Making version

.gitignore

@@ -16,4 +16,5 @@ cubin/
 tempSol.json
 src/rslts/*.json
 main.pdf
+src/utilities/js*
 *~*

README.md

@@ -4,6 +4,7 @@
 
 - [HeterogeneousSwept1D](#heterogeneousswept1d)
     - [Dependencies:](#dependencies)
+- [To Run](#to-run)
 
 <!-- /TOC -->
 
@@ -13,8 +14,20 @@ Python files in main folder are entry points.
 
 ## Dependencies:
 
-------------------------------
+[jsoncpp](https://github.com/open-source-parsers/jsoncpp) - jsoncpp.sh in main folder will download, amalgamate and put in utilities folder.
 
-My dependents include:
+[palettable](http://www.palettable.io) - to run some of the plotting scripts
 
-------------------------------
+CUDA 7.5 or greater
+MPI 2
+
+# To Run
+
+make with makefile in src folder
+
+Run from src folder 
+mpirun -np [nprocs] ./bin/[executable] [scheme (C or S) for classic or swept] [path to json with run specifications (eg. tests folder)] [path to output folder] [additional options as tokens eg. tpb 32 or lx 85]
+
+python script stdExp will run standard experiment.
+
+json files in result folder are coded [s or t for solution or timing[problem eg Euler or Heat][_rank]or[S or C for swept or classic]]

runtools/result_help.py

@@ -138,10 +138,12 @@ def annotatePlot(self, fh, ax):
             fh.subplots_adjust(bottom=0.08, right=0.85, top=0.9, 
                                 wspace=0.15, hspace=0.25)
 
-    def savePlot(self, fh, plotpath):
+    def savePlot(self, fh, plotpath, shw=False):
 
         plotfile = op.join(plotpath, self.plotname + self.ext)
         fh.savefig(plotfile, dpi=200, bbox_inches="tight")
+        if shw:
+            plt.show()
 
     def gifify(self, plotpath, fh, ax):
 

src/decomposition/classic.h

@@ -30,7 +30,7 @@ void classicStepCPU(states *state, int numx, int tstep)
     {
         stepUpdate(state, k, tstep);
     }
-    // if (tstep % 5000 == 5) cout << numx << " " << ranks[1] << " " << printout(state, 0) << " " << printout(state+1, 0) << " " << tstep << " " << printout(state+numx-1, 0) << " " << printout(state + numx, 0) << endl; 
+    if ((tstep>120 && tstep <125) || tstep<6 ) cout << ranks[1] << " " << printout(state, 0) << " " << printout(state+1, 0) << " " << tstep << " " << printout(state+numx-10, 0) << " " << printout(state + numx, 0) << endl; 
 }
 
 // void classicDPass(double *putSt, double *getSt, int tstep)

src/decomposition/swept.h

@@ -184,6 +184,7 @@ void wholeDiamondCPU(states *state, int tstep)
         }
     tnow++;
     }
+    if(!ranks[1]) cout << tnow << " " << tstep << endl;
 }
 
 void splitDiamondCPU(states *state, int tstep)
@@ -200,13 +201,13 @@ void splitDiamondCPU(states *state, int tstep)
             if (n == cGlob.ht)
             {
                 ssLeft[0] = state[n-1], ssLeft[1] = state[n];
-                stepUpdate(ssLeft, n, tnow);
+                stepUpdate(&ssLeft[0], n, tnow);
                 state[n] = ssLeft[1];
             }
             else if (n == cGlob.htp)
             {
                 ssRight[1] = state[n], ssRight[2] = state[n+1];
-                stepUpdate(ssRight, n, tnow);
+                stepUpdate(&ssRight[0], n, tnow);
                 state[n] = ssRight[1];
             }
             else
@@ -224,13 +225,13 @@ void splitDiamondCPU(states *state, int tstep)
             if (n == cGlob.ht)
             {
                 ssLeft[0] = state[n-1], ssLeft[1] = state[n];
-                stepUpdate(ssLeft, n, tnow);
+                stepUpdate(&ssLeft[0], n, tnow);
                 state[n] = ssLeft[1];
             }
             else if (n == cGlob.htp)
             {
                 ssRight[1] = state[n], ssRight[2] = state[n+1];
-                stepUpdate(ssRight, n, tnow);
+                stepUpdate(&ssRight[0], n, tnow);
                 state[n] = ssRight[1];
             }
             else
@@ -329,6 +330,8 @@ double sweptWrapper(states **state, ivec xpts, ivec alen, int *tstep)
 
         cout << "UP " << endl;
 
+        MPI_Send(state[2] + 5, 1, struct_type, 0, 0, MPI_COMM_WORLD);
+
         // ------------ Pass Edges ------------ // 
         // -- FRONT TO BACK -- //
 
