Skip to content

[runtime_lib] Add padding to avoid false sharing and preallocate some space to reduce reallocation #152

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: swarm_backend
Choose a base branch
from
Open

[runtime_lib] Add padding to avoid false sharing and preallocate some space to reduce reallocation #152

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
49 changes: 35 additions & 14 deletions src/runtime_lib/infra_swarm/list.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,10 +8,33 @@
namespace swarm_runtime {
class VertexFrontierList {
public:
int32_t max_num_elems;
int32_t current_level; // Let's have this point to the last inserted frontier. If VFL is empty, then this is -1.
const int32_t max_num_elems;
private:
char _padding_0[SWARM_CACHE_LINE - sizeof(max_num_elems)];

std::vector<swarm::UnorderedQueue<int>> frontiers;
public:
int32_t current_level = -1; // Let's have this point to the last inserted frontier. If VFL is empty, then this is -1.
private:
char _padding_1[SWARM_CACHE_LINE - sizeof(current_level)];

public:
struct alignas(SWARM_CACHE_LINE) paddedUnorderedQueue {
swarm::UnorderedQueue<int> q;
char _padding[SWARM_CACHE_LINE - sizeof(q)];
};
static_assert(sizeof(paddedUnorderedQueue) == SWARM_CACHE_LINE, "");

std::vector<paddedUnorderedQueue> frontiers;

VertexFrontierList(int32_t max_elems) : max_num_elems(max_elems) {
// Ensure reallocation (which is expensive) is rare, by allocation space
// in large chunks. Note: do NOT call frontiers.reserve() anywhere else.
// Instead, let push_back() or emplace_back() do all the work: they will
// double the allocated capacity when needed, ensuring that reallocations
// stay rare. As no application uses many instances of VertexFrontierList,
// allocating 16k cachelines (1 MB) is acceptable.
frontiers.reserve(16 * 1024);
}

void extend_frontier_list() {
for (int i = 0; i < 10; i++) {
Expand All @@ -21,9 +44,7 @@ class VertexFrontierList {
};

VertexFrontierList create_new_vertex_frontier_list(int32_t max_elems) {
VertexFrontierList vl;
vl.max_num_elems = max_elems;
vl.current_level = -1; // Points at the current head. If there is one frontier in the VFL, then this will be 0.
VertexFrontierList vl(max_elems);
vl.extend_frontier_list();
return vl;
}
Expand All @@ -32,7 +53,7 @@ VertexFrontierList create_new_vertex_frontier_list(int32_t max_elems) {
void builtin_insert(VertexFrontierList &v1, VertexFrontier &frontier) {
v1.current_level++;
for (int i = 0; i < frontier.elems.size(); i++) {
v1.frontiers[v1.current_level].push(frontier[i]);
v1.frontiers[v1.current_level].q.push(frontier[i]);
}
if (v1.current_level >= v1.frontiers.size() - 1) {
v1.extend_frontier_list();
Expand All @@ -42,27 +63,27 @@ void builtin_insert(VertexFrontierList &v1, VertexFrontier &frontier) {
template <typename T>
static void builtin_insert(VertexFrontierList &v1, swarm::UnorderedQueue<T> *frontier) {
v1.current_level++;

v1.frontiers[v1.current_level] = *frontier;
if (v1.current_level >= v1.frontiers.size() - 2) {
v1.frontiers[v1.current_level].q = *frontier;

if (v1.current_level >= v1.frontiers.size() - 1) {
v1.extend_frontier_list();
}
}

// insert, given an explicit round to insert into.
void builtin_insert(VertexFrontierList &v1, int vertex, int round) {
v1.frontiers[round].push(vertex);
v1.frontiers[round].q.push(vertex);
}

// pop the last frontier. This is pointed to by current_level.
void builtin_retrieve(VertexFrontierList &v1, VertexFrontier &frontier) {
if (v1.current_level < 0) {
return;
}
int total = v1.frontiers[v1.current_level].startMaterialize();
int total = v1.frontiers[v1.current_level].q.startMaterialize();
frontier.elems.resize(total);
int32_t* a = frontier.elems.data();
v1.frontiers[v1.current_level].finishMaterialize(total, a);
v1.frontiers[v1.current_level].q.finishMaterialize(total, a);
frontier.num_elems = total;

v1.current_level--;
Expand All @@ -74,7 +95,7 @@ static void builtin_retrieve(VertexFrontierList &v1, swarm::UnorderedQueue<T> *f
if (v1.current_level < 0) {
return;
}
*frontier = std::move(v1.frontiers[v1.current_level]);
*frontier = std::move(v1.frontiers[v1.current_level].q);
/*
frontier->clear();
int total = v1.frontiers[v1.current_level].startMaterialize();
Expand Down