static float
draw(GLuint *q, int iters)
{
float start_time, end_time;
glUseProgram(prog);
glUniform1i(iters_loc, iters);
start_time = get_time();
if (test == TIMESTAMP) {
glQueryCounter(q[0], GL_TIMESTAMP);
} else {
glBeginQuery(GL_TIME_ELAPSED, q[0]);
}
piglit_draw_rect(-1, -1, 2, 2);
if (test == TIMESTAMP) {
glQueryCounter(q[1], GL_TIMESTAMP);
} else {
glEndQuery(GL_TIME_ELAPSED);
}
/* This glFinish() is important, since this is used in a
* timing loop.
*/
glFinish();
end_time = get_time();
return end_time - start_time;
}
int main(int /*argc*/, char ** /*argv*/) {
BaseApp app;
ProgramObject program;
auto mainWindow = app.getMainWindow();
std::string prefix = app.getResourceDir() + "Shaders/Examples/e06_ModelLoader/";
PerspectiveCamera cam;
OrbitManipulator manipulator(&cam);
manipulator.setupCallbacks(app);
NodeShared root;
GLuint query[2];
app.addInitCallback([&]() {
auto vs = compileShader(GL_VERTEX_SHADER, Loader::text(prefix + "phong.vert"));
auto fs = compileShader(GL_FRAGMENT_SHADER, Loader::text(prefix + "phong.frag"));
program = createProgram(vs, fs);
root = Loader::scene(app.getResourceDir() + "Models/sponza/sponza.fbx");
glCreateQueries(GL_TIMESTAMP, 2, query);
SDL_GL_SetSwapInterval(0);
});
app.addResizeCallback([&](int w, int h) {
glViewport(0, 0, w, h);
cam.setAspect(float(w) / float(h));
});
app.addDrawCallback([&]() {
glQueryCounter(query[0], GL_TIMESTAMP);
glClearColor(0.2, 0.2, 0.2, 1);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glEnable(GL_DEPTH_TEST);
//bunny
program.use();
program.setMatrix4fv("p", value_ptr(cam.getProjection()));
program.setMatrix4fv("v", value_ptr(cam.getView()));
drawNode(program, root);
glQueryCounter(query[1], GL_TIMESTAMP);
GLuint64 time1, time2;
glGetQueryObjectui64v(query[0], GL_QUERY_RESULT, &time1);
glGetQueryObjectui64v(query[1], GL_QUERY_RESULT, &time2);
std::string s = "fps: " + std::to_string(1e9 / (time2 - time1)) + " (" + std::to_string(ImGui::GetIO().Framerate) + ")";
label(s, 0, 0, 300, 100);
});
return app.run();
}
void Query::begin()
{
glBeginQuery(target, id);
if (target == GL_TIME_ELAPSED) {
glQueryCounter(id, GL_TIMESTAMP);
}
}
void GLTimerQuery::begin()
{
glQueryCounter(mQueryStartObj, GL_TIMESTAMP);
setActive(true);
mEndIssued = false;
}
void GLTimerQuery::end()
{
glQueryCounter(mQueryEndObj, GL_TIMESTAMP);
mEndIssued = true;
mFinalized = false;
}
const Renderer::ProfileInfo& OpenGLRenderer::popDebugGroup() {
#ifdef RW_GRAPHICS_STATS
if (ogl_ext_KHR_debug) {
glPopDebugGroup();
currentDebugDepth--;
RW_ASSERT(currentDebugDepth >= 0);
ProfileInfo& prof = profileInfo[currentDebugDepth];
glQueryCounter(debugQuery, GL_TIMESTAMP);
GLuint64 current_time;
glGetQueryObjectui64v(debugQuery, GL_QUERY_RESULT, ¤t_time);
prof.duration = current_time - prof.timerStart;
// Add counters to the parent group
if (currentDebugDepth > 0) {
ProfileInfo& p = profileInfo[currentDebugDepth - 1];
p.draws += prof.draws;
p.buffers += prof.buffers;
p.primitives += prof.primitives;
p.textures += prof.textures;
p.uploads += prof.uploads;
}
return prof;
}
#endif
return profileInfo[0];
}
double GpuTimer::elapsed()
{
glQueryCounter(inflightPairs.back().endQuery, GL_TIMESTAMP);
// Query just issued is not going to be available just yet.
