Skip to content

Commit 5cf687e

Browse files
JeanLucPonsJeanLucPons
committed
Updated test
1 parent 31eb0fe commit 5cf687e

File tree

1 file changed

+141
-26
lines changed

1 file changed

+141
-26
lines changed

‎atgpu/main.cpp

+141-26
Original file line numberDiff line numberDiff line change
@@ -39,6 +39,22 @@ AT_FLOAT *createGrid(AT_FLOAT x1,AT_FLOAT y1,AT_FLOAT x2,AT_FLOAT y2,uint32_t nb
3939

4040
}
4141

42+
AT_FLOAT *createArc(AT_FLOAT radius,AT_FLOAT startAngle,AT_FLOAT endAngle,uint32_t nbParticles) {
43+
44+
uint32_t rinSize = nbParticles * 6 * sizeof(AT_FLOAT);
45+
AT_FLOAT* rin = (AT_FLOAT*)malloc(rinSize);
46+
memset(rin,0,rinSize);
47+
48+
for(int i=0;i<nbParticles;i++) {
49+
AT_FLOAT angle = startAngle + (endAngle-startAngle)*(AT_FLOAT)i/(AT_FLOAT)(nbParticles-1);
50+
RINPTR(i)[0] = radius*cos(angle);
51+
RINPTR(i)[2] = radius*sin(angle);
52+
}
53+
54+
return rin;
55+
56+
}
57+
4258
void printGPUInfo() {
4359

4460
try {
@@ -51,43 +67,130 @@ void printGPUInfo() {
5167

5268
}
5369

54-
int main(int argc,char **arv) {
70+
Lattice *loadLattice(string latticeName,SymplecticIntegrator& integrator,int gpu,double stepSize=0.0) {
5571

56-
printGPUInfo();
57-
58-
SymplecticIntegrator integrator(4);
5972
CppInterface *dI = new CppInterface();
6073
AbstractInterface::setHandler(dI);
6174
vector<CppObject> elements;
62-
int DEVID = 1;
6375

76+
REPRLoader *loader = new REPRLoader(latticeName);
77+
loader->parseREPR(elements);
78+
if( stepSize!=0.0 ) {
79+
// Adjust NumIntSteps to be as close as possible to stepSize
80+
for(auto & element : elements) {
81+
try {
82+
// For element that has NumIntSteps attribute
83+
string &stepStr = element.getField("NumIntSteps");
84+
string &lengthStr = element.getField("Length");
85+
double length = stod(lengthStr);
86+
int nstep = (int)((length / stepSize)+0.5);
87+
if(nstep<1) nstep = 1;
88+
element.addField("NumIntSteps", to_string(nstep));
89+
} catch (string&) {
90+
}
91+
}
92+
}
93+
delete loader;
6494

65-
try {
66-
//REPRLoader *loader = new REPRLoader("Z:/tmp/pons/at/test/lattice/betamodel_radon.repr");
67-
REPRLoader *loader = new REPRLoader("/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
68-
//REPRLoader *loader = new REPRLoader("/segfs/tmp/pons/lattice/simple_ebs.repr");
69-
loader->parseREPR(elements);
70-
} catch (string& errStr) {
71-
cout << "Parse failed: " << errStr << endl;
72-
exit(0);
95+
Lattice *l = new Lattice(0, integrator, 6e9, gpu);
96+
double t0 = AbstractGPU::get_ticks();
97+
for (auto &element: elements) {
98+
dI->setObject(&element);
99+
l->addElement();
73100
}
101+
l->generateGPUKernel();
102+
double t1 = AbstractGPU::get_ticks();
103+
cout << "Ring build: " << (t1 - t0) * 1000.0 << "ms" << endl;
104+
105+
t0 = AbstractGPU::get_ticks();
106+
l->fillGPUMemory();
107+
t1 = AbstractGPU::get_ticks();
108+
cout << "GPU lattice loading: " << (t1 - t0) * 1000.0 << "ms" << endl;
109+
110+
return l;
111+
112+
}
113+
114+
void integratorTest(int gpu,string latticeName) {
115+
116+
SymplecticIntegrator integrator(4);
117+
118+
npy::npy_data<double> refPoints = npy::read_npy<double>("/segfs/tmp/pons/at/test/data/ref_arc_yfr_10000.npy");
74119

75120
try {
76121

77-
Lattice *l = new Lattice(0, integrator, 6e9, DEVID);
78-
double t0 = AbstractGPU::get_ticks();
79-
for (auto &element: elements) {
80-
dI->setObject(&element);
81-
l->addElement();
122+
for(int integ=1;integ<7;integ++) {
123+
124+
integrator.setType(integ);
125+
Lattice *l = loadLattice(latticeName,integrator,gpu,0.01);
126+
127+
uint32_t nbTurn = 1;
128+
uint32_t nbPart = 256;
129+
uint32_t refs[] = {l->getNbElement()};
130+
uint32_t nbRef = sizeof(refs) / sizeof(uint32_t);
131+
uint64_t routSize = nbTurn * nbPart * nbRef * 6 * sizeof(AT_FLOAT);
132+
AT_FLOAT *rout = (AT_FLOAT *) malloc(routSize);
133+
134+
// Choose an arc close to 1mm where unexpected tune drift is observed when step size in too small (EBS lattice)
135+
AT_FLOAT *rin = createArc(0.001,M_PI/2.0,-M_PI/2.0,nbPart);
136+
137+
l->run(nbTurn, nbPart, rin, rout, nbRef, refs, 0, nullptr, nullptr, nullptr, nullptr);
138+
139+
double err = 0;
140+
double max = 0;
141+
int pMax;
142+
int cMax;
143+
for(int p=0;p<256;p++) {
144+
AT_FLOAT *P = ROUTPTR(p, 0, nbTurn - 1);
145+
err += SQR(P[0] - refPoints.data[p*6+0]);
146+
err += SQR(P[1] - refPoints.data[p*6+1]);
147+
err += SQR(P[2] - refPoints.data[p*6+2]);
148+
err += SQR(P[3] - refPoints.data[p*6+3]);
149+
err += SQR(P[4] - refPoints.data[p*6+4]);
150+
err += SQR(P[5] - refPoints.data[p*6+5]);
151+
if( abs(P[0] - refPoints.data[p*6+0])>max ) {max = abs(P[0] - refPoints.data[p*6+0]);pMax=p;cMax=0;}
152+
if( abs(P[1] - refPoints.data[p*6+1])>max ) {max = abs(P[1] - refPoints.data[p*6+1]);pMax=p;cMax=1;}
153+
if( abs(P[2] - refPoints.data[p*6+2])>max ) {max = abs(P[2] - refPoints.data[p*6+2]);pMax=p;cMax=2;}
154+
if( abs(P[3] - refPoints.data[p*6+3])>max ) {max = abs(P[3] - refPoints.data[p*6+3]);pMax=p;cMax=3;}
155+
if( abs(P[4] - refPoints.data[p*6+4])>max ) {max = abs(P[4] - refPoints.data[p*6+4]);pMax=p;cMax=4;}
156+
if( abs(P[5] - refPoints.data[p*6+5])>max ) {max = abs(P[5] - refPoints.data[p*6+5]);pMax=p;cMax=5;}
157+
}
158+
err = sqrt(err) / (6.0*256.0);
159+
AT_FLOAT *P = ROUTPTR(pMax, 0, nbTurn - 1);
160+
cout << "[" << integ << "]" << err << " max=" << max << " @" << P[cMax] << " " << abs(max/P[cMax]) << endl;
161+
162+
/*
163+
if( integ==4 ) {
164+
// Save ref
165+
npy::npy_data_ptr<double> d;
166+
d.data_ptr = (double *) rout;
167+
d.shape = {6, nbPart, nbRef, nbTurn};
168+
d.fortran_order = true;
169+
npy::write_npy("/segfs/tmp/pons/at/test/data/ref_arc_yfr_10000.npy", d);
170+
}
171+
*/
172+
173+
free(rout);
174+
free(rin);
175+
delete l;
176+
82177
}
83-
l->generateGPUKernel();
84-
double t1 = AbstractGPU::get_ticks();
85-
cout << "Ring build: " << (t1 - t0) * 1000.0 << "ms" << endl;
86178

87-
t0 = AbstractGPU::get_ticks();
88-
l->fillGPUMemory();
89-
t1 = AbstractGPU::get_ticks();
90-
cout << "GPU lattice loading: " << (t1 - t0) * 1000.0 << "ms" << endl;
179+
} catch (string& errStr) {
180+
string err = "Fail: " + errStr;
181+
cout << "Error: " << err << endl;
182+
}
183+
184+
}
185+
186+
void performanceTest(int gpu,string latticeName) {
187+
188+
double t0,t1;
189+
SymplecticIntegrator integrator(4);
190+
191+
try {
192+
193+
Lattice *l = loadLattice(latticeName,integrator,gpu);
91194

92195
string gpuName = l->getGPUContext()->name();
93196
cout << "Running test on " << gpuName << " (" << l->getGPUContext()->coreNumber() << " units)"
@@ -108,6 +211,7 @@ int main(int argc,char **arv) {
108211
uint32_t nbRef = sizeof(refs) / sizeof(uint32_t);
109212
uint32_t starts[] = {0,100,200,300,400,500,600,700};
110213
uint32_t nbStride = sizeof(starts) / sizeof(uint32_t);
214+
//uint32_t nbStride = 0;
111215

112216
uint64_t routSize = nbTurn * nbPart * nbRef * 6 * sizeof(AT_FLOAT);
113217
AT_FLOAT *rout = (AT_FLOAT *) malloc(routSize);
@@ -168,10 +272,21 @@ int main(int argc,char **arv) {
168272
delete l;
169273

170274
} catch (string& errStr) {
171-
string err = "at_gpupass() failed: " + errStr;
275+
string err = "Fail: " + errStr;
172276
cout << "Error: " << err << endl;
173277
}
174278

279+
}
280+
281+
int main(int argc,char **arv) {
282+
283+
//printGPUInfo();
284+
285+
int DEVID = 0;
286+
//performanceTest(DEVID,"Z:/tmp/pons/at/test/lattice/betamodel_radon.repr");
287+
performanceTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
288+
//performanceTest(DEVID,"/segfs/tmp/pons/lattice/simple_ebs.repr");
289+
//integratorTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
175290

176291
return 0;
177292

0 commit comments

Comments
 (0)