File:  [gforth] / gforth / prof-inline.fs
Revision 1.1: download - view: text, annotated - select for diffs
Fri Sep 3 14:04:47 2004 UTC (17 years ago) by anton
Branches: MAIN
CVS tags: HEAD
made a separate profiling file for the inliner measurements
profile.fs now needs a cleanup to get rid of some of this junk

    1: \ get some data on potential (partial) inlining
    2: 
    3: \ Copyright (C) 2004 Free Software Foundation, Inc.
    4: 
    5: \ This file is part of Gforth.
    6: 
    7: \ Gforth is free software; you can redistribute it and/or
    8: \ modify it under the terms of the GNU General Public License
    9: \ as published by the Free Software Foundation; either version 2
   10: \ of the License, or (at your option) any later version.
   11: 
   12: \ This program is distributed in the hope that it will be useful,
   13: \ but WITHOUT ANY WARRANTY; without even the implied warranty of
   14: \ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   15: \ GNU General Public License for more details.
   16: 
   17: \ You should have received a copy of the GNU General Public License
   18: \ along with this program; if not, write to the Free Software
   19: \ Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
   20: 
   21: 
   22: \ relies on some Gforth internals
   23: 
   24: \ !! assumption: each file is included only once; otherwise you get
   25: \ the counts for just one of the instances of the file.  This can be
   26: \ fixed by making sure that every source position occurs only once as
   27: \ a profile point.
   28: 
   29: true constant count-calls? \ do some profiling of colon definitions etc.
   30: 
   31: \ for true COUNT-CALLS?:
   32: 
   33: \ What data do I need for evaluating the effectiveness of (partial) inlining?
   34: 
   35: \ static and dynamic counts of everything:
   36: 
   37: \ original BB length (histogram and average)
   38: \ BB length with partial inlining (histogram and average)
   39: \   since we cannot partially inline library calls, we use a parameter
   40: \   that represents the amount of partial inlining we can expect there.
   41: \ number of tail calls (original and after partial inlining)
   42: \ number of calls (original and after partial inlining)
   43: \ reason for BB end: call, return, execute, branch
   44: 
   45: \ how many static calls are there to a word?  How many of the dynamic
   46: \ calls call just a single word?
   47: 
   48: struct
   49:     cell%    field profile-next
   50:     cell% 2* field profile-count
   51:     cell% 2* field profile-sourcepos
   52:     cell%    field profile-char \ character position in line
   53:     count-calls? [if]
   54: 	cell% field profile-colondef? \ is this a colon definition start
   55: 	cell% field profile-calls \ static calls to the colon def
   56: 	cell% field profile-straight-line \ may contain calls, but no other CF
   57: 	cell% field profile-calls-from \ static calls in the colon def
   58:     [endif]
   59: end-struct profile% \ profile point
   60: 
   61: variable profile-points \ linked list of profile%
   62: 0 profile-points !
   63: variable next-profile-point-p \ the address where the next pp will be stored
   64: profile-points next-profile-point-p !
   65: count-calls? [if]
   66:     variable last-colondef-profile \ pointer to the pp of last colon definition
   67: [endif]
   68:     
   69: : new-profile-point ( -- addr )
   70:     profile% %alloc >r
   71:     0. r@ profile-count 2!
   72:     current-sourcepos r@ profile-sourcepos 2!
   73:     >in @ r@ profile-char !
   74:     [ count-calls? ] [if]
   75: 	r@ profile-colondef? off
   76: 	0 r@ profile-calls !
   77: 	r@ profile-straight-line on
   78: 	0 r@ profile-calls-from !
   79:     [endif]
   80:     0 r@ profile-next !
   81:     r@ next-profile-point-p @ !
   82:     r@ profile-next next-profile-point-p !
   83:     r> ;
   84: 
   85: : print-profile ( -- )
   86:     profile-points @ begin
   87: 	dup while
   88: 	    dup >r
   89: 	    r@ profile-sourcepos 2@ .sourcepos ." :"
   90: 	    r@ profile-char @ 0 .r ." : "
   91: 	    r@ profile-count 2@ 0 d.r cr
   92: 	    r> profile-next @
   93:     repeat
   94:     drop ;
   95: 
   96: : print-profile-coldef ( -- )
   97:     profile-points @ begin
   98: 	dup while
   99: 	    dup >r
  100: 	    r@ profile-colondef? @ if
  101: 		r@ profile-sourcepos 2@ .sourcepos ." :"
  102: 		r@ profile-char @ 3 .r ." : "
  103: 		r@ profile-count 2@ 10 d.r
  104: 		r@ profile-straight-line @ space 2 .r
  105: 		r@ profile-calls @ 4 .r
  106: 		cr
  107: 	    endif
  108: 	    r> profile-next @
  109:     repeat
  110:     drop ;
  111: 
  112: : dinc ( profilep -- )
  113:     \ increment double pointed to by d-addr
  114:     profile-count dup 2@ 1. d+ rot 2! ;
  115: 
  116: : profile-this ( -- )
  117:     new-profile-point POSTPONE literal POSTPONE dinc ;
  118: 
  119: \ Various words trigger PROFILE-THIS.  In order to avoid getting
  120: \ several calls to PROFILE-THIS from a compiling word (like ?EXIT), we
  121: \ just wait until the next word is parsed by the text interpreter (in
  122: \ compile state) and call PROFILE-THIS only once then.  The whole
  123: \ BEFORE-WORD hooking etc. is there for this.
