ENH: Expose the NumPy C SIMD vectorization interface "NPYV" to Python

  '_simd' is a new module to bring the NumPy C SIMD vectorization interface "NPYV"

  The module is designed to be extremely flexible so that it can accommodate any kind
  intrinsics, also to generate a python interface almost similar to the C interface.

  The main purpose of this module is to test NPYV intrinsics in python,
  but still can be used as an effective solution in designing SIMD kernels.

  Also add a new command-line argument `--simd-test` to control of targeted CPU features
  for the `_simd` module.

Co-authored-by: Matti Picus <matti.picus@gmail.com>
Co-authored-by: Eric Wieser <wieser.eric@gmail.com>

2 files changed, 18 insertions, 2 deletions

diff --git a/numpy/distutils/command/build.py b/numpy/distutils/command/build.py
index 60ba4c917..6025586cd 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 SSE41 SSE42 XOP (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_ext.py b/numpy/distutils/command/build_ext.py
index 1a881c56a..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: