1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/prof/prof.cpp	Sat Mar 03 10:31:27 2012 -0600
     1.3 @@ -0,0 +1,405 @@
     1.4 +// VisualBoyAdvance - Nintendo Gameboy/GameboyAdvance (TM) emulator.
     1.5 +// Copyright (C) 1999-2003 Forgotten
     1.6 +// Copyright (C) 2004 Forgotten and the VBA development team
     1.7 +
     1.8 +// This program is free software; you can redistribute it and/or modify
     1.9 +// it under the terms of the GNU General Public License as published by
    1.10 +// the Free Software Foundation; either version 2, or(at your option)
    1.11 +// any later version.
    1.12 +//
    1.13 +// This program is distributed in the hope that it will be useful,
    1.14 +// but WITHOUT ANY WARRANTY; without even the implied warranty of
    1.15 +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    1.16 +// GNU General Public License for more details.
    1.17 +//
    1.18 +// You should have received a copy of the GNU General Public License
    1.19 +// along with this program; if not, write to the Free Software Foundation,
    1.20 +// Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
    1.21 +
    1.22 +// adapted from gmon.c
    1.23 +/*-
    1.24 + * Copyright (c) 1991, 1998 The Regents of the University of California.
    1.25 + * All rights reserved.
    1.26 + *
    1.27 + * Redistribution and use in source and binary forms, with or without
    1.28 + * modification, are permitted provided that the following conditions
    1.29 + * are met:
    1.30 + * 1. Redistributions of source code must retain the above copyright
    1.31 + *    notice, this list of conditions and the following disclaimer.
    1.32 + * 2. Redistributions in binary form must reproduce the above copyright
    1.33 + *    notice, this list of conditions and the following disclaimer in the
    1.34 + *    documentation and/or other materials provided with the distribution.
    1.35 + * 3. [rescinded 22 July 1999]
    1.36 + * 4. Neither the name of the University nor the names of its contributors
    1.37 + *    may be used to endorse or promote products derived from this software
    1.38 + *    without specific prior written permission.
    1.39 + *
    1.40 + * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
    1.41 + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
    1.42 + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
    1.43 + * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
    1.44 + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
    1.45 + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
    1.46 + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
    1.47 + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
    1.48 + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
    1.49 + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
    1.50 + * SUCH DAMAGE.
    1.51 + */
    1.52 +
    1.53 +#include <stdlib.h>
    1.54 +#include <stdio.h>
    1.55 +#include <memory.h>
    1.56 +
    1.57 +#include "gmon.h"
    1.58 +#include "gmon_out.h"
    1.59 +
    1.60 +#include "../common/System.h"
    1.61 +#include "../gba/GBA.h"
    1.62 +#include "../gba/GBAGlobals.h"
    1.63 +#include "../NLS.h"
    1.64 +
    1.65 +/*
    1.66 + * froms is actually a bunch of unsigned shorts indexing tos
    1.67 + */
    1.68 +static int  profiling = 3;
    1.69 +static unsigned short *froms;
    1.70 +static struct tostruct *tos = 0;
    1.71 +static long  tolimit = 0;
    1.72 +static u32  s_lowpc = 0;
    1.73 +static u32  s_highpc = 0;
    1.74 +static unsigned long s_textsize = 0;
