BUG, SIMD: use scalar cmul on bad Apple clang x86_64 (numpy#24828)

* BUG, SIMD: use scalar cmul on bad Apple clang x86_64

Apple clang 14.0.0 outputs code with non-deterministic behaviour for the
AVX2-accelerated `multiply` ufunc on `complex64` and `complex128` for
x86_64 on macOS with AVX2 enabled.  This bug is fixed by Apple clang
14.0.3, but 14.0.0 is still commonly the available toolchain on CI
images.

In order to not output unsound code, this simply skips the SIMD version
of the ufunc when using an affected compiler.

* MAINT: Formatting cleanup.

---------

Co-authored-by: Charles Harris <[email protected]>

numpy/core/src/umath/loops_arithm_fp.dispatch.c.src

@@ -367,6 +367,16 @@ NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_@kind@)
     }
 #endif
 #if @VECTOR@
+    // Certain versions of Apple clang (commonly used in CI images) produce
+    // non-deterministic output in the mul path with AVX2 enabled on x86_64.
+    // Work around by scalarising.
+    #if @is_mul@ \
+            && defined(NPY_CPU_AMD64) && defined(__clang__) \
+            && defined(__apple_build_version__) \
+            && __apple_build_version__ >= 14000000 \
+            && __apple_build_version__ < 14030000
+        goto loop_scalar;
+    #endif  // end affected Apple clang.
     if (is_mem_overlap(b_src0, b_ssrc0, b_dst, b_sdst, len) ||
         is_mem_overlap(b_src1, b_ssrc1, b_dst, b_sdst, len) ||
         b_sdst  % sizeof(@ftype@) != 0 || b_sