Merge pull request #19098 from seiko2plus/issue_19084

BUG, SIMD: Fix detect host/native CPU features on ICC during compile-time

20 files changed, 284 insertions, 22 deletions

diff --git a/numpy/distutils/ccompiler_opt.py b/numpy/distutils/ccompiler_opt.py
index ae1e6a180..47d07ad4a 100644
--- a/numpy/distutils/ccompiler_opt.py
+++ b/numpy/distutils/ccompiler_opt.py
@@ -552,7 +552,7 @@ class _Distutils:
             ccompiler = self._ccompiler
         return ccompiler.compile(sources, extra_postargs=flags, **kwargs)
 
-    def dist_test(self, source, flags):
+    def dist_test(self, source, flags, macros=[]):
         """Return True if 'CCompiler.compile()' able to compile
         a source file with certain flags.
         """
@@ -569,7 +569,7 @@ class _Distutils:
         test = False
         try:
             self.dist_compile(
-                [source], flags, output_dir=self.conf_tmp_path
+                [source], flags, macros=macros, output_dir=self.conf_tmp_path
             )
             test = True
         except CompileError as e:
@@ -1172,20 +1172,23 @@ class _Feature:
 
         self.feature_is_cached = True
 
-    def feature_names(self, names=None, force_flags=None):
+    def feature_names(self, names=None, force_flags=None, macros=[]):
         """
         Returns a set of CPU feature names that supported by platform and the **C** compiler.
 
         Parameters
         ----------
-        'names': sequence or None, optional
+        names: sequence or None, optional
             Specify certain CPU features to test it against the **C** compiler.
             if None(default), it will test all current supported features.
             **Note**: feature names must be in upper-case.
 
-        'force_flags': list or None, optional
-            If None(default), default compiler flags for every CPU feature will be used
-            during the test.
+        force_flags: list or None, optional
+            If None(default), default compiler flags for every CPU feature will
+            be used during the test.
+
+        macros : list of tuples, optional
+            A list of C macro definitions.
         """
         assert(
             names is None or (
@@ -1198,7 +1201,9 @@ class _Feature:
             names = self.feature_supported.keys()
         supported_names = set()
         for f in names:
-            if self.feature_is_supported(f, force_flags=force_flags):
+            if self.feature_is_supported(
+                f, force_flags=force_flags, macros=macros
+            ):
                 supported_names.add(f)
         return supported_names
 
@@ -1433,20 +1438,23 @@ class _Feature:
         return self.cc_normalize_flags(flags)
 
     @_Cache.me
-    def feature_test(self, name, force_flags=None):
+    def feature_test(self, name, force_flags=None, macros=[]):
         """
         Test a certain CPU feature against the compiler through its own
         check file.
 
         Parameters
         ----------
-        'name': str
+        name: str
             Supported CPU feature name.
 
-        'force_flags': list or None, optional
+        force_flags: list or None, optional
             If None(default), the returned flags from `feature_flags()`
             will be used.
-       """
+
+        macros : list of tuples, optional
+            A list of C macro definitions.
+        """
         if force_flags is None:
             force_flags = self.feature_flags(name)
 
@@ -1462,24 +1470,29 @@ class _Feature:
         if not os.path.exists(test_path):
             self.dist_fatal("feature test file is not exist", test_path)
 
-        test = self.dist_test(test_path, force_flags + self.cc_flags["werror"])
+        test = self.dist_test(
+            test_path, force_flags + self.cc_flags["werror"], macros=macros
+        )
         if not test:
             self.dist_log("testing failed", stderr=True)
         return test
 
     @_Cache.me
-    def feature_is_supported(self, name, force_flags=None):
+    def feature_is_supported(self, name, force_flags=None, macros=[]):
         """
         Check if a certain CPU feature is supported by the platform and compiler.
 
         Parameters
         ----------
-        'name': str
+        name: str
             CPU feature name in uppercase.
 
-        'force_flags': list or None, optional
-            If None(default), default compiler flags for every CPU feature will be used
-            during test.
+        force_flags: list or None, optional
+            If None(default), default compiler flags for every CPU feature will
+            be used during test.
+
+        macros : list of tuples, optional
+            A list of C macro definitions.
         """
         assert(name.isupper())
         assert(force_flags is None or isinstance(force_flags, list))
@@ -1487,9 +1500,9 @@ class _Feature:
         supported = name in self.feature_supported
         if supported:
             for impl in self.feature_implies(name):
-                if not self.feature_test(impl, force_flags):
+                if not self.feature_test(impl, force_flags, macros=macros):
                     return False
-            if not self.feature_test(name, force_flags):
+            if not self.feature_test(name, force_flags, macros=macros):
                 return False
         return supported
 
@@ -1812,7 +1825,9 @@ class _Parse:
                     self.dist_fatal(arg_name,
                         "native option isn't supported by the compiler"
                     )
-                features_to = self.feature_names(force_flags=native)
+                features_to = self.feature_names(
+                    force_flags=native, macros=[("DETECT_FEATURES", 1)]
+                )
             elif TOK == "MAX":
                 features_to = self.feature_supported.keys()
             elif TOK == "MIN":
diff --git a/numpy/distutils/checks/cpu_avx.c b/numpy/distutils/checks/cpu_avx.c
index cee4f36ab..26ae18466 100644
--- a/numpy/distutils/checks/cpu_avx.c
+++ b/numpy/distutils/checks/cpu_avx.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __AVX__
+        #error "HOST/ARCH doesn't support AVX"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx2.c b/numpy/distutils/checks/cpu_avx2.c
index 15b6c919b..ddde868f1 100644
--- a/numpy/distutils/checks/cpu_avx2.c
+++ b/numpy/distutils/checks/cpu_avx2.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __AVX2__
+        #error "HOST/ARCH doesn't support AVX2"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_clx.c b/numpy/distutils/checks/cpu_avx512_clx.c
index 4baa8fea0..81edcd067 100644
--- a/numpy/distutils/checks/cpu_avx512_clx.c
+++ b/numpy/distutils/checks/cpu_avx512_clx.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __AVX512VNNI__
+        #error "HOST/ARCH doesn't support CascadeLake AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_cnl.c b/numpy/distutils/checks/cpu_avx512_cnl.c
index f2ff3725e..5799f122b 100644
--- a/numpy/distutils/checks/cpu_avx512_cnl.c
+++ b/numpy/distutils/checks/cpu_avx512_cnl.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__AVX512VBMI__) || !defined(__AVX512IFMA__)
+        #error "HOST/ARCH doesn't support CannonLake AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_icl.c b/numpy/distutils/checks/cpu_avx512_icl.c
index 085b947e0..3cf44d731 100644
--- a/numpy/distutils/checks/cpu_avx512_icl.c
+++ b/numpy/distutils/checks/cpu_avx512_icl.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__AVX512VPOPCNTDQ__) || !defined(__AVX512BITALG__) || !defined(__AVX512VPOPCNTDQ__)
+        #error "HOST/ARCH doesn't support IceLake AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_knl.c b/numpy/distutils/checks/cpu_avx512_knl.c
index 10ba52bcc..b3f4f6976 100644
--- a/numpy/distutils/checks/cpu_avx512_knl.c
+++ b/numpy/distutils/checks/cpu_avx512_knl.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__AVX512ER__) || !defined(__AVX512PF__)
+        #error "HOST/ARCH doesn't support Knights Landing AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_knm.c b/numpy/distutils/checks/cpu_avx512_knm.c
index d03b0fe8b..2c426462b 100644
--- a/numpy/distutils/checks/cpu_avx512_knm.c
+++ b/numpy/distutils/checks/cpu_avx512_knm.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__AVX5124FMAPS__) || !defined(__AVX5124VNNIW__) || !defined(__AVX512VPOPCNTDQ__)
+        #error "HOST/ARCH doesn't support Knights Mill AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512_skx.c b/numpy/distutils/checks/cpu_avx512_skx.c
index 047618762..8840efb7e 100644
--- a/numpy/distutils/checks/cpu_avx512_skx.c
+++ b/numpy/distutils/checks/cpu_avx512_skx.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__AVX512VL__) || !defined(__AVX512BW__) || !defined(__AVX512DQ__)
+        #error "HOST/ARCH doesn't support SkyLake AVX512 features"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512cd.c b/numpy/distutils/checks/cpu_avx512cd.c
index 52f4c7f8b..5e29c79e3 100644
--- a/numpy/distutils/checks/cpu_avx512cd.c
+++ b/numpy/distutils/checks/cpu_avx512cd.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __AVX512CD__
+        #error "HOST/ARCH doesn't support AVX512CD"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_avx512f.c b/numpy/distutils/checks/cpu_avx512f.c
index 22d861471..d0eb7b1ad 100644
--- a/numpy/distutils/checks/cpu_avx512f.c
+++ b/numpy/distutils/checks/cpu_avx512f.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __AVX512F__
+        #error "HOST/ARCH doesn't support AVX512F"
+    #endif
+#endif
+
 #include <immintrin.h>
 
 int main(int argc, char **argv)
diff --git a/numpy/distutils/checks/cpu_f16c.c b/numpy/distutils/checks/cpu_f16c.c
index 678c582e4..fdf36cec5 100644
--- a/numpy/distutils/checks/cpu_f16c.c
+++ b/numpy/distutils/checks/cpu_f16c.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __F16C__
+        #error "HOST/ARCH doesn't support F16C"
+    #endif
+#endif
+
 #include <emmintrin.h>
 #include <immintrin.h>
 
