Major build cleanup.

.clang-format

@@ -0,0 +1,6 @@
+Language:        Cpp
+BasedOnStyle:  Google
+AllowShortFunctionsOnASingleLine: None
+AllowShortIfStatementsOnASingleLine: false
+AllowShortLoopsOnASingleLine: false
+PenaltyBreakComment: 100

.gitignore

@@ -36,7 +36,7 @@ nosetests.xml
 
 # Project specific
 tsne/bh_sne.cpp
-tsne/bh_sne_3d.cpp
 tsne/bh_sne_src/bh_tsne
 *.pkl.gz
 .cache
+.pytest_cache/

.travis.yml

@@ -1,37 +1,22 @@
 language: python
 
 python:
-  - "2.7"
-  # - "3.4"
+- "2.7"
+compiler: clang
+env:
+- CC="clang" CXX="clang++" LDFLAGS="-fuse-ld=lld"
 
+dist: xenial
 addons:
   apt:
     packages:
       - build-essential
-      - libatlas-base-dev
-
-sudo: false
+      - liblapack3
 
 install:
-  - if [[ "$TRAVIS_PYTHON_VERSION" == "2.7" ]]; then
-      wget http://repo.continuum.io/miniconda/Miniconda-latest-Linux-x86_64.sh -O miniconda.sh;
-    else
-      wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh;
-    fi
-  - bash miniconda.sh -b -p $HOME/miniconda
-  - export PATH="$HOME/miniconda/bin:$PATH"
-  - conda config --set always_yes yes --set changeps1 no
-  - conda update -q conda
-  - conda info -a
-
-  - conda create -q -n test-environment python=$TRAVIS_PYTHON_VERSION
-  - source activate test-environment
-  - conda install numpy cython scipy pytest scikit-learn
-  - python setup.py install
-  - source deactivate
-
+  - pip install numpy==1.14.2 cython==0.28.2 pytest==3.5.1 scikit-learn==0.19.1 scipy==1.0.1
 script:
   - pwd
-  - source activate test-environment
+  - python setup.py install
   - cd tsne/tests
   - py.test -s -vv

Makefile

@@ -1,13 +1,39 @@
-build:
-	python setup.py build_ext --inplace
+.PHONY: build
 
-install:
-	python setup.py build_ext --inplace
-	python setup.py install
+build: setup.py tsne/bh_sne.pyx \
+       $(wildcard tsne/bh_sne_src/*.h) \
+       $(wildcard tsne/bh_sne_src/*.cpp)
+	python $< build_ext --inplace
 
-sdist:
-	python setup.py sdist
+PROFILE_LOCATION=$(shell pwd)/profile
+
+$(PROFILE_LOCATION)/bh_sne.profdata: PROFILE=generate
+$(PROFILE_LOCATION)/bh_sne.profdata: setup.py Makefile tsne/bh_sne.pyx \
+                               $(wildcard tsne/bh_sne_src/*.h) \
+                               $(wildcard tsne/bh_sne_src/*.cpp)
+	rm -rf build tsne/*.so *.so $(PROFILE_LOCATION)/*.profraw
+	PROFILE=generate python $< build_ext --inplace
+	mkdir -p $(PROFILE_LOCATION)
+	cd tsne/tests && LLVM_PROFILE_FILE="$(PROFILE_LOCATION)/%p-%m.profraw" py.test
+	llvm-profdata merge -output=$@ $(PROFILE_LOCATION)/*.profraw
+	rm -rf tsne/*.so *.so build tsne/*.o
+
+build_pgo: setup.py $(PROFILE_LOCATION)/bh_sne.profdata
+	PROFILE="$(PROFILE_LOCATION)/bh_sne.profdata" python $< build_ext --inplace
+
+install: setup.py build
+	python $< install
+
+sdist: setup.py tsne/bh_sne.pyx \
+       $(wildcard tsne/bh_sne_src/*.h) \
+       $(wildcard tsne/bh_sne_src/*.cpp)
+	python $< sdist
+
+test: build
+	cd tsne/tests && time py.test -s -vv
 
 clean:
-	rm -rf *.pyc *.so build/ bh_sne.cpp
-	rm -rf tsne/*.pyc tsne/*.so tsne/build/ tsne/bh_sne.cpp
+	make -C tsne/bh_sne_src clean
+	rm -rf *.pyc *.so build/ profile/ bh_sne.cpp
+	rm -rf tsne/*.pyc tsne/*.so tsne/*.o tsne/build/ tsne/bh_sne.cpp
+	rm -rf .pytest_cache/

README.md

@@ -23,7 +23,6 @@ Requirements
 * [numpy](numpy.scipy.org) > =1.7.1
 * [scipy](http://www.scipy.org/) >= 0.12.0
 * [cython](cython.org) >= 0.19.1
-* [cblas](http://www.netlib.org/blas/) or [openblas](https://github.com/xianyi/OpenBLAS). Tested version is v0.2.5 and v0.2.6 (not necessary for OSX).
 
