Add auto-dispatch specification to Intel packages (#11697)

This PR adds the ability to specify the auto-dispatch targets that can
be used by the Intel compilers. The `-ax` flag will be written to the
respective compiler configuration files. This ability is very handy when
wanting to build optimized builds for various architectures. This PR
does not set any optimization flags, however.

3 files changed, 59 insertions, 1 deletions

diff --git a/lib/spack/spack/build_systems/intel.py b/lib/spack/spack/build_systems/intel.py
index 73cdcd8eaf..0936eef07c 100644
--- a/lib/spack/spack/build_systems/intel.py
+++ b/lib/spack/spack/build_systems/intel.py
@@ -25,11 +25,11 @@ from spack.util.executable import Executable
 from spack.util.prefix import Prefix
 from spack.build_environment import dso_suffix
 
-
 # A couple of utility functions that might be useful in general. If so, they
 # should really be defined elsewhere, unless deemed heretical.
 # (Or na"ive on my part).
 
+
 def debug_print(msg, *args):
     '''Prints a message (usu. a variable) and the callers' names for a couple
     of stack frames.
@@ -115,6 +115,14 @@ class IntelPackage(PackageBase):
         'intel-mpi@5.1:5.99':         2016,
     }
 
+    # Below is the list of possible values for setting auto dispatch functions
+    # for the Intel compilers. Using these allows for the building of fat
+    # binaries that will detect the CPU SIMD capabilities at run time and
+    # activate the appropriate extensions.
+    auto_dispatch_options = ('COMMON-AVX512', 'MIC-AVX512', 'CORE-AVX512',
+                             'CORE-AVX2', 'CORE-AVX-I', 'AVX', 'SSE4.2',
+                             'SSE4.1', 'SSSE3', 'SSE3', 'SSE2')
+
     @property
     def license_required(self):
         # The Intel libraries are provided without requiring a license as of
@@ -1242,6 +1250,30 @@ class IntelPackage(PackageBase):
                 fh.write('-Xlinker -rpath={0}\n'.format(compilers_lib_dir))
 
     @run_after('install')
+    def configure_auto_dispatch(self):
+        if self._has_compilers:
+            if ('auto_dispatch=none' in self.spec):
+                return
+
+            compilers_bin_dir = self.component_bin_dir('compiler')
+
+            for compiler_name in 'icc icpc ifort'.split():
+                f = os.path.join(compilers_bin_dir, compiler_name)
+                if not os.path.isfile(f):
+                    raise InstallError(
+                        'Cannot find compiler command to configure '
+                        'auto_dispatch:\n\t' + f)
+
+                ad = []
+                for x in IntelPackage.auto_dispatch_options:
+                    if 'auto_dispatch={0}'.format(x) in self.spec:
+                        ad.append(x)
+
+                compiler_cfg = os.path.abspath(f + '.cfg')
+                with open(compiler_cfg, 'a') as fh:
+                    fh.write('-ax{0}\n'.format(','.join(ad)))
+
+    @run_after('install')
     def filter_compiler_wrappers(self):
         if (('+mpi' in self.spec or self.provides('mpi')) and
                 '~newdtags' in self.spec):
diff --git a/var/spack/repos/builtin/packages/intel-parallel-studio/package.py b/var/spack/repos/builtin/packages/intel-parallel-studio/package.py
index 73e63d7662..409e642000 100644
--- a/var/spack/repos/builtin/packages/intel-parallel-studio/package.py
+++ b/var/spack/repos/builtin/packages/intel-parallel-studio/package.py
@@ -129,6 +129,13 @@ class IntelParallelStudio(IntelPackage):
         multi=False
     )
 
+    auto_dispatch_options = IntelPackage.auto_dispatch_options
+    variant(
+        'auto_dispatch',
+        values=any_combination_of(*auto_dispatch_options),
+        description='Enable generation of multiple auto-dispatch code paths'
+    )
+
     # Components available in all editions
     variant('daal', default=True,
             description='Install the Intel DAAL libraries')
@@ -186,6 +193,12 @@ class IntelParallelStudio(IntelPackage):
     conflicts('+daal',      when='@cluster.0:cluster.2015.7')
     conflicts('+daal',      when='@composer.0:composer.2015.7')
 
+    # MacOS does not support some of the auto dispatch settings
+    conflicts('auto_dispatch=SSE2', 'platform=darwin',
+              msg='SSE2 is not supported on MacOS')
+    conflicts('auto_dispatch=SSE3', 'platform=darwin target=x86_64',
+              msg='SSE3 is not supported on MacOS x86_64')
+
     def setup_dependent_environment(self, *args):
         # Handle in callback, conveying client's compilers in additional arg.
         # CAUTION - DUP code in:
diff --git a/var/spack/repos/builtin/packages/intel/package.py b/var/spack/repos/builtin/packages/intel/package.py
index 5f723b010f..a611c09fb7 100644
--- a/var/spack/repos/builtin/packages/intel/package.py
+++ b/var/spack/repos/builtin/packages/intel/package.py
@@ -43,5 +43,18 @@ class Intel(IntelPackage):
 
     variant('rpath', default=True, description='Add rpath to .cfg files')
 
+    auto_dispatch_options = IntelPackage.auto_dispatch_options
+    variant(
+        'auto_dispatch',
+        values=any_combination_of(*auto_dispatch_options),
+        description='Enable generation of multiple auto-dispatch code paths'
+    )
+
+    # MacOS does not support some of the auto dispatch settings
+    conflicts('auto_dispatch=SSE2', 'platform=darwin',
+              msg='SSE2 is not supported on MacOS')
+    conflicts('auto_dispatch=SSE3', 'platform=darwin target=x86_64',
+              msg='SSE3 is not supported on MacOS x86_64')
+
     # Since the current package is a subset of 'intel-parallel-studio',
     # all remaining Spack actions are handled in the package class.