cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Overloaded max/min for int3 and removed old comments

Browse Source
This commit is contained in:
jpekkila
2019-07-04 16:24:08 +03:00
parent e1d545b0eb
commit a53e0a170d
2 changed files with 20 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
src/core/astaroth.cu
View File

@@ -188,8 +188,8 @@ acLoadWithOffset(const AcMesh& host_mesh, const int3& src, const int num_vertice
const int3 s0 = src; const int3 s0 = src;
const int3 s1 = gridIdx3d(grid, gridIdx(grid, s0) + num_vertices); const int3 s1 = gridIdx3d(grid, gridIdx(grid, s0) + num_vertices);
const int3 da = (int3){max(s0.x, d0.x), max(s0.y, d0.y), max(s0.z, d0.z)}; const int3 da = max(s0, d0);
const int3 db = (int3){min(s1.x, d1.x), min(s1.y, d1.y), min(s1.z, d1.z)}; const int3 db = min(s1, d1);
/* /*
printf("Device %d\n", i); printf("Device %d\n", i);
printf("\ts0: "); printInt3(s0); printf("\n"); printf("\ts0: "); printInt3(s0); printf("\n");
@@ -202,8 +202,8 @@ acLoadWithOffset(const AcMesh& host_mesh, const int3& src, const int num_vertice
*/ */
if (db.z >= da.z) { if (db.z >= da.z) {
const int copy_cells = gridIdx(subgrid, db) - gridIdx(subgrid, da); const int copy_cells = gridIdx(subgrid, db) - gridIdx(subgrid, da);
const int3 da_local = (int3){ // DECOMPOSITION OFFSET HERE
da.x, da.y, da.z - i * grid.n.z / num_devices}; // DECOMPOSITION OFFSET HERE const int3 da_local = (int3){da.x, da.y, da.z - i * grid.n.z / num_devices};
// printf("\t\tcopy %d cells to local index ", copy_cells); printInt3(da_local); // printf("\t\tcopy %d cells to local index ", copy_cells); printInt3(da_local);
// printf("\n"); // printf("\n");
copyMeshToDevice(devices[i], STREAM_PRIMARY, host_mesh, da, da_local, copy_cells); copyMeshToDevice(devices[i], STREAM_PRIMARY, host_mesh, da, da_local, copy_cells);
@@ -224,27 +224,14 @@ acStoreWithOffset(const int3& src, const int num_vertices, AcMesh* host_mesh)
const int3 s0 = src; const int3 s0 = src;
const int3 s1 = gridIdx3d(grid, gridIdx(grid, s0) + num_vertices); const int3 s1 = gridIdx3d(grid, gridIdx(grid, s0) + num_vertices);
const int3 da = (int3){max(s0.x, d0.x), max(s0.y, d0.y), max(s0.z, d0.z)}; const int3 da = max(s0, d0);
const int3 db = (int3){min(s1.x, d1.x), min(s1.y, d1.y), min(s1.z, d1.z)}; const int3 db = min(s1, d1);
/*
printf("Device %d\n", i);
printf("\ts0: "); printInt3(s0); printf("\n");
printf("\td0: "); printInt3(d0); printf("\n");
printf("\tda: "); printInt3(da); printf("\n");
printf("\tdb: "); printInt3(db); printf("\n");
printf("\td1: "); printInt3(d1); printf("\n");
printf("\ts1: "); printInt3(s1); printf("\n");
printf("\t-> %s to device %d\n", db.z >= da.z ? "Copy" : "Do not copy", i);
*/
if (db.z >= da.z) { if (db.z >= da.z) {
const int copy_cells = gridIdx(subgrid, db) - gridIdx(subgrid, da); const int copy_cells = gridIdx(subgrid, db) - gridIdx(subgrid, da);
const int3 da_local = (int3){ // DECOMPOSITION OFFSET HERE
da.x, da.y, da.z - i * grid.n.z / num_devices}; // DECOMPOSITION OFFSET HERE const int3 da_local = (int3){da.x, da.y, da.z - i * grid.n.z / num_devices};
// printf("\t\tcopy %d cells from local index ", copy_cells); printInt3(da_local);
// printf("\n");
copyMeshToHost(devices[i], STREAM_PRIMARY, da_local, da, copy_cells, host_mesh); copyMeshToHost(devices[i], STREAM_PRIMARY, da_local, da, copy_cells, host_mesh);
} }
// printf("\n");
} }
acBoundcondStep(); // TODO note: this is not the most efficient way to do things acBoundcondStep(); // TODO note: this is not the most efficient way to do things
return AC_SUCCESS; return AC_SUCCESS;

12
src/core/math_utils.h
View File

@@ -42,6 +42,18 @@ min(const T& a, const T& b)
return a < b ? a : b; return a < b ? a : b;
} }
static inline const int3
max(const int3& a, const int3& b)
{
return (int3){max(a.x, b.x), max(a.y, b.y), max(a.z, b.z)};
}
static inline const int3
min(const int3& a, const int3& b)
{
return (int3){min(a.x, b.x), min(a.y, b.y), min(a.z, b.z)};
}
template <class T> template <class T>
static inline const T static inline const T
sum(const T& a, const T& b) sum(const T& a, const T& b)