SET_IP} in the VM instruction. This enables an optimization that
892: is not yet implemented in the vmgen-ex code (but in Gforth). The
893: typical application is in conditional VM branches:
894:
895: @example
896: if (branch_condition) @{
897: SET_IP(target); INST_TAIL; /* now this INST_TAIL is necessary */
898: @}
899: SUPER_CONTINUE;
900: @end example
901:
902: @end table
903:
904: Note that Vmgen is not smart about C-level tokenization, comments,
905: strings, or conditional compilation, so it will interpret even a
906: commented-out SUPER_END as ending a basic block (or, e.g.,
907: @samp{RESET_IP;} as @samp{SET_IP;}). Conversely, Vmgen requires the literal
908: presence of these strings; Vmgen will not see them if they are hiding in
909: a C preprocessor macro.
910:
911:
912: @c --------------------------------------------------------------------
913: @node C Code restrictions, , C Code Macros, Simple instructions
914: @subsection C Code restrictions
915: @cindex C code restrictions
916: @cindex restrictions on C code
917: @cindex assumptions about C code
918:
919: @cindex accessing stack (pointer)
920: @cindex stack pointer, access
921: @cindex instruction pointer, access
922: Vmgen generates code and performs some optimizations under the
923: assumption that the user-supplied C code does not access the stack
924: pointers or stack items, and that accesses to the instruction pointer
925: only occur through special macros. In general you should heed these
926: restrictions. However, if you need to break these restrictions, read
927: the following.
928:
929: Accessing a stack or stack pointer directly can be a problem for several
930: reasons:
931: @cindex stack caching, restriction on C code
932: @cindex superinstructions, restrictions on components
933:
934: @itemize @bullet
935:
936: @item
937: Vmgen optionally supports caching the top-of-stack item in a local
938: variable (that is allocated to a register). This is the most frequent
939: source of trouble. You can deal with it either by not using
940: top-of-stack caching (slowdown factor 1-1.4, depending on machine), or
941: by inserting flushing code (e.g., @samp{IF_spTOS(sp[...] = spTOS);}) at
942: the start and reloading code (e.g., @samp{IF_spTOS(spTOS = sp[0])}) at
943: the end of problematic C code. Vmgen inserts a stack pointer update
944: before the start of the user-supplied C code, so the flushing code has
945: to use an index that corrects for that. In the future, this flushing
946: may be done automatically by mentioning a special string in the C code.
947: @c sometimes flushing and/or reloading unnecessary
948:
949: @item
950: The Vmgen-erated code loads the stack items from stack-pointer-indexed
951: memory into variables before the user-supplied C code, and stores them
952: from variables to stack-pointer-indexed memory afterwards. If you do
953: any writes to the stack through its stack pointer in your C code, it
954: will not affect the variables, and your write may be overwritten by the
955: stores after the C code. Similarly, a read from a stack using a stack
956: pointer will not reflect computations of stack items in the same VM
957: instruction.
958:
959: @item
960: Superinstructions keep stack items in variables across the whole
961: superinstruction. So you should not include VM instructions, that
962: access a stack or stack pointer, as components of superinstructions
963: (@pxref{VM profiler}).
964:
965: @end itemize
966:
967: You should access the instruction pointer only through its special
968: macros (@samp{IP}, @samp{SET_IP}, @samp{IPTOS}); this ensure that these
969: macros can be implemented in several ways for best performance.
970: @samp{IP} points to the next instruction, and @samp{IPTOS} is its
971: contents.
972:
973:
974: @c --------------------------------------------------------------------
975: @node Superinstructions, Register Machines, Simple instructions, Input File Format
976: @section Superinstructions
977: @cindex superinstructions, defining
978: @cindex defining superinstructions
979:
980: Note: don't invest too much work in (static) superinstructions; a future
981: version of Vmgen will support dynamic superinstructions (see Ian
982: Piumarta and Fabio Riccardi, @cite{Optimizing Direct Threaded Code by
983: Selective Inlining}, PLDI'98), and static superinstructions have much
984: less benefit in that context (preliminary results indicate only a factor
985: 1.1 speedup).
986:
987: Here is an example of a superinstruction definition:
988:
989: @example
990: lit_sub = lit sub
991: @end example
992:
993: @code{lit_sub} is the name of the superinstruction, and @code{lit} and
994: @code{sub} are its components. This superinstruction performs the same
995: action as the sequence @code{lit} and @code{sub}. It is generated
996: automatically by the VM code generation functions whenever that sequence
997: occurs, so if you want to use this superinstruction, you just need to
998: add this definition (and even that can be partially automatized,
999: @pxref{VM profiler}).
1000:
1001: @cindex prefixes of superinstructions
1002: Vmgen requires that the component instructions are simple instructions
1003: defined before superinstructions using the components. Currently, Vmgen
1004: also requires that all the subsequences at the start of a
1005: superinstruction (prefixes) must be defined as superinstruction before
1006: the superinstruction. I.e., if you want to define a superinstruction
1007:
1008: @example
1009: foo4 = load add sub mul
1010: @end example
1011:
1012: you first have to define @code{load}, @code{add}, @code{sub} and
1013: @code{mul}, plus
1014:
1015: @example
1016: foo2 = load add
1017: foo3 = load add sub
1018: @end example
1019:
1020: Here, @code{sumof4} is the longest prefix of @code{sumof5}, and @code{sumof3}
1021: is the longest prefix of @code{sumof4}.
1022:
1023: Note that Vmgen assumes that only the code it generates accesses stack
1024: pointers, the instruction pointer, and various stack items, and it
1025: performs optimizations based on this assumption. Therefore, VM
1026: instructions where your C code changes the instruction pointer should
1027: only be used as last component; a VM instruction where your C code
1028: accesses a stack pointer should not be used as component at all. Vmgen
1029: does not check these restrictions, they just result in bugs in your
1030: interpreter.
1031:
1032: @c -------------------------------------------------------------------
1033: @node Register Machines, , Superinstructions, Input File Format
1034: @section Register Machines
1035: @cindex Register VM
1036: @cindex Superinstructions for register VMs
1037: @cindex tracing of register VMs
1038:
1039: If you want to implement a register VM rather than a stack VM with
1040: Vmgen, there are two ways to do it: Directly and through
1041: superinstructions.
1042:
1043: If you use the direct way, you define instructions that take the
1044: register numbers as immediate arguments, like this:
1045:
1046: @example
1047: add3 ( #src1 #src2 #dest -- )
1048: reg[dest] = reg[src1]+reg[src2];
1049: @end example
1050:
1051: A disadvantage of this method is that during tracing you only see the
1052: register numbers, but not the register contents. Actually, with an
1053: appropriate definition of @code{printarg_src} (@pxref{VM engine}), you
1054: can print the values of the source registers on entry, but you cannot
1055: print the value of the destination register on exit.
1056:
1057: If you use superinstructions to define a register VM, you define simple
1058: instructions that use a stack, and then define superinstructions that
1059: have no overall stack effect, like this:
1060:
1061: @example
1062: loadreg ( #src -- n )
1063: n = reg[src];
1064:
1065: storereg ( n #dest -- )
1066: reg[dest] = n;
1067:
1068: adds ( n1 n2 -- n )
1069: n = n1+n2;
1070:
1071: add3 = loadreg loadreg adds storereg
1072: @end example
1073:
1074: An advantage of this method is that you see the values and not just the
1075: register numbers in tracing. A disadvantage of this method is that
1076: currently you cannot generate superinstructions directly, but only
1077: through generating a sequence of simple instructions (we might change
1078: this in the future if there is demand).
1079:
1080: Could the register VM support be improved, apart from the issues
1081: mentioned above? It is hard to see how to do it in a general way,
1082: because there are a number of different designs that different people
1083: mean when they use the term @emph{register machine} in connection with
1084: VM interpreters. However, if you have ideas or requests in that
1085: direction, please let me know (@pxref{Contact}).
1086:
1087: @c ********************************************************************
1088: @node Error messages, Using the generated code, Input File Format, Top
1089: @chapter Error messages
1090: @cindex error messages
1091:
1092: These error messages are created by Vmgen:
1093:
1094: @table @code
1095:
1096: @cindex @code{# can only be on the input side} error
1097: @item # can only be on the input side
1098: You have used an instruction-stream prefix (usually @samp{#}) after the
1099: @samp{--} (the output side); you can only use it before (the input
1100: side).
1101:
1102: @cindex @code{prefix for this combination must be defined earlier} error
1103: @item the prefix for this combination must be defined earlier
1104: You have defined a superinstruction (e.g. @code{abc = a b c}) without
1105: defining its direct prefix (e.g., @code{ab = a b}),
1106: @xref{Superinstructions}.
1107:
1108: @cindex @code{sync line syntax} error
1109: @item sync line syntax
1110: If you are using a preprocessor (e.g., @command{m4}) to generate Vmgen
1111: input code, you may want to create @code{#line} directives (aka sync
1112: lines). This error indicates that such a line is not in th syntax
1113: expected by Vmgen (this should not happen; please report the offending
1114: line in a bug report).
1115:
1116: @cindex @code{syntax error, wrong char} error
1117: @cindex syntax error, wrong char
1118: A syntax error. If you do not see right away where the error is, it may
1119: be helpful to check the following: Did you put an empty line in a VM
1120: instruction where the C code is not delimited by braces (then the empty
1121: line ends the VM instruction)? If you used brace-delimited C code, did
1122: you put the delimiting braces (and only those) at the start of the line,
1123: without preceding white space? Did you forget a delimiting brace?
1124:
1125: @cindex @code{too many stacks} error
1126: @item too many stacks
1127: Vmgen currently supports 3 stacks (plus the instruction stream); if you
1128: need more, let us know.
1129:
1130: @cindex @code{unknown prefix} error
1131: @item unknown prefix
1132: The stack item does not match any defined type prefix (after stripping
1133: away any stack prefix). You should either declare the type prefix you
1134: want for that stack item, or use a different type prefix
1135:
1136: @item @code{unknown primitive} error
1137: @item unknown primitive
1138: You have used the name of a simple VM instruction in a superinstruction
1139: definition without defining the simple VM instruction first.
1140:
1141: @end table
1142:
1143: In addition, the C compiler can produce errors due to code produced by
1144: Vmgen; e.g., you need to define type cast functions.