// We'll return the oldest available result.
// On startup, there won't be any available.
if (inflightPairs.size() == 1)
return 0.;
// Check if a previously issued query is available. Since startQuery
// is issued before endQuery, we check endQuery only.
QueryPair qp = inflightPairs.front();
GLuint queryAvailable = GL_FALSE;
glGetQueryObjectuiv(qp.endQuery, GL_QUERY_RESULT_AVAILABLE, &queryAvailable);
if (queryAvailable) {
availablePairs.push(qp);
inflightPairs.pop();
GLuint64 startTime, endTime;
glGetQueryObjectui64v(qp.startQuery, GL_QUERY_RESULT, &startTime);
glGetQueryObjectui64v(qp.endQuery, GL_QUERY_RESULT, &endTime);
double duration = double(endTime-startTime) * 0.001 * 0.001; // Milliseconds.
return duration;
}
return 0.;
}
void
beginProfile(trace::Call &call, bool isDraw) {
/* Create call query */
CallQuery query;
query.isDraw = isDraw;
query.call = call.no;
query.sig = call.sig;
query.program = glretrace::currentContext ? glretrace::currentContext->activeProgram : 0;
/* GPU profiling only for draw calls */
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]);
}
}
callQueries.push_back(query);
/* CPU profiling for all calls */
if (retrace::profilingCpuTimes) {
callQueries.back().cpuStart = os::getTime();
}
}
enum piglit_result
piglit_display(void)
{
int test_pass = 1;
GLuint query;
GLint result;
piglit_ortho_projection(piglit_width, piglit_height, GL_FALSE);
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
/* Guaranteed to be random, see: http://xkcd.com/221
*/
query = 3243;
test_pass &= is_query_matches(query, 0, "with un-generated name");
glGenQueries(1, &query);
test_pass &= is_query_matches(query, 0, "after glGenQueries");
glQueryCounter(query, GL_TIMESTAMP);
test_pass &= is_query_matches(query, 1, "after glQueryCounter");
/* Do a little drawing at least */
glColor3ub(0x00, 0xff, 0x00);
piglit_draw_rect(0, 0, piglit_width, piglit_height);
glGetQueryObjectiv(query, GL_QUERY_RESULT, &result);
test_pass &= is_query_matches(query, 1, "after glGetQueryObjectiv");
glDeleteQueries(1, &query);
test_pass &= is_query_matches(query, 0, "after glDeleteQueries");
piglit_present_results();
return test_pass ? PIGLIT_PASS : PIGLIT_FAIL;
}
void GPUTimer::beginQuery( uint32 frameID )
{
if( !glExt::ARB_timer_query ) return;
ASSERT( !_activeQuery );
if( _queryFrame != frameID )
{
if( !updateResults() ) return;
_queryFrame = frameID;
_numQueries = 0;
}
// Create new query pair if necessary
uint32 queryObjs[2];
if( _numQueries++ * 2 == _queryPool.size() )
{
glGenQueries( 2, queryObjs );
_queryPool.push_back( queryObjs[0] );
_queryPool.push_back( queryObjs[1] );
}
else
{
queryObjs[0] = _queryPool[(_numQueries - 1) * 2];
}
_activeQuery = true;
glQueryCounter( queryObjs[0], GL_TIMESTAMP );
}
void Profile::time(std::string name)
{
assert(!CHECKERROR);
if (ready)
{
//create query struct for "name" if there isn't one
QueryMap::iterator q = queries.find(name);
if (q == queries.end())
{
std::pair<QueryMap::iterator, bool> ret;
ret = queries.insert(make_pair(name, Query()));
q = ret.first;
}
//if "name" has already been set after begin(), ignore this call
if (q->second.timeDiff > 0)
return;
//attach this timediff to "name"
q->second.timeDiff = current;
queryOrder.push_back(q->first);
//move to the next query
CHECKERROR;
glQueryCounter(getNextQuery(), GL_TIMESTAMP);
//NOTE: if this causes an error it's probably because you're using AMD
//my guess is AMD won't allow two querys at the same time even of differing types
CHECKERROR;
}
}
// Create a new profile section
void VSProfileLib::createNewSection(std::string &name, pTime w, bool profileGL) {
section s;
#if VSPL_PROFILE == VSPL_PROFILE_CPU_AND_GPU
s.profileGL = profileGL;
#else
s.profileGL = false;
#endif
s.parent = (sCurrLevel > 0 ?