    1.75 +
    1.76 +static int ssiz;
    1.77 +static char *sbuf;
    1.78 +static int s_scale;
    1.79 +
    1.80 +static int hz = 0;
    1.81 +static int hist_num_bins = 0;
    1.82 +static char hist_dimension[16] = "seconds";
    1.83 +static char hist_dimension_abbrev = 's';
    1.84 +
    1.85 +/* see profil(2) where this is describe (incorrectly) */
    1.86 +#define  SCALE_1_TO_1 0x10000L
    1.87 +
    1.88 +void profPut32(char *b, u32 v)
    1.89 +{
    1.90 +  b[0] = v & 255;
    1.91 +  b[1] = (v >> 8) & 255;
    1.92 +  b[2] = (v >> 16) & 255;
    1.93 +  b[3] = (v >> 24) & 255;
    1.94 +}
    1.95 +
    1.96 +void profPut16(char *b, u16 v)
    1.97 +{
    1.98 +  b[0] = v & 255;
    1.99 +  b[1] = (v >> 8) & 255;
   1.100 +}
   1.101 +
   1.102 +int profWrite8(FILE *f, u8 b)
   1.103 +{
   1.104 +  if(fwrite(&b, 1, 1, f) != 1)
   1.105 +    return 1;
   1.106 +  return 0;
   1.107 +}
   1.108 +
   1.109 +int profWrite32(FILE *f, u32 v)
   1.110 +{
   1.111 +  char buf[4];
   1.112 +
   1.113 +  profPut32(buf, v);
   1.114 +  if(fwrite(buf, 1, 4, f) != 4)
   1.115 +    return 1;
   1.116 +  return 0;
   1.117 +}
   1.118 +
   1.119 +int profWrite(FILE *f, char *buf, unsigned int n)
   1.120 +{
   1.121 +  if(fwrite(buf, 1, n, f) != n)
   1.122 +    return 1;
   1.123 +  return 0;
   1.124 +}
   1.125 +
   1.126 +/* Control profiling;
   1.127 +   profiling is what mcount checks to see if
   1.128 +   all the data structures are ready.  */
   1.129 +
   1.130 +void profControl(int mode)
   1.131 +{
   1.132 +  if (mode) {
   1.133 +    /* start */
   1.134 +#ifdef PROFILING
   1.135 +    cpuProfil(sbuf, ssiz, (u32)s_lowpc, s_scale);
   1.136 +#endif
   1.137 +    profiling = 0;
   1.138 +  } else {
   1.139 +    /* stop */
   1.140 +#ifdef PROFILING
   1.141 +    cpuProfil(NULL, 0, 0, 0);
   1.142 +#endif
   1.143 +    profiling = 3;
   1.144 +  }
   1.145 +}
   1.146 +
   1.147 +
   1.148 +#define MSG N_("No space for profiling buffer(s)\n")
   1.149 +
   1.150 +void profStartup(u32 lowpc, u32 highpc)
   1.151 +{
   1.152 +  int monsize;
   1.153 +  char *buffer;
   1.154 +  int o;
   1.155 +  
   1.156 +  /*
   1.157 +   * round lowpc and highpc to multiples of the density we're using
   1.158 +   * so the rest of the scaling (here and in gprof) stays in ints.
   1.159 +   */
   1.160 +  lowpc = ROUNDDOWN(lowpc, HISTFRACTION*sizeof(HISTCOUNTER));
   1.161 +  s_lowpc = lowpc;
   1.162 +  highpc = ROUNDUP(highpc, HISTFRACTION*sizeof(HISTCOUNTER));
   1.163 +  s_highpc = highpc;
   1.164 +  s_textsize = highpc - lowpc;
   1.165 +  monsize = (s_textsize / HISTFRACTION);
   1.166 +  buffer = (char *)calloc(1, 2*monsize );
   1.167 +  if ( buffer == NULL ) {
   1.168 +    systemMessage(0, MSG);
   1.169 +    return;
   1.170 +  }
   1.171 +  froms = (unsigned short *) calloc(1, 4*s_textsize / HASHFRACTION );
   1.172 +  if ( froms == NULL ) {
   1.173 +    systemMessage(0, MSG);
   1.174 +    free(buffer);
   1.175 +    buffer = NULL;
   1.176 +    return;
   1.177 +  }
   1.178 +  tolimit = s_textsize * ARCDENSITY / 100;
   1.179 +  if ( tolimit < MINARCS ) {
   1.180 +    tolimit = MINARCS;
   1.181 +  } else if ( tolimit > 65534 ) {
   1.182 +    tolimit = 65534;
   1.183 +  }
   1.184 +  tos = (struct tostruct *) calloc(1, tolimit * sizeof( struct tostruct ) );
   1.185 +  if ( tos == NULL ) {
   1.186 +    systemMessage(0, MSG);
   1.187 +    
   1.188 +    free(buffer);
   1.189 +    buffer = NULL;
   1.190 +    
   1.191 +    free(froms);
   1.192 +    froms = NULL;
   1.193 +    
   1.194 +    return;
   1.195 +  }
   1.196 +  tos[0].link = 0;
   1.197 +  sbuf = buffer;
   1.198 +  ssiz = monsize;
   1.199 +  if ( monsize <= 0 )
   1.200 +    return;
   1.201 +  o = highpc - lowpc;
   1.202 +  if( monsize < o )
   1.203 +    s_scale = (int)(( (float) monsize / o ) * SCALE_1_TO_1);
   1.204 +  else
   1.205 +    s_scale = SCALE_1_TO_1;
   1.206 +  profControl(1);
   1.207 +}
   1.208 +
   1.209 +void profCleanup()
   1.210 +{
   1.211 +  FILE *fd;
   1.212 +  int fromindex;
   1.213 +  int endfrom;
   1.214 +  u32 frompc;
   1.215 +  int toindex;
   1.216 +  struct gmon_hdr ghdr;
   1.217 +  
   1.218 +  profControl(0);
   1.219 +  fd = fopen( "gmon.out" , "wb" );
   1.220 +  if ( fd == NULL ) {
   1.221 +    systemMessage( 0, "mcount: gmon.out" );
   1.222 +    return;
   1.223 +  }
   1.224 +
   1.225 +  memcpy(&ghdr.cookie[0], GMON_MAGIC, 4);
   1.226 +  profPut32((char *)ghdr.version, GMON_VERSION);
   1.227 +
   1.228 +  if(fwrite(&ghdr, sizeof(ghdr), 1, fd) != 1) {
   1.229 +    systemMessage(0, "mcount: gmon.out header");
   1.230 +    fclose(fd);
   1.231 +    return;
   1.232 +  }
   1.233 +
   1.234 +  if(hz == 0)
   1.235 +    hz = 100;
   1.236 +
   1.237 +  hist_num_bins = ssiz;
   1.238 +  
   1.239 +  if(profWrite8(fd, GMON_TAG_TIME_HIST) ||
   1.240 +     profWrite32(fd, (u32)s_lowpc) ||
   1.241 +     profWrite32(fd, (u32)s_highpc) ||
   1.242 +     profWrite32(fd, hist_num_bins) ||
   1.243 +     profWrite32(fd, hz) ||
   1.244 +     profWrite(fd, hist_dimension, 15) ||
   1.245 +     profWrite(fd, &hist_dimension_abbrev, 1)) {
   1.246 +    systemMessage(0, "mcount: gmon.out hist");
   1.247 +    fclose(fd);
   1.248 +    return;
   1.249 +  }
   1.250 +  u16 *hist_sample = (u16 *)sbuf;
   1.251 +
   1.252 +  u16 count;
   1.253 +  int i;
   1.254 +  
   1.255 +  for(i = 0; i < hist_num_bins; ++i) {
   1.256 +    profPut16((char *)&count, hist_sample[i]);
   1.257 +
   1.258 +    if(fwrite(&count, sizeof(count), 1, fd) != 1) {
   1.259 +      systemMessage(0, "mcount: gmon.out sample");
   1.260 +      fclose(fd);
   1.261 +      return;
   1.262 +    }
   1.263 +  }
   1.264 +  
   1.265 +  endfrom = s_textsize / (HASHFRACTION * sizeof(*froms));
   1.266 +  for ( fromindex = 0 ; fromindex < endfrom ; fromindex++ ) {
   1.267 +    if ( froms[fromindex] == 0 ) {
   1.268 +      continue;
   1.269 +    }
   1.270 +    frompc = s_lowpc + (fromindex * HASHFRACTION * sizeof(*froms));
   1.271 +    for (toindex=froms[fromindex]; toindex!=0; toindex=tos[toindex].link) {
   1.272 +      if(profWrite8(fd, GMON_TAG_CG_ARC) ||
   1.273 +         profWrite32(fd, (u32)frompc) ||
   1.274 +         profWrite32(fd, (u32)tos[toindex].selfpc) ||
   1.275 +         profWrite32(fd, tos[toindex].count)) {
   1.276 +        systemMessage(0, "mcount: arc");
   1.277 +        fclose(fd);
   1.278 +        return;
   1.279 +      }
   1.280 +    }
   1.281 +  }
   1.282 +  fclose(fd);
   1.283 +}
   1.284 +
   1.285 +void profCount()
   1.286 +{
   1.287 +  register u32 selfpc;
   1.288 +  register unsigned short *frompcindex;
   1.289 +  register struct tostruct *top;
   1.290 +  register struct tostruct *prevtop;
   1.291 +  register long toindex;
   1.292 +
   1.293 +  /*
   1.294 +   * find the return address for mcount,
   1.295 +   * and the return address for mcount's caller.
   1.296 +   */
   1.297 +  
   1.298 +  /* selfpc = pc pushed by mcount call.
   1.299 +     This identifies the function that was just entered.  */
   1.300 +  selfpc = (u32) reg[14].I;
   1.301 +  /* frompcindex = pc in preceding frame.
