Microarchitecture specific optimizations for LLVM (#13250)

* Added architecture specific optimization flags for Clang / LLVM
* Disallow compiler optimizations for mixed toolchains
    * We emit a warning when building for a mixed toolchain
* Fixed issues with suffixed versions of compilers; Apple's Clang will, 
    for the time being, fall back on x86-64 for every compilation.

9 files changed, 292 insertions, 10 deletions

diff --git a/lib/spack/llnl/util/cpu/microarchitecture.py b/lib/spack/llnl/util/cpu/microarchitecture.py
index 319eec28c7..e14e1a8c66 100644
--- a/lib/spack/llnl/util/cpu/microarchitecture.py
+++ b/lib/spack/llnl/util/cpu/microarchitecture.py
@@ -224,7 +224,8 @@ class Microarchitecture(object):
             version, _, suffix = version.partition('-')
 
             # If the suffixes are not all equal there's no match
-            if suffix != min_suffix or suffix != max_suffix:
+            if ((suffix != min_suffix and min_version) or
+                (suffix != max_suffix and max_version)):
                 return False
 
             # Assume compiler versions fit into semver
diff --git a/lib/spack/llnl/util/cpu/microarchitectures.json b/lib/spack/llnl/util/cpu/microarchitectures.json
index 88e2bb665f..02f507fa6d 100644
--- a/lib/spack/llnl/util/cpu/microarchitectures.json
+++ b/lib/spack/llnl/util/cpu/microarchitectures.json
@@ -60,7 +60,12 @@
             "name": "x86-64",
             "flags": "-march={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": ":",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu=generic"
+        }
       }
     },
     "nocona": {
@@ -76,6 +81,11 @@
         "gcc": {
           "versions": "4:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -92,6 +102,11 @@
         "gcc": {
           "versions": "4:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -118,7 +133,12 @@
             "name": "corei7",
             "flags": "-march={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "westmere": {
@@ -139,6 +159,11 @@
         "gcc": {
           "versions": "4.9:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -168,7 +193,12 @@
             "name": "corei7-avx",
             "flags": "-march={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "ivybridge": {
@@ -199,7 +229,12 @@
             "name": "core-avx-i",
             "flags": "-march={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "haswell": {
@@ -235,7 +270,12 @@
             "name": "core-avx2",
             "flags": "-march={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "broadwell": {
@@ -266,6 +306,11 @@
         "gcc": {
           "versions": "4.9:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -300,6 +345,11 @@
         "gcc": {
           "versions": "5.3:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -337,6 +387,12 @@
           "versions": "5.1:",
           "name": "knl",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "knl",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -378,6 +434,12 @@
           "name": "skylake-avx512",
           "versions": "5.3:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "skylake-avx512",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -421,6 +483,11 @@
         "gcc": {
           "versions": "8:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -462,6 +529,11 @@
         "gcc": {
           "versions": "9:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "8.0:",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -518,7 +590,20 @@
           "name": "icelake-client",
           "versions": "8:",
           "flags": "-march={name} -mtune={name}"
-        }
+        },
+        "clang": [
+          {
+            "versions": "7.0:",
+            "name": "icelake-client",
+            "family": "x86-64",
+            "flags": "-march={family} -mcpu={name}"
+          },
+          {
+            "versions": "6.0:6.9",
+            "family": "x86-64",
+            "flags": "-march={family} -mcpu={name}"
+          }
+        ]
       }
     },
     "bulldozer": {
@@ -545,6 +630,12 @@
           "name": "bdver1",
           "versions": "4.6:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "bdver1",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -576,6 +667,12 @@
           "name": "bdver2",
           "versions": "4.7:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "bdver2",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -608,6 +705,12 @@
           "name": "bdver3",
           "versions": "4.8:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "bdver3",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -643,6 +746,12 @@
           "name": "bdver4",
           "versions": "4.9:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "name": "bdver4",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -681,6 +790,12 @@
           "name": "znver1",
           "versions": "6:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "4.0:",
+          "name": "znver1",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -720,6 +835,12 @@
           "name": "znver2",
           "versions": "9:",
           "flags": "-march={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "9.0:",
+          "name": "znver2",
+          "family": "x86-64",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -732,6 +853,11 @@
           "name": "powerpc64",
           "versions": "4:",
           "flags": "-mcpu={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": ":",
+          "family": "ppc64",
+          "flags": "-march={family} -mcpu=generic"
         }
       }
     },
@@ -744,6 +870,12 @@
         "gcc": {
           "versions": "4.5:",
           "flags": "-mcpu={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "ppc64",
+          "name": "pwr7",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -763,7 +895,13 @@
             "warnings": "Using GCC 4.8 to optimize for Power 8 might not work if you are not on Red Hat Enterprise Linux 7, where a custom backport of the feature has been done. Upstream support from GCC starts in version 4.9",
             "flags": "-mcpu={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "ppc64",
+          "name": "pwr8",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "power9": {
@@ -775,6 +913,12 @@
         "gcc": {
           "versions": "6:",
           "flags": "-mcpu={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "ppc64",
+          "name": "pwr9",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -787,6 +931,11 @@
           "name": "powerpc64le",
           "versions": "4:",
           "flags": "-mcpu={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": ":",
+          "family": "ppc64le",
+          "flags": "-march={family} -mcpu=generic"
         }
       }
     },
@@ -808,7 +957,13 @@
             "name": "power8",
             "flags": "-mcpu={name} -mtune={name}"
           }
-        ]
+        ],
+        "clang": {
+          "versions": "3.9:",
+          "family": "ppc64le",
+          "name": "pwr8",
+          "flags": "-march={family} -mcpu={name}"
+        }
       }
     },
     "power9le": {
@@ -821,6 +976,12 @@
           "name": "power9",
           "versions": "6:",
           "flags": "-mcpu={name} -mtune={name}"
+        },
+        "clang": {
+          "versions": "3.9:",
+          "family": "ppc64le",
+          "name": "pwr9",
+          "flags": "-march={family} -mcpu={name}"
         }
       }
     },
@@ -832,6 +993,11 @@
         "gcc": {
           "versions": "4:",
           "flags": "-march=armv8-a -mtune=generic"
+        },
+        "clang": {
+          "versions": ":",
+          "family": "aarch64",
+          "flags": "-march={family} -mcpu=generic"
         }
       }
     },
@@ -840,6 +1006,11 @@
       "vendor": "generic",
       "features": [],
       "compilers": {
+        "clang": {
+          "versions": ":",
+          "family": "arm",
+          "flags": "-march={family} -mcpu=generic"
+        }
       }
     },
     "ppc": {
diff --git a/lib/spack/spack/architecture.py b/lib/spack/spack/architecture.py
index 53cbc44ca5..dd99cfa1e3 100644
--- a/lib/spack/spack/architecture.py
+++ b/lib/spack/spack/architecture.py
@@ -58,6 +58,7 @@ will be responsible for compiler detection.
 """
 import functools
 import inspect
+import warnings
 
 import six
 
@@ -184,6 +185,21 @@ class Target(object):
         return cpu_flag in self.microarchitecture
 
     def optimization_flags(self, compiler):
+        """Returns the flags needed to optimize for this target using
+        the compiler passed as argument.
+
+        Args:
+            compiler (CompilerSpec or Compiler): object that contains both the
+                name and the version of the compiler we want to use
+        """
+        if isinstance(compiler, spack.compiler.Compiler):
+            if spack.compilers.is_mixed_toolchain(compiler):
+                msg = ('microarchitecture specific optimizations are not '
+                       'supported yet on mixed compiler toolchains [check'
+                       ' {0.name}@{0.version} for further details]')
+                warnings.warn(msg.format(compiler))
+                return ''
+
         return self.microarchitecture.optimization_flags(
             compiler.name, str(compiler.version)
         )
diff --git a/lib/spack/spack/compilers/__init__.py b/lib/spack/spack/compilers/__init__.py
index 206aa17e1b..bc2e4df391 100644
--- a/lib/spack/spack/compilers/__init__.py
+++ b/lib/spack/spack/compilers/__init__.py
@@ -676,6 +676,44 @@ def make_compiler_list(detected_versions):
     return compilers
 
 
+def is_mixed_toolchain(compiler):
+    """Returns True if the current compiler is a mixed toolchain,
+    False otherwise.
+
+    Args:
+        compiler (Compiler): a valid compiler object
+    """
+    cc = os.path.basename(compiler.cc or '')
+    cxx = os.path.basename(compiler.cxx or '')
+    f77 = os.path.basename(compiler.f77 or '')
+    fc = os.path.basename(compiler.fc or '')
+
+    toolchains = set()
+    for compiler_cls in all_compiler_types():
+        # Inspect all the compiler toolchain we know. If a compiler is the
+        # only compiler supported there it belongs to that toolchain.
+        def name_matches(name, name_list):
+            # This is such that 'gcc' matches variations
+            # like 'ggc-9' etc that are found in distros
+            name, _, _ = name.partition('-')
+            return len(name_list) == 1 and name and name in name_list
+
+        if any([
+            name_matches(cc, compiler_cls.cc_names),
+            name_matches(cxx, compiler_cls.cxx_names),
+            name_matches(f77, compiler_cls.f77_names),
+            name_matches(fc, compiler_cls.fc_names)
+        ]):
+            tty.debug("[TOOLCHAIN] MATCH {0}".format(compiler_cls.__name__))
+            toolchains.add(compiler_cls.__name__)
+
+    if len(toolchains) > 1:
+        tty.debug("[TOOLCHAINS] {0}".format(toolchains))
+        return True
+
+    return False
+
+
 class InvalidCompilerConfigurationError(spack.error.SpackError):
 
     def __init__(self, compiler_spec):
diff --git a/lib/spack/spack/test/architecture.py b/lib/spack/spack/test/architecture.py
index b36219aa76..7e120000d4 100644
--- a/lib/spack/spack/test/architecture.py
+++ b/lib/spack/spack/test/architecture.py
@@ -169,3 +169,21 @@ def test_target_container_semantic(cpu_flag, target_name):
 def test_arch_spec_container_semantic(item, architecture_str):
     architecture = spack.spec.ArchSpec(architecture_str)
     assert item in architecture
+
+
+@pytest.mark.parametrize('compiler_spec,target_name,expected_flags', [
+    # Check compilers with version numbers from a single toolchain
+    ('gcc@4.7.2', 'haswell', '-march=core-avx2 -mtune=core-avx2'),
+    # Check mixed toolchains
+    ('clang@8.0.0', 'broadwell', ''),
+    # Check clang compilers with 'apple' suffix
+    ('clang@9.1.0-apple', 'x86_64', '-march=x86-64 -mcpu=generic')
+])
+@pytest.mark.filterwarnings("ignore:microarchitecture specific")
+def test_optimization_flags(
+        compiler_spec, target_name, expected_flags, config
+):
+    target = spack.architecture.Target(target_name)
+    compiler = spack.compilers.compilers_for_spec(compiler_spec).pop()
+    opt_flags = target.optimization_flags(compiler)
+    assert opt_flags == expected_flags
diff --git a/lib/spack/spack/test/compilers.py b/lib/spack/spack/test/compilers.py
index dc104225f1..ac107170b0 100644
--- a/lib/spack/spack/test/compilers.py
+++ b/lib/spack/spack/test/compilers.py
@@ -469,3 +469,11 @@ def test_cce_version_detection(version_str, expected_version):
 def test_fj_version_detection(version_str, expected_version):
     version = spack.compilers.fj.Fj.extract_version_from_output(version_str)
     assert version == expected_version
+
+
+@pytest.mark.parametrize('compiler_spec,expected_result', [
+    ('gcc@4.7.2', False), ('clang@3.3', False), ('clang@8.0.0', True)
+])
+def test_detecting_mixed_toolchains(compiler_spec, expected_result, config):
+    compiler = spack.compilers.compilers_for_spec(compiler_spec).pop()
+    assert spack.compilers.is_mixed_toolchain(compiler) is expected_result
diff --git a/lib/spack/spack/test/concretize.py b/lib/spack/spack/test/concretize.py
index 5243ba7681..f0c8e93bcf 100644
--- a/lib/spack/spack/test/concretize.py
+++ b/lib/spack/spack/test/concretize.py
@@ -585,7 +585,9 @@ class TestConcretize(object):
 
     @pytest.mark.parametrize('spec, best_achievable', [
         ('mpileaks%gcc@4.8', 'haswell'),
-        ('mpileaks%gcc@5.3.0', 'skylake_avx512')
+        ('mpileaks%gcc@5.3.0', 'skylake_avx512'),
+        # Apple's clang always falls back to x86-64 for now
+        ('mpileaks%clang@9.1.0-apple', 'x86_64')
     ])
     def test_adjusting_default_target_based_on_compiler(
             self, spec, best_achievable, current_host
diff --git a/lib/spack/spack/test/data/compilers.yaml b/lib/spack/spack/test/data/compilers.yaml
index d7092945d5..f619c8d351 100644
--- a/lib/spack/spack/test/data/compilers.yaml
+++ b/lib/spack/spack/test/data/compilers.yaml
@@ -114,3 +114,24 @@ compilers:
       cflags: -O3
       cxxflags: -O3
     modules: 'None'
+- compiler:
+    spec: clang@8.0.0
+    operating_system: redhat7
+    paths:
+      cc: /path/to/clang-8
+      cxx: /path/to/clang++-8
+      f77: /path/to/gfortran-9
+      fc: /path/to/gfortran-9
+    flags:
+      cflags: -O3
+      cxxflags: -O3
+    modules: 'None'
+- compiler:
+    spec: clang@9.1.0-apple
+    operating_system: elcapitan
+    paths:
+      cc: /path/to/clang
+      cxx: /path/to/clang++
+      f77: None
+      fc: None
+    modules: 'None'
diff --git a/lib/spack/spack/test/llnl/util/cpu.py b/lib/spack/spack/test/llnl/util/cpu.py
index 9c65825a52..5713580a72 100644
--- a/lib/spack/spack/test/llnl/util/cpu.py
+++ b/lib/spack/spack/test/llnl/util/cpu.py
@@ -191,6 +191,7 @@ def test_target_json_schema():
 
 
 @pytest.mark.parametrize('target_name,compiler,version,expected_flags', [
+    # Test GCC
     ('x86_64', 'gcc', '4.9.3', '-march=x86-64 -mtune=generic'),
     ('x86_64', 'gcc', '4.2.0', '-march=x86-64 -mtune=generic'),
     ('x86_64', 'gcc', '4.1.1', '-march=x86-64 -mtune=x86-64'),
@@ -198,6 +199,12 @@ def test_target_json_schema():
     ('nehalem', 'gcc', '4.9.3', '-march=nehalem -mtune=nehalem'),
     ('nehalem', 'gcc', '4.8.5', '-march=corei7 -mtune=corei7'),
     ('sandybridge', 'gcc', '4.8.5', '-march=corei7-avx -mtune=corei7-avx'),
+    # Test Clang / LLVM
+    ('sandybridge', 'clang', '3.9.0', '-march=x86-64 -mcpu=sandybridge'),
+    ('icelake', 'clang', '6.0.0', '-march=x86-64 -mcpu=icelake'),
+    ('icelake', 'clang', '8.0.0', '-march=x86-64 -mcpu=icelake-client'),
+    ('zen2', 'clang', '9.0.0', '-march=x86-64 -mcpu=znver2'),
+    ('power9le', 'clang', '8.0.0', '-march=ppc64le -mcpu=pwr9'),
     # Test that an unknown compiler returns an empty string
     ('sandybridge', 'unknown', '4.8.5', ''),
 ])