sLevels[sCurrLevel-1].cursor : -1);
s.name = name;
s.calls = 1;
s.totalTime = 0;
s.totalQueryTime = 0;
sLevels[sCurrLevel].cursor++;
if (profileGL) {
pair p;
glGenQueries(2, p.queries);
glQueryCounter(p.queries[0], GL_TIMESTAMP);
s.queriesGL[sBackBuffer].push_back(p);
}
GetTicks(&(s.startTime));
s.wastedTime = s.startTime - w;
sLevels[sCurrLevel].sec.push_back(s);
}
void GPUTimer::endQuery()
{
if( _activeQuery )
{
glQueryCounter( _queryPool[_numQueries * 2 - 1], GL_TIMESTAMP );
_activeQuery = false;
}
}
inline Profiler::Slot Profiler::beginSection( const char* name, GPUInterface* gpuif )
{
GLuint queryFrame = m_numFrames % FRAME_DELAY;
Slot slot = m_frameEntries++;
if (slot >= m_entries.size()){
grow((unsigned int)(m_entries.size() * 2));
if (gpuif){
gpuif->TimerGrow( getRequiredTimers() );
}
}
if (m_entries[slot].name != name ||
m_entries[slot].gpuif != gpuif )
{
m_entries[slot].name = name;
m_entries[slot].gpuif = gpuif;
m_resetDelay = CONFIG_DELAY;
}
int level = m_level++;
m_entries[slot].level = level;
m_entries[slot].splitter = false;
#ifdef SUPPORT_NVTOOLSEXT
{
nvtxEventAttributes_t eventAttrib = {0};
eventAttrib.version = NVTX_VERSION;
eventAttrib.size = NVTX_EVENT_ATTRIB_STRUCT_SIZE;
eventAttrib.colorType = NVTX_COLOR_ARGB;
unsigned char color[4];
color[0] = 255;
color[1] = 0;
color[2] = slot % 2 ? 127 : 255;
color[3] = 255;
color[2] -= level * 16;
color[3] -= level * 16;
eventAttrib.color = *(uint32_t*)(color);
eventAttrib.messageType = NVTX_MESSAGE_TYPE_ASCII;
eventAttrib.message.ascii = name;
nvtxRangePushEx(&eventAttrib);
}
#endif
if (gpuif){
gpuif->TimerSetup( getTimerIdx(slot,queryFrame,true) );
}
else{
glQueryCounter(m_entries[slot].queries[queryFrame],GL_TIMESTAMP);
}
m_entries[slot].deltas[queryFrame] = -getMicroSeconds();
return slot;
}
void GPUTimer::endQuery()
{
#if !defined (PLATFORM_ES2)
if( _activeQuery )
{
glQueryCounter( _queryPool[_numQueries * 2 - 1], GL_TIMESTAMP );
_activeQuery = false;
}
#endif
}
void scene_buildings::gpu_draw(base::frame_context * const ctx){
base::hptimer timer;
timer.start();
if (base::cfg().no_cull){
glDisable(GL_CULL_FACE);
}
glUseProgram(_prg);
// this is only needed on AMD cards due to driver bug wich needs
// to have attr0 array anabled
//base::set_attr0_vbo_amd_wa();
glQueryCounter(ctx->_time_queries[0], GL_TIMESTAMP);
glUniformMatrix4fv(_prg_mvp, 1, GL_FALSE, glm::value_ptr(ctx->_mvp));
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, _indices_vbo);
glUniform1i(_prg_tb_blocks, 0);
glActiveTexture(GL_TEXTURE0);
// bind scene texture buffers
for (uint32 i = 0; i < _tiles.size(); i++){
int x = _tiles[i]._tile_pos.x;
int y = _tiles[i]._tile_pos.y;
glUniform2i(_prg_tile_offset, x,y);
glUniform1ui(_prg_total_count, _tiles[i]._blocks_count);
glBindTexture(GL_TEXTURE_BUFFER, _tiles[i]._blocks_tb);
glDrawElementsInstanced(GL_TRIANGLES,
10 * 3 * base::cfg().blocks_per_idc,
GL_UNSIGNED_INT,
0,
(_tiles[i]._blocks_count + base::cfg().blocks_per_idc - 1) / base::cfg().blocks_per_idc);
}
ctx->_cpu_render_time = timer.elapsed_time();
glQueryCounter(ctx->_time_queries[1], GL_TIMESTAMP);
}
PIGLIT_GL_TEST_CONFIG_END
static GLint64
get_gpu_time_via_query(GLuint q)
{
GLint64 time;
glQueryCounter(q, GL_TIMESTAMP);
glGetQueryObjecti64v(q, GL_QUERY_RESULT, &time);
return time;
}
void MetricBackend_opengl::beginQuery(QueryBoundary boundary) {
// GPU related
if (glQueriesNeeded[boundary]) {
std::array<GLuint, QUERY_LIST_END> query;
glGenQueries(QUERY_LIST_END, query.data());
if (metrics[METRIC_GPU_START].profiled[boundary] ||
(metrics[METRIC_GPU_DURATION].profiled[boundary] && supportsTimestamp))
{
glQueryCounter(query[QUERY_GPU_START], GL_TIMESTAMP);
}
if (metrics[METRIC_GPU_DURATION].profiled[boundary] && !supportsTimestamp) {
glBeginQuery(GL_TIME_ELAPSED, query[QUERY_GPU_DURATION]);
}
if (metrics[METRIC_GPU_PIXELS].profiled[boundary]) {
glBeginQuery(GL_SAMPLES_PASSED, query[QUERY_OCCLUSION]);
}
queries[boundary].push(std::move(query));
}
// CPU related
if (metrics[METRIC_CPU_START].profiled[boundary] ||
metrics[METRIC_CPU_DURATION].profiled[boundary])
{
cpuStart[boundary] = getCurrentTime();
if (metrics[METRIC_CPU_START].profiled[boundary]) {
int64_t time = cpuStart[boundary] * cpuTimeScale - baseTime;
data[METRIC_CPU_START][boundary]->addData(boundary, time);
}
}
if (metrics[METRIC_CPU_VSIZE_START].profiled[boundary] ||
metrics[METRIC_CPU_VSIZE_DURATION].profiled[boundary])
{
vsizeStart[boundary] = os::getVsize();
if (metrics[METRIC_CPU_VSIZE_START].profiled[boundary]) {
int64_t time = vsizeStart[boundary];
data[METRIC_CPU_VSIZE_START][boundary]->addData(boundary, time);
}
}
if (metrics[METRIC_CPU_RSS_START].profiled[boundary] ||
metrics[METRIC_CPU_RSS_DURATION].profiled[boundary])
{
rssStart[boundary] = os::getRss();
if (metrics[METRIC_CPU_RSS_START].profiled[boundary]) {
int64_t time = rssStart[boundary];
data[METRIC_CPU_RSS_START][boundary]->addData(boundary, time);
}
}
queryInProgress[boundary] = true;
// DRAWCALL is a CALL
if (boundary == QUERY_BOUNDARY_DRAWCALL) beginQuery(QUERY_BOUNDARY_CALL);
}
bool render()
{
glm::ivec2 WindowSize(this->getWindowSize());
glm::mat4 Projection = glm::perspective(glm::pi<float>() * 0.25f, 4.0f / 3.0f, 0.1f, 100.0f);
glm::mat4 Model = glm::mat4(1.0f);
glm::mat4 MVP = Projection * this->view() * Model;
glQueryCounter(QueryName[query::BEGIN], GL_TIMESTAMP);
glViewport(0, 0, WindowSize.x, WindowSize.y);
glClearBufferfv(GL_COLOR, 0, &glm::vec4(0.0f, 0.0f, 0.0f, 1.0f)[0]);
glUseProgram(ProgramName);
glUniformMatrix4fv(UniformMVP, 1, GL_FALSE, &MVP[0][0]);
glBindVertexArray(VertexArrayName);
glDrawArrays(GL_TRIANGLES, 0, VertexCount);
glQueryCounter(QueryName[query::END], GL_TIMESTAMP);
