portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Memory manager debug Felix case

Browse Source
This commit is contained in:
Peter Boyle 2022-12-01 00:25:04 -05:00
parent 4a359fa9e9
commit f8c70545a0
5 changed files with 196 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
Grid/perfmon/PerfCount.cc
View File

@ -27,10 +27,13 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
/* END LEGAL */
#include <Grid/GridCore.h>
#include <Grid/perfmon/PerfCount.h>
#include <Grid/perfmon/Timer.h>
#include <Grid/perfmon/PerfCount.h>
NAMESPACE_BEGIN(Grid);
GridTimePoint theProgramStart = GridClock::now();
#define CacheControl(L,O,R) ((PERF_COUNT_HW_CACHE_##L)|(PERF_COUNT_HW_CACHE_OP_##O<<8)| (PERF_COUNT_HW_CACHE_RESULT_##R<<16))
#define RawConfig(A,B) (A<<8|B)
const PerformanceCounter::PerformanceCounterConfig PerformanceCounter::PerformanceCounterConfigs [] = {

22
Grid/perfmon/Timer.h
View File

@ -35,17 +35,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
NAMESPACE_BEGIN(Grid)
// Dress the output; use std::chrono
// C++11 time facilities better?
inline double usecond(void) {
struct timeval tv;
#ifdef TIMERS_ON
gettimeofday(&tv,NULL);
#endif
return 1.0*tv.tv_usec + 1.0e6*tv.tv_sec;
}
typedef std::chrono::system_clock GridClock;
//typedef std::chrono::system_clock GridClock;
typedef std::chrono::high_resolution_clock GridClock;
typedef std::chrono::time_point<GridClock> GridTimePoint;
typedef std::chrono::seconds GridSecs;
@ -53,6 +44,15 @@ typedef std::chrono::milliseconds GridMillisecs;
typedef std::chrono::microseconds GridUsecs;
typedef std::chrono::microseconds GridTime;
extern GridTimePoint theProgramStart;
// Dress the output; use std::chrono
// C++11 time facilities better?
inline double usecond(void) {
auto usecs = std::chrono::duration_cast<GridUsecs>(GridClock::now()-theProgramStart);
return 1.0*usecs.count();
}
inline std::ostream& operator<< (std::ostream & stream, const GridSecs & time)
{
stream << time.count()<<" s";

70
Grid/perfmon/Tracing.h Normal file
View File

@ -0,0 +1,70 @@
#pragma once
NAMESPACE_BEGIN(Grid);
#ifdef GRID_TRACING_NVTX
#include <nvToolsExt.h>
class GridTracer {
public:
GridTracer(const char* name) {
nvtxRangePushA(name);
}
~GridTracer() {
nvtxRangePop();
}
};
inline void tracePush(const char *name) { nvtxRangePushA(name); }
inline void tracePop(const char *name) { nvtxRangePop(); }
inline int traceStart(const char *name) { }
inline void traceStop(int ID) { }
#endif
#ifdef GRID_TRACING_ROCTX
#include <roctracer/roctx.h>
class GridTracer {
public:
GridTracer(const char* name) {
roctxRangePushA(name);
std::cout << "roctxRangePush "<<name<<std::endl;
}
~GridTracer() {
roctxRangePop();
std::cout << "roctxRangePop "<<std::endl;
}
};
inline void tracePush(const char *name) { roctxRangePushA(name); }
inline void tracePop(const char *name) { roctxRangePop(); }
inline int traceStart(const char *name) { roctxRangeStart(name); }
inline void traceStop(int ID) { roctxRangeStop(ID); }
#endif
#ifdef GRID_TRACING_TIMER
class GridTracer {
public:
const char *name;
double elapsed;
GridTracer(const char* _name) {
name = _name;
elapsed=-usecond();
}
~GridTracer() {
elapsed+=usecond();
std::cout << GridLogTracing << name << " took " <<elapsed<< " us" <<std::endl;
}
};
inline void tracePush(const char *name) { }
inline void tracePop(const char *name) { }
inline int traceStart(const char *name) { return 0; }
inline void traceStop(int ID) { }
#endif
#ifdef GRID_TRACING_NONE
#define GRID_TRACE(name)
inline void tracePush(const char *name) { }
inline void tracePop(const char *name) { }
inline int traceStart(const char *name) { return 0; }
inline void traceStop(int ID) { }
#else
#define GRID_TRACE(name) GridTracer uniq_name_using_macros##__COUNTER__(name);
#endif
NAMESPACE_END(Grid);

2
systems/mac-arm/config-command-mpi
View File

@ -1 +1 @@
CXX=mpicxx-openmpi-mp CXXFLAGS=-I/opt/local/include/ LDFLAGS=-L/opt/local/lib/ ../../configure --enable-simd=GEN --enable-debug --enable-comms=mpi
CXX=mpicxx-openmpi-mp CXXFLAGS=-I/opt/local/include/ LDFLAGS=-L/opt/local/lib/ ../../configure --enable-simd=GEN --enable-debug --enable-comms=mpi --enable-unified=no

110
tests/core/Test_memory_manager.cc Normal file
View File

@ -0,0 +1,110 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_memory_manager.cc
Copyright (C) 2022
Author: Peter Boyle <pboyle@bnl.gov>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/Grid.h>
using namespace std;
using namespace Grid;
void MemoryTest(GridCartesian * FGrid,int N);
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
int N=100;
for(int i=0;i<N;i++){
std::cout << "============================"<<std::endl;
std::cout << "Epoch "<<i<<"/"<<N<<std::endl;
std::cout << "============================"<<std::endl;
MemoryTest(UGrid,256);
MemoryManager::Print();
AUDIT();
}
Grid_finalize();
}
void MemoryTest(GridCartesian * FGrid, int N)
{
LatticeComplexD zero(FGrid); zero=Zero();
std::vector<LatticeComplexD> A(N,zero);//FGrid);
std::vector<ComplexD> B(N,ComplexD(0.0)); // Update sequentially on host
for(int v=0;v<N;v++) A[v] = Zero();
uint64_t counter = 0;
for(int epoch = 0;epoch<10000;epoch++){
int v = random() %N; // Which vec
int w = random() %2; // Write or read
int e = random() %3; // expression or for loop
int dev= random() %2; // On device?
// int e=1;
ComplexD zc = counter++;
if ( w ) {
B[v] = B[v] + zc;
if ( e == 0 ) {
A[v] = A[v] + zc - A[v] + A[v];
} else {
if ( dev ) {
autoView(A_v,A[v],AcceleratorWrite);
accelerator_for(ss,FGrid->oSites(),1,{
A_v[ss] = A_v[ss] + zc;
});
} else {
autoView(A_v,A[v],CpuWrite);
thread_for(ss,FGrid->oSites(),{
A_v[ss] = A_v[ss] + zc;
});
}
}
} else {
if ( e == 0 ) {
A[v] = A[v] + A[v] - A[v];
} else {
if ( dev ) {
autoView(A_v,A[v],AcceleratorRead);
accelerator_for(ss,FGrid->oSites(),1,{
assert(B[v]==A_v[ss]()()().getlane(0));
});
// std::cout << "["<<v<<"] checked on GPU"<<B[v]<<std::endl;
} else {
autoView(A_v,A[v],CpuRead);
thread_for(ss,FGrid->oSites(),{
assert(B[v]==A_v[ss]()()().getlane(0));
});
// std::cout << "["<<v<<"] checked on CPU"<<B[v]<<std::endl;
}
}
}
}
}