Param (#205)

* test first step * typo * changed template * changed template * debug * debug * added matrixParam to AddForcingTermsKernelDriver * changed header * debug * debug * debug * test * passing vector_matrix directly * debug * implemented simple class * add std::vector * debug * ahhh * passing 3 matrix data as a group * deleted const * debug * debug * debug * debug * debug * added matrix data in a group for AddJacobianFunction * debug * store d_rate_constants in struct pointer * using cudaMallocManaged * using cudaMallocManaged * using cudaMallocManaged * device strcut * get all pointers into struct for AddJacobian * debug * debug * passed? * passing struct of processSet to kernelDriver * clean up * removed header * group processSet data members into struct * minor fix * minor fix * get rid of micm from struct * debug * debug * minor fix * minor fix * minor fix * clean up * debug * test * clean up * clean up * clean up * clean up * test * test out for storing pointer to variable * store reused device pointer to variable in kernel * store reused device pointer to variable in kernel * added struct for sparseMatrix * added struct for sparseMatrix * sparse matrix in struct
NCAR · Sep 5, 2023 · 12b01eb · 12b01eb
1 parent d1b34a2
commit 12b01eb
Show file tree

Hide file tree

Showing 5 changed files with 224 additions and 190 deletions.
diff --git a/include/micm/process/cuda_process_set.cuh b/include/micm/process/cuda_process_set.cuh
@@ -1,40 +1,17 @@
 #pragma once
+#include <micm/util/cuda_param.hpp>
 
 namespace micm
 {
   namespace cuda
   {
-   std::chrono::nanoseconds AddForcingTermsKernelDriver(
-        const double* rate_constants_data,
-        const double* state_variables_data,
-        double* forcing_data,
-        size_t n_grids,
-        size_t n_reactions,
-        size_t n_species,
-        const size_t* number_of_reactants,
-        const size_t* reactant_ids,
-        size_t reactant_ids_size,
-        const size_t* number_of_products,
-        const size_t* product_ids,
-        size_t product_ids_size,
-        const double* yields,
-        size_t yields_size);
+    std::chrono::nanoseconds AddForcingTermsKernelDriver(
+        CUDAMatrixParam& matrixParam,
+        CUDAProcessSetParam& processSet); 
 
     std::chrono::nanoseconds AddJacobianTermsKernelDriver(
-        const double* rate_constants,
-        const double* state_variables,
-        size_t n_grids,
-        size_t n_reactions,
-        size_t n_species,
-        double* jacobian,
-        size_t jacobian_size,
-        const size_t* number_of_reactants, 
-        const size_t* reactant_ids,
-        size_t reactant_ids_size,
-        const size_t* number_of_products,
-        const double* yields,
-        size_t yields_size,
-        const size_t* jacobian_flat_ids,
-        size_t jacobian_flat_ids_size);
+        CUDAMatrixParam& matrixParam, 
+        CUDASparseMatrixParam& sparseMatrix, 
+        CUDAProcessSetParam& processSet);
   }  // namespace cuda
 }  // namespace micm
diff --git a/include/micm/process/cuda_process_set.hpp b/include/micm/process/cuda_process_set.hpp
@@ -4,6 +4,7 @@
 #pragma once
 
