#include "glproc.hpp"
#include "glstate.hpp"
#include "glretrace.hpp"
+#include "os_time.hpp"
+/* Synchronous debug output may reduce performance however,
+ * without it the callNo in the callback may be inaccurate
+ * as the callback may be called at any time.
+ */
+#define DEBUG_OUTPUT_SYNCHRONOUS 0
namespace glretrace {
bool insideList = false;
bool insideGlBeginEnd = false;
+struct CallQuery
+{
+ GLuint ids[3];
+ unsigned call;
+ GLuint program;
+ const trace::FunctionSig *sig;
+ int64_t cpuStart;
+ int64_t cpuEnd;
+};
+
+static bool supportsElapsed = true;
+static bool supportsTimestamp = true;
+static bool supportsOcclusion = true;
+static bool supportsDebugOutput = true;
+
+static bool firstFrame = true;
+static std::list<CallQuery> callQueries;
+
+static void APIENTRY
+debugOutputCallback(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* message, GLvoid* userParam);
void
checkGlError(trace::Call &call) {
os << "\n";
}
-struct CallQuery
-{
- GLuint ids[2];
- unsigned call;
- const trace::FunctionSig *sig;
-};
-
-static const int maxActiveCallQueries = 256;
-static std::list<CallQuery> callQueries;
-static bool firstFrame = true;
+static inline GLint64
+getGpuTimestamp() {
+ GLuint query = 0;
+ GLint64 timestamp = 0;
-static GLuint64
-getTimestamp() {
- GLuint query;
- GLuint64 timestamp;
-
- glGenQueries(1, &query);
-
- glQueryCounter(query, GL_TIMESTAMP);
- glGetQueryObjectui64vEXT(query, GL_QUERY_RESULT, ×tamp);
-
- glDeleteQueries(1, &query);
+ if (retrace::profilingGpuTimes && supportsTimestamp) {
+ glGenQueries(1, &query);
+ glQueryCounter(query, GL_TIMESTAMP);
+ glGetQueryObjecti64vEXT(query, GL_QUERY_RESULT, ×tamp);
+ glDeleteQueries(1, &query);
+ }
return timestamp;
}
+static inline GLint64
+getCpuTimestamp() {
+ if (retrace::profilingCpuTimes) {
+ return os::getTime();
+ } else {
+ return 0;
+ }
+}
+
static void
completeCallQuery(CallQuery& query) {
/* Get call start and duration */
- GLuint64 timestamp, duration;
- glGetQueryObjectui64vEXT(query.ids[0], GL_QUERY_RESULT, ×tamp);
- glGetQueryObjectui64vEXT(query.ids[1], GL_QUERY_RESULT, &duration);
- glDeleteQueries(2, query.ids);
+ GLint64 gpuStart = 0, gpuDuration = 0, cpuDuration = 0, samples = 0;
+
+ if (retrace::profilingGpuTimes) {
+ if (supportsTimestamp) {
+ glGetQueryObjecti64vEXT(query.ids[0], GL_QUERY_RESULT, &gpuStart);
+ }
+
+ glGetQueryObjecti64vEXT(query.ids[1], GL_QUERY_RESULT, &gpuDuration);
+ }
+
+ if (retrace::profilingCpuTimes) {
+ cpuDuration = query.cpuEnd - query.cpuStart;
+ }
+
+ if (retrace::profilingPixelsDrawn) {
+ glGetQueryObjecti64vEXT(query.ids[2], GL_QUERY_RESULT, &samples);
+ }
+
+ glDeleteQueries(3, query.ids);
/* Add call to profile */
- retrace::profiler.addCall(query.call, query.sig->name, timestamp, duration);
+ retrace::profiler.addCall(query.call, query.sig->name, query.program, samples, gpuStart, gpuDuration, query.cpuStart, cpuDuration);
}
void
-beginProfileGPU(trace::Call &call) {
- if (firstFrame) {
- frame_start();
+flushQueries() {