GLint AvailableBegin = GL_FALSE;
glGetQueryObjectiv(QueryName[query::BEGIN], GL_QUERY_RESULT_AVAILABLE, &AvailableBegin);
GLint AvailableEnd = GL_FALSE;
glGetQueryObjectiv(QueryName[query::END], GL_QUERY_RESULT_AVAILABLE, &AvailableEnd);
// The OpenGL implementations will wait for the query if it's not available
GLint64 TimeBegin = 0, TimeEnd = 0;
glGetQueryObjecti64v(QueryName[query::BEGIN], GL_QUERY_RESULT, &TimeBegin);
glGetQueryObjecti64v(QueryName[query::END], GL_QUERY_RESULT, &TimeEnd);
//glGetInteger64v(GL_TIMESTAMP, &TimeBegin);
//glGetInteger64v(GL_TIMESTAMP, &TimeEnd);
fprintf(stdout, "%d, %d / Time stamp: %f ms \r", AvailableBegin, AvailableEnd, (TimeEnd - TimeBegin) / 1000.f / 1000.f);
return TimeEnd - TimeBegin > 0;
}
void GPU_Profile::create_timer()
{
u32 timer = 0;
#ifdef GLSL4_API
glGenQueries(1, &timer);
glQueryCounter(timer, GL_TIMESTAMP);
#endif
datas.push_back(ProfileInfo(timer, frame, draw));
#ifdef ENABLE_MARKER
dump(true);
#endif
}
void Profile::begin()
{
assert(!CHECKERROR);
if (ready)
{
ready = false;
}
if (queries.size())
{
//if the last query is done, they are all done
GLint available;
glGetQueryObjectiv(queryObjs[current-1], GL_QUERY_RESULT_AVAILABLE, &available);
if (available)
{
//get the results
timeStamps.resize(queryObjs.size());
for (int i = 0; i < (int)queryObjs.size(); ++i)
{
CHECKERROR;
GLuint64 result;
glGetQueryObjectui64vEXT(queryObjs[i], GL_QUERY_RESULT, &result);
timeStamps[i] = result;
CHECKERROR;
}
//append the differences to the samples list
for (QueryMap::iterator it = queries.begin(); it != queries.end(); ++it)
{
if (it->second.timeDiff < 1) continue;
GLuint64 timediff = timeStamps[it->second.timeDiff] - timeStamps[it->second.timeDiff-1];
it->second.times.push_back(timediff);
if (it->second.times.size() > PROFILE_SAMPLES)
it->second.times.pop_front();
it->second.timeDiff = -1;
}
//start the next set of samples
ready = true;
restartQueries();
}
}
else
ready = true;
if (ready)
{
glQueryCounter(getNextQuery(), GL_TIMESTAMP);
}
}
// accumulates the time spent in the section
void VSProfileLib::accumulate() {
section *s;
pTime t,t2;
GetTicks(&t);
s = &(sLevels[sCurrLevel].sec[sLevels[sCurrLevel].cursor]);
if (s->profileGL) {
glQueryCounter(s->queriesGL[sBackBuffer][s->queriesGL[sBackBuffer].size()-1].queries[1], GL_TIMESTAMP);
}
// to measure wasted time when accumulating
GetTicks(&t2);
s->wastedTime += (t2-t);
s->totalTime += (t - s->startTime);
}
// updates a profile section
void VSProfileLib::updateSection(int cur, pTime w) {
section *s;
s = &(sLevels[sCurrLevel].sec[cur]);
s->calls++;
sLevels[sCurrLevel].cursor = cur;
if (s->profileGL) {
pair p;
glGenQueries(2, p.queries);
glQueryCounter(p.queries[0], GL_TIMESTAMP);
s->queriesGL[sBackBuffer].push_back(p);
}
GetTicks(&s->startTime);
s->wastedTime += s->startTime - w;