 #include <micm/process/process_set.hpp>
+#include <micm/util/cuda_param.hpp>
 
 #ifdef USE_CUDA
 #  include <micm/process/cuda_process_set.cuh>
@@ -51,21 +52,27 @@ namespace micm
       const MatrixPolicy<double>& state_variables,
       MatrixPolicy<double>& forcing) const
   {
+    CUDAMatrixParam matrixParam;
+    matrixParam.rate_constants = rate_constants.AsVector().data(); 
+    matrixParam.state_variables = state_variables.AsVector().data(); 
+    matrixParam.forcing = forcing.AsVector().data(); 
+    matrixParam.n_grids = rate_constants.size(); 
+    matrixParam.n_reactions = rate_constants[0].size(); 
+    matrixParam.n_species = state_variables[0].size();
+
+    CUDAProcessSetParam processSet; 
+    processSet.number_of_reactants = number_of_reactants_.data(); 
+    processSet.reactant_ids = reactant_ids_.data(); 
+    processSet.reactant_ids_size = reactant_ids_.size(); 
+    processSet.number_of_products = number_of_products_.data(); 
+    processSet.product_ids = product_ids_.data(); 
+    processSet.product_ids_size = product_ids_.size(); 
+    processSet.yields = yields_.data(); 
+    processSet.yields_size = yields_.size(); 
+
     std::chrono::nanoseconds kernel_duration = micm::cuda::AddForcingTermsKernelDriver(
-        rate_constants.AsVector().data(),
-        state_variables.AsVector().data(),
-        forcing.AsVector().data(),
-        rate_constants.size(),
-        rate_constants[0].size(),
-        state_variables[0].size(),
-        number_of_reactants_.data(),
-        reactant_ids_.data(),
-        reactant_ids_.size(),
-        number_of_products_.data(),
-        product_ids_.data(),
-        product_ids_.size(),
-        yields_.data(),
-        yields_.size());
+        matrixParam,
+        processSet);
     return kernel_duration;  // time performance of kernel function
   }
   template<template<class> class MatrixPolicy, template<class> class SparseMatrixPolicy>
@@ -75,22 +82,31 @@ namespace micm
       const MatrixPolicy<double>& state_variables,
       SparseMatrixPolicy<double>& jacobian) const
   {
+    CUDAMatrixParam matrixParam; 
+    matrixParam.rate_constants = rate_constants.AsVector().data(); 
+    matrixParam.state_variables = state_variables.AsVector().data(); 
+    matrixParam.n_grids = rate_constants.size(); 
+    matrixParam.n_reactions = rate_constants[0].size(); 
+    matrixParam.n_species = state_variables[0].size(); 
+
+    CUDASparseMatrixParam sparseMatrix; 
+    sparseMatrix.jacobian = jacobian.AsVector().data(); 
+    sparseMatrix.jacobian_size = jacobian.AsVector().size(); 
+
+    CUDAProcessSetParam processSet; 
+    processSet.number_of_reactants = number_of_reactants_.data(); 
+    processSet.reactant_ids = reactant_ids_.data(); 
+    processSet.reactant_ids_size = reactant_ids_.size(); 
+    processSet.number_of_products = number_of_products_.data(); 
+    processSet.yields = yields_.data(); 
+    processSet.yields_size = yields_.size(); 
+    processSet.jacobian_flat_ids = jacobian_flat_ids_.data(); 
+    processSet.jacobian_flat_ids_size = jacobian_flat_ids_.size(); 
+
     std::chrono::nanoseconds kernel_duration = micm::cuda::AddJacobianTermsKernelDriver(
-        rate_constants.AsVector().data(),
-        state_variables.AsVector().data(),
-        rate_constants.size(),      // n_grids
-        rate_constants[0].size(),   // n_reactions
-        state_variables[0].size(),  // n_species
-        jacobian.AsVector().data(),
-        jacobian.AsVector().size(),
-        number_of_reactants_.data(),
-        reactant_ids_.data(),
-        reactant_ids_.size(),
-        number_of_products_.data(),
-        yields_.data(),
-        yields_.size(),
-        jacobian_flat_ids_.data(),
-        jacobian_flat_ids_.size());
+        matrixParam, 
+        sparseMatrix, 
+        processSet);
     return kernel_duration;  // time performance of kernel function
   }
 }  // namespace micm

diff --git a/include/micm/util/cuda_param.hpp b/include/micm/util/cuda_param.hpp
@@ -0,0 +1,30 @@
+#ifndef CUDA_PARAM_HPP
+#define CUDA_PARAM_HPP
+  //member data of class CUDAProcessSet grouped in struct passing to kernel driver function 
+  struct CUDAProcessSetParam{
+    const size_t* number_of_reactants; 
+    const size_t* reactant_ids; 
+    size_t reactant_ids_size; 
+    const size_t* number_of_products; 
+    const size_t* product_ids; 
+    size_t product_ids_size; 
+    const double* yields; 
+    size_t yields_size; 
+    const size_t* jacobian_flat_ids; 
+    size_t jacobian_flat_ids_size; 
+}; 
+//different matrix data grouped in struct passing to kernel driver function 
+    struct CUDAMatrixParam{
+     const double* rate_constants; 
+     const double* state_variables; 
+     double* forcing; 
+     size_t n_grids; 
+     size_t n_reactions; 
+     size_t n_species; 
+}; 
+//sparseMatrix data grouped in struct passing to kernel driver function 
+struct CUDASparseMatrixParam{
+   double* jacobian; 
+   size_t jacobian_size; 
+};
+#endif