1145:
1146: @c ********************************************************************
1147: @node Using the generated code, Hints, Error messages, Top
1148: @chapter Using the generated code
1149: @cindex generated code, usage
1150: @cindex Using vmgen-erated code
1151:
1152: The easiest way to create a working VM interpreter with Vmgen is
1153: probably to start with @file{vmgen-ex}, and modify it for your purposes.
1154: This chapter explains what the various wrapper and generated files do.
1155: It also contains reference-manual style descriptions of the macros,
1156: variables etc. used by the generated code, and you can skip that on
1157: first reading.
1158:
1159: @menu
1160: * VM engine:: Executing VM code
1161: * VM instruction table::
1162: * VM code generation:: Creating VM code (in the front-end)
1163: * Peephole optimization:: Creating VM superinstructions
1164: * VM disassembler:: for debugging the front end
1165: * VM profiler:: for finding worthwhile superinstructions
1166: @end menu
1167:
1168: @c --------------------------------------------------------------------
1169: @node VM engine, VM instruction table, Using the generated code, Using the generated code
1170: @section VM engine
1171: @cindex VM instruction execution
1172: @cindex engine
1173: @cindex executing VM code
1174: @cindex @file{engine.c}
1175: @cindex @file{-vm.i} output file
1176:
1177: The VM engine is the VM interpreter that executes the VM code. It is
1178: essential for an interpretive system.
1179:
1180: Vmgen supports two methods of VM instruction dispatch: @emph{threaded
1181: code} (fast, but gcc-specific), and @emph{switch dispatch} (slow, but
1182: portable across C compilers); you can use conditional compilation
1183: (@samp{defined(__GNUC__)}) to choose between these methods, and our
1184: example does so.
1185:
1186: For both methods, the VM engine is contained in a C-level function.
1187: Vmgen generates most of the contents of the function for you
1188: (@file{@var{name}-vm.i}), but you have to define this function, and
1189: macros and variables used in the engine, and initialize the variables.
1190: In our example the engine function also includes
1191: @file{@var{name}-labels.i} (@pxref{VM instruction table}).
1192:
1193: @cindex tracing VM code
1194: @cindex superinstructions and tracing
1195: In addition to executing the code, the VM engine can optionally also
1196: print out a trace of the executed instructions, their arguments and
1197: results. For superinstructions it prints the trace as if only component
1198: instructions were executed; this allows to introduce new
1199: superinstructions while keeping the traces comparable to old ones
1200: (important for regression tests).
1201:
1202: It costs significant performance to check in each instruction whether to
1203: print tracing code, so we recommend producing two copies of the engine:
1204: one for fast execution, and one for tracing. See the rules for
1205: @file{engine.o} and @file{engine-debug.o} in @file{vmgen-ex/Makefile}
1206: for an example.
1207:
1208: The following macros and variables are used in @file{@var{name}-vm.i}:
1209:
1210: @table @code
1211:
1212: @findex LABEL
1213: @item LABEL(@var{inst_name})
1214: This is used just before each VM instruction to provide a jump or
1215: @code{switch} label (the @samp{:} is provided by Vmgen). For switch
1216: dispatch this should expand to @samp{case @var{label}:}; for
1217: threaded-code dispatch this should just expand to @samp{@var{label}:}.
1218: In either case @var{label} is usually the @var{inst_name} with some
1219: prefix or suffix to avoid naming conflicts.
1220:
1221: @findex LABEL2
1222: @item LABEL2(@var{inst_name})
1223: This will be used for dynamic superinstructions; at the moment, this
1224: should expand to nothing.
1225:
1226: @findex NAME
1227: @item NAME(@var{inst_name_string})
1228: Called on entering a VM instruction with a string containing the name of
1229: the VM instruction as parameter. In normal execution this should be
1230: expand to nothing, but for tracing this usually prints the name, and
1231: possibly other information (several VM registers in our example).
1232:
1233: @findex DEF_CA
1234: @item DEF_CA
1235: Usually empty. Called just inside a new scope at the start of a VM
1236: instruction. Can be used to define variables that should be visible
1237: during every VM instruction. If you define this macro as non-empty, you
1238: have to provide the finishing @samp{;} in the macro.
1239:
1240: @findex NEXT_P0
1241: @findex NEXT_P1
1242: @findex NEXT_P2
1243: @item NEXT_P0 NEXT_P1 NEXT_P2
1244: The three parts of instruction dispatch. They can be defined in
1245: different ways for best performance on various processors (see
1246: @file{engine.c} in the example or @file{engine/threaded.h} in Gforth).
1247: @samp{NEXT_P0} is invoked right at the start of the VM instruction (but
1248: after @samp{DEF_CA}), @samp{NEXT_P1} right after the user-supplied C
1249: code, and @samp{NEXT_P2} at the end. The actual jump has to be
1250: performed by @samp{NEXT_P2} (if you would do it earlier, important parts
1251: of the VM instruction would not be executed).
1252:
1253: The simplest variant is if @samp{NEXT_P2} does everything and the other
1254: macros do nothing. Then also related macros like @samp{IP},
1255: @samp{SET_IP}, @samp{IP}, @samp{INC_IP} and @samp{IPTOS} are very
1256: straightforward to define. For switch dispatch this code consists just
1257: of a jump to the dispatch code (@samp{goto next_inst;} in our example);
1258: for direct threaded code it consists of something like
1259: @samp{(@{cfa=*ip++; goto *cfa;@})}.
1260:
1261: Pulling code (usually the @samp{cfa=*ip++;}) up into @samp{NEXT_P1}
1262: usually does not cause problems, but pulling things up into
1263: @samp{NEXT_P0} usually requires changing the other macros (and, at least
1264: for Gforth on Alpha, it does not buy much, because the compiler often
1265: manages to schedule the relevant stuff up by itself). An even more
1266: extreme variant is to pull code up even further, into, e.g., NEXT_P1 of
1267: the previous VM instruction (prefetching, useful on PowerPCs).
1268:
1269: @findex INC_IP
1270: @item INC_IP(@var{n})
1271: This increments @code{IP} by @var{n}.
1272:
1273: @findex SET_IP
1274: @item SET_IP(@var{target})
1275: This sets @code{IP} to @var{target}.
1276:
1277: @cindex type cast macro
1278: @findex vm_@var{A}2@var{B}
1279: @item vm_@var{A}2@var{B}(a,b)
1280: Type casting macro that assigns @samp{a} (of type @var{A}) to @samp{b}
1281: (of type @var{B}). This is mainly used for getting stack items into
1282: variables and back. So you need to define macros for every combination
1283: of stack basic type (@code{Cell} in our example) and type-prefix types
1284: used with that stack (in both directions). For the type-prefix type,
1285: you use the type-prefix (not the C type string) as type name (e.g.,
1286: @samp{vm_Cell2i}, not @samp{vm_Cell2Cell}). In addition, you have to
1287: define a vm_@var{X}2@var{X} macro for the stack's basic type @var{X}
1288: (used in superinstructions).
1289:
1290: @cindex instruction stream, basic type
1291: The stack basic type for the predefined @samp{inst-stream} is
1292: @samp{Cell}. If you want a stack with the same item size, making its
1293: basic type @samp{Cell} usually reduces the number of macros you have to
1294: define.
1295:
1296: @cindex unions in type cast macros
1297: @cindex casts in type cast macros
1298: @cindex type casting between floats and integers
1299: Here our examples differ a lot: @file{vmgen-ex} uses casts in these
1300: macros, whereas @file{vmgen-ex2} uses union-field selection (or
1301: assignment to union fields). Note that casting floats into integers and
1302: vice versa changes the bit pattern (and you do not want that). In this
1303: case your options are to use a (temporary) union, or to take the address
1304: of the value, cast the pointer, and dereference that (not always
1305: possible, and sometimes expensive).
1306:
1307: @findex vm_two@var{A}2@var{B}
1308: @findex vm_@var{B}2two@var{A}
1309: @item vm_two@var{A}2@var{B}(a1,a2,b)
1310: @item vm_@var{B}2two@var{A}(b,a1,a2)
1311: Type casting between two stack items (@code{a1}, @code{a2}) and a
1312: variable @code{b} of a type that takes two stack items. This does not
1313: occur in our small examples, but you can look at Gforth for examples
1314: (see @code{vm_twoCell2d} in @file{engine/forth.h}).
1315:
1316: @cindex stack pointer definition
1317: @cindex instruction pointer definition
1318: @item @var{stackpointer}
1319: For each stack used, the stackpointer name given in the stack
1320: declaration is used. For a regular stack this must be an l-expression;
1321: typically it is a variable declared as a pointer to the stack's basic
1322: type. For @samp{inst-stream}, the name is @samp{IP}, and it can be a
1323: plain r-value; typically it is a macro that abstracts away the
1324: differences between the various implementations of @code{NEXT_P*}.
1325:
1326: @cindex top of stack caching
1327: @cindex stack caching
1328: @cindex TOS
1329: @findex IPTOS
1330: @item @var{stackpointer}TOS
1331: The top-of-stack for the stack pointed to by @var{stackpointer}. If you
1332: are using top-of-stack caching for that stack, this should be defined as
1333: variable; if you are not using top-of-stack caching for that stack, this
1334: should be a macro expanding to @samp{@var{stackpointer}[0]}. The stack
1335: pointer for the predefined @samp{inst-stream} is called @samp{IP}, so
1336: the top-of-stack is called @samp{IPTOS}.
1337:
1338: @findex IF_@var{stackpointer}TOS
1339: @item IF_@var{stackpointer}TOS(@var{expr})
1340: Macro for executing @var{expr}, if top-of-stack caching is used for the
1341: @var{stackpointer} stack. I.e., this should do @var{expr} if there is
1342: top-of-stack caching for @var{stackpointer}; otherwise it should do
1343: nothing.
1344:
1345: @findex SUPER_END
1346: @item SUPER_END
1347: This is used by the VM profiler (@pxref{VM profiler}); it should not do
1348: anything in normal operation, and call @code{vm_count_block(IP)} for
1349: profiling.
1350:
1351: @findex SUPER_CONTINUE
1352: @item SUPER_CONTINUE
1353: This is just a hint to Vmgen and does nothing at the C level.
