Merge pull request #455 from kubagalecki/kernel-arg-refactor

Constant memory refactor
CFD-GO · Oct 23, 2023 · 57dd196 · 57dd196
2 parents 365770c + 0890776
commit 57dd196
Show file tree

Hide file tree

Showing 22 changed files with 667 additions and 652 deletions.
diff --git a/src/CartLatticeAccess.hpp.Rt b/src/CartLatticeAccess.hpp.Rt
@@ -262,21 +262,23 @@ class CartLatticeAccess {
   const z_t z;
   const flag_t nt;
   const range_int<0,1,0,1> nx, ny, nz;
+  const CartLatticeContainer* container;
   public:
-  CudaDeviceFunction CartLatticeAccess(const int& x_, const int& y_, const int& z_) :
+  CudaDeviceFunction CartLatticeAccess(const int& x_, const int& y_, const int& z_, const CartLatticeContainer& container_) :
     x(x_),y(y_),z(z_),
-    nt(fetchNodeType(constContainer, x, y, z)),
-    nx(constContainer.nx),ny(constContainer.ny),nz(constContainer.nz)
+    nt(fetchNodeType(container_, x, y, z)),
+    nx(container_.nx),ny(container_.ny),nz(container_.nz),
+    container(&container_)
     { }
-  CudaDeviceFunction real_t getX() const { return constContainer.px + x; }
-  CudaDeviceFunction real_t getY() const { return constContainer.py + y; }
-  CudaDeviceFunction real_t getZ() const { return constContainer.pz + z; }
+  CudaDeviceFunction real_t getX() const { return container->px + x; }
+  CudaDeviceFunction real_t getY() const { return container->py + y; }
+  CudaDeviceFunction real_t getZ() const { return container->pz + z; }
   CudaDeviceFunction flag_t getNodeType() const { return nt; }
 
   CudaDeviceFunction inline cut_t getQ(const int& d) const  {
-    if (constContainer.Q == NULL) return NO_CUT;
+    if (!container->Q) return NO_CUT;
     size_t i = ((((size_t)d)*nz+z)*ny+y)*nx+x;
-    return constContainer.Q[i];
+    return container->Q[i];
   }
 
 <?R for (f in rows(Fields)) { ?>
@@ -299,7 +301,7 @@ template <class dx_t, class dy_t, class dz_t>
 CudaDeviceFunction real_t CartLatticeAccess< x_t, y_t, z_t >::load_<?%s f$nicename ?> (const dx_t & dx, const dy_t & dy, const dz_t & dz) const
 {
   storage_t ret; <?R
-  con = make.context("constContainer.in");
+  con = make.context("container->in");
   p = PV(c("x","y","z"));
   dp = PV(c("dx","dy","dz"));
   if (f$minx == f$maxx) dp[1] = f$minx
@@ -318,7 +320,7 @@ template <class N>
 CudaDeviceFunction void CartLatticeAccess< x_t, y_t, z_t >::pop<?%s s$suffix ?>(N & node) const
 {
 	storage_t val; <?R
-  con = make.context("constContainer.in",pocket=TRUE);
+  con = make.context("container->in",pocket=TRUE);
   for (d in rows(Density)[s$load.densities]) {
     f = rows(Fields)[[match(d$field, Fields$name)]]
     dp = c(-d$dx, -d$dy, -d$dz)
@@ -332,7 +334,7 @@ template <class N>
 CudaDeviceFunction void CartLatticeAccess< x_t, y_t, z_t >::push<?%s s$suffix ?>(N & node) const
 {
   storage_t val; <?R
-  con = make.context("constContainer.out",pocket=TRUE);
+  con = make.context("container->out",pocket=TRUE);
   for (f in rows(Fields)[s$save.fields]) { ?>
   val = <?%s real_to_storage(paste("node",f$name,sep="."),f) ?>; <?R
     con=save.field("val", f, p, con)
@@ -344,7 +346,7 @@ template < class x_t, class y_t, class z_t >
 template <class N>
 CudaDeviceFunction void CartLatticeAccess< x_t, y_t, z_t >::push<?%s s$suffix ?>_adj(N & node) const
 { <?R
-  con = make.context("constContainer.adjout",pocket=TRUE);
+  con = make.context("container->adjout",pocket=TRUE);
   for (d in rows(Density)[s$load.densities]) {
     f = rows(Fields)[[match(d$field, Fields$name)]]
     dp = c(-d$dx, -d$dy, -d$dz)
@@ -357,7 +359,7 @@ template < class x_t, class y_t, class z_t >
 template <class N>
 CudaDeviceFunction void CartLatticeAccess< x_t, y_t, z_t >::pop<?%s s$suffix ?>_adj(N & node) const
 { <?R
-  con = make.context("constContainer.adjin",pocket=TRUE);
+  con = make.context("container->adjin",pocket=TRUE);
   for (f in rows(Fields)[s$save.fields]) {
     val = paste("node",f$adjoint_name,sep=".")
     con = field.access( d=val, f, p, pattern="put", access="getsum", MContext=con)