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