 [Anaconda](http://continuum.io/downloads) is recommended.
 

setup.py

@@ -7,6 +7,7 @@
 """
 
 import sys
+import os
 import platform
 
 from distutils.core import setup
@@ -28,35 +29,45 @@
 
     if v2 >= 10:
         # More than 10.10
-        extra_compile_args=['-I/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/Headers']
+        extra_compile_args = [
+            '-I/System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/Headers']
     else:
-        extra_compile_args=['-I/System/Library/Frameworks/vecLib.framework/Headers']
+        extra_compile_args = [
+            '-I/System/Library/Frameworks/vecLib.framework/Headers']
 
     ext_modules = [Extension(name='bh_sne',
-                   sources=['tsne/bh_sne_src/sptree.cpp', 'tsne/bh_sne_src/tsne.cpp', 'tsne/bh_sne.pyx'],
-                   include_dirs=[numpy.get_include(), 'tsne/bh_sne_src/'],
-                   extra_compile_args=extra_compile_args + ['-ffast-math', '-O3'],
-                   extra_link_args=['-Wl,-framework', '-Wl,Accelerate', '-lcblas'],
-                   language='c++')]
+                             sources=['tsne/bh_sne_src/sptree.cpp',
+                                      'tsne/bh_sne_src/tsne.cpp', 'tsne/bh_sne.pyx'],
+                             include_dirs=[
+                                 numpy.get_include(), 'tsne/bh_sne_src/'],
+                             extra_compile_args=extra_compile_args +
+                             ['-ffast-math', '-O3', '-std=c++14'],
+                             extra_link_args=['-Wl,-framework',
+                                              '-Wl,Accelerate'],
+                             language='c++')]
 
 else:
     # LINUX
-
+    opt_flags = ['-msse3', '-O3', '-flto']
+    if 'PROFILE' in os.environ:
+        if os.environ['PROFILE'] == 'generate':
+            opt_flags.append('-fprofile-instr-generate')
+        elif os.path.isfile(os.environ['PROFILE']):
+            opt_flags.append('-fprofile-instr-use='+os.environ['PROFILE'])
+    ldflags = list(opt_flags)
+    if 'LDFLAGS' in os.environ and os.environ['LDFLAGS']:
+        ldflags.extend(os.environ['LDFLAGS'].split(' '))
+    ldflags.extend(['-Wl,-Bdynamic,--as-needed', '-lgcc_s'])
     ext_modules = [Extension(name='bh_sne',
-                   sources=['tsne/bh_sne_src/sptree.cpp', 'tsne/bh_sne_src/tsne.cpp', 'tsne/bh_sne.pyx'],
-                   include_dirs=[numpy.get_include(), '/usr/local/include', 'tsne/bh_sne_src/'],
-                   library_dirs=['/usr/local/lib', '/usr/lib64/atlas'],
-                   extra_compile_args=['-msse3', '-O3', '-fPIC', '-w', '-ffast-math'],
-                   extra_link_args=['-Wl,-Bdynamic,--as-needed', '-lgcc_s'],
-                   language='c++'),
-
-                   Extension(name='bh_sne_3d',
-                   sources=['tsne/bh_sne_src/sptree.cpp', 'tsne/bh_sne_src/tsne.cpp', 'tsne/bh_sne_3d.pyx'],
-                   include_dirs=[numpy.get_include(), '/usr/local/include', 'tsne/bh_sne_src/'],
-                   library_dirs=['/usr/local/lib', '/usr/lib64/atlas'],
-                   extra_compile_args=['-msse3', '-O3', '-fPIC', '-w', '-ffast-math', '-DTSNE3D'],
-                   extra_link_args=['-Wl,-Bdynamic,--as-needed', '-lgcc_s'],
-                   language='c++')]
+                             sources=['tsne/bh_sne_src/sptree.cpp',
+                                      'tsne/bh_sne_src/tsne.cpp', 'tsne/bh_sne.pyx'],
+                             include_dirs=[
+                                 numpy.get_include(), 'tsne/bh_sne_src/'],
+                             extra_compile_args=opt_flags +
+                             ['-Wall', '-fPIC', '-std=c++14', '-w'],
+                             extra_link_args=ldflags,
+                             language='c++'),
+                   ]
 
 ext_modules = cythonize(ext_modules)
 
@@ -77,4 +88,4 @@
       packages=find_packages(),
       ext_modules=ext_modules,
       install_requires=required
-)
+      )

tsne/__init__.py

@@ -1,10 +1,11 @@
 # encoding: utf-8
 from __future__ import division
+
 import numpy as np
 import scipy.linalg as la
 import sys
 from bh_sne import BH_SNE
-from bh_sne_3d import BH_SNE_3D
+
 
 def bh_sne(data, pca_d=None, d=2, perplexity=30., theta=0.5,
            random_state=None, copy_data=False, init=None,
@@ -79,17 +80,16 @@ def bh_sne(data, pca_d=None, d=2, perplexity=30., theta=0.5,
     if mom_switch_iter is None:
         mom_switch_iter = 250
 
