unsigned call;
GLuint program;
const trace::FunctionSig *sig;
- uint64_t start;
- uint64_t duration;
+ 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;
os << "\n";
}
-static GLuint64
+static inline GLint64
getGpuTimestamp() {
GLuint query = 0;
- GLuint64 timestamp = 0;
+ GLint64 timestamp = 0;
if (retrace::profilingGpuTimes && supportsTimestamp) {
glGenQueries(1, &query);
glQueryCounter(query, GL_TIMESTAMP);
- glGetQueryObjectui64vEXT(query, GL_QUERY_RESULT, ×tamp);
+ glGetQueryObjecti64vEXT(query, GL_QUERY_RESULT, ×tamp);
glDeleteQueries(1, &query);
}
return timestamp;
}
-static GLuint64
+static inline GLint64
getCpuTimestamp() {
if (retrace::profilingCpuTimes) {
return os::getTime();
static void
completeCallQuery(CallQuery& query) {
/* Get call start and duration */
- GLuint64 timestamp = 0, duration = 0, samples = 0;
+ GLint64 gpuStart = 0, gpuDuration = 0, cpuDuration = 0, samples = 0;
if (retrace::profilingGpuTimes) {
if (supportsTimestamp) {
- glGetQueryObjectui64vEXT(query.ids[0], GL_QUERY_RESULT, ×tamp);
+ glGetQueryObjecti64vEXT(query.ids[0], GL_QUERY_RESULT, &gpuStart);
}
- if (supportsElapsed) {
- glGetQueryObjectui64vEXT(query.ids[1], GL_QUERY_RESULT, &duration);
- }
+ glGetQueryObjecti64vEXT(query.ids[1], GL_QUERY_RESULT, &gpuDuration);
+ }
+
+ if (retrace::profilingCpuTimes) {
+ cpuDuration = query.cpuEnd - query.cpuStart;
}
- if (retrace::profilingPixelsDrawn && supportsOcclusion) {
- glGetQueryObjectui64vEXT(query.ids[2], GL_QUERY_RESULT, &samples);
+ 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, query.program, samples, timestamp, duration, query.start, query.duration);
+ retrace::profiler.addCall(query.call, query.sig->name, query.program, samples, gpuStart, gpuDuration, query.cpuStart, cpuDuration);
}
void
}
void
-beginProfile(trace::Call &call) {
+beginProfile(trace::Call &call, bool isDraw) {
if (firstFrame) {
frame_start();
}
query.sig = call.sig;
query.program = glretrace::currentContext ? glretrace::currentContext->activeProgram : 0;
- glGenQueries(3, query.ids);
+ if (isDraw) {
+ glGenQueries(3, query.ids);
- if (retrace::profilingGpuTimes) {
- if (supportsTimestamp) {
- glQueryCounter(query.ids[0], GL_TIMESTAMP);
- }
+ if (retrace::profilingGpuTimes) {
+ if (supportsTimestamp) {
+ glQueryCounter(query.ids[0], GL_TIMESTAMP);
+ }
- if (supportsElapsed) {
glBeginQuery(GL_TIME_ELAPSED, query.ids[1]);
}
- }
- if (retrace::profilingPixelsDrawn && supportsOcclusion) {
- glBeginQuery(GL_SAMPLES_PASSED, query.ids[2]);
+ if (retrace::profilingPixelsDrawn) {
+ glBeginQuery(GL_SAMPLES_PASSED, query.ids[2]);
+ }
}
if (retrace::profilingCpuTimes) {
- query.start = os::getTime();
+ query.cpuStart = getCpuTimestamp();
}
callQueries.push_back(query);
}
void
-endProfile(trace::Call &call) {
+endProfile(trace::Call &call, bool isDraw) {
if (retrace::profilingCpuTimes) {
CallQuery& query = callQueries.back();
- query.duration = os::getTime() - query.start;
- }
+ query.cpuEnd = getCpuTimestamp();
- if (retrace::profilingGpuTimes && supportsElapsed) {
- glEndQuery(GL_TIME_ELAPSED);
+ 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 (retrace::profilingPixelsDrawn && supportsOcclusion) {
- glEndQuery(GL_SAMPLES_PASSED);
+ if (isDraw) {
+ if (retrace::profilingGpuTimes) {
+ glEndQuery(GL_TIME_ELAPSED);
+ }
+
+ if (retrace::profilingPixelsDrawn) {
+ glEndQuery(GL_SAMPLES_PASSED);
+ }
}
}
void
initContext() {
- /* Check for extension support */
- const char* extensions = (const char*)glGetString(GL_EXTENSIONS);
+ GLint64 gpuTime = 0, cpuTime = 0;
+ const char* extensions;
+
+ extensions = (const char*)glGetString(GL_EXTENSIONS);
- 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);
+ /* 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) {
exit(-1);
}
- if (retrace::debug) {
- bool supportsDebugOutput = glws::checkExtension("GL_ARB_debug_output", extensions);
+ if (retrace::debug && supportsDebugOutput) {
+ glDebugMessageCallbackARB(&debugOutputCallback, currentContext);
- if (supportsDebugOutput) {
- glDebugMessageCallbackARB(&debugOutputCallback, currentContext);
+ if (DEBUG_OUTPUT_SYNCHRONOUS) {
+ glEnable(GL_DEBUG_OUTPUT_SYNCHRONOUS_ARB);
+ }
+ }
- 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);
}
}
firstFrame = false;
if (retrace::profiling) {
- retrace::profiler.addFrameStart(retrace::frameNo, getGpuTimestamp(), getCpuTimestamp());
+ int64_t gpuStart = getGpuTimestamp();
+ int64_t cpuStart = getCpuTimestamp();
+ retrace::profiler.addFrameStart(retrace::frameNo, gpuStart, cpuStart);
}
}
flushQueries();
/* Indicate end of current frame */
- retrace::profiler.addFrameEnd(getGpuTimestamp(), getCpuTimestamp());
+ int64_t gpuEnd = getGpuTimestamp();
+ int64_t cpuEnd = getCpuTimestamp();
+ retrace::profiler.addFrameEnd(gpuEnd, cpuEnd);
}
retrace::frameComplete(call);