diff --git a/numpy/distutils/checks/cpu_fma3.c b/numpy/distutils/checks/cpu_fma3.c
index 2f879c3b3..bfeef22b5 100644
--- a/numpy/distutils/checks/cpu_fma3.c
+++ b/numpy/distutils/checks/cpu_fma3.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__FMA__) && !defined(__AVX2__)
+        #error "HOST/ARCH doesn't support FMA3"
+    #endif
+#endif
+
 #include <xmmintrin.h>
 #include <immintrin.h>
 
diff --git a/numpy/distutils/checks/cpu_popcnt.c b/numpy/distutils/checks/cpu_popcnt.c
index 540c98dab..813c461f0 100644
--- a/numpy/distutils/checks/cpu_popcnt.c
+++ b/numpy/distutils/checks/cpu_popcnt.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env vr `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #if !defined(__SSE4_2__) && !defined(__POPCNT__)
+        #error "HOST/ARCH doesn't support POPCNT"
+    #endif
+#endif
+
 #ifdef _MSC_VER
     #include <nmmintrin.h>
 #else
@@ -11,7 +24,7 @@ int main(int argc, char **argv)
     unsigned long long a = *((unsigned long long*)argv[argc-1]);
     unsigned int b = *((unsigned int*)argv[argc-2]);
 
-#if defined(_M_X64) || defined(__x86_64__)    
+#if defined(_M_X64) || defined(__x86_64__)
     a = _mm_popcnt_u64(a);
 #endif
     b = _mm_popcnt_u32(b);
diff --git a/numpy/distutils/checks/cpu_sse.c b/numpy/distutils/checks/cpu_sse.c
index bb98bf63c..602b74e7b 100644
--- a/numpy/distutils/checks/cpu_sse.c
+++ b/numpy/distutils/checks/cpu_sse.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSE__
+        #error "HOST/ARCH doesn't support SSE"
+    #endif
+#endif
+
 #include <xmmintrin.h>
 
 int main(void)
diff --git a/numpy/distutils/checks/cpu_sse2.c b/numpy/distutils/checks/cpu_sse2.c
index 658afc9b4..33826a9ed 100644
--- a/numpy/distutils/checks/cpu_sse2.c
+++ b/numpy/distutils/checks/cpu_sse2.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSE2__
+        #error "HOST/ARCH doesn't support SSE2"
+    #endif
+#endif
+
 #include <emmintrin.h>
 
 int main(void)
diff --git a/numpy/distutils/checks/cpu_sse3.c b/numpy/distutils/checks/cpu_sse3.c
index aece1e601..d47c20f74 100644
--- a/numpy/distutils/checks/cpu_sse3.c
+++ b/numpy/distutils/checks/cpu_sse3.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSE3__
+        #error "HOST/ARCH doesn't support SSE3"
+    #endif
+#endif
+
 #include <pmmintrin.h>
 
 int main(void)
diff --git a/numpy/distutils/checks/cpu_sse41.c b/numpy/distutils/checks/cpu_sse41.c
index bfdb9feac..7c80238a3 100644
--- a/numpy/distutils/checks/cpu_sse41.c
+++ b/numpy/distutils/checks/cpu_sse41.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSE4_1__
+        #error "HOST/ARCH doesn't support SSE41"
+    #endif
+#endif
+
 #include <smmintrin.h>
 
 int main(void)
diff --git a/numpy/distutils/checks/cpu_sse42.c b/numpy/distutils/checks/cpu_sse42.c
index 24f5d93fe..f60e18f3c 100644
--- a/numpy/distutils/checks/cpu_sse42.c
+++ b/numpy/distutils/checks/cpu_sse42.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSE4_2__
+        #error "HOST/ARCH doesn't support SSE42"
+    #endif
+#endif
+
 #include <smmintrin.h>
 
 int main(void)
diff --git a/numpy/distutils/checks/cpu_ssse3.c b/numpy/distutils/checks/cpu_ssse3.c
index ad0abc1e6..fde390d6a 100644
--- a/numpy/distutils/checks/cpu_ssse3.c
+++ b/numpy/distutils/checks/cpu_ssse3.c
@@ -1,3 +1,16 @@
+#if defined(DETECT_FEATURES) && defined(__INTEL_COMPILER)
+    /*
+     * Unlike GCC and CLANG, Intel Compiler exposes all supported intrinsics,
+     * whether or not the build options for those features are specified.
+     * Therefore, we must test #definitions of CPU features when option native/host
+     * is enabled via `--cpu-baseline` or through env var `CFLAGS` otherwise
+     * the test will be broken and leads to enable all possible features.
+     */
+    #ifndef __SSSE3__
+        #error "HOST/ARCH doesn't support SSSE3"
+    #endif
+#endif
+
 #include <tmmintrin.h>
 
 int main(void)