Merge pull request #16304 from seiko2plus/issue_16302

TST, MAINT: Fix detecting and testing armhf features

2 files changed, 34 insertions, 9 deletions

diff --git a/numpy/core/src/common/npy_cpu_features.c.src b/numpy/core/src/common/npy_cpu_features.c.src
index e676932c9..bd4743905 100644
--- a/numpy/core/src/common/npy_cpu_features.c.src
+++ b/numpy/core/src/common/npy_cpu_features.c.src
@@ -349,15 +349,17 @@ npy__cpu_init_features_linux(void)
     if ((hwcap2 & NPY__HWCAP2_AES)  || (hwcap2 & NPY__HWCAP2_SHA1)  ||
         (hwcap2 & NPY__HWCAP2_SHA2) || (hwcap2 & NPY__HWCAP2_PMULL) ||
         (hwcap2 & NPY__HWCAP2_CRC32))
+    {
+        hwcap = hwcap2;
 #else
     if (1)
-#endif
     {
         if (!(hwcap & (NPY__HWCAP_FP | NPY__HWCAP_ASIMD))) {
             // Is this could happen? maybe disabled by kernel
             // BTW this will break the baseline of AARCH64
             return 1;
         }
+#endif
         npy__cpu_have[NPY_CPU_FEATURE_FPHP]       = (hwcap & NPY__HWCAP_FPHP)     != 0;
         npy__cpu_have[NPY_CPU_FEATURE_ASIMDHP]    = (hwcap & NPY__HWCAP_ASIMDHP)  != 0;
         npy__cpu_have[NPY_CPU_FEATURE_ASIMDDP]    = (hwcap & NPY__HWCAP_ASIMDDP)  != 0;
diff --git a/numpy/core/tests/test_cpu_features.py b/numpy/core/tests/test_cpu_features.py
index 3b5cb3157..337b7330c 100644
--- a/numpy/core/tests/test_cpu_features.py
+++ b/numpy/core/tests/test_cpu_features.py
@@ -16,23 +16,36 @@ class AbstractTest(object):
     def test_features(self):
         self.load_flags()
         for gname, features in self.features_groups.items():
-            test_features = [self.features_map.get(f, f) in self.features_flags for f in features]
+            test_features = [self.cpu_have(f) for f in features]
             assert_equal(__cpu_features__.get(gname), all(test_features))
 
         for feature_name in self.features:
-            map_name = self.features_map.get(feature_name, feature_name)
-            cpu_have = map_name in self.features_flags
+            cpu_have = self.cpu_have(feature_name)
             npy_have = __cpu_features__.get(feature_name)
             assert_equal(npy_have, cpu_have)
 
-    def load_flags_proc(self, magic_key):
+    def cpu_have(self, feature_name):
+        map_names = self.features_map.get(feature_name, feature_name)
+        if isinstance(map_names, str):
+            return map_names in self.features_flags
+        for f in map_names:
+            if f in self.features_flags:
+                return True
+        return False
+
+    def load_flags_cpuinfo(self, magic_key):
+        self.features_flags = self.get_cpuinfo_item(magic_key)
+
+    def get_cpuinfo_item(self, magic_key):
+        values = set()
         with open('/proc/cpuinfo') as fd:
             for line in fd:
                 if not line.startswith(magic_key):
                     continue
                 flags_value = [s.strip() for s in line.split(':', 1)]
                 if len(flags_value) == 2:
-                    self.features_flags = self.features_flags.union(flags_value[1].upper().split())
+                    values = values.union(flags_value[1].upper().split())
+        return values
 
     def load_flags_auxv(self):
         import subprocess
@@ -75,7 +88,7 @@ class Test_X86_Features(AbstractTest):
         AVX5124FMAPS="AVX512_4FMAPS", AVX5124VNNIW="AVX512_4VNNIW", AVX512VPOPCNTDQ="AVX512_VPOPCNTDQ",
     )
     def load_flags(self):
-        self.load_flags_proc("flags")
+        self.load_flags_cpuinfo("flags")
 
 is_power = re.match("^(powerpc|ppc)64", machine, re.IGNORECASE)
 @pytest.mark.skipif(not is_linux or not is_power, reason="Only for Linux and Power")
@@ -97,8 +110,18 @@ class Test_ARM_Features(AbstractTest):
         NEON_VFPV4 = ["NEON", "VFPV4"],
     )
     def load_flags(self):
-        self.load_flags_proc("Features")
-        if re.match("^(aarch64|AARCH64)", platform.machine()):
+        self.load_flags_cpuinfo("Features")
+        arch = self.get_cpuinfo_item("CPU architecture")
+        # in case of mounting virtual filesystem of aarch64 kernel
+        is_rootfs_v8 = int('0'+next(iter(arch))) > 7 if arch else 0
+        if  re.match("^(aarch64|AARCH64)", machine) or is_rootfs_v8:
             self.features_map = dict(
                 NEON="ASIMD", HALF="ASIMD", VFPV4="ASIMD"
             )
+        else:
+            self.features_map = dict(
+                # ELF auxiliary vector and /proc/cpuinfo on Linux kernel(armv8 aarch32)
+                # doesn't provide information about ASIMD, so we assume that ASIMD is supported
+                # if the kernel reports any one of the following ARM8 features.
+                ASIMD=("AES", "SHA1", "SHA2", "PMULL", "CRC32")
+            )