   1.302 +     This identifies the caller of the function just entered.  */
   1.303 +  frompcindex = (unsigned short *) reg[12].I;
   1.304 +  /*
   1.305 +   * check that we are profiling
   1.306 +   * and that we aren't recursively invoked.
   1.307 +   */
   1.308 +  if (profiling) {
   1.309 +    goto out;
   1.310 +  }
   1.311 +  profiling++;
   1.312 +  /*
   1.313 +   * check that frompcindex is a reasonable pc value.
   1.314 +   * for example: signal catchers get called from the stack,
   1.315 +   *   not from text space.  too bad.
   1.316 +   */
   1.317 +  frompcindex = (unsigned short *) ((long) frompcindex - (long) s_lowpc);
   1.318 +  if ((unsigned long) frompcindex > s_textsize) {
   1.319 +    goto done;
   1.320 +  }
   1.321 +  frompcindex =
   1.322 +    &froms[((long) frompcindex) / (HASHFRACTION * sizeof(*froms))];
   1.323 +  toindex = *frompcindex;
   1.324 +  if (toindex == 0) {
   1.325 +    /*
   1.326 +     * first time traversing this arc
   1.327 +     */
   1.328 +    toindex = ++tos[0].link;
   1.329 +    if (toindex >= tolimit) {
   1.330 +      goto overflow;
   1.331 +    }
   1.332 +    *frompcindex = (unsigned short)toindex;
   1.333 +    top = &tos[toindex];
   1.334 +    top->selfpc = selfpc;
   1.335 +    top->count = 1;
   1.336 +    top->link = 0;
   1.337 +    goto done;
   1.338 +  }
   1.339 +  top = &tos[toindex];
   1.340 +  if (top->selfpc == selfpc) {
   1.341 +    /*
   1.342 +     * arc at front of chain; usual case.
   1.343 +     */
   1.344 +    top->count++;
   1.345 +    goto done;
   1.346 +  }
   1.347 +  /*
   1.348 +   * have to go looking down chain for it.
   1.349 +   * top points to what we are looking at,
   1.350 +   * prevtop points to previous top.
   1.351 +   * we know it is not at the head of the chain.
   1.352 +   */
   1.353 +  for (; /* goto done */; ) {
   1.354 +    if (top->link == 0) {
   1.355 +      /*
   1.356 +       * top is end of the chain and none of the chain
   1.357 +       * had top->selfpc == selfpc.
   1.358 +       * so we allocate a new tostruct
   1.359 +       * and link it to the head of the chain.
   1.360 +       */
   1.361 +      toindex = ++tos[0].link;
   1.362 +      if (toindex >= tolimit) {
   1.363 +        goto overflow;
   1.364 +      }
   1.365 +      top = &tos[toindex];
   1.366 +      top->selfpc = selfpc;
   1.367 +      top->count = 1;
   1.368 +      top->link = *frompcindex;
   1.369 +      *frompcindex = (unsigned short)toindex;
   1.370 +      goto done;
   1.371 +    }
   1.372 +    /*
   1.373 +     * otherwise, check the next arc on the chain.
   1.374 +     */
   1.375 +    prevtop = top;
   1.376 +    top = &tos[top->link];
   1.377 +    if (top->selfpc == selfpc) {
   1.378 +      /*
   1.379 +       * there it is.
   1.380 +       * increment its count
   1.381 +       * move it to the head of the chain.
   1.382 +       */
   1.383 +      top->count++;
   1.384 +      toindex = prevtop->link;
   1.385 +      prevtop->link = top->link;
   1.386 +      top->link = *frompcindex;
   1.387 +      *frompcindex = (unsigned short)toindex;
   1.388 +      goto done;
   1.389 +    }
   1.390 +    
   1.391 +  }
   1.392 + done:
   1.393 +  profiling--;
   1.394 +  /* and fall through */
   1.395 + out:
   1.396 +  return;  /* normal return restores saved registers */
   1.397 +  
   1.398 + overflow:
   1.399 +  profiling++; /* halt further profiling */
   1.400 +#define TOLIMIT "mcount: tos overflow\n"
   1.401 +  systemMessage(0, TOLIMIT);
   1.402 +  goto out;
   1.403 +}
   1.404 +
   1.405 +void profSetHertz(int h)
   1.406 +{
   1.407 +  hz = h;
   1.408 +}