diff --git a/runtime/domain/block_forest.cpp b/runtime/domain/block_forest.cpp
index 61b7818b81b01977ea7c9dfec57a294bb6ae799f..44d30197437bef211a450f6e6ebfc59d0ff0e04d 100644
--- a/runtime/domain/block_forest.cpp
+++ b/runtime/domain/block_forest.cpp
@@ -27,6 +27,7 @@ void BlockForest::updateNeighborhood() {
ranks.clear();
naabbs.clear();
+ aabb_offsets.clear();
aabbs.clear();
for(auto& iblock: *forest) {
@@ -59,6 +60,7 @@ void BlockForest::updateNeighborhood() {
auto rank = nbh.first;
auto aabb_list = nbh.second;
ranks.push_back((int) rank);
+ aabb_offsets.push_back(this->total_aabbs);
naabbs.push_back((int) aabb_list.size());
for(auto &aabb: aabb_list) {
@@ -80,7 +82,7 @@ void BlockForest::updateNeighborhood() {
void BlockForest::copyRuntimeArray(const std::string& name, void *dest, const int size) {
void *src = name.compare('ranks') ? ranks.data() :
name.compare('naabbs') ? vec_naabbs.data() :
- name.compare('rank_offsets') ? offsets :
+ name.compare('aabb_offsets') ? aabb_offsets.data() :
name.compare('aabbs') ? vec_aabbs.data() :
name.compare('subdom') ? subdom;
@@ -102,7 +104,7 @@ void BlockForest::updateWeights(PairsSimulation *ps, int nparticles) {
ps, aabb.xMin(), aabb.xMax(), aabb.yMin(), aabb.yMax(), aabb.zMin(), aabb.zMax(),
&(block_info.computationalWeight), &(block_info.communicationWeight));
- for(uint_t branch = 0; branch < 8; ++branch) {
+ for(int branch = 0; branch < 8; ++branch) {
const auto b_id = BlockID(block->getId(), branch);
const auto b_aabb = forest->getAABBFromBlockId(b_id);
auto& b_info = info[b_id];
@@ -117,16 +119,16 @@ void BlockForest::updateWeights(PairsSimulation *ps, int nparticles) {
auto block = static_cast<blockforest::Block *>(&iblock);
auto& block_info = info[block->getId()];
- for(uint_t neigh = 0; neigh < block->getNeighborhoodSize(); ++neigh) {
+ for(int neigh = 0; neigh < block->getNeighborhoodSize(); ++neigh) {
bs.sendBuffer(block->getNeighborProcess(neigh)) <<
blockforest::InfoCollection::value_type(block->getId(), block_info);
}
- for(uint_t branch = 0; branch < 8; ++branch) {
+ for(int branch = 0; branch < 8; ++branch) {
const auto b_id = BlockID(block->getId(), branch);
auto& b_info = info[b_id];
- for(uint_t neigh = 0; neigh < block->getNeighborhoodSize(); ++neigh) {
+ for(int neigh = 0; neigh < block->getNeighborhoodSize(); ++neigh) {
bs.sendBuffer(block->getNeighborProcess(neigh)) <<
blockforest::InfoCollection::value_type(b_id, b_info);
}
@@ -145,27 +147,27 @@ void BlockForest::updateWeights(PairsSimulation *ps, int nparticles) {
}
}
-Vector3<uint_t> BlockForest::getBlockConfig(uint_t num_processes, uint_t nx, uint_t ny, uint_t nz) {
- const uint_t bx_factor = 1;
- const uint_t by_factor = 1;
- const uint_t bz_factor = 1;
- const uint_t ax = nx * ny;
- const uint_t ay = nx * nz;
- const uint_t az = ny * nz;
+Vector3<int> BlockForest::getBlockConfig(int num_processes, int nx, int ny, int nz) {
+ const int bx_factor = 1;
+ const int by_factor = 1;
