git.cworth.org Git - apitrace/blob - retrace/glretrace_main.cpp

   1 /**************************************************************************
   2  *
   3  * Copyright 2011 Jose Fonseca
   4  * Copyright (C) 2013 Intel Corporation. All rights reversed.
   5  * Author: Shuang He <shuang.he@intel.com>
   6  * All Rights Reserved.
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a copy
   9  * of this software and associated documentation files (the "Software"), to deal
  10  * in the Software without restriction, including without limitation the rights
  11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  12  * copies of the Software, and to permit persons to whom the Software is
  13  * furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included in
  16  * all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  21  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  24  * THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28
  29 #include <string.h>
  30
  31 #include "retrace.hpp"
  32 #include "glproc.hpp"
  33 #include "glstate.hpp"
  34 #include "glretrace.hpp"
  35 #include "os_time.hpp"
  36 #include "os_memory.hpp"
  37
  38 /* Synchronous debug output may reduce performance however,
  39  * without it the callNo in the callback may be inaccurate
  40  * as the callback may be called at any time.
  41  */
  42 #define DEBUG_OUTPUT_SYNCHRONOUS 0
  43
  44 namespace glretrace {
  45
  46 bool insideList = false;
  47 bool insideGlBeginEnd = false;
  48 bool supportsARBShaderObjects = false;
  49
  50 enum {
  51     GPU_START = 0,
  52     GPU_DURATION,
  53     OCCLUSION,
  54     NUM_QUERIES,
  55 };
  56
  57 struct CallQuery
  58 {
  59     GLuint ids[NUM_QUERIES];
  60     unsigned call;
  61     bool isDraw;
  62     GLuint program;
  63     const trace::FunctionSig *sig;
  64     int64_t cpuStart;
  65     int64_t cpuEnd;
  66     int64_t vsizeStart;
  67     int64_t vsizeEnd;
  68     int64_t rssStart;
  69     int64_t rssEnd;
  70 };
  71
  72 static bool supportsElapsed = true;
  73 static bool supportsTimestamp = true;
  74 static bool supportsOcclusion = true;
  75 static bool supportsDebugOutput = true;
  76
  77 static std::list<CallQuery> callQueries;
  78
  79 static void APIENTRY
  80 debugOutputCallback(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* message, GLvoid* userParam);
  81
  82 void
  83 checkGlError(trace::Call &call) {
  84     GLenum error = glGetError();
  85     while (error != GL_NO_ERROR) {
  86         std::ostream & os = retrace::warning(call);
  87
  88         os << "glGetError(";
  89         os << call.name();
  90         os << ") = ";
  91
  92         switch (error) {
  93         case GL_INVALID_ENUM:
  94             os << "GL_INVALID_ENUM";
  95             break;
  96         case GL_INVALID_VALUE:
  97             os << "GL_INVALID_VALUE";
  98             break;
  99         case GL_INVALID_OPERATION:
 100             os << "GL_INVALID_OPERATION";
 101             break;
 102         case GL_STACK_OVERFLOW:
 103             os << "GL_STACK_OVERFLOW";
 104             break;
 105         case GL_STACK_UNDERFLOW:
 106             os << "GL_STACK_UNDERFLOW";
 107             break;
 108         case GL_OUT_OF_MEMORY:
 109             os << "GL_OUT_OF_MEMORY";
 110             break;
 111         case GL_INVALID_FRAMEBUFFER_OPERATION:
 112             os << "GL_INVALID_FRAMEBUFFER_OPERATION";
 113             break;
 114         case GL_TABLE_TOO_LARGE:
 115             os << "GL_TABLE_TOO_LARGE";
 116             break;
 117         default:
 118             os << error;
 119             break;
 120         }
 121         os << "\n";
 122
 123         error = glGetError();
 124     }
 125 }
 126
 127 static inline int64_t
 128 getCurrentTime(void) {
 129     if (retrace::profilingGpuTimes && supportsTimestamp) {
 130         /* Get the current GL time without stalling */
 131         GLint64 timestamp = 0;