1354:
1355: @findex VM_DEBUG
1356: @item VM_DEBUG
1357: If this is defined, the tracing code will be compiled in (slower
1358: interpretation, but better debugging). Our example compiles two
1359: versions of the engine, a fast-running one that cannot trace, and one
1360: with potential tracing and profiling.
1361:
1362: @findex vm_debug
1363: @item vm_debug
1364: Needed only if @samp{VM_DEBUG} is defined. If this variable contains
1365: true, the VM instructions produce trace output. It can be turned on or
1366: off at any time.
1367:
1368: @findex vm_out
1369: @item vm_out
1370: Needed only if @samp{VM_DEBUG} is defined. Specifies the file on which
1371: to print the trace output (type @samp{FILE *}).
1372:
1373: @findex printarg_@var{type}
1374: @item printarg_@var{type}(@var{value})
1375: Needed only if @samp{VM_DEBUG} is defined. Macro or function for
1376: printing @var{value} in a way appropriate for the @var{type}. This is
1377: used for printing the values of stack items during tracing. @var{Type}
1378: is normally the type prefix specified in a @code{type-prefix} definition
1379: (e.g., @samp{printarg_i}); in superinstructions it is currently the
1380: basic type of the stack.
1381:
1382: @end table
1383:
1384:
1385: @c --------------------------------------------------------------------
1386: @node VM instruction table, VM code generation, VM engine, Using the generated code
1387: @section VM instruction table
1388: @cindex instruction table
1389: @cindex opcode definition
1390: @cindex labels for threaded code
1391: @cindex @code{vm_prim}, definition
1392: @cindex @file{-labels.i} output file
1393:
1394: For threaded code we also need to produce a table containing the labels
1395: of all VM instructions. This is needed for VM code generation
1396: (@pxref{VM code generation}), and it has to be done in the engine
1397: function, because the labels are not visible outside. It then has to be
1398: passed outside the function (and assigned to @samp{vm_prim}), to be used
1399: by the VM code generation functions.
1400:
1401: This means that the engine function has to be called first to produce
1402: the VM instruction table, and later, after generating VM code, it has to
1403: be called again to execute the generated VM code (yes, this is ugly).
1404: In our example program, these two modes of calling the engine function
1405: are differentiated by the value of the parameter ip0 (if it equals 0,
1406: then the table is passed out, otherwise the VM code is executed); in our
1407: example, we pass the table out by assigning it to @samp{vm_prim} and
1408: returning from @samp{engine}.
1409:
1410: In our example (@file{vmgen-ex/engine.c}), we also build such a table for
1411: switch dispatch; this is mainly done for uniformity.
1412:
1413: For switch dispatch, we also need to define the VM instruction opcodes
1414: used as case labels in an @code{enum}.
1415:
1416: For both purposes (VM instruction table, and enum), the file
1417: @file{@var{name}-labels.i} is generated by Vmgen. You have to define
1418: the following macro used in this file:
1419:
1420: @table @code
1421:
1422: @findex INST_ADDR
1423: @item INST_ADDR(@var{inst_name})
1424: For switch dispatch, this is just the name of the switch label (the same
1425: name as used in @samp{LABEL(@var{inst_name})}), for both uses of
1426: @file{@var{name}-labels.i}. For threaded-code dispatch, this is the
1427: address of the label defined in @samp{LABEL(@var{inst_name})}); the
1428: address is taken with @samp{&&} (@pxref{Labels as Values, , Labels as
1429: Values, gcc.info, GNU C Manual}).
1430:
1431: @end table
1432:
1433:
1434: @c --------------------------------------------------------------------
1435: @node VM code generation, Peephole optimization, VM instruction table, Using the generated code
1436: @section VM code generation
1437: @cindex VM code generation
1438: @cindex code generation, VM
1439: @cindex @file{-gen.i} output file
1440:
1441: Vmgen generates VM code generation functions in @file{@var{name}-gen.i}
1442: that the front end can call to generate VM code. This is essential for
1443: an interpretive system.
1444:
1445: @findex gen_@var{inst}
1446: For a VM instruction @samp{x ( #a b #c -- d )}, Vmgen generates a
1447: function with the prototype
1448:
1449: @example
1450: void gen_x(Inst **ctp, a_type a, c_type c)
1451: @end example
1452:
1453: The @code{ctp} argument points to a pointer to the next instruction.
1454: @code{*ctp} is increased by the generation functions; i.e., you should
1455: allocate memory for the code to be generated beforehand, and start with
1456: *ctp set at the start of this memory area. Before running out of
1457: memory, allocate a new area, and generate a VM-level jump to the new
1458: area (this overflow handling is not implemented in our examples).
1459:
1460: @cindex immediate arguments, VM code generation
1461: The other arguments correspond to the immediate arguments of the VM
1462: instruction (with their appropriate types as defined in the
1463: @code{type_prefix} declaration.
1464:
1465: The following types, variables, and functions are used in
1466: @file{@var{name}-gen.i}:
1467:
1468: @table @code
1469:
1470: @findex Inst
1471: @item Inst
1472: The type of the VM instruction; if you use threaded code, this is
1473: @code{void *}; for switch dispatch this is an integer type.
1474:
1475: @cindex @code{vm_prim}, use
1476: @item vm_prim
1477: The VM instruction table (type: @code{Inst *}, @pxref{VM instruction table}).