  124: 
  125: \ The reason that we do this is because we use the source position for
  126: \ the profiling information, and there's only one source position for
  127: \ ?EXIT.  If we used the threaded code position instead, we would see
  128: \ that ?EXIT compiles to several threaded-code words, and could use
  129: \ different profile points for them.  However, usually dealing with
  130: \ the source is more practical.
  131: 
  132: \ Another benefit is that we can ask for profiling anywhere in a
  133: \ control-flow word (even before it compiles its own stuff).
  134: 
  135: \ Potential problem: Consider "COMPILING ] [" where COMPILING compiles
  136: \ a whole colon definition (and triggers our profiler), but during the
  137: \ compilation of the colon definition there is no parsing.  Afterwards
  138: \ you get interpret state at first (no profiling, either), but after
  139: \ the "]" you get parsing in compile state, and PROFILE-THIS gets
  140: \ called (and compiles code that is never executed).  It would be
  141: \ better if we had a way of knowing whether we are in a colon def or
  142: \ not (and used that knowledge instead of STATE).
  143: 
  144: Defer before-word-profile ( -- )
  145: ' noop IS before-word-profile
  146: 
  147: : before-word1 ( -- )
  148:     before-word-profile defers before-word ;
  149: 
  150: ' before-word1 IS before-word
  151: 
  152: : profile-this-compiling ( -- )
  153:     state @ if
  154: 	profile-this
  155: 	['] noop IS before-word-profile
  156:     endif ;
  157: 
  158: : cock-profiler ( -- )
  159:     \ as in cock the gun - pull the trigger
  160:     ['] profile-this-compiling IS before-word-profile
  161:     [ count-calls? ] [if] \ we are at a non-colondef profile point
  162: 	last-colondef-profile @ profile-straight-line off
  163:     [endif]
  164: ;
  165: 
  166: : hook-profiling-into ( "name" -- )
  167:     \ make (deferred word) "name" call cock-profiler, too
  168:     ' >body >r :noname
  169:     POSTPONE cock-profiler
  170:     r@ @ compile, \ old hook behaviour
  171:     POSTPONE ;
  172:     r> ! ; \ change hook behaviour
  173: 
  174: : note-execute ( -- )
  175:     \ end of BB due to execute
  176: ;
  177: 
  178: : note-call ( addr -- )
  179:     \ addr is the body address of a called colon def or does handler
  180:     dup 3 cells + @ ['] dinc >body = if
  181: 	1 over  cell+ @ profile-calls +!
  182:     endif
  183:     drop ;
  184:     
  185: : prof-compile, ( xt -- )
  186:     dup >does-code if
  187: 	dup >does-code note-call
  188:     then
  189:     dup >code-address CASE
  190: 	docol:   OF dup >body note-call ENDOF
  191: 	dodefer: OF note-execute ENDOF
  192: 	dofield: OF >body @ ['] lit+ peephole-compile, , EXIT ENDOF
  193: 	\ dofield: OF >body @ POSTPONE literal ['] + peephole-compile, EXIT ENDOF
  194: 	\ code words and ;code-defined words (code words could be optimized):
  195: 	dup in-dictionary? IF drop POSTPONE literal ['] execute peephole-compile, EXIT THEN
  196:     ENDCASE
  197:     DEFERS compile, ;
  198: 
  199: \ hook-profiling-into then-like
  200: \ \ hook-profiling-into if-like    \ subsumed by other-control-flow
  201: \ \ hook-profiling-into ahead-like \ subsumed by other-control-flow
  202: \ hook-profiling-into other-control-flow
  203: \ hook-profiling-into begin-like
  204: \ hook-profiling-into again-like
  205: \ hook-profiling-into until-like
  206: 
  207: : :-hook-profile ( -- )
  208:     defers :-hook
  209:     next-profile-point-p @
  210:     profile-this
  211:     @ dup last-colondef-profile !
  212:     profile-colondef? on ;
  213: 
  214: ' :-hook-profile IS :-hook
  215: ' prof-compile, IS compile,

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