+ const int bz_factor = 1;
+ const int ax = nx * ny;
+ const int ay = nx * nz;
+ const int az = ny * nz;
- uint_t bestsurf = 2 * (ax + ay + az);
- uint_t x = 1;
- uint_t y = 1;
- uint_t z = 1;
+ int bestsurf = 2 * (ax + ay + az);
+ int x = 1;
+ int y = 1;
+ int z = 1;
- for(uint_t i = 1; i < num_processes; ++i) {
+ for(int i = 1; i < num_processes; ++i) {
if(num_processes % i == 0) {
- const uint_t rem_yz = num_processes / i;
+ const int rem_yz = num_processes / i;
- for(uint_t j = 1; j < rem_yz; ++j) {
+ for(int j = 1; j < rem_yz; ++j) {
if(rem_yz % j == 0) {
- const uint_t k = rem_yz / j;
- const uint_t surf = (ax / i / j) + (ay / i / k) + (az / j / k);
+ const int k = rem_yz / j;
+ const int surf = (ax / i / j) + (ay / i / k) + (az / j / k);
if(surf < bestsurf) {
x = i, y = j, z = k;
@@ -176,13 +178,13 @@ Vector3<uint_t> BlockForest::getBlockConfig(uint_t num_processes, uint_t nx, uin
}
}
- return Vector3<uint_t>(x * bx_factor, y * by_factor, z * bz_factor);
+ return Vector3<int>(x * bx_factor, y * by_factor, z * bz_factor);
}
-uint_t BlockForest::getInitialRefinementLevel(uint_t num_processes) {
- uint_t splitFactor = 8;
- uint_t blocks = splitFactor;
- uint_t refinementLevel = 1;
+int BlockForest::getInitialRefinementLevel(int num_processes) {
+ int splitFactor = 8;
+ int blocks = splitFactor;
+ int refinementLevel = 1;
while(blocks < num_processes) {
refinementLevel++;
@@ -230,7 +232,7 @@ void BlockForest::initialize(int *argc, char ***argv) {
math::AABB domain(xmin, ymin, zmin, xmax, ymax, zmax);
int gridsize[3] = {32, 32, 32};
auto procs = mpiManager->numProcesses();
- auto block_config = use_load_balancing ? Vector3<uint_t>(1, 1, 1) : getBlockConfig(procs, gridsize[0], gridsize[1], gridsize[2]);
+ auto block_config = use_load_balancing ? Vector3<int>(1, 1, 1) : getBlockConfig(procs, gridsize[0], gridsize[1], gridsize[2]);
auto ref_level = use_load_balancing ? getInitialRefinementLevel(procs) : 0;
forest = blockforest::createBlockForest(
diff --git a/runtime/domain/block_forest.hpp b/runtime/domain/block_forest.hpp
index 7a447c10e8eb2d90aa1ff70b7cb631f132d31c7a..6facb7d9e4d7f1e2a72935a8bdab67e060c87771 100644
--- a/runtime/domain/block_forest.hpp
+++ b/runtime/domain/block_forest.hpp
@@ -13,10 +13,11 @@ class BlockForest : public DomainPartitioner {
private:
std::shared_ptr<BlockForest> forest;
blockforest::InfoCollection info;
- std::map<uint_t, std::vector<math::AABB>> neighborhood;
- std::map<uint_t, std::vector<BlockID>> blocks_pushed;
+ std::map<int, std::vector<math::AABB>> neighborhood;
+ std::map<int, std::vector<BlockID>> blocks_pushed;
std::vector<int> ranks;
std::vector<int> naabbs;
+ std::vector<int> aabb_offsetss;
std::vector<double> aabbs;
real_t *subdom;
int world_size, rank, nranks, total_aabbs;
@@ -44,8 +45,8 @@ public:
void initializeWorkloadBalancer();
void updateNeighborhood();
void updateWeights(real_t *position, int nparticles);
- Vector3<uint_t> getBlockConfig(uint_t num_processes, uint_t nx, uint_t ny, uint_t nz);
- uint_t getInitialRefinementLevel(uint_t num_processes);
+ Vector3<int> getBlockConfig(int num_processes, int nx, int ny, int nz);
+ int getInitialRefinementLevel(int num_processes);
void setBoundingBox();
void rebalance();
diff --git a/src/pairs/sim/domain_partitioning.py b/src/pairs/sim/domain_partitioning.py
index a6173004d422d8f2d06bb84c125df5597821b440..dae2d28fc0a518741ecfd5797e240ed9877adabd 100644
--- a/src/pairs/sim/domain_partitioning.py
+++ b/src/pairs/sim/domain_partitioning.py
@@ -78,7 +78,7 @@ class BlockForest:
self.aabb_capacity = sim.add_var('aabb_capacity', Types.Int32)
self.ranks = sim.add_static_array('ranks', [self.nranks_capacity], Types.Int32)
self.naabbs = sim.add_static_array('naabbs', [self.nranks_capacity], Types.Int32)
- self.offsets = sim.add_static_array('rank_offsets', [self.nranks_capacity], Types.Int32)
+ self.aabb_offsets = sim.add_static_array('aabb_offsets', [self.nranks_capacity], Types.Int32)
self.aabbs = sim.add_static_array('aabbs', [self.aabb_capacity, 6], Types.Real)
self.subdom = sim.add_static_array('subdom', [sim.ndims() * 2], Types.Real)
@@ -105,7 +105,7 @@ class BlockForest:
Assign(self.sim, self.nranks_capacity, self.nranks + 10)
self.ranks.realloc()
self.naabbs.realloc()
- self.offsets.realloc()
+ self.aabb_offsets.realloc()
for _ in Filter(self.sim, self.aabb_capacity < self.naabbs):
Assign(self.sim, self.aabb_capacity, self.naabbs + 20)
@@ -114,8 +114,7 @@ class BlockForest:
Call_Void(self.sim, "pairs->copyRuntimeArray", ['ranks', self.ranks, self.nranks])
Call_Void(self.sim, "pairs->copyRuntimeArray", ['naabbs', self.naabbs, self.nranks])
- Call_Void(self.sim, "pairs->copyRuntimeArray", ['rank_offsets', self.rank_offsets, self.nranks])
- Call_Void(self.sim, "pairs->copyRuntimeArray", ['pbc', self.pbc, self.naabbs * 3])
+ Call_Void(self.sim, "pairs->copyRuntimeArray", ['aabb_offsets', self.aabb_offsets, self.nranks])
Call_Void(self.sim, "pairs->copyRuntimeArray", ['aabbs', self.aabbs, self.naabbs * 6])
Call_Void(self.sim, "pairs->copyRuntimeArray", ['subdom', self.subdom, self.sim.ndims() * 2])
@@ -123,12 +122,12 @@ class BlockForest:
# Particles with one of the following flags are ignored
flags_to_exclude = (Flags.Infinite | Flags.Global)
- for i in For(self.sim, 0, self.sim.nlocal):
- particle_flags = self.sim.particle_flags
+ for r in For(self.sim, 0, self.nranks):
+ for i in For(self.sim, 0, self.sim.nlocal):
+ particle_flags = self.sim.particle_flags
- for _ in Filter(self.sim, ScalarOp.cmp(particle_flags[i] & flags_to_exclude, 0)):
- for r in For(self.sim, 0, self.nranks):
- for aabb_id in For(self.sim, self.offsets[r], self.offsets[r] + self.naabbs[r]):
+ for _ in Filter(self.sim, ScalarOp.cmp(particle_flags[i] & flags_to_exclude, 0)):
+ for aabb_id in For(self.sim, self.aabb_offsets[r], self.aabb_offsets[r] + self.naabbs[r]):
full_cond = None
pbc_shifts = []