 132         glGetInteger64v(GL_TIMESTAMP, &timestamp);
 133         return timestamp;
 134     } else {
 135         return os::getTime();
 136     }
 137 }
 138
 139 static inline int64_t
 140 getTimeFrequency(void) {
 141     if (retrace::profilingGpuTimes && supportsTimestamp) {
 142         return 1000000000;
 143     } else {
 144         return os::timeFrequency;
 145     }
 146 }
 147
 148 static inline void
 149 getCurrentVsize(int64_t& vsize) {
 150     vsize = os::getVsize();
 151 }
 152
 153 static inline void
 154 getCurrentRss(int64_t& rss) {
 155     rss = os::getRss();
 156 }
 157
 158 static void
 159 completeCallQuery(CallQuery& query) {
 160     /* Get call start and duration */
 161     int64_t gpuStart = 0, gpuDuration = 0, cpuDuration = 0, pixels = 0, vsizeDuration = 0, rssDuration = 0;
 162
 163     if (query.isDraw) {
 164         if (retrace::profilingGpuTimes) {
 165             if (supportsTimestamp) {
 166                 glGetQueryObjecti64vEXT(query.ids[GPU_START], GL_QUERY_RESULT, &gpuStart);
 167             }
 168
 169             glGetQueryObjecti64vEXT(query.ids[GPU_DURATION], GL_QUERY_RESULT, &gpuDuration);
 170         }
 171
 172         if (retrace::profilingPixelsDrawn) {
 173             glGetQueryObjecti64vEXT(query.ids[OCCLUSION], GL_QUERY_RESULT, &pixels);
 174         }
 175
 176     } else {
 177         pixels = -1;
 178     }
 179
 180     if (retrace::profilingCpuTimes) {
 181         double cpuTimeScale = 1.0E9 / getTimeFrequency();
 182         cpuDuration = (query.cpuEnd - query.cpuStart) * cpuTimeScale;
 183         query.cpuStart *= cpuTimeScale;
 184     }
 185
 186     if (retrace::profilingMemoryUsage) {
 187         vsizeDuration = query.vsizeEnd - query.vsizeStart;
 188         rssDuration = query.rssEnd - query.rssStart;
 189     }
 190
 191     glDeleteQueries(NUM_QUERIES, query.ids);
 192
 193     /* Add call to profile */
 194     retrace::profiler.addCall(query.call, query.sig->name, query.program, pixels, gpuStart, gpuDuration, query.cpuStart, cpuDuration, query.vsizeStart, vsizeDuration, query.rssStart, rssDuration);
 195 }
 196
 197 void
 198 flushQueries() {
 199     for (std::list<CallQuery>::iterator itr = callQueries.begin(); itr != callQueries.end(); ++itr) {
 200         completeCallQuery(*itr);
 201     }
 202
 203     callQueries.clear();
 204 }
 205
 206 void
 207 beginProfile(trace::Call &call, bool isDraw) {
 208     glretrace::Context *currentContext = glretrace::getCurrentContext();
 209
 210     /* Create call query */
 211     CallQuery query;
 212     query.isDraw = isDraw;
 213     query.call = call.no;
 214     query.sig = call.sig;
 215     query.program = currentContext ? currentContext->activeProgram : 0;
 216
 217     glGenQueries(NUM_QUERIES, query.ids);
 218
 219     /* GPU profiling only for draw calls */
 220     if (isDraw) {
 221         if (retrace::profilingGpuTimes) {
 222             if (supportsTimestamp) {
 223                 glQueryCounter(query.ids[GPU_START], GL_TIMESTAMP);
 224             }
 225
 226             glBeginQuery(GL_TIME_ELAPSED, query.ids[GPU_DURATION]);
 227         }
 228
 229         if (retrace::profilingPixelsDrawn) {
 230             glBeginQuery(GL_SAMPLES_PASSED, query.ids[OCCLUSION]);
 231         }
 232     }
 233
 234     callQueries.push_back(query);
 235
 236     /* CPU profiling for all calls */
 237     if (retrace::profilingCpuTimes) {
 238         CallQuery& query = callQueries.back();
 239         query.cpuStart = getCurrentTime();
 240     }
 241
 242     if (retrace::profilingMemoryUsage) {
 243         CallQuery& query = callQueries.back();
 244         query.vsizeStart = os::getVsize();
 245         query.rssStart = os::getRss();
 246     }
 247 }
 248
 249 void
 250 endProfile(trace::Call &call, bool isDraw) {
 251
 252     /* CPU profiling for all calls */
 253     if (retrace::profilingCpuTimes) {
 254         CallQuery& query = callQueries.back();
 255         query.cpuEnd = getCurrentTime();
 256     }