+ for (std::list<CallQuery>::iterator itr = callQueries.begin(); itr != callQueries.end(); ++itr) {
+ completeCallQuery(*itr);
}
- /* Ensure we don't have TOO many queries waiting for results */
- if (callQueries.size() >= maxActiveCallQueries) {
- completeCallQuery(callQueries.front());
- callQueries.pop_front();
+ callQueries.clear();
+}
+
+void
+beginProfile(trace::Call &call, bool isDraw) {
+ if (firstFrame) {
+ frame_start();
}
/* Create call query */
CallQuery query;
query.call = call.no;
query.sig = call.sig;
+ query.program = glretrace::currentContext ? glretrace::currentContext->activeProgram : 0;
+
+ if (isDraw) {
+ glGenQueries(3, query.ids);
+
+ if (retrace::profilingGpuTimes) {
+ if (supportsTimestamp) {
+ glQueryCounter(query.ids[0], GL_TIMESTAMP);
+ }
+
+ glBeginQuery(GL_TIME_ELAPSED, query.ids[1]);
+ }
+
+ if (retrace::profilingPixelsDrawn) {
+ glBeginQuery(GL_SAMPLES_PASSED, query.ids[2]);
+ }
+ }
- glGenQueries(2, query.ids);
- glQueryCounter(query.ids[0], GL_TIMESTAMP);
- glBeginQuery(GL_TIME_ELAPSED, query.ids[1]);
+ if (retrace::profilingCpuTimes) {
+ query.cpuStart = getCpuTimestamp();
+ }
callQueries.push_back(query);
}
void
-endProfileGPU(trace::Call &call) {
- glEndQuery(GL_TIME_ELAPSED);
+endProfile(trace::Call &call, bool isDraw) {
+ if (retrace::profilingCpuTimes) {
+ CallQuery& query = callQueries.back();
+ query.cpuEnd = getCpuTimestamp();
+
+ if (!isDraw) {
+ retrace::profiler.addCall(query.call, query.sig->name, query.program, -1, 0, 0, query.cpuStart, query.cpuEnd - query.cpuStart);
+ callQueries.pop_back();
+ }
+ }
+
+ if (isDraw) {
+ if (retrace::profilingGpuTimes) {
+ glEndQuery(GL_TIME_ELAPSED);
+ }
+
+ if (retrace::profilingPixelsDrawn) {
+ glEndQuery(GL_SAMPLES_PASSED);
+ }
+ }
+}
+
+void
+initContext() {
+ GLint64 gpuTime = 0, cpuTime = 0;
+ const char* extensions;
+
+ extensions = (const char*)glGetString(GL_EXTENSIONS);
+
+ /* Ensure we have adequate extension support */
+ supportsTimestamp = glws::checkExtension("GL_ARB_timer_query", extensions);
+ supportsElapsed = glws::checkExtension("GL_EXT_timer_query", extensions) || supportsTimestamp;
+ supportsOcclusion = glws::checkExtension("GL_ARB_occlusion_query", extensions);
+ supportsDebugOutput = glws::checkExtension("GL_ARB_debug_output", extensions);
+
+ if (retrace::profilingGpuTimes) {
+ if (!supportsTimestamp && !supportsElapsed) {
+ std::cout << "Error: Cannot run profile, GL_EXT_timer_query extension is not supported." << std::endl;
+ exit(-1);
+ }
+
+ GLint bits = 0;
+ glGetQueryiv(GL_TIME_ELAPSED, GL_QUERY_COUNTER_BITS, &bits);
+
+ if (!bits) {
+ std::cout << "Error: Cannot run profile, GL_QUERY_COUNTER_BITS == 0." << std::endl;
+ exit(-1);
+ }
+ }
+
+ if (retrace::profilingPixelsDrawn && !supportsOcclusion) {
+ std::cout << "Error: Cannot run profile, GL_ARB_occlusion_query extension is not supported." << std::endl;
+ exit(-1);
+ }
+
+ if (retrace::debug && supportsDebugOutput) {
+ glDebugMessageCallbackARB(&debugOutputCallback, currentContext);
+
+ if (DEBUG_OUTPUT_SYNCHRONOUS) {
+ glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS_ARB);
+ }
+ }
+
+ /* Sync the gpu and cpu start times */
+ if (!retrace::profiler.hasBaseTimes()) {
+ if (retrace::profilingGpuTimes) {
+ gpuTime = getGpuTimestamp();
+ }
+
+ if (retrace::profilingCpuTimes) {
+ cpuTime = getCpuTimestamp();
+ }
+
+ retrace::profiler.setBaseTimes(gpuTime, cpuTime);
+ }
}
void
frame_start() {
firstFrame = false;
- if (retrace::profileGPU) {
- retrace::profiler.addFrameStart(retrace::frameNo, getTimestamp());
+ if (retrace::profiling) {
+ int64_t gpuStart = getGpuTimestamp();
+ int64_t cpuStart = getCpuTimestamp();
+ retrace::profiler.addFrameStart(retrace::frameNo, gpuStart, cpuStart);
}
}
void
frame_complete(trace::Call &call) {
- if (retrace::profileGPU) {
+ if (retrace::profiling) {
/* Complete any remaining queries */
- for (std::list<CallQuery>::iterator itr = callQueries.begin(); itr != callQueries.end(); ++itr) {
- completeCallQuery(*itr);
- }
-
- callQueries.clear();
+ flushQueries();
/* Indicate end of current frame */
- retrace::profiler.addFrameEnd(getTimestamp());
+ int64_t gpuEnd = getGpuTimestamp();
+ int64_t cpuEnd = getCpuTimestamp();
+ retrace::profiler.addFrameEnd(gpuEnd, cpuEnd);
}
retrace::frameComplete(call);
}
}
+static const char*
+getDebugOutputSource(GLenum source) {
+ switch(source) {
+ case GL_DEBUG_SOURCE_API_ARB:
+ return "API";
+ case GL_DEBUG_SOURCE_WINDOW_SYSTEM_ARB:
+ return "Window System";
+ case GL_DEBUG_SOURCE_SHADER_COMPILER_ARB:
+ return "Shader Compiler";
+ case GL_DEBUG_SOURCE_THIRD_PARTY_ARB:
+ return "Third Party";
+ case GL_DEBUG_SOURCE_APPLICATION_ARB:
+ return "Application";
+ case GL_DEBUG_SOURCE_OTHER_ARB:
+ default:
+ return "";
+ }
+}
+
+static const char*
+getDebugOutputType(GLenum type) {
+ switch(type) {
+ case GL_DEBUG_TYPE_ERROR_ARB:
+ return "error";
+ case GL_DEBUG_TYPE_DEPRECATED_BEHAVIOR_ARB:
+ return "deprecated behaviour";
+ case GL_DEBUG_TYPE_UNDEFINED_BEHAVIOR_ARB:
+ return "undefined behaviour";
+ case GL_DEBUG_TYPE_PORTABILITY_ARB:
+ return "portability issue";
+ case GL_DEBUG_TYPE_PERFORMANCE_ARB:
+ return "performance issue";
+ case GL_DEBUG_TYPE_OTHER_ARB:
+ default:
+ return "unknown issue";
+ }
+}
+
+static const char*
+getDebugOutputSeverity(GLenum severity) {
+ switch(severity) {
+ case GL_DEBUG_SEVERITY_HIGH_ARB:
+ return "High";
+ case GL_DEBUG_SEVERITY_MEDIUM_ARB:
+ return "Medium";
+ case GL_DEBUG_SEVERITY_LOW_ARB:
+ return "Low";
+ default:
+ return "usnknown";
+ }
+}
+
+static void APIENTRY
+debugOutputCallback(GLenum source, GLenum type, GLuint id, GLenum severity, GLsizei length, const GLchar* message, GLvoid* userParam) {
+ std::cerr << retrace::callNo << ": ";
+ std::cerr << "glDebugOutputCallback: ";
+ std::cerr << getDebugOutputSeverity(severity) << " severity ";
+ std::cerr << getDebugOutputSource(source) << " " << getDebugOutputType(type);
+ std::cerr << " " << id;
+ std::cerr << ", " << message;
+ std::cerr << std::endl;
+}
+
} /* namespace glretrace */
void
retrace::flushRendering(void) {
+ glretrace::flushQueries();
glFlush();
}