[BACK]Return to profile.fs CVS log [TXT] [DIR] Up to [gforth] / gforth

Diff for /gforth/profile.fs between versions 1.1 and 1.5

version 1.1, 2004/08/18 15:04:20 version 1.5, 2007/12/31 19:02:24
Line 1 Line 1
 \ count execution of control-flow edges  \ count execution of control-flow edges
   
 \ Copyright (C) 2004 Free Software Foundation, Inc.  \ Copyright (C) 2004,2007 Free Software Foundation, Inc.
   
 \ This file is part of Gforth.  \ This file is part of Gforth.
   
 \ Gforth is free software; you can redistribute it and/or  \ Gforth is free software; you can redistribute it and/or
 \ modify it under the terms of the GNU General Public License  \ modify it under the terms of the GNU General Public License
 \ as published by the Free Software Foundation; either version 2  \ as published by the Free Software Foundation, either version 3
 \ of the License, or (at your option) any later version.  \ of the License, or (at your option) any later version.
   
 \ This program is distributed in the hope that it will be useful,  \ This program is distributed in the hope that it will be useful,
Line 15 Line 15
 \ GNU General Public License for more details.  \ GNU General Public License for more details.
   
 \ You should have received a copy of the GNU General Public License  \ You should have received a copy of the GNU General Public License
 \ along with this program; if not, write to the Free Software  \ along with this program. If not, see http://www.gnu.org/licenses/.
 \ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.  
   
   
 \ relies on some Gforth internals  \ relies on some Gforth internals
Line 26 Line 25
 \ fixed by making sure that every source position occurs only once as  \ fixed by making sure that every source position occurs only once as
 \ a profile point.  \ a profile point.
   
   true constant count-calls? \ do some profiling of colon definitions etc.
   
   \ for true COUNT-CALLS?:
   
   \ What data do I need for evaluating the effectiveness of (partial) inlining?
   
   \ static and dynamic counts of everything:
   
   \ original BB length (histogram and average)
   \ BB length with partial inlining (histogram and average)
   \   since we cannot partially inline library calls, we use a parameter
   \   that represents the amount of partial inlining we can expect there.
   \ number of tail calls (original and after partial inlining)
   \ number of calls (original and after partial inlining)
   \ reason for BB end: call, return, execute, branch
   
   \ how many static calls are there to a word?  How many of the dynamic
   \ calls call just a single word?
   
 struct  struct
     cell%    field profile-next      cell%    field profile-next
     cell% 2* field profile-count      cell% 2* field profile-count
     cell% 2* field profile-sourcepos      cell% 2* field profile-sourcepos
     cell%    field profile-char \ character position in line      cell%    field profile-char \ character position in line
       count-calls? [if]
           cell% field profile-colondef? \ is this a colon definition start
           cell% field profile-calls \ static calls to the colon def
           cell% field profile-straight-line \ may contain calls, but no other CF
           cell% field profile-calls-from \ static calls in the colon def
       [endif]
 end-struct profile% \ profile point  end-struct profile% \ profile point
   
 variable profile-points \ linked list of profile%  variable profile-points \ linked list of profile%
 0 profile-points !  0 profile-points !
   variable next-profile-point-p \ the address where the next pp will be stored
   profile-points next-profile-point-p !
   count-calls? [if]
       variable last-colondef-profile \ pointer to the pp of last colon definition
   [endif]
       
 : new-profile-point ( -- addr )  : new-profile-point ( -- addr )
     profile% %alloc >r      profile% %alloc >r
     0. r@ profile-count 2!      0. r@ profile-count 2!
     current-sourcepos r@ profile-sourcepos 2!      current-sourcepos r@ profile-sourcepos 2!
     >in @ r@ profile-char !      >in @ r@ profile-char !
     profile-points @ r@ profile-next !      [ count-calls? ] [if]
     r@ profile-points !          r@ profile-colondef? off
           0 r@ profile-calls !
           r@ profile-straight-line on
           0 r@ profile-calls-from !
       [endif]
       0 r@ profile-next !
       r@ next-profile-point-p @ !
       r@ profile-next next-profile-point-p !
     r> ;      r> ;
   