1478:
1479: @findex gen_inst
1480: @item gen_inst(Inst **ctp, Inst i)
1481: This function compiles the instruction @code{i}. Take a look at it in
1482: @file{vmgen-ex/peephole.c}. It is trivial when you don't want to use
1483: superinstructions (just the last two lines of the example function), and
1484: slightly more complicated in the example due to its ability to use
1485: superinstructions (@pxref{Peephole optimization}).
1486:
1487: @findex genarg_@var{type_prefix}
1488: @item genarg_@var{type_prefix}(Inst **ctp, @var{type} @var{type_prefix})
1489: This compiles an immediate argument of @var{type} (as defined in a
1490: @code{type-prefix} definition). These functions are trivial to define
1491: (see @file{vmgen-ex/support.c}). You need one of these functions for
1492: every type that you use as immediate argument.
1493:
1494: @end table
1495:
1496: @findex BB_BOUNDARY
1497: In addition to using these functions to generate code, you should call
1498: @code{BB_BOUNDARY} at every basic block entry point if you ever want to
1499: use superinstructions (or if you want to use the profiling supported by
1500: Vmgen; but this support is also useful mainly for selecting
1501: superinstructions). If you use @code{BB_BOUNDARY}, you should also
1502: define it (take a look at its definition in @file{vmgen-ex/mini.y}).
1503:
1504: You do not need to call @code{BB_BOUNDARY} after branches, because you
1505: will not define superinstructions that contain branches in the middle
1506: (and if you did, and it would work, there would be no reason to end the
1507: superinstruction at the branch), and because the branches announce
1508: themselves to the profiler.
1509:
1510:
1511: @c --------------------------------------------------------------------
1512: @node Peephole optimization, VM disassembler, VM code generation, Using the generated code
1513: @section Peephole optimization
1514: @cindex peephole optimization
1515: @cindex superinstructions, generating
1516: @cindex @file{peephole.c}
1517: @cindex @file{-peephole.i} output file
1518:
1519: You need peephole optimization only if you want to use
1520: superinstructions. But having the code for it does not hurt much if you
1521: do not use superinstructions.
1522:
1523: A simple greedy peephole optimization algorithm is used for
1524: superinstruction selection: every time @code{gen_inst} compiles a VM
1525: instruction, it checks if it can combine it with the last VM instruction
1526: (which may also be a superinstruction resulting from a previous peephole
1527: optimization); if so, it changes the last instruction to the combined
1528: instruction instead of laying down @code{i} at the current @samp{*ctp}.
1529:
1530: The code for peephole optimization is in @file{vmgen-ex/peephole.c}.
1531: You can use this file almost verbatim. Vmgen generates
1532: @file{@var{file}-peephole.i} which contains data for the peephoile
1533: optimizer.
1534:
1535: @findex init_peeptable
1536: You have to call @samp{init_peeptable()} after initializing
1537: @samp{vm_prim}, and before compiling any VM code to initialize data
1538: structures for peephole optimization. After that, compiling with the VM
1539: code generation functions will automatically combine VM instructions
1540: into superinstructions. Since you do not want to combine instructions
1541: across VM branch targets (otherwise there will not be a proper VM
1542: instruction to branch to), you have to call @code{BB_BOUNDARY}
1543: (@pxref{VM code generation}) at branch targets.
1544:
1545:
1546: @c --------------------------------------------------------------------
1547: @node VM disassembler, VM profiler, Peephole optimization, Using the generated code
1548: @section VM disassembler
1549: @cindex VM disassembler
1550: @cindex disassembler, VM code
1551: @cindex @file{disasm.c}
1552: @cindex @file{-disasm.i} output file
1553:
1554: A VM code disassembler is optional for an interpretive system, but
1555: highly recommended during its development and maintenance, because it is
1556: very useful for detecting bugs in the front end (and for distinguishing
1557: them from VM interpreter bugs).
1558:
1559: Vmgen supports VM code disassembling by generating
1560: @file{@var{file}-disasm.i}. This code has to be wrapped into a
1561: function, as is done in @file{vmgen-ex/disasm.c}. You can use this file
1562: almost verbatim. In addition to @samp{vm_@var{A}2@var{B}(a,b)},
1563: @samp{vm_out}, @samp{printarg_@var{type}(@var{value})}, which are
1564: explained above, the following macros and variables are used in
1565: @file{@var{file}-disasm.i} (and you have to define them):
1566:
1567: @table @code
1568:
1569: @item ip
1570: This variable points to the opcode of the current VM instruction.
1571:
1572: @cindex @code{IP}, @code{IPTOS} in disassmbler
1573: @item IP IPTOS
1574: @samp{IPTOS} is the first argument of the current VM instruction, and
1575: @samp{IP} points to it; this is just as in the engine, but here
1576: @samp{ip} points to the opcode of the VM instruction (in contrast to the
1577: engine, where @samp{ip} points to the next cell, or even one further).
1578:
1579: @findex VM_IS_INST
1580: @item VM_IS_INST(Inst i, int n)
1581: Tests if the opcode @samp{i} is the same as the @samp{n}th entry in the
1582: VM instruction table.
