Added ExactDriftPass ExactMultipolePass ExactMultipoleRadPass

Author: JeanLucPons

atgpu/BndMPoleSymplectic4Pass.cpp

@@ -1,7 +1,7 @@
 # AT GPU debugging project
 
-set(_CUDA OFF)
-set(_OPENCL ON)
+set(_CUDA ON)
+set(_OPENCL OFF)
 
 cmake_minimum_required(VERSION 3.13)
 project(AT_GPU)
@@ -16,8 +16,11 @@ set(SOURCE_FILES
         PassMethodFactory.cpp
         IdentityPass.cpp
         DriftPass.cpp
+        ExactDriftPass.cpp
         StrMPoleSymplectic4Pass.cpp
+        ExactMultipolePass.cpp
         StrMPoleSymplectic4RadPass.cpp
+        ExactMultipoleRadPass.cpp
         BndMPoleSymplectic4Pass.cpp
         BndMPoleSymplectic4RadPass.cpp
         CavityPass.cpp
@@ -57,7 +60,7 @@ if(MSVC)
   include_directories($ENV{CUDA_PATH}/include ${OCL_SDK})
   link_directories($ENV{CUDA_PATH}/lib/x64)
 else()
-  set(CMAKE_CXX_FLAGS "-O2 ${GPU_CXX_FLAGS}")
+  set(CMAKE_CXX_FLAGS "-g ${GPU_CXX_FLAGS}")
   set(SOURCE_FILES ${SOURCE_FILES} PyInterface.cpp)
   set(PYTHON_INCL /usr/include/python3.8)
   set(PYTHON_LIB /usr/lib/python3.8)

atgpu/CMakeLists.txt

@@ -11,7 +11,7 @@ typedef double AT_FLOAT;
 
 // Type of lattice element
 // The define must be the corresponding class name in uppercase
-#define NB_PASSMETHOD_TYPE           8
+#define NB_PASSMETHOD_TYPE           11
 #define IDENTITYPASS                 0
 #define DRIFTPASS                    1
 #define BNDMPOLESYMPLECTIC4PASS      2
@@ -20,6 +20,9 @@ typedef double AT_FLOAT;
 #define STRMPOLESYMPLECTIC4RADPASS   5
 #define CAVITYPASS                   6
 #define RFCAVITYPASS                 7
+#define EXACTDRIFTPASS               8
+#define EXACTMULTIPOLEPASS           9
+#define EXACTMULTIPOLERADPASS        10
 
 // Structure alignement
 #if defined(__CUDACC__) // NVCC

atgpu/Element.h

@@ -230,7 +230,7 @@ void Lattice::generateGPUKernel() {
   factory.generatePassMethodsCalls(code);
   code.append("    }\n");
   checkLostParticle(code);
-  //code.append("    printf(\"%d: %f %f %f %f %f %f\\n\",elem,r6[0],r6[1],r6[2],r6[3],r6[4],r6[5]);\n");
+  //code.append("    printf(\"%d %d: %f %f %f %f %f %f\\n\",turn,elem,r6[0],r6[1],r6[2],r6[3],r6[4],r6[5]);\n");
   code.append("    elem++;\n");
   code.append("  }\n");
 

atgpu/Lattice.cpp

@@ -1,11 +1,11 @@
 #include "PassMethodFactory.h"
 #include "DriftPass.h"
+#include "ExactDriftPass.h"
+#include "ExactMultipoleRadPass.h"
 #include "StrMPoleSymplectic4RadPass.h"
 #include "BndMPoleSymplectic4RadPass.h"
 #include "RFCavityPass.h"
 #include "AbstractGPU.h"
-#include <string.h>
-#include <iostream>
 #include <algorithm>
 
 using namespace std;
@@ -35,6 +35,9 @@ PassMethodFactory::PassMethodFactory(SymplecticIntegrator& integrator) noexcept:
   passMethodInfos[STRMPOLESYMPLECTIC4RADPASS] = REGISTER_PASS(StrMPoleSymplectic4RadPass);
   passMethodInfos[CAVITYPASS] = REGISTER_PASS(CavityPass);
   passMethodInfos[RFCAVITYPASS] = REGISTER_PASS(RFCavityPass);
+  passMethodInfos[EXACTDRIFTPASS] = REGISTER_PASS(ExactDriftPass);
+  passMethodInfos[EXACTMULTIPOLEPASS] = REGISTER_PASS(ExactMultipolePass);
+  passMethodInfos[EXACTMULTIPOLERADPASS] = REGISTER_PASS(ExactMultipoleRadPass);
 
 }
 

atgpu/PassMethodFactory.cpp

@@ -225,12 +225,12 @@ void performanceTest(int gpu,string latticeName) {
       l->run(nbTurn, nbPart, rin, rout, nbRef, refs, nbStride, starts, lostAtTurn, lostAtElem, lostAtCoord,false);
       t1 = AbstractGPU::get_ticks();
 
-      //int pIdx = 0;
-      //cout << "lostAtTurn[" << pIdx << "]" << lostAtTurn[pIdx] << endl;
-      //cout << "lostAtElem[" << pIdx << "]" << lostAtElem[pIdx] << endl;
-      //cout << "lostAtCoord[" << pIdx << "]" << lostAtCoord[pIdx*6+0] << "," << lostAtCoord[pIdx*6+1] << endl;
+      AT_FLOAT *P = &rin[114*6];
+      cout << "[" << 114 << "] " << P[0] << " " << P[1] << " " << P[2] << " " << P[3] << " " << P[4] << " " << P[5] << endl;
+      cout << "Lost turn:" << lostAtTurn[114] << endl;
+      cout << "Lost elem:" << lostAtElem[114] << endl;
+      cout << "Lost coord:" << lostAtCoord[114*6] << "," << lostAtCoord[114*6+2] << endl;
 
-      AT_FLOAT *P;
       uint32_t strideSize = nbPart / nbStride;
       for(int stride=0; stride < nbStride; stride++) {
         P = ROUTPTR(stride*strideSize, 0, nbTurn - 1);
@@ -284,8 +284,9 @@ int main(int argc,char **arv) {
 
   int DEVID = 0;
   //performanceTest(DEVID,"Z:/tmp/pons/at/test/lattice/betamodel_radon.repr");
-  performanceTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
-  //performanceTest(DEVID,"/segfs/tmp/pons/lattice/simple_ebs.repr");
+  //performanceTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
+  //performanceTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_exact.repr");
+  performanceTest(DEVID,"/segfs/tmp/pons/at/test/lattice/simple_ebs_radon.repr");
   //integratorTest(DEVID,"/segfs/tmp/pons/at/test/lattice/betamodel_radon.repr");
 
   return 0;

atgpu/main.cpp

@@ -138,6 +138,7 @@ def select_omp():
 
 c_pass_methods = glob.glob(join(integrator_src_orig, '*Pass.c'))
 cpp_pass_methods = glob.glob(join(integrator_src_orig, '*Pass.cc'))
+gpu_pass_methods = glob.glob(join('atgpu', '*Pass.cpp'))
 diffmatrix_source = join(diffmatrix_orig, 'findmpoleraddiffmatrix.c')
 at_source = join('pyat', 'at.c')
 
@@ -192,20 +193,13 @@ def cpp_integrator_ext(pass_method):
     extra_compile_args=cflags
 )
 
-gpusource = [join('atgpu', 'AbstractGPU.cpp'),
-             join('atgpu', 'AbstractInterface.cpp'),
-             join('atgpu', 'PyInterface.cpp'),
-             join('atgpu', 'Lattice.cpp'),
-             join('atgpu', 'PassMethodFactory.cpp'),
-             join('atgpu', 'SymplecticIntegrator.cpp'),
-             join('atgpu', 'IdentityPass.cpp'),
-             join('atgpu', 'DriftPass.cpp'),
-             join('atgpu', 'StrMPoleSymplectic4Pass.cpp'),
-             join('atgpu', 'BndMPoleSymplectic4Pass.cpp'),
-             join('atgpu', 'StrMPoleSymplectic4RadPass.cpp'),
-             join('atgpu', 'BndMPoleSymplectic4RadPass.cpp'),
-             join('atgpu', 'CavityPass.cpp'),
-             join('atgpu', 'RFCavityPass.cpp')]
+gpusource = (gpu_pass_methods +
+           [join('atgpu', 'AbstractGPU.cpp'),
+            join('atgpu', 'AbstractInterface.cpp'),
+            join('atgpu', 'PyInterface.cpp'),
+            join('atgpu', 'Lattice.cpp'),
+            join('atgpu', 'PassMethodFactory.cpp'),
+            join('atgpu', 'SymplecticIntegrator.cpp')])
 
 cudaext = Extension(
     name='at.tracking.gpu',