Optimize C++ implementation

.github/workflows/cibuildwheel.yml

@@ -192,7 +192,7 @@ jobs:
 
 
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
         with:
           submodules: recursive
       - uses: actions/setup-python@v5
@@ -201,29 +201,32 @@ jobs:
           python-version: '3.10'
       - name: Set up QEMU
         if: runner.os == 'Linux'
-        uses: docker/setup-qemu-action@v1
+        uses: docker/setup-qemu-action@v3
         with:
           platforms: all
       - name: Build wheels
         uses: pypa/cibuildwheel@v2.17.0
         env:
           CIBW_MANYLINUX_X86_64_IMAGE: ${{ matrix.manylinux_image }}
           CIBW_BUILD: cp${{ matrix.python }}-${{ matrix.platform_id }}
-          CIBW_BEFORE_BUILD: pip install pybind11
+          CIBW_BEFORE_BUILD: |
+            pip install pybind11
+            sudo apt-get update && sudo apt-get install -y zlib1g-dev liblzma-dev
           CIBW_TEST_COMMAND: pytest {project}/tests -vv
           CIBW_TEST_REQUIRES: pytest numpy
           CIBW_BUILD_VERBOSITY: 1
           CIBW_ARCHS: ${{ matrix.arch }}
           MACOSX_DEPLOYMENT_TARGET: ${{ matrix.macosx_deployment_target }}
-      - uses: actions/upload-artifact@v2
+      - uses: actions/upload-artifact@v4
         with:
+          name: artifact-${{ matrix.os }}-py${{ matrix.python }}
           path: ./wheelhouse/*.whl
 
   build_sdist:
     name: Build source distribution
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v4
         with:
           submodules: recursive
       - uses: actions/setup-python@v5
@@ -232,8 +235,9 @@ jobs:
           python-version: '3.10'
       - name: Build sdist
         run: python setup.py sdist
-      - uses: actions/upload-artifact@v2
+      - uses: actions/upload-artifact@v4
         with:
+          name: artifact-sdist
           path: dist/*.tar.gz
 
   upload_pypi:
@@ -242,9 +246,9 @@ jobs:
     # upload to PyPI on every tag starting with 'v'
     if: github.event_name == 'push' && startsWith(github.event.ref, 'refs/tags/v')
     steps:
-      - uses: actions/download-artifact@v2
+      - uses: actions/download-artifact@v4
         with:
-          name: artifact
+          pattern: artifact-*
           path: dist
       - uses: pypa/gh-action-pypi-publish@v1.4.2
         with:

cpp/prtree.h

@@ -77,8 +77,10 @@ auto list_list_to_arrays(vec<vec<T>> out_ll)
   }
   vec<T> out;
   out.reserve(sum);
-  for (const auto &v : out_ll)
-    out.insert(out.end(), v.begin(), v.end());
+  for (auto &v : out_ll)
+    out.insert(out.end(),
+                std::make_move_iterator(v.begin()),
+                std::make_move_iterator(v.end()));
 
   return make_tuple(
       std::move(as_pyarray(out_s)),
@@ -104,14 +106,14 @@ static const float REBUILD_THRE = 1.25;
 #define unlikely(x) (x)
 #endif
 
-std::string compress(std::string &data)
+inline std::string compress(const std::string &data)
 {
   std::string output;
   snappy::Compress(data.data(), data.size(), &output);
   return output;
 }
 
-std::string decompress(std::string &data)
+inline std::string decompress(const std::string &data)
 {
   std::string output;
   snappy::Uncompress(data.data(), data.size(), &output);
@@ -702,25 +704,22 @@ class PRTreeElement
 template <class T, int B = 6, int D = 2>
 void bfs(const std::function<void(std::unique_ptr<PRTreeLeaf<T, B, D>> &)> &func, vec<PRTreeElement<T, B, D>> &flat_tree, const BB<D> target)
 {
-  queue<size_t> que;
-  auto qpush_if_intersect = [&](const size_t &i)
+  vec<size_t> que;
+  que.reserve(flat_tree.size());
+  auto qpush_if_intersect = [&](size_t i)
   {
     PRTreeElement<T, B, D> &r = flat_tree[i];
-    // std::cout << "i " << (long int) i << " : " << (bool) r.leaf << std::endl;
     if (r(target))
     {
-      // std::cout << " is pushed" << std::endl;
-      que.emplace(i);
+      que.push_back(i);
     }
   };
 
-  // std::cout << "size: " << flat_tree.size() << std::endl;
   qpush_if_intersect(0);
-  while (!que.empty())
+  size_t qhead = 0;
+  while (qhead < que.size())
   {
-    size_t idx = que.front();
-    // std::cout << "idx: " << (long int) idx << std::endl;
-    que.pop();
+    size_t idx = que[qhead++];
     PRTreeElement<T, B, D> &elem = flat_tree[idx];
 
     if (elem.leaf)
@@ -733,7 +732,8 @@ void bfs(const std::function<void(std::unique_ptr<PRTreeLeaf<T, B, D>> &)> &func
       for (size_t offset = 0; offset < B; offset++)
       {
         size_t jdx = idx * B + offset + 1;
-        qpush_if_intersect(jdx);
+        if (jdx < flat_tree.size())
+          qpush_if_intersect(jdx);
       }
     }
   }

requirements-dev.txt

@@ -1,3 +1,3 @@
 pytest==7.1.2
 pybind11==2.9.0
-cmake==3.22.4
+cmake==3.22.4

requirements.txt

@@ -1 +1 @@
-numpy>=1.16,<2.0
+numpy>=1.16,<2.0

run_profile.sh

@@ -1,3 +1,4 @@
+#!/usr/bin/env bash
 set -e
 
 rm -rf build dist .pytest_cache
@@ -8,4 +9,4 @@ python docs/run_profile.py
 so_path=src/python_prtree/PRTree.cpython-310-x86_64-linux-gnu.so
 google-pprof --callgrind $so_path build.prof > cg_build.prof
 google-pprof --callgrind $so_path find_all.prof > cg_find_all.prof
-google-pprof --callgrind $so_path insert.prof > cg_insert.prof
+google-pprof --callgrind $so_path insert.prof > cg_insert.prof

run_test.sh

@@ -1,3 +1,4 @@
+#!/usr/bin/env bash
 set -e
 
 rm -rf build dist .pytest_cache