BUG: Fix build on ppc64 when the baseline set to Power9 or higher

Backport of #24806. This backport was already made in #25083, but that didn't make use of the linux_qemu.yml action, so this tweaks it a bit.
numpy · charris · Nov 11, 2023 · Aug 21, 2023 · Aug 24, 2023 · Nov 10, 2023
commit ba1e504629460c91cdfa7cdc32c9182c5829f49d
diff --git a/.github/workflows/linux_qemu.yml b/.github/workflows/linux_qemu.yml
@@ -23,6 +23,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
 
+permissions:
+  contents: read
+
 jobs:
   linux_qemu:
     if: "github.repository == 'numpy/numpy'"
@@ -48,6 +51,13 @@ jobs:
               "-Dallow-noblas=true",
               "test_kind or test_multiarray or test_simd or test_umath or test_ufunc",
           ]
+          - [
+              "ppc64le - baseline(Power9)",
+              "powerpc64le-linux-gnu",
+              "ppc64le/ubuntu:22.04",
+              "-Dallow-noblas=true -Dcpu-baseline=vsx3",
+              "test_kind or test_multiarray or test_simd or test_umath or test_ufunc",
+          ]
           - [
               "s390x",
               "s390x-linux-gnu",
@@ -74,7 +84,7 @@ jobs:
 
     name: "${{ matrix.BUILD_PROP[0] }}"
     steps:
-    - uses: actions/checkout@c85c95e3d7251135ab7dc9ce3241c5835cc595a9 # v3.5.3
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
       with:
         submodules: recursive
         fetch-depth: 0

diff --git a/meson_cpu/ppc64/meson.build b/meson_cpu/ppc64/meson.build
@@ -33,7 +33,7 @@ VSX3 = mod_features.new(
 VSX4 = mod_features.new(
   'VSX4', 4, implies: VSX3, args: {'val': '-mcpu=power10', 'match': '.*[mcpu=|vsx].*'},
   detect: {'val': 'VSX4', 'match': 'VSX.*'},
-  test_code: files(source_root + '/numpy/distutils/checks/cpu_vsx3.c')[0],
+  test_code: files(source_root + '/numpy/distutils/checks/cpu_vsx4.c')[0],
   extra_tests: {
     'VSX4_MMA': files(source_root + '/numpy/distutils/checks/extra_vsx4_mma.c')[0]
   }

diff --git a/numpy/core/src/common/half.hpp b/numpy/core/src/common/half.hpp
@@ -47,7 +47,7 @@ class Half final {
     /// Default constructor. initialize nothing.
     Half() = default;
 
-    /// Constract from float
+    /// Construct from float
     /// If there are no hardware optimization available, rounding will always
     /// be set to ties to even.
     explicit Half(float f)
@@ -118,7 +118,7 @@ class Half final {
     #endif
     }
 
-    /// Returns a new Half constracted from the IEEE 754 binary16.
+    /// Returns a new Half constructed from the IEEE 754 binary16.
     static constexpr Half FromBits(uint16_t bits)
     {
         Half h{};
@@ -131,7 +131,7 @@ class Half final {
         return bits_;
     }
 
-    /// @name Comparison operators (orderd)
+    /// @name Comparison operators (ordered)
     /// @{
     constexpr bool operator==(Half r) const
     {
@@ -155,7 +155,7 @@ class Half final {
     }
     /// @}
 
-    /// @name Comparison operators (unorderd)
+    /// @name Comparison operators (unordered)
     /// @{
     constexpr bool operator!=(Half r) const
     {

diff --git a/numpy/core/tests/test_half.py b/numpy/core/tests/test_half.py
@@ -274,17 +274,17 @@ def test_half_correctness(self):
         if len(a32_fail) != 0:
             bad_index = a32_fail[0]
             assert_equal(self.finite_f32, a_manual,
-                 "First non-equal is half value %x -> %g != %g" %
-                            (self.finite_f16[bad_index],
+                 "First non-equal is half value 0x%x -> %g != %g" %
+                            (a_bits[bad_index],
                              self.finite_f32[bad_index],
                              a_manual[bad_index]))
 
         a64_fail = np.nonzero(self.finite_f64 != a_manual)[0]
         if len(a64_fail) != 0:
             bad_index = a64_fail[0]
             assert_equal(self.finite_f64, a_manual,
-                 "First non-equal is half value %x -> %g != %g" %
-                            (self.finite_f16[bad_index],
+                 "First non-equal is half value 0x%x -> %g != %g" %
+                            (a_bits[bad_index],
                              self.finite_f64[bad_index],
                              a_manual[bad_index]))
 
@@ -327,7 +327,8 @@ def test_half_funcs(self):
         a = np.array([0, 0, -1, -1/1e20, 0, 2.0**-24, 7.629e-6], dtype=float16)
         assert_equal(a.nonzero()[0],
                      [2, 5, 6])
-        a = a.byteswap().newbyteorder()
+        a = a.byteswap()
+        a = a.view(a.dtype.newbyteorder())
         assert_equal(a.nonzero()[0],
                      [2, 5, 6])
 

diff --git a/numpy/distutils/ccompiler_opt.py b/numpy/distutils/ccompiler_opt.py
@@ -1309,7 +1309,7 @@ def feature_names(self, names=None, force_flags=None, macros=[]):
     def feature_is_exist(self, name):
         """
         Returns True if a certain feature is exist and covered within
-        `_Config.conf_features`.
+        ``_Config.conf_features``.
 
         Parameters
         ----------