}
void OpenGLRenderer::pushDebugGroup(const std::string& title) {
#ifdef RW_GRAPHICS_STATS
if (ogl_ext_KHR_debug) {
glPushDebugGroup(GL_DEBUG_SOURCE_APPLICATION, 0, -1, title.c_str());
ProfileInfo& prof = profileInfo[currentDebugDepth];
prof.buffers = prof.draws = prof.textures = prof.uploads =
prof.primitives = 0;
glQueryCounter(debugQuery, GL_TIMESTAMP);
glGetQueryObjectui64v(debugQuery, GL_QUERY_RESULT, &prof.timerStart);
currentDebugDepth++;
RW_ASSERT(currentDebugDepth < MAX_DEBUG_DEPTH);
}
#else
RW_UNUSED(title);
#endif
}
void MetricBackend_opengl::endQuery(QueryBoundary boundary) {
if (queryInProgress[boundary]) {
// CPU related
if (metrics[METRIC_CPU_DURATION].profiled[boundary])
{
cpuEnd[boundary] = getCurrentTime();
int64_t time = (cpuEnd[boundary] - cpuStart[boundary]) * cpuTimeScale;
data[METRIC_CPU_DURATION][boundary]->addData(boundary, time);
}
if (metrics[METRIC_CPU_VSIZE_DURATION].profiled[boundary])
{
vsizeEnd[boundary] = os::getVsize();
int64_t time = vsizeEnd[boundary] - vsizeStart[boundary];
data[METRIC_CPU_VSIZE_DURATION][boundary]->addData(boundary, time);
}
if (metrics[METRIC_CPU_RSS_DURATION].profiled[boundary])
{
rssEnd[boundary] = os::getRss();
int64_t time = rssEnd[boundary] - rssStart[boundary];
data[METRIC_CPU_RSS_DURATION][boundary]->addData(boundary, time);
}
// GPU related
if (glQueriesNeeded[boundary]) {
std::array<GLuint, QUERY_LIST_END> &query = queries[boundary].back();
if (metrics[METRIC_GPU_DURATION].profiled[boundary] && supportsTimestamp) {
// GL_TIME_ELAPSED cannot be used in nested queries
// so prefer this if timestamps are supported
glQueryCounter(query[QUERY_GPU_DURATION], GL_TIMESTAMP);
}
if (metrics[METRIC_GPU_PIXELS].profiled[boundary]) {
glEndQuery(GL_SAMPLES_PASSED);
}
}
queryInProgress[boundary] = false;
}
// DRAWCALL is a CALL
if (boundary == QUERY_BOUNDARY_DRAWCALL) endQuery(QUERY_BOUNDARY_CALL);
// clear queries after each frame
if (boundary == QUERY_BOUNDARY_FRAME && glQueriesNeededAnyBoundary) {
processQueries();
}
}
void
beginProfile(trace::Call &call, bool isDraw) {
glretrace::Context *currentContext = glretrace::getCurrentContext();
/* Create call query */
CallQuery query;
query.isDraw = isDraw;
query.call = call.no;
query.sig = call.sig;
query.program = currentContext ? currentContext->activeProgram : 0;
glGenQueries(NUM_QUERIES, query.ids);
/* GPU profiling only for draw calls */
if (isDraw) {
if (retrace::profilingGpuTimes) {
if (supportsTimestamp) {
glQueryCounter(query.ids[GPU_START], GL_TIMESTAMP);
}
glBeginQuery(GL_TIME_ELAPSED, query.ids[GPU_DURATION]);
}
if (retrace::profilingPixelsDrawn) {
glBeginQuery(GL_SAMPLES_PASSED, query.ids[OCCLUSION]);
}
}
callQueries.push_back(query);
/* CPU profiling for all calls */
if (retrace::profilingCpuTimes) {
CallQuery& query = callQueries.back();
query.cpuStart = getCurrentTime();