1583:
1584: @end table
1585:
1586:
1587: @c --------------------------------------------------------------------
1588: @node VM profiler, , VM disassembler, Using the generated code
1589: @section VM profiler
1590: @cindex VM profiler
1591: @cindex profiling for selecting superinstructions
1592: @cindex superinstructions and profiling
1593: @cindex @file{profile.c}
1594: @cindex @file{-profile.i} output file
1595:
1596: The VM profiler is designed for getting execution and occurence counts
1597: for VM instruction sequences, and these counts can then be used for
1598: selecting sequences as superinstructions. The VM profiler is probably
1599: not useful as profiling tool for the interpretive system. I.e., the VM
1600: profiler is useful for the developers, but not the users of the
1601: interpretive system.
1602:
1603: The output of the profiler is: for each basic block (executed at least
1604: once), it produces the dynamic execution count of that basic block and
1605: all its subsequences; e.g.,
1606:
1607: @example
1608: 9227465 lit storelocal
1609: 9227465 storelocal branch
1610: 9227465 lit storelocal branch
1611: @end example
1612:
1613: I.e., a basic block consisting of @samp{lit storelocal branch} is
1614: executed 9227465 times.
1615:
1616: @cindex @file{stat.awk}
1617: @cindex @file{seq2rule.awk}
1618: This output can be combined in various ways. E.g.,
1619: @file{vmgen-ex/stat.awk} adds up the occurences of a given sequence wrt
1620: dynamic execution, static occurence, and per-program occurence. E.g.,
1621:
1622: @example
1623: 2 16 36910041 loadlocal lit
1624: @end example
1625:
1626: @noindent
1627: indicates that the sequence @samp{loadlocal lit} occurs in 2 programs,
1628: in 16 places, and has been executed 36910041 times. Now you can select
1629: superinstructions in any way you like (note that compile time and space
1630: typically limit the number of superinstructions to 100--1000). After
1631: you have done that, @file{vmgen/seq2rule.awk} turns lines of the form
1632: above into rules for inclusion in a Vmgen input file. Note that this
1633: script does not ensure that all prefixes are defined, so you have to do
1634: that in other ways. So, an overall script for turning profiles into
1635: superinstructions can look like this:
1636:
1637: @example
1638: awk -f stat.awk fib.prof test.prof|
1639: awk '$3>=10000'| #select sequences
1640: fgrep -v -f peephole-blacklist| #eliminate wrong instructions
1641: awk -f seq2rule.awk| #turn into superinstructions
1642: sort -k 3 >mini-super.vmg #sort sequences
1643: @end example
1644:
1645: Here the dynamic count is used for selecting sequences (preliminary
1646: results indicate that the static count gives better results, though);
1647: the third line eliminates sequences containing instructions that must not
1648: occur in a superinstruction, because they access a stack directly. The
1649: dynamic count selection ensures that all subsequences (including
1650: prefixes) of longer sequences occur (because subsequences have at least
1651: the same count as the longer sequences); the sort in the last line
1652: ensures that longer superinstructions occur after their prefixes.
1653:
1654: But before using this, you have to have the profiler. Vmgen supports its
1655: creation by generating @file{@var{file}-profile.i}; you also need the
1656: wrapper file @file{vmgen-ex/profile.c} that you can use almost verbatim.
1657:
1658: @cindex @code{SUPER_END} in profiling
1659: @cindex @code{BB_BOUNDARY} in profiling
1660: The profiler works by recording the targets of all VM control flow
1661: changes (through @code{SUPER_END} during execution, and through
1662: @code{BB_BOUNDARY} in the front end), and counting (through
1663: @code{SUPER_END}) how often they were targeted. After the program run,
1664: the numbers are corrected such that each VM basic block has the correct
1665: count (entering a block without executing a branch does not increase the
1666: count, and the correction fixes that), then the subsequences of all
1667: basic blocks are printed. To get all this, you just have to define
1668: @code{SUPER_END} (and @code{BB_BOUNDARY}) appropriately, and call
1669: @code{vm_print_profile(FILE *file)} when you want to output the profile
1670: on @code{file}.
1671:
1672: @cindex @code{VM_IS_INST} in profiling
1673: The @file{@var{file}-profile.i} is similar to the disassembler file, and
1674: it uses variables and functions defined in @file{vmgen-ex/profile.c},
1675: plus @code{VM_IS_INST} already defined for the VM disassembler
1676: (@pxref{VM disassembler}).
1677:
1678: @c **********************************************************
1679: @node Hints, The future, Using the generated code, Top
1680: @chapter Hints
1681: @cindex hints
1682:
1683: @menu
1684: * Floating point:: and stacks
1685: @end menu
1686:
1687: @c --------------------------------------------------------------------
1688: @node Floating point, , Hints, Hints
1689: @section Floating point
1690:
1691: How should you deal with floating point values? Should you use the same
1692: stack as for integers/pointers, or a different one? This section
1693: discusses this issue with a view on execution speed.
1694:
1695: The simpler approach is to use a separate floating-point stack. This
1696: allows you to choose FP value size without considering the size of the
1697: integers/pointers, and you avoid a number of performance problems. The
1698: main downside is that this needs an FP stack pointer (and that may not
1699: fit in the register file on the 386 arhitecture, costing some
1700: performance, but comparatively little if you take the other option into
1701: account). If you use a separate FP stack (with stack pointer @code{fp}),
1702: using an fpTOS is helpful on most machines, but some spill the fpTOS
1703: register into memory, and fpTOS should not be used there.
