Skip to content

Commit

Permalink
[prof] in CODEGEN, fix clang format for timermap.h, check_sa.cpp, tim…
Browse files Browse the repository at this point in the history 
…er.h, counters.cc
  • Loading branch information
@valassi
valassi committed Aug 19, 2024
1 parent 7af63a2 commit 8335edd
Show file tree
Hide file tree
Showing 4 changed files with 61 additions and 57 deletions.
4 changes: 2 additions & 2 deletions ...X/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/check_sa.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -652,7 +652,7 @@ main( int argc, char** argv )

// *** STOP THE NEW OLD-STYLE TIMER FOR MATRIX ELEMENTS (WAVEFUNCTIONS) ***
wv3acount += timermap.stop(); // calc only
wavecount += wv3acount; // calc plus copy
wavecount += wv3acount; // calc plus copy

#ifdef MGONGPUCPP_GPUIMPL
if( !bridge )
Expand Down Expand Up @@ -736,7 +736,7 @@ main( int argc, char** argv )
wavetimes[iiter] = wavecounts[iiter] * secPerCount;
wv3atimes[iiter] = wv3acounts[iiter] * secPerCount;
}

// === STEP 8 ANALYSIS
// --- 8a Analysis: compute stats after the loop
const std::string statKey = "8a CompStat";
Expand Down
32 changes: 21 additions & 11 deletions ...X/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/counters.cc
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,11 +59,13 @@ extern "C"
if( getenv( "CUDACPP_RUNTIME_USECHRONOTIMERS" ) ) usechronotimers = true;
for( int icounter = 0; icounter < NCOUNTERSMAX + 3; icounter++ )
{
array_tags[icounter] = ""; // ensure that this is initialized to ""
array_tags[icounter] = ""; // ensure that this is initialized to ""
array_istesttimer[icounter] = false; // ensure that this is initialized to false
}
if( usechronotimers ) program_chronotimer.start();
else program_rdtsctimer.start();
if( usechronotimers )
program_chronotimer.start();
else
program_rdtsctimer.start();
return;
}

Expand Down Expand Up @@ -112,8 +114,10 @@ extern "C"
throw std::runtime_error( sstr.str() );
}
array_counters[icounter] += *pnevt;
if( usechronotimers ) array_chronotimers[icounter].start();
else array_rdtsctimers[icounter].start();
if( usechronotimers )
array_chronotimers[icounter].start();
else
array_rdtsctimers[icounter].start();
return;
}

Expand All @@ -129,20 +133,26 @@ extern "C"
sstr << "ERROR! counter #" << icounter << " does not exist";
throw std::runtime_error( sstr.str() );
}
if( usechronotimers ) array_chronotimers[icounter].stop();
else array_rdtsctimers[icounter].stop();
if( usechronotimers )
array_chronotimers[icounter].stop();
else
array_rdtsctimers[icounter].stop();
return;
}

void counters_finalise_()
{
using namespace counters;
// Dump program counters
if( usechronotimers ) program_chronotimer.stop();
else program_rdtsctimer.stop();
if( usechronotimers )
program_chronotimer.stop();
else
program_rdtsctimer.stop();
float program_totaltime = ( usechronotimers ? program_chronotimer.getTotalDurationSeconds() : program_rdtsctimer.getTotalDurationSeconds() );
if( usechronotimers ) printf( " [COUNTERS] *** USING STD::CHRONO TIMERS ***\n" );
else printf( " [COUNTERS] *** USING RDTSC-BASED TIMERS ***\n" );
if( usechronotimers )
printf( " [COUNTERS] *** USING STD::CHRONO TIMERS ***\n" );
else
printf( " [COUNTERS] *** USING RDTSC-BASED TIMERS ***\n" );
printf( " [COUNTERS] PROGRAM TOTAL : %9.4fs\n", program_totaltime );
if( disablecalltimers ) return;
// Extract time duration from all timers
Expand Down
55 changes: 21 additions & 34 deletions epochX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/timer.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,13 +21,13 @@ namespace mgOnGpu
{

// ---------------------------------------------------------------------------

// ChronoTimer: default ("old") timers based on std::chrono clocks
// With respect to the original Timer class, this uses a new implementation with nanosecond counts
// With respect to the original Timer class, this also uses a new API with explicit start/stop
// Template argument T can be any of high_resolution_clock, steady_clock, system_clock
// See https://www.modernescpp.com/index.php/the-three-clocks
// See https://codereview.stackexchange.com/questions/196245/extremely-simple-timer-class-in-c
// See https://codereview.stackexchange.com/questions/196245/extremely-simple-timer-class-in-c
template<typename T>
class ChronoTimer
{
Expand All @@ -50,8 +50,7 @@ namespace mgOnGpu
};

template<typename T>
inline
ChronoTimer<T>::ChronoTimer()
inline ChronoTimer<T>::ChronoTimer()
: m_totalDuration()
, m_started( false )
, m_startTime()
Expand All @@ -62,8 +61,7 @@ namespace mgOnGpu
}

