src/gallium/auxiliary/util/p_debug_prof.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /**
  29  * @file
  30  * Poor-man profiling.
  31  *
  32  * @author José Fonseca <jrfonseca@tungstengraphics.com>
  33  *
  34  * @sa http://blogs.msdn.com/joshpoley/archive/2008/03/12/poor-man-s-profiler.aspx
  35  * @sa http://www.johnpanzer.com/aci_cuj/index.html
  36  */
  37
  38 #include "pipe/p_config.h"
  39
  40 #if defined(PROFILE) && defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
  41
  42 #include <windows.h>
  43 #include <winddi.h>
  44
  45 #include "pipe/p_debug.h"
  46 #include "util/u_string.h"
  47
  48
  49 #define PROFILE_TABLE_SIZE (1024*1024)
  50 #define FILE_NAME_SIZE 256
  51
  52 struct debug_profile_entry
  53 {
  54    uintptr_t caller;
  55    uintptr_t callee;
  56    uint64_t samples;
  57 };
  58
  59 static unsigned long enabled = 0;
  60
  61 static WCHAR wFileName[FILE_NAME_SIZE] = L"\\??\\c:\\00000000.prof";
  62 static ULONG_PTR iFile = 0;
  63
  64 static struct debug_profile_entry *table = NULL;
  65 static unsigned long free_table_entries = 0;
  66 static unsigned long max_table_entries = 0;
  67
  68 uint64_t start_stamp = 0;
  69 uint64_t end_stamp = 0;
  70
  71
  72 static void
  73 debug_profile_entry(uintptr_t caller, uintptr_t callee, uint64_t samples)
  74 {
  75    unsigned hash = ( caller + callee ) & PROFILE_TABLE_SIZE - 1;
  76
  77    while(1) {
  78       if(table[hash].caller == 0 && table[hash].callee == 0) {
  79          table[hash].caller = caller;
  80          table[hash].callee = callee;
  81          table[hash].samples = samples;
  82          --free_table_entries;
  83          break;
  84       }
  85       else if(table[hash].caller == caller && table[hash].callee == callee) {
  86          table[hash].samples += samples;
  87          break;
  88       }
  89       else {
  90          ++hash;
  91       }
  92    }
  93 }
  94
  95
  96 static uintptr_t caller_stack[1024];
  97 static unsigned last_caller = 0;
  98
  99
 100 static int64_t delta(void) {
 101    int64_t result = end_stamp - start_stamp;
 102    if(result > UINT64_C(0xffffffff))
 103       result = 0;
 104    return result;
 105 }
 106
 107
 108 static void __cdecl
 109 debug_profile_enter(uintptr_t callee)
 110 {
 111    uintptr_t caller = last_caller ? caller_stack[last_caller - 1] : 0;
 112
 113    if (caller)
 114       debug_profile_entry(caller, 0, delta());
 115    debug_profile_entry(caller, callee, 1);
 116    caller_stack[last_caller++] = callee;
 117 }
 118
 119
 120 static void __cdecl
 121 debug_profile_exit(uintptr_t callee)
 122 {
 123    debug_profile_entry(callee, 0, delta());
 124    if(last_caller)
 125       --last_caller;
 126 }
 127
 128
 129 /**
 130  * Called at the start of every method or function.
 131  *
 132  * @sa http://msdn.microsoft.com/en-us/library/c63a9b7h.aspx
 133  */
 134 void __declspec(naked) __cdecl
 135 _penter(void) {
 136    _asm {
 137       push eax
 138       mov eax, [enabled]
 139       test eax, eax
 140       jz skip
 141
 142       push edx
 143
 144       rdtsc
 145       mov dword ptr [end_stamp], eax
 146       mov dword ptr [end_stamp+4], edx
 147
 148       xor eax, eax
 149       mov [enabled], eax
 150
 151       mov eax, [esp+8]
 152
 153       push ebx
 154       push ecx
 155       push ebp
 156       push edi
 157       push esi
 158
 159       push eax
 160       call debug_profile_enter
 161       add esp, 4
 162
 163       pop esi
 164       pop edi
 165       pop ebp
 166       pop ecx
 167       pop ebx
 168
 169       mov eax, 1
 170       mov [enabled], eax
 171
 172       rdtsc
 173       mov dword ptr [start_stamp], eax
 174       mov dword ptr [start_stamp+4], edx
 175
 176       pop edx
 177 skip:
 178       pop eax
 179       ret
 180    }
 181 }
 182
 183
 184 /**
 185  * Called at the end of Calls the end of every method or function.
