Merge branch 'master' into fix-issue-10244

5 files changed, 119 insertions, 76 deletions

diff --git a/numpy/distutils/command/autodist.py b/numpy/distutils/command/autodist.py
index 8f6436004..b72d0cab1 100644
--- a/numpy/distutils/command/autodist.py
+++ b/numpy/distutils/command/autodist.py
@@ -46,15 +46,16 @@ def check_restrict(cmd):
     return ''
 
 
-def check_compiler_gcc4(cmd):
-    """Return True if the C compiler is GCC 4.x."""
+def check_compiler_gcc(cmd):
+    """Check if the compiler is GCC."""
+
     cmd._check_compiler()
     body = textwrap.dedent("""
         int
         main()
         {
-        #if (! defined __GNUC__) || (__GNUC__ < 4)
-        #error gcc >= 4 required
+        #if (! defined __GNUC__)
+        #error gcc required
         #endif
             return 0;
         }
@@ -62,6 +63,30 @@ def check_compiler_gcc4(cmd):
     return cmd.try_compile(body, None, None)
 
 
+def check_gcc_version_at_least(cmd, major, minor=0, patchlevel=0):
+    """
+    Check that the gcc version is at least the specified version."""
+
+    cmd._check_compiler()
+    version = '.'.join([str(major), str(minor), str(patchlevel)])
+    body = textwrap.dedent("""
+        int
+        main()
+        {
+        #if (! defined __GNUC__) || (__GNUC__ < %(major)d) || \\
+                (__GNUC_MINOR__ < %(minor)d) || \\
+                (__GNUC_PATCHLEVEL__ < %(patchlevel)d)
+        #error gcc >= %(version)s required
+        #endif
+            return 0;
+        }
+        """)
+    kw = {'version': version, 'major': major, 'minor': minor,
+          'patchlevel': patchlevel}
+
+    return cmd.try_compile(body % kw, None, None)
+
+
 def check_gcc_function_attribute(cmd, attribute, name):
     """Return True if the given function attribute is supported."""
     cmd._check_compiler()
diff --git a/numpy/distutils/command/build.py b/numpy/distutils/command/build.py
index 60ba4c917..a4fda537d 100644
--- a/numpy/distutils/command/build.py
+++ b/numpy/distutils/command/build.py
@@ -22,6 +22,8 @@ class build(old_build):
          "specify a list of dispatched CPU optimizations"),
         ('disable-optimization', None,
          "disable CPU optimized code(dispatch,simd,fast...)"),
+        ('simd-test=', None,
+         "specify a list of CPU optimizations to be tested against NumPy SIMD interface"),
         ]
 
     help_options = old_build.help_options + [
@@ -36,6 +38,16 @@ class build(old_build):
         self.cpu_baseline = "min"
         self.cpu_dispatch = "max -xop -fma4" # drop AMD legacy features by default
         self.disable_optimization = False
+        """
+        the '_simd' module is a very large. Adding more dispatched features
+        will increase binary size and compile time. By default we minimize
+        the targeted features to those most commonly used by the NumPy SIMD interface(NPYV),
+        NOTE: any specified features will be ignored if they're:
+            - part of the baseline(--cpu-baseline)
+            - not part of dispatch-able features(--cpu-dispatch)
+            - not supported by compiler or platform
+        """
+        self.simd_test = "BASELINE SSE2 SSE42 XOP FMA4 (FMA3 AVX2) AVX512F AVX512_SKX VSX VSX2 VSX3 NEON ASIMD"
 
     def finalize_options(self):
         build_scripts = self.build_scripts
diff --git a/numpy/distutils/command/build_clib.py b/numpy/distutils/command/build_clib.py
index 87345adbc..a0db6f31f 100644
--- a/numpy/distutils/command/build_clib.py
+++ b/numpy/distutils/command/build_clib.py
@@ -259,57 +259,56 @@ class build_clib(old_build_clib):
         if requiref90:
             self.mkpath(module_build_dir)
 
-        dispatch_objects = []
-        if not self.disable_optimization:
-            dispatch_sources  = [
-                c_sources.pop(c_sources.index(src))
-                for src in c_sources[:] if src.endswith(".dispatch.c")
-            ]
-            if dispatch_sources:
-                if not self.inplace:
-                    build_src = self.get_finalized_command("build_src").build_src
-                else:
-                    build_src = None
-                dispatch_objects = self.compiler_opt.try_dispatch(
-                    dispatch_sources,
-                    output_dir=self.build_temp,
-                    src_dir=build_src,
-                    macros=macros,
-                    include_dirs=include_dirs,
-                    debug=self.debug,
-                    extra_postargs=extra_postargs
-                )
-            extra_args_baseopt = extra_postargs + self.compiler_opt.cpu_baseline_flags()
-        else:
-            extra_args_baseopt = extra_postargs
-            macros.append(("NPY_DISABLE_OPTIMIZATION", 1))
-
         if compiler.compiler_type == 'msvc':
             # this hack works around the msvc compiler attributes
             # problem, msvc uses its own convention :(
             c_sources += cxx_sources
             cxx_sources = []
 
+        # filtering C dispatch-table sources when optimization is not disabled,
+        # otherwise treated as normal sources.
+        copt_c_sources = []
+        copt_baseline_flags = []
+        copt_macros = []
+        if not self.disable_optimization:
+            copt_build_src = None if self.inplace else self.get_finalized_command("build_src").build_src
+            copt_c_sources = [
+                c_sources.pop(c_sources.index(src))
+                for src in c_sources[:] if src.endswith(".dispatch.c")
+            ]
+            copt_baseline_flags = self.compiler_opt.cpu_baseline_flags()
+        else:
+            copt_macros.append(("NPY_DISABLE_OPTIMIZATION", 1))
+
         objects = []
+        if copt_c_sources:
+            log.info("compiling C dispatch-able sources")
+            objects += self.compiler_opt.try_dispatch(copt_c_sources,
+                                                      output_dir=self.build_temp,
+                                                      src_dir=copt_build_src,
+                                                      macros=macros + copt_macros,
+                                                      include_dirs=include_dirs,
+                                                      debug=self.debug,
+                                                      extra_postargs=extra_postargs)
+
         if c_sources:
             log.info("compiling C sources")
-            objects = compiler.compile(c_sources,
-                                       output_dir=self.build_temp,
-                                       macros=macros,
-                                       include_dirs=include_dirs,
-                                       debug=self.debug,
-                                       extra_postargs=extra_args_baseopt)
-        objects.extend(dispatch_objects)
+            objects += compiler.compile(c_sources,
+                                        output_dir=self.build_temp,
+                                        macros=macros + copt_macros,
+                                        include_dirs=include_dirs,
+                                        debug=self.debug,
+                                        extra_postargs=extra_postargs + copt_baseline_flags)
 
         if cxx_sources:
             log.info("compiling C++ sources")
             cxx_compiler = compiler.cxx_compiler()
             cxx_objects = cxx_compiler.compile(cxx_sources,
                                                output_dir=self.build_temp,
-                                               macros=macros,
+                                               macros=macros + copt_macros,
                                                include_dirs=include_dirs,
                                                debug=self.debug,
-                                               extra_postargs=extra_postargs)
+                                               extra_postargs=extra_postargs + copt_baseline_flags)
             objects.extend(cxx_objects)
 
         if f_sources or fmodule_sources:
diff --git a/numpy/distutils/command/build_ext.py b/numpy/distutils/command/build_ext.py
index b6557fcf6..ca6f8bcd2 100644
--- a/numpy/distutils/command/build_ext.py
+++ b/numpy/distutils/command/build_ext.py
@@ -19,8 +19,7 @@ from numpy.distutils.misc_util import (
     has_cxx_sources, has_f_sources, is_sequence
 )
 from numpy.distutils.command.config_compiler import show_fortran_compilers
-from numpy.distutils.ccompiler_opt import new_ccompiler_opt
-
+from numpy.distutils.ccompiler_opt import new_ccompiler_opt, CCompilerOpt
 
 class build_ext (old_build_ext):
 
@@ -39,6 +38,8 @@ class build_ext (old_build_ext):
          "specify a list of dispatched CPU optimizations"),
         ('disable-optimization', None,
          "disable CPU optimized code(dispatch,simd,fast...)"),
+        ('simd-test=', None,
+         "specify a list of CPU optimizations to be tested against NumPy SIMD interface"),
     ]
 
     help_options = old_build_ext.help_options + [
@@ -56,6 +57,7 @@ class build_ext (old_build_ext):
         self.cpu_baseline = None
         self.cpu_dispatch = None
         self.disable_optimization = None
+        self.simd_test = None
 
     def finalize_options(self):
         if self.parallel:
@@ -87,7 +89,9 @@ class build_ext (old_build_ext):
                                         ('cpu_baseline', 'cpu_baseline'),
                                         ('cpu_dispatch', 'cpu_dispatch'),
                                         ('disable_optimization', 'disable_optimization'),
+                                        ('simd_test', 'simd_test')
                                   )
+        CCompilerOpt.conf_target_groups["simd_test"] = self.simd_test
 
     def run(self):
         if not self.extensions:
@@ -406,52 +410,49 @@ class build_ext (old_build_ext):
 
         include_dirs = ext.include_dirs + get_numpy_include_dirs()
 
