Minor changes:

- removed unused header inclusion
- update comments
- removed unnecessary cast

opm/simulators/linalg/bda/opencl/BILU0.cpp

@@ -24,7 +24,6 @@
 #include <opm/common/ErrorMacros.hpp>
 #include <dune/common/timer.hh>
 
-#include <opm/simulators/linalg/bda/BdaSolver.hpp>
 #include <opm/simulators/linalg/bda/opencl/BILU0.hpp>
 #include <opm/simulators/linalg/bda/opencl/ChowPatelIlu.hpp>
 #include <opm/simulators/linalg/bda/opencl/openclKernels.hpp>
@@ -152,8 +151,9 @@ bool BILU0<block_size>::analyze_matrix(BlockedMatrix *mat, BlockedMatrix *jacMat
     if (opencl_ilu_parallel) {
         err |= queue->enqueueWriteBuffer(s.rowIndices, CL_FALSE, 0, Nb * sizeof(unsigned), fromOrder.data(), nullptr, &events[2]);
     } else {
-        // rowsPerColorPrefix is misused here
-        // s.rowIndices[i] == i must hold
+        // fromOrder is not initialized, so use something else to fill s.rowIndices
+        // s.rowIndices[i] == i must hold, since every rowidx is mapped to itself (i.e. no actual mapping)
+        // rowsPerColorPrefix is misused here, it contains an increasing sequence (0, 1, 2, ...)
         err |= queue->enqueueWriteBuffer(s.rowIndices, CL_FALSE, 0, Nb * sizeof(unsigned), rowsPerColorPrefix.data(), nullptr, &events[2]);
     }
 

opm/simulators/linalg/bda/opencl/BILU0.hpp

@@ -68,10 +68,12 @@ private:
     bool opencl_ilu_parallel;
 
     typedef struct {
-        cl::Buffer invDiagVals;
-        cl::Buffer diagIndex;
-        cl::Buffer rowsPerColor;
-        cl::Buffer rowIndices;
+        cl::Buffer invDiagVals;    // nnz values of diagonal blocks of the matrix, inverted
+        cl::Buffer diagIndex;      // index of diagonal block of each row, used to differentiate between lower and upper triangular part
+        cl::Buffer rowsPerColor;   // number of rows for every color
+        cl::Buffer rowIndices;     // mapping every row to another index
+                                   // after mapping, all rows that are processed in parallel are contiguous
+                                   // equal to the contents of fromOrder
 #if CHOW_PATEL
         cl::Buffer Lvals, Lcols, Lrows;
         cl::Buffer Uvals, Ucols, Urows;

opm/simulators/linalg/bda/opencl/openclSolverBackend.cpp

@@ -476,7 +476,7 @@ void openclSolverBackend<block_size>::copy_system_to_gpu() {
     int sum = 0;
     for (int i = 0; i < Nb; ++i) {
         int size_row = mat->rowPointers[i + 1] - mat->rowPointers[i];
-        memcpy(vals_contiguous.data() + sum, reinterpret_cast<double*>(mat->nnzValues) + sum, size_row * sizeof(double) * block_size * block_size);
+        memcpy(vals_contiguous.data() + sum, mat->nnzValues + sum, size_row * sizeof(double) * block_size * block_size);
         sum += size_row * block_size * block_size;
     }
     err = queue->enqueueWriteBuffer(d_Avals, CL_TRUE, 0, sizeof(double) * nnz, vals_contiguous.data(), nullptr, &events[0]);
@@ -513,7 +513,7 @@ void openclSolverBackend<block_size>::update_system_on_gpu() {
     int sum = 0;
     for (int i = 0; i < Nb; ++i) {
         int size_row = mat->rowPointers[i + 1] - mat->rowPointers[i];
-        memcpy(vals_contiguous.data() + sum, reinterpret_cast<double*>(mat->nnzValues) + sum, size_row * sizeof(double) * block_size * block_size);
+        memcpy(vals_contiguous.data() + sum, mat->nnzValues + sum, size_row * sizeof(double) * block_size * block_size);
         sum += size_row * block_size * block_size;
     }
     err = queue->enqueueWriteBuffer(d_Avals, CL_TRUE, 0, sizeof(double) * nnz, vals_contiguous.data(), nullptr, &events[0]);