1704:
1705: The other approach is to share one stack (pointed to by, say, @code{sp})
1706: between integer/pointer and floating-point values. This is ok if you do
1707: not use @code{spTOS}. If you do use @code{spTOS}, the compiler has to
1708: decide whether to put that variable into an integer or a floating point
1709: register, and the other type of operation becomes quite expensive on
1710: most machines (because moving values between integer and FP registers is
1711: quite expensive). If a value of one type has to be synthesized out of
1712: two values of the other type (@code{double} types), things are even more
1713: interesting.
1714:
1715: One way around this problem would be to not use the @code{spTOS}
1716: supported by Vmgen, but to use explicit top-of-stack variables (one for
1717: integers, one for FP values), and having a kind of accumulator+stack
1718: architecture (e.g., Ocaml bytecode uses this approach); however, this is
1719: a major change, and it's ramifications are not completely clear.
1720:
1721: @c **********************************************************
1722: @node The future, Changes, Hints, Top
1723: @chapter The future
1724: @cindex future ideas
1725:
1726: We have a number of ideas for future versions of Gforth. However, there
1727: are so many possible things to do that we would like some feedback from
1728: you. What are you doing with Vmgen, what features are you missing, and
1729: why?
1730:
1731: One idea we are thinking about is to generate just one @file{.c} file
1732: instead of letting you copy and adapt all the wrapper files (you would
1733: still have to define stuff like the type-specific macros, and stack
1734: pointers etc. somewhere). The advantage would be that, if we change the
1735: wrapper files between versions, you would not need to integrate your
1736: changes and our changes to them; Vmgen would also be easier to use for
1737: beginners. The main disadvantage of that is that it would reduce the
1738: flexibility of Vmgen a little (well, those who like flexibility could
1739: still patch the resulting @file{.c} file, like they are now doing for
1740: the wrapper files). In any case, if you are doing things to the wrapper
1741: files that would cause problems in a generated-@file{.c}-file approach,
1742: please let us know.
1743:
1744: @c **********************************************************
1745: @node Changes, Contact, The future, Top
1746: @chapter Changes
1747: @cindex Changes from old versions
1748:
1749: Use-visible changes between 0.5.9-20010501 and 0.5.9-20020822:
1750:
1751: There is now a manual (in info, HTML, Postscript, or plain text format).
1752:
1753: There is the vmgen-ex2 variant of the vmgen-ex example; the new
1754: variant uses a union type instead of lots of casting.
1755:
1756: Both variants of the example can now be compiled with an ANSI C compiler
1757: (using switch dispatch and losing quite a bit of performance); tested
1758: with @command{lcc}.
1759:
1760: Users of the gforth-0.5.9-20010501 version of Vmgen need to change
1761: several things in their source code to use the current version. I
1762: recommend keeping the gforth-0.5.9-20010501 version until you have
1763: completed the change (note that you can have several versions of Gforth
1764: installed at the same time). I hope to avoid such incompatible changes
1765: in the future.
1766:
1767: The required changes are:
1768:
1769: @table @code
1770:
1771: @cindex @code{TAIL;}, changes
1772: @item TAIL;
1773: has been renamed into @code{INST_TAIL;} (less chance of an accidental
1774: match).
1775:
1776: @cindex @code{vm_@var{A}2@var{B}}, changes
1777: @item vm_@var{A}2@var{B}
1778: now takes two arguments.
1779:
1780: @cindex @code{vm_two@var{A}2@var{B}}, changes
1781: @item vm_two@var{A}2@var{B}(b,a1,a2);
1782: changed to vm_two@var{A}2@var{B}(a1,a2,b) (note the absence of the @samp{;}).
1783:
1784: @end table
1785:
1786: Also some new macros have to be defined, e.g., @code{INST_ADDR}, and
1787: @code{LABEL}; some macros have to be defined in new contexts, e.g.,
1788: @code{VM_IS_INST} is now also needed in the disassembler.
1789:
1790: @c *********************************************************
1791: @node Contact, Copying This Manual, Changes, Top
1792: @chapter Contact
1793:
1794: To report a bug, use
1795: @url{https://savannah.gnu.org/bugs/?func=addbug&group_id=2672}.
1796:
1797: For discussion on Vmgen (e.g., how to use it), use the mailing list
1798: @email{bug-vmgen@@mail.freesoftware.fsf.org} (use
1799: @url{http://mail.gnu.org/mailman/listinfo/help-vmgen} to subscribe).
1800:
1801: You can find vmgen information at
1802: @url{http://www.complang.tuwien.ac.at/anton/vmgen/}.
1803:
1804: @c ***********************************************************
1805: @node Copying This Manual, Index, Contact, Top
1806: @appendix Copying This Manual
1807:
1808: @menu
1809: * GNU Free Documentation License:: License for copying this manual.
1810: @end menu
1811:
1812: @include fdl.texi
1813:
1814:
1815: @node Index, , Copying This Manual, Top
1816: @unnumbered Index
1817:
1818: @printindex cp
1819:
1820: @bye
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to vmgen.texi CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [gforth] / gforth / doc