-    if d == 2:
-        tsne = BH_SNE()
-    elif d == 3:
-        tsne = BH_SNE_3D()
-    else:
+    if d != 2 and d != 3:
         raise Exception("TSNE dimensions must be 2 or 3")
 
-    Y = tsne.run(X, N, X.shape[1], d, perplexity, theta, seed, init=init, use_init=use_init,
-                 max_iter=max_iter, stop_lying_iter=stop_lying_iter, mom_switch_iter=mom_switch_iter)
+    Y = BH_SNE.run(X, d,
+                   perplexity, theta,
+                   seed, init=init, use_init=use_init,
+                   max_iter=max_iter, stop_lying_iter=stop_lying_iter, mom_switch_iter=mom_switch_iter)
     return Y
 
+
 from ._version import get_versions
 __version__ = get_versions()['version']
 del get_versions

tsne/bh_sne.pyx

@@ -5,24 +5,33 @@ cimport cython
 from libcpp cimport bool
 
 cdef extern from "tsne.h":
-    cdef cppclass TSNE:
-        TSNE()
-        void run(double* X, int N, int D, double* Y, int no_dims, double perplexity, double theta, int rand_seed, bool skip_random_init, double *init, bool use_init, int max_iter, int stop_lying_iter, int mom_switch_iter)
+    void _c_run "run" (
+        double* X, int N, int D, double* Y,
+        int no_dims, double perplexity,
+        double theta, int rand_seed,
+        bool skip_random_init, double *init, bool use_init,
+        int max_iter, int stop_lying_iter, int mom_switch_iter) nogil
 
 cdef class BH_SNE:
-    cdef TSNE* thisptr # hold a C++ instance
-
-    def __cinit__(self):
-        self.thisptr = new TSNE()
-
-    def __dealloc__(self):
-        del self.thisptr
-
     @cython.boundscheck(False)
     @cython.wraparound(False)
-    def run(self, X, N, D, d, perplexity, theta, seed, init, use_init, max_iter, stop_lying_iter, mom_switch_iter):
-        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] _X = np.ascontiguousarray(X)
-        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] _init = np.ascontiguousarray(init)
-        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] Y = np.zeros((N, d), dtype=np.float64)
-        self.thisptr.run(&_X[0,0], N, D, &Y[0,0], d, perplexity, theta, seed, False, &_init[0,0], use_init, max_iter, stop_lying_iter, mom_switch_iter)
+    @staticmethod
+    def run(X, int d,
+            double perplexity, double theta,
+            int seed, init, bool use_init,
+            int max_iter, int stop_lying_iter, int mom_switch_iter):
+        cdef int N = X.shape[0]
+        cdef int D = X.shape[1]
+        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] _X = np.ascontiguousarray(
+                X, dtype=np.float64)
+        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] _init = np.ascontiguousarray(
+                init, dtype=np.float64)
+        cdef np.ndarray[np.float64_t, ndim=2, mode='c'] Y = np.zeros(
+                (N, d), dtype=np.float64, order='C')
+        with nogil:
+            _c_run(&_X[0,0], N, D, &Y[0,0],
+                   d, perplexity,
+                   theta, seed,
+                   False, &_init[0,0], use_init,
+                   max_iter, stop_lying_iter, mom_switch_iter)
         return Y

tsne/bh_sne_src/Makefile

@@ -2,25 +2,22 @@
 #CFLAGS = -march=haswell -ffast-math -O3 -Rpass=loop-vectorize -Rpass-missed=loop-vectorize  -Rpass-analysis=loop-vectorize
 #CFLAGS = -march=haswell -ffast-math -O3
 
-CXX = g++
-CFLAGS = -ffast-math -O3
+CXX?=g++
+CFLAGS?=-ffast-math -O3 -Wall -std=c++14
 
-all: bh_tsne bh_tsne_3d
+all: bh_tsne
 
-bh_tsne: tsne.o sptree.o
-	$(CXX) $(CFLAGS) tsne.o sptree.o -o bh_tsne
-
-bh_tsne_3d: tsne_3d.o sptree.o
-	$(CXX) $(CFLAGS) tsne_3d.o sptree.o -o bh_tsne_3d
+bh_tsne: main.o tsne.o sptree.o
+	$(CXX) $(CFLAGS) $(LDFLAGS) $^ -o $@
 
 sptree.o: sptree.cpp sptree.h
-	$(CXX) $(CFLAGS) -c sptree.cpp
+	$(CXX) $(CFLAGS) -c $<
 
 tsne.o: tsne.cpp tsne.h sptree.h vptree.h
-	$(CXX) $(CFLAGS) -c tsne.cpp
+	$(CXX) $(CFLAGS) -c $<
 
-tsne_3d.o: tsne.cpp tsne.h sptree.h vptree.h
-	$(CXX) $(CFLAGS) -DTSNE3D -c tsne.cpp
+main.o: main.cpp tsne.h
+	$(CXX) $(CFLAGS) -c $<
 
 clean:
-	rm -Rf *.o bh_tsne bh_tsne_3d
+	rm -Rf *.o bh_tsne