 257
 258     /* GPU profiling only for draw calls */
 259     if (isDraw) {
 260         if (retrace::profilingGpuTimes) {
 261             glEndQuery(GL_TIME_ELAPSED);
 262         }
 263
 264         if (retrace::profilingPixelsDrawn) {
 265             glEndQuery(GL_SAMPLES_PASSED);
 266         }
 267     }
 268
 269     if (retrace::profilingMemoryUsage) {
 270         CallQuery& query = callQueries.back();
 271         query.vsizeEnd = os::getVsize();
 272         query.rssEnd = os::getRss();
 273     }
 274 }
 275
 276 void
 277 initContext() {
 278     glretrace::Context *currentContext = glretrace::getCurrentContext();
 279
 280     /* Ensure we have adequate extension support */
 281     assert(currentContext);
 282     supportsTimestamp   = currentContext->hasExtension("GL_ARB_timer_query");
 283     supportsElapsed     = currentContext->hasExtension("GL_EXT_timer_query") || supportsTimestamp;
 284     supportsOcclusion   = currentContext->hasExtension("GL_ARB_occlusion_query");
 285     supportsDebugOutput = currentContext->hasExtension("GL_ARB_debug_output");
 286     supportsARBShaderObjects = currentContext->hasExtension("GL_ARB_shader_objects");
 287
 288     /* Check for timer query support */
 289     if (retrace::profilingGpuTimes) {
 290         if (!supportsTimestamp && !supportsElapsed) {
 291             std::cout << "Error: Cannot run profile, GL_EXT_timer_query extension is not supported." << std::endl;
 292             exit(-1);
 293         }
 294
 295         GLint bits = 0;
 296         glGetQueryiv(GL_TIME_ELAPSED, GL_QUERY_COUNTER_BITS, &bits);
 297
 298         if (!bits) {
 299             std::cout << "Error: Cannot run profile, GL_QUERY_COUNTER_BITS == 0." << std::endl;
 300             exit(-1);
 301         }
 302     }
 303
 304     /* Check for occlusion query support */
 305     if (retrace::profilingPixelsDrawn && !supportsOcclusion) {
 306         std::cout << "Error: Cannot run profile, GL_ARB_occlusion_query extension is not supported." << std::endl;
 307         exit(-1);
 308     }
 309
 310     /* Setup debug message call back */
 311     if (retrace::debug && supportsDebugOutput) {
 312         glretrace::Context *currentContext = glretrace::getCurrentContext();
 313         glDebugMessageCallbackARB(&debugOutputCallback, currentContext);
 314
 315         if (DEBUG_OUTPUT_SYNCHRONOUS) {
 316             glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS_ARB);
 317         }
 318     }
 319
 320     /* Sync the gpu and cpu start times */
 321     if (retrace::profilingCpuTimes || retrace::profilingGpuTimes) {
 322         if (!retrace::profiler.hasBaseTimes()) {
 323             double cpuTimeScale = 1.0E9 / getTimeFrequency();
 324             GLint64 currentTime = getCurrentTime() * cpuTimeScale;
 325             retrace::profiler.setBaseCpuTime(currentTime);
 326             retrace::profiler.setBaseGpuTime(currentTime);
 327         }
 328     }
 329
 330     if (retrace::profilingMemoryUsage) {
 331         GLint64 currentVsize, currentRss;
 332         getCurrentVsize(currentVsize);
 333         retrace::profiler.setBaseVsizeUsage(currentVsize);
 334         getCurrentRss(currentRss);
 335         retrace::profiler.setBaseRssUsage(currentRss);
 336     }
 337 }
 338
 339 void
 340 frame_complete(trace::Call &call) {
 341     if (retrace::profiling) {
 342         /* Complete any remaining queries */
 343         flushQueries();
 344
 345         /* Indicate end of current frame */
 346         retrace::profiler.addFrameEnd();
 347     }
 348
 349     retrace::frameComplete(call);
 350
 351     glretrace::Context *currentContext = glretrace::getCurrentContext();
 352     if (!currentContext) {
 353         return;
 354     }
 355
 356     assert(currentContext->drawable);
 357     if (retrace::debug && !currentContext->drawable->visible) {
 358         retrace::warning(call) << "could not infer drawable size (glViewport never called)\n";
 359     }
 360 }