@@ -560,6 +563,13 @@ double sweptWrapper(states **state, ivec xpts, ivec alen, int *tstep)
     }
     else
     {
+        states tmp;
+        if (!ranks[1]) 
+        {
+            MPI_Recv(&tmp, 1, struct_type, 3, MPI_ANY_TAG,  MPI_COMM_WORLD, MPI_STATUS_IGNORE);
+            cout << printout(&tmp, 2) << " " << printout(state[0]+1, 2) << endl;
+        }
+
         const int xc = cGlob.xcpu, xcp = xc+1;
         for (int k=0; k<cGlob.cBks; k++)
         { 

src/equations/euler.h

@@ -65,8 +65,8 @@ struct eqConsts {
 
 //---------------// 
 struct states {
-    REALthree Q[2]; // Full Step, Midpoint step state variables
     REAL Pr; // Pressure ratio
+    REALthree Q[2]; // Full Step, Midpoint step state variables
 };
 
 std::string outVars[NVARS] = {"DENSITY", "VELOCITY", "ENERGY", "PRESSURE"}; //---------------// 
@@ -193,7 +193,7 @@ __host__ inline states icond(double xs, double lx)
     int side = (xs > HALF*lx);
     s.Q[0] = hBounds[side];
     s.Q[1] = hBounds[side];
-    s.Pr = 0.0;
+    s.Pr = -10.0;
     return s;
 }
 
@@ -247,12 +247,11 @@ __host__ void mpi_type(MPI_Datatype *dtype)
     MPI_Type_create_struct(3, n, disp, typs, &vtype);
     MPI_Type_commit(&vtype);
 
-    typs[0] = vtype;
-    typs[2] = vtype;
-    disp[1] = 3*sizeof(vtype);
-    disp[2] = 4*sizeof(REAL);
+    int n2[2] = {1, 2};
+    MPI_Datatype typs2[2] = {MPI_R, vtype}; 
+    MPI_Aint disp2[2] = {0, sizeof(REAL)};
 
-    MPI_Type_create_struct(3, n, disp, typs, dtype);
+    MPI_Type_create_struct(2, n2, disp2, typs2, dtype);
     MPI_Type_commit(dtype);
 
     MPI_Type_free(&vtype);
@@ -291,7 +290,7 @@ __device__ __host__
 __forceinline__
 REALthree limitor(REALthree qH, REALthree qN, REAL pRatio)
 {   
-    if(QNAN(pRatio) || pRatio<0)  
+    if(QNAN(pRatio) || (pRatio<1.0e-8))  
     {
         return qH;
     }
@@ -374,12 +373,13 @@ void eulerStep(states *state, int idx, int tstep)
 __device__ __host__ 
 void stepUpdate(states *state, int idx, int tstep)
 {
+    int ts = DIVMOD(tstep);
     if (tstep & 1) //Odd - Rslt is 0 for even numbers
     {
-        pressureRatio(state, idx, DIVMOD(tstep));
+        pressureRatio(state, idx, ts);
     }
     else
     {
-        eulerStep(state, idx, DIVMOD(tstep));
+        eulerStep(state, idx, ts);
     }
 }

src/equations/heat.h

@@ -169,11 +169,7 @@ __host__ void initialState(jsons inJs, states *inl, int idx, int xst)
 
 __host__ void mpi_type(MPI_Datatype *dtype)
 { 
-    MPI_Datatype typs[2] = {MPI_R, MPI_R};
-    int n[2] = {1};
-    MPI_Aint disp[2] = {0, sizeof(REAL)};
-
-    MPI_Type_create_struct(2, n, disp, typs, dtype);
+    MPI_Type_contiguous(2, MPI_R, dtype);
     MPI_Type_commit(dtype);
 }
 

src/jpyparse.py

@@ -42,9 +42,9 @@
 fg, axi = plt.subplots(sp[0], sp[1])
 jdf.metaparse(meta)
 jdf.plotResult(fg, axi)
-jdf.savePlot(fg, resultpath)
+jdf.savePlot(fg, resultpath, shw=True)
 dff = jdf.ddf
 ddfk = list(dff.keys())
 dsam = dff[ddfk[0]]
 dsk = dsam.columns.values.tolist()
-
+

src/solmain.cu

@@ -117,8 +117,8 @@ int main(int argc, char *argv[])
     }
     else
     {
-        state = new states* [1];
-        cudaCheckError(cudaHostAlloc((void **) &state[0], xalloc * cGlob.szState, cudaHostAllocDefault));
+        state = new states*[1];
+        state[0] = new states[xalloc * cGlob.szState];
     }
 
     for (int i=0; i<nrows; i++)
@@ -128,7 +128,6 @@ int main(int argc, char *argv[])
     }
 
     // If you have selected scheme I, it will only initialize and output the initial values.
-    if (!ranks[1])	std::cout << "Initial Values Instantiated." << std::endl;
 
     if (scheme.compare("I"))
     {
@@ -207,17 +206,22 @@ int main(int argc, char *argv[])
     soljson << solution;
     soljson.close();
 
-    endMPI();
-
-    for (int k=0; k<nrows; k++)
-    {
-        cudaFreeHost(state[k]);
-    }
-    delete[] state;   
     if (cGlob.hasGpu)
     {
+        for (int k=0; k<3; k++)cudaFreeHost(state[k]);
         cudaDeviceSynchronize();
         cudaDeviceReset();
     }
+    else
+    {   
+        delete[] state[0];
+    }
+    delete[] state;   
+    for (int k=0; k<nrows; k++)
+    {
+        std::cout << nrows << " " << ranks[1] << " " << cGlob.hasGpu << endl; 
+    }
+
+    endMPI();
     return 0;
 }