 186  *
 187  * @sa http://msdn.microsoft.com/en-us/library/xc11y76y.aspx
 188  */
 189 void __declspec(naked) __cdecl
 190 _pexit(void) {
 191    _asm {
 192       push eax
 193       mov eax, [enabled]
 194       test eax, eax
 195       jz skip
 196
 197       push edx
 198
 199       rdtsc
 200       mov dword ptr [end_stamp], eax
 201       mov dword ptr [end_stamp+4], edx
 202
 203       xor eax, eax
 204       mov [enabled], eax
 205
 206       mov eax, [esp+8]
 207
 208       push ebx
 209       push ecx
 210       push ebp
 211       push edi
 212       push esi
 213
 214       push eax
 215       call debug_profile_exit
 216       add esp, 4
 217
 218       pop esi
 219       pop edi
 220       pop ebp
 221       pop ecx
 222       pop ebx
 223
 224       mov eax, 1
 225       mov [enabled], eax
 226
 227       rdtsc
 228       mov dword ptr [start_stamp], eax
 229       mov dword ptr [start_stamp+4], edx
 230
 231       pop edx
 232 skip:
 233       pop eax
 234       ret
 235    }
 236 }
 237
 238
 239 /**
 240  * Reference function for calibration.
 241  */
 242 void __declspec(naked)
 243 __debug_profile_reference(void) {
 244    _asm {
 245       call _penter
 246       call _pexit
 247       ret
 248    }
 249 }
 250
 251
 252 void
 253 debug_profile_start(void)
 254 {
 255    WCHAR *p;
 256
 257    // increment starting from the less significant digit
 258    p = &wFileName[14];
 259    while(1) {
 260       if(*p == '9') {
 261          *p-- = '0';
 262       }
 263       else {
 264          *p += 1;
 265          break;
 266       }
 267    }
 268
 269    table = EngMapFile(wFileName,
 270                       PROFILE_TABLE_SIZE*sizeof(struct debug_profile_entry),
 271                       &iFile);
 272    if(table) {
 273       unsigned i;
 274
 275       free_table_entries = max_table_entries = PROFILE_TABLE_SIZE;
 276       memset(table, 0, PROFILE_TABLE_SIZE*sizeof(struct debug_profile_entry));
 277
 278       table[0].caller = (uintptr_t)&__debug_profile_reference;
 279       table[0].callee = 0;
 280       table[0].samples = 0;
 281       --free_table_entries;
 282
 283       _asm {
 284          push edx
 285          push eax
 286
 287          rdtsc
 288          mov dword ptr [start_stamp], eax
 289          mov dword ptr [start_stamp+4], edx
 290
 291          pop edx
 292          pop eax
 293       }
 294
 295       last_caller = 0;
 296
 297       enabled = 1;
 298
 299       for(i = 0; i < 8; ++i) {
 300          _asm {
 301             call __debug_profile_reference
 302          }
 303       }
 304    }
 305 }
 306
 307
 308 void
 309 debug_profile_stop(void)
 310 {
 311    enabled = 0;
 312
 313    if(iFile)
 314       EngUnmapFile(iFile);
 315    iFile = 0;
 316    table = NULL;
 317    free_table_entries = max_table_entries = 0;
 318 }
 319
 320 #endif /* PROFILE */