 : dinc ( d-addr -- )  
     \ increment double pointed to by d-addr  
     dup 2@ 1. d+ rot 2! ;  
   
 : profile-this ( -- )  
     new-profile-point profile-count POSTPONE literal POSTPONE dinc ;  
   
 : profile-:-hook ( -- )  
     defers :-hook profile-this ;  
   
 : print-profile ( -- )  : print-profile ( -- )
     profile-points @ begin      profile-points @ begin
         dup while          dup while
Line 66  variable profile-points \ linked list of Line 92  variable profile-points \ linked list of
     repeat      repeat
     drop ;      drop ;
   
 ' profile-:-hook is :-hook  : print-profile-coldef ( -- )
       profile-points @ begin
           dup while
               dup >r
               r@ profile-colondef? @ if
                   r@ profile-sourcepos 2@ .sourcepos ." :"
                   r@ profile-char @ 0 .r ." : "
                   r@ profile-count 2@ 0 d.r
                   r@ profile-straight-line @ space .
                   cr
               endif
               r> profile-next @
       repeat
       drop ;
   
   
   : dinc ( d-addr -- )
       \ increment double pointed to by d-addr
       dup 2@ 1. d+ rot 2! ;
   
   : profile-this ( -- )
       new-profile-point profile-count POSTPONE literal POSTPONE dinc ;
   
   \ Various words trigger PROFILE-THIS.  In order to avoid getting
   \ several calls to PROFILE-THIS from a compiling word (like ?EXIT), we
   \ just wait until the next word is parsed by the text interpreter (in
   \ compile state) and call PROFILE-THIS only once then.  The whole
   \ BEFORE-WORD hooking etc. is there for this.
   
   \ The reason that we do this is because we use the source position for
   \ the profiling information, and there's only one source position for
   \ ?EXIT.  If we used the threaded code position instead, we would see
   \ that ?EXIT compiles to several threaded-code words, and could use
   \ different profile points for them.  However, usually dealing with
   \ the source is more practical.
   
   \ Another benefit is that we can ask for profiling anywhere in a
   \ control-flow word (even before it compiles its own stuff).
   
   \ Potential problem: Consider "COMPILING ] [" where COMPILING compiles
   \ a whole colon definition (and triggers our profiler), but during the
   \ compilation of the colon definition there is no parsing.  Afterwards
   \ you get interpret state at first (no profiling, either), but after
   \ the "]" you get parsing in compile state, and PROFILE-THIS gets
   \ called (and compiles code that is never executed).  It would be
   \ better if we had a way of knowing whether we are in a colon def or
   \ not (and used that knowledge instead of STATE).
   
   Defer before-word-profile ( -- )
   ' noop IS before-word-profile
   
   : before-word1 ( -- )
       before-word-profile defers before-word ;
   
   ' before-word1 IS before-word
   
   : profile-this-compiling ( -- )
       state @ if
           profile-this
           ['] noop IS before-word-profile
       endif ;
   
   : cock-profiler ( -- )
       \ as in cock the gun - pull the trigger
       ['] profile-this-compiling IS before-word-profile
       [ count-calls? ] [if] \ we are at a non-colondef profile point
           last-colondef-profile @ profile-straight-line off
       [endif]
   ;
   
   : hook-profiling-into ( "name" -- )
       \ make (deferred word) "name" call cock-profiler, too
       ' >body >r :noname
       POSTPONE cock-profiler
       r@ @ compile, \ old hook behaviour
       POSTPONE ;
       r> ! ; \ change hook behaviour
   
   hook-profiling-into then-like
   \ hook-profiling-into if-like    \ subsumed by other-control-flow
   \ hook-profiling-into ahead-like \ subsumed by other-control-flow
   hook-profiling-into other-control-flow
   hook-profiling-into begin-like
   hook-profiling-into again-like
   hook-profiling-into until-like
   
   count-calls? [if]
       : :-hook-profile ( -- )
           defers :-hook
           next-profile-point-p @
           profile-this
           @ dup last-colondef-profile !
           profile-colondef? on ;
   
       ' :-hook-profile IS :-hook
   [else]
       hook-profiling-into exit-like
       hook-profiling-into :-hook
   [endif]
Removed from v.1.1  
changed lines
  Added in v.1.5

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