Merge pull request #197 from vaexio/superhash

Superhash: binby, groupby, unique, value_counts and xarray support

.gitmodules

@@ -4,3 +4,9 @@
 [submodule "packages/vaex-core/vendor/string-view-lite"]
 	path = packages/vaex-core/vendor/string-view-lite
 	url = https://github.com/martinmoene/string-view-lite
+[submodule "packages/vaex-core/vendor/hopscotch-map"]
+	path = packages/vaex-core/vendor/hopscotch-map
+	url = https://github.com/Tessil/hopscotch-map
+[submodule "packages/vaex-core/vendor/flat_hash_map"]
+	path = packages/vaex-core/vendor/flat_hash_map
+	url = https://github.com/skarupke/flat_hash_map

.travis.yml

@@ -42,10 +42,7 @@ before_install:
   - source deactivate
   - source activate test-environment
   - which pip
-  - pip install pybind11
-  # - conda install -c conda-forge pandas kapteyn # these extra installs should disappear
-  # - conda create --name dev --clone test
-#  - pip install -r requirements.txt
+  - pip install -r requirements.txt
 install:
   - source activate test-environment
   - (cd packages/vaex-core; pip install -v .)

appveyor.yml

@@ -20,9 +20,9 @@ install:
   - conda config --set always_yes yes --set changeps1 no
   - conda update -q conda
   - conda info -a
-  - "conda create -q -n test-environment -c conda-forge python=%PYTHON_VERSION% numpy scipy pyqt matplotlib pyopengl h5py numexpr astropy tornado cython pandas runipy cython pytest numba pyarrow>=0.12 graphviz python-graphviz pcre"
+  - "conda create -q -n test-environment -c conda-forge python=%PYTHON_VERSION% numpy scipy pyqt matplotlib pyopengl h5py numexpr astropy tornado cython pandas runipy cython pytest numba pyarrow graphviz python-graphviz pcre"
   - activate test-environment
-  - pip install pybind11
+  - pip install "numpy>=1.13" "pyarrow>=0.12"
   - pip install -r requirements.txt
   - pushd packages\vaex-core   && pip install . && popd
   - pushd packages\vaex-hdf5   && pip install . && popd

packages/vaex-core/setup.py

@@ -63,15 +63,17 @@ def __str__(self):
 if platform.system().lower() == 'windows':
     extra_compile_args = ["/EHsc"]
 else:
+    # TODO: maybe enable these flags for non-wheel/conda builds? ["-mtune=native", "-march=native"]
     extra_compile_args = ["-std=c++11", "-mfpmath=sse", "-O3", "-funroll-loops"]
+    extra_compile_args.append("-g")
 if sys.platform == 'darwin':
     extra_compile_args.append("-mmacosx-version-min=10.9")
 
 # on windows (Conda-forge builds), the dirname is an absolute path
 extension_vaexfast = Extension("vaex.vaexfast", [os.path.relpath(os.path.join(dirname, "src/vaexfast.cpp"))],
                                include_dirs=[get_numpy_include()],
                                extra_compile_args=extra_compile_args)
-extension_strings = Extension("vaex.strings", [os.path.relpath(os.path.join(dirname, "src/strings.cpp"))],
+extension_strings = Extension("vaex.superstrings", [os.path.relpath(os.path.join(dirname, "src/strings.cpp"))],
                                include_dirs=[
                                    get_numpy_include(),
                                    get_pybind_include(),
@@ -88,10 +90,34 @@ def __str__(self):
                                extra_compile_args=extra_compile_args,
                                libraries=['pcre', 'pcrecpp']
                                )
-extension_superutils = Extension("vaex.superutils", [os.path.relpath(os.path.join(dirname, "src/superutils.cpp"))],
-                               include_dirs=[get_numpy_include(), get_pybind_include(),
-                               get_pybind_include(user=True)],
-                               extra_compile_args=extra_compile_args)
+extension_superutils = Extension("vaex.superutils", [
+        os.path.relpath(os.path.join(dirname, "src/hash_object.cpp")),
+        os.path.relpath(os.path.join(dirname, "src/hash_primitives.cpp")),
+        os.path.relpath(os.path.join(dirname, "src/superutils.cpp")),
+        os.path.relpath(os.path.join(dirname, "src/hash_string.cpp")),
+    ],
+    include_dirs=[
+        get_numpy_include(), get_pybind_include(),
+        get_pybind_include(user=True),
+        'vendor/flat_hash_map',
+        'vendor/sparse-map/include',
+        'vendor/hopscotch-map/include',
+        'vendor/string-view-lite/include'
+    ],
+    extra_compile_args=extra_compile_args)
+
+extension_superagg = Extension("vaex.superagg", [
+        os.path.relpath(os.path.join(dirname, "src/superagg.cpp")),
+    ],
+    include_dirs=[
+        get_numpy_include(), get_pybind_include(),
+        get_pybind_include(user=True),
+        'vendor/flat_hash_map',
+        'vendor/sparse-map/include',
+        'vendor/hopscotch-map/include',
+        'vendor/string-view-lite/include'
+    ],
+    extra_compile_args=extra_compile_args)
 
 setup(name=name + '-core',
       version=version,
@@ -104,7 +130,7 @@ def __str__(self):
       license=license,
       package_data={'vaex': ['test/files/*.fits', 'test/files/*.vot', 'test/files/*.hdf5']},
       packages=['vaex', 'vaex.core', 'vaex.file', 'vaex.test', 'vaex.ext', 'vaex.misc'],
-      ext_modules=[extension_vaexfast] if on_rtd else [extension_vaexfast, extension_strings, extension_superutils],
+      ext_modules=[extension_vaexfast] if on_rtd else [extension_vaexfast, extension_strings, extension_superutils, extension_superagg],
       zip_safe=False,
       entry_points={
           'console_scripts': ['vaex = vaex.__main__:main'],

packages/vaex-core/src/hash.hpp

@@ -0,0 +1,21 @@
+// #include "flat_hash_map.hpp"
+// #include "unordered_map.hpp"
+#include "tsl/hopscotch_set.h"
+#include "tsl/hopscotch_map.h"
+
+namespace vaex {
+
+template<class Key, class Value, class Hash=std::hash<Key>, class Compare=std::equal_to<Key>>
+// using hashmap = ska::flat_hash_map<Key, Value, Hash, Compare>;
+using hashmap = tsl::hopscotch_map<Key, Value, Hash, Compare>;
+// template<class Key,  class Hash, class Compare>
+// using hashset = tsl::hopscotch_set<Key, Hash, Compare>;
+
+// we cannot modify .second, instead use .value()
+// see https://github.com/Tessil/hopscotch-map
+template<class I, class V>
+inline void set_second(I& it, V &&value) {
+    it.value() = value;
+}
+
+}