}
if (retrace::profilingMemoryUsage) {
CallQuery& query = callQueries.back();
query.vsizeStart = os::getVsize();
query.rssStart = os::getRss();
}
}
static void
getCurrentTimes(int64_t& cpuTime, int64_t& gpuTime) {
GLuint query;
if (retrace::profilingGpuTimes && supportsTimestamp) {
glGenQueries(1, &query);
glQueryCounter(query, GL_TIMESTAMP);
glGetQueryObjecti64vEXT(query, GL_QUERY_RESULT, &gpuTime);
} else {
gpuTime = 0;
}
if (retrace::profilingCpuTimes) {
cpuTime = os::getTime();
} else {
cpuTime = 0;
}
if (retrace::profilingGpuTimes && supportsTimestamp) {
glDeleteQueries(1, &query);
}
}
inline void Profiler::endSection( Slot slot )
{
GLuint queryFrame = m_numFrames % FRAME_DELAY;
m_entries[slot].deltas[queryFrame] += getMicroSeconds();
if (m_entries[slot].gpuif){
m_entries[slot].gpuif->TimerSetup( getTimerIdx(slot,queryFrame,false) );
}
else{
glQueryCounter(m_entries[slot].queries[queryFrame + FRAME_DELAY],GL_TIMESTAMP);
#if NV_TIMER_FLUSH
glFlush();
#endif
}
#ifdef SUPPORT_NVTOOLSEXT
nvtxRangePop();
#endif
m_level--;
}
static HRESULT wined3d_timestamp_query_ops_issue(struct wined3d_query *query, DWORD flags)
{
struct wined3d_device *device = query->device;
const struct wined3d_gl_info *gl_info = &device->adapter->gl_info;
TRACE("query %p, flags %#x.\n", query, flags);
if (gl_info->supported[ARB_TIMER_QUERY])
{
struct wined3d_timestamp_query *tq = query->extendedData;
struct wined3d_context *context;
if (flags & WINED3DISSUE_BEGIN)
{
WARN("Ignoring WINED3DISSUE_BEGIN with a TIMESTAMP query.\n");
}
if (flags & WINED3DISSUE_END)
{
if (tq->context)
context_free_timestamp_query(tq);
context = context_acquire(query->device, NULL);
context_alloc_timestamp_query(context, tq);
GL_EXTCALL(glQueryCounter(tq->id, GL_TIMESTAMP));
checkGLcall("glQueryCounter()");
context_release(context);
}
}
else
{
ERR("Timestamp queries not supported.\n");
}
if (flags & WINED3DISSUE_END)
query->state = QUERY_SIGNALLED;
return WINED3D_OK;
}
void GpuTimer::start()
{
QueryPair qp;
if (availablePairs.empty()) {
glGenQueries(1, &qp.startQuery);
glGenQueries(1, &qp.endQuery);
}
else {
qp = availablePairs.front();
availablePairs.pop(); // Reuse old OpenGL ids, no need for additional glGenQueries.
}
inflightPairs.push(qp);
// Create a fence and insert it into the command stream. This fence will be signaled
// when the GPU has completed all previously issued commands. glWaitSync will defer
// subsequently issued commands (without blocking CPU) until the fence signals.
// This prevents overlapped execution of OpenGL calls and should make timing measurements
// more precise. For more, see e.g. OpenGL Insights.
GLsync sync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
glWaitSync(sync, 0, GL_TIMEOUT_IGNORED);
glDeleteSync(sync); // Flag for deletion.
glQueryCounter(qp.startQuery, GL_TIMESTAMP);
}