-        dispatch_objects = []
+        # filtering C dispatch-table sources when optimization is not disabled,
+        # otherwise treated as normal sources.
+        copt_c_sources = []
+        copt_baseline_flags = []
+        copt_macros = []
         if not self.disable_optimization:
-            dispatch_sources  = [
+            copt_build_src = None if self.inplace else self.get_finalized_command("build_src").build_src
+            copt_c_sources = [
                 c_sources.pop(c_sources.index(src))
                 for src in c_sources[:] if src.endswith(".dispatch.c")
             ]
-            if dispatch_sources:
-                if not self.inplace:
-                    build_src = self.get_finalized_command("build_src").build_src
-                else:
-                    build_src = None
-                dispatch_objects = self.compiler_opt.try_dispatch(
-                    dispatch_sources,
-                    output_dir=output_dir,
-                    src_dir=build_src,
-                    macros=macros,
-                    include_dirs=include_dirs,
-                    debug=self.debug,
-                    extra_postargs=extra_args,
-                    **kws
-                )
-            extra_args_baseopt = extra_args + self.compiler_opt.cpu_baseline_flags()
+            copt_baseline_flags = self.compiler_opt.cpu_baseline_flags()
         else:
-            extra_args_baseopt = extra_args
-            macros.append(("NPY_DISABLE_OPTIMIZATION", 1))
+            copt_macros.append(("NPY_DISABLE_OPTIMIZATION", 1))
 
         c_objects = []
+        if copt_c_sources:
+            log.info("compiling C dispatch-able sources")
+            c_objects += self.compiler_opt.try_dispatch(copt_c_sources,
+                                                        output_dir=output_dir,
+                                                        src_dir=copt_build_src,
+                                                        macros=macros + copt_macros,
+                                                        include_dirs=include_dirs,
+                                                        debug=self.debug,
+                                                        extra_postargs=extra_args,
+                                                        **kws)
         if c_sources:
             log.info("compiling C sources")
-            c_objects = self.compiler.compile(c_sources,
-                                              output_dir=output_dir,
-                                              macros=macros,
-                                              include_dirs=include_dirs,
-                                              debug=self.debug,
-                                              extra_postargs=extra_args_baseopt,
-                                              **kws)
-        c_objects.extend(dispatch_objects)
-
+            c_objects += self.compiler.compile(c_sources,
+                                               output_dir=output_dir,
+                                               macros=macros + copt_macros,
+                                               include_dirs=include_dirs,
+                                               debug=self.debug,
+                                               extra_postargs=extra_args + copt_baseline_flags,
+                                               **kws)
         if cxx_sources:
             log.info("compiling C++ sources")
             c_objects += cxx_compiler.compile(cxx_sources,
                                               output_dir=output_dir,
-                                              macros=macros,
+                                              macros=macros + copt_macros,
                                               include_dirs=include_dirs,
                                               debug=self.debug,
-                                              extra_postargs=extra_args,
+                                              extra_postargs=extra_args + copt_baseline_flags,
                                               **kws)
 
         extra_postargs = []
@@ -559,7 +560,7 @@ class build_ext (old_build_ext):
         unlinkable_fobjects = list(unlinkable_fobjects)
 
         # Expand possible fake static libraries to objects
-        for lib in list(libraries):
+        for lib in libraries:
             for libdir in library_dirs:
                 fake_lib = os.path.join(libdir, lib + '.fobjects')
                 if os.path.isfile(fake_lib):
diff --git a/numpy/distutils/command/config.py b/numpy/distutils/command/config.py
index e54a54449..60881f4a3 100644
--- a/numpy/distutils/command/config.py
+++ b/numpy/distutils/command/config.py
@@ -20,9 +20,10 @@ from numpy.distutils.mingw32ccompiler import generate_manifest
 from numpy.distutils.command.autodist import (check_gcc_function_attribute,
                                               check_gcc_function_attribute_with_intrinsics,
                                               check_gcc_variable_attribute,
+                                              check_gcc_version_at_least,
                                               check_inline,
                                               check_restrict,
-                                              check_compiler_gcc4)
+                                              check_compiler_gcc)
 
 LANG_EXT['f77'] = '.f'
 LANG_EXT['f90'] = '.f90'
@@ -416,9 +417,9 @@ class config(old_config):
         otherwise."""
         return check_restrict(self)
 
-    def check_compiler_gcc4(self):
-        """Return True if the C compiler is gcc >= 4."""
-        return check_compiler_gcc4(self)
+    def check_compiler_gcc(self):
+        """Return True if the C compiler is gcc"""
+        return check_compiler_gcc(self)
 
     def check_gcc_function_attribute(self, attribute, name):
         return check_gcc_function_attribute(self, attribute, name)
@@ -431,6 +432,11 @@ class config(old_config):
     def check_gcc_variable_attribute(self, attribute):
         return check_gcc_variable_attribute(self, attribute)
 
+    def check_gcc_version_at_least(self, major, minor=0, patchlevel=0):
+        """Return True if the GCC version is greater than or equal to the
+        specified version."""
+        return check_gcc_version_at_least(self, major, minor, patchlevel)
+
     def get_output(self, body, headers=None, include_dirs=None,
                    libraries=None, library_dirs=None,
                    lang="c", use_tee=None):