 361
 362 static const char*
 363 getDebugOutputSource(GLenum source) {
 364     switch(source) {
 365     case GL_DEBUG_SOURCE_API_ARB:
 366         return "API";
 367     case GL_DEBUG_SOURCE_WINDOW_SYSTEM_ARB:
 368         return "Window System";
 369     case GL_DEBUG_SOURCE_SHADER_COMPILER_ARB:
 370         return "Shader Compiler";
 371     case GL_DEBUG_SOURCE_THIRD_PARTY_ARB:
 372         return "Third Party";
 373     case GL_DEBUG_SOURCE_APPLICATION_ARB:
 374         return "Application";
 375     case GL_DEBUG_SOURCE_OTHER_ARB:
 376     default:
 377         return "";
 378     }
 379 }
 380
 381 static const char*
 382 getDebugOutputType(GLenum type) {
 383     switch(type) {
 384     case GL_DEBUG_TYPE_ERROR_ARB:
 385         return "error";
 386     case GL_DEBUG_TYPE_DEPRECATED_BEHAVIOR_ARB:
 387         return "deprecated behaviour";
 388     case GL_DEBUG_TYPE_UNDEFINED_BEHAVIOR_ARB:
 389         return "undefined behaviour";
 390     case GL_DEBUG_TYPE_PORTABILITY_ARB:
 391         return "portability issue";
 392     case GL_DEBUG_TYPE_PERFORMANCE_ARB:
 393         return "performance issue";
 394     case GL_DEBUG_TYPE_OTHER_ARB:
 395     default:
 396         return "unknown issue";
 397     }
 398 }
 399
 400 static const char*
 401 getDebugOutputSeverity(GLenum severity) {
 402     switch(severity) {
 403     case GL_DEBUG_SEVERITY_HIGH_ARB:
 404         return "High";
 405     case GL_DEBUG_SEVERITY_MEDIUM_ARB:
 406         return "Medium";
 407     case GL_DEBUG_SEVERITY_LOW_ARB:
 408         return "Low";
 409     default:
 410         return "usnknown";
 411     }
 412 }
 413
 414 static void APIENTRY
 415 debugOutputCallback(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* message, GLvoid* userParam) {
 416     std::cerr << retrace::callNo << ": ";
 417     std::cerr << "glDebugOutputCallback: ";
 418     std::cerr << getDebugOutputSeverity(severity) << " severity ";
 419     std::cerr << getDebugOutputSource(source) << " " << getDebugOutputType(type);
 420     std::cerr << " " << id;
 421     std::cerr << ", " << message;
 422     std::cerr << std::endl;
 423 }
 424
 425 } /* namespace glretrace */
 426
 427
 428 class GLDumper : public retrace::Dumper {
 429 public:
 430     image::Image *
 431     getSnapshot(void) {
 432         if (!glretrace::getCurrentContext()) {
 433             return NULL;
 434         }
 435         return glstate::getDrawBufferImage();
 436     }
 437
 438     bool
 439     dumpState(std::ostream &os) {
 440         glretrace::Context *currentContext = glretrace::getCurrentContext();
 441         if (glretrace::insideGlBeginEnd ||
 442             !currentContext) {
 443             return false;
 444         }
 445         glstate::dumpCurrentContext(os);
 446         return true;
 447     }
 448 };
 449
 450 static GLDumper glDumper;
 451
 452
 453 void
 454 retrace::setUp(void) {
 455     glws::init();
 456     dumper = &glDumper;
 457 }
 458
 459
 460 void
 461 retrace::addCallbacks(retrace::Retracer &retracer)
 462 {
 463     retracer.addCallbacks(glretrace::gl_callbacks);
 464     retracer.addCallbacks(glretrace::glx_callbacks);
 465     retracer.addCallbacks(glretrace::wgl_callbacks);
 466     retracer.addCallbacks(glretrace::cgl_callbacks);
 467     retracer.addCallbacks(glretrace::egl_callbacks);
 468 }
 469
 470
 471 void
 472 retrace::flushRendering(void) {
 473     glretrace::Context *currentContext = glretrace::getCurrentContext();
 474     if (currentContext) {
 475         glretrace::flushQueries();
 476         glFlush();
 477     }
 478 }
 479
 480 void
 481 retrace::waitForInput(void) {
 482     glretrace::Context *currentContext = glretrace::getCurrentContext();
 483     if (currentContext) {
 484         glretrace::flushQueries();
 485         glFlush();
 486     }
 487     while (glws::processEvents()) {
 488         os::sleep(100*1000);
 489     }
 490 }
 491
 492 void
 493 retrace::cleanUp(void) {
 494     glws::cleanup();
 495 }