template<typename T>
inline
void
inline void
ChronoTimer<T>::start()
{
assert( !m_started );
Expand All @@ -72,8 +70,7 @@ namespace mgOnGpu
}

template<typename T>
inline
void
inline void
ChronoTimer<T>::stop()
{
assert( m_started );
Expand All @@ -82,32 +79,29 @@ namespace mgOnGpu
}

template<typename T>
inline
uint64_t
inline uint64_t
ChronoTimer<T>::getCountsSinceStart() const
{
return getDurationSinceStart().count();
}

template<typename T>
inline
typename ChronoTimer<T>::DURATION
ChronoTimer<T>::getDurationSinceStart() const
typename ChronoTimer<T>::DURATION
ChronoTimer<T>::getDurationSinceStart() const
{
return T::now() - m_startTime;
}

template<typename T>
inline
float
inline float
ChronoTimer<T>::secondsPerCount() const
{
return (float)RATIO::num / RATIO::den;
}

template<typename T>
inline
float
inline float
ChronoTimer<T>::getTotalDurationSeconds()
{
assert( !m_started );
Expand All @@ -116,7 +110,7 @@ namespace mgOnGpu
}

// ---------------------------------------------------------------------------

// RdtscTimer: faster ("new") *EXPERIMENTAL* timers based on rdtsc
// The rdtsc() call is derived from the TSCNS class (https://github.com/MengRao/tscns)
// The conversion of rdtsc counts to seconds is calibrated on the average frequency during the timer lifetime
Expand All @@ -143,8 +137,7 @@ namespace mgOnGpu
uint64_t m_ctorCount;
};

inline
uint64_t
inline uint64_t
RdtscTimer::rdtsc()
{
#if defined( __x86_64__ )
Expand All @@ -154,8 +147,7 @@ namespace mgOnGpu
#endif
}

inline
RdtscTimer::RdtscTimer()
inline RdtscTimer::RdtscTimer()
: m_totalDuration( 0 )
, m_started( false )
, m_startCount( 0 )
Expand All @@ -166,33 +158,29 @@ namespace mgOnGpu
m_ctorCount = rdtsc();
}

inline
void
inline void
RdtscTimer::start()
{
assert( !m_started );
m_started = true;
m_startCount = rdtsc();
}

inline
void
inline void
RdtscTimer::stop()
{
assert( m_started );
m_started = false;
m_totalDuration += getCountsSinceStart();
}

inline
uint64_t
inline uint64_t
RdtscTimer::getCountsSinceStart() const
{
return rdtsc() - m_startCount;
}

inline
float
inline float
RdtscTimer::secondsPerCount()
{
m_ctorTimer.stop();
Expand All @@ -201,8 +189,7 @@ namespace mgOnGpu
return secPerCount;
}

inline
float
inline float
RdtscTimer::getTotalDurationSeconds()
{
assert( !m_started );
Expand Down
27 changes: 17 additions & 10 deletions ...hX/cudacpp/CODEGEN/PLUGIN/CUDACPP_SA_OUTPUT/madgraph/iolibs/template_files/gpu/timermap.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -52,8 +52,10 @@ namespace mgOnGpu
// Switch to a new partition
if( !m_started )
{
if( m_useChronoTimers ) m_chronoTimer.start();
else m_rdtscTimer.start();
if( m_useChronoTimers )
m_chronoTimer.start();
else
m_rdtscTimer.start();
m_started = true;
}
m_active = key;
Expand All @@ -75,11 +77,15 @@ namespace mgOnGpu
uint64_t last = 0;
if( m_active != "" )
{
if( m_useChronoTimers ) last = m_chronoTimer.getCountsSinceStart();
else last = m_rdtscTimer.getCountsSinceStart();
if( m_useChronoTimers )
last = m_chronoTimer.getCountsSinceStart();
else
last = m_rdtscTimer.getCountsSinceStart();
m_partitionTotalCounts[m_active] += last;
if( m_useChronoTimers ) m_chronoTimer.stop();
else m_rdtscTimer.stop();
if( m_useChronoTimers )
m_chronoTimer.stop();
else
m_rdtscTimer.stop();
m_started = false;
}
m_active = "";
Expand All @@ -92,9 +98,11 @@ namespace mgOnGpu
// Return timer calibration (at this point in time for rdtsc, constant in time for chrono)
float secondsPerCount()
{
if( m_useChronoTimers ) return m_chronoTimer.secondsPerCount();
else return m_rdtscTimer.secondsPerCount();
}
if( m_useChronoTimers )
return m_chronoTimer.secondsPerCount();
else
return m_rdtscTimer.secondsPerCount();
}

// Dump the overall results
void dump( std::ostream& ostr = std::cout, bool json = false )
Expand Down Expand Up @@ -189,7 +197,6 @@ namespace mgOnGpu
std::map<std::string, uint32_t> m_partitionIds;
bool m_useChronoTimers;
bool m_started; // when the timer is stopped, it must be explicitly restarted

};

}
Expand Down

0 comments on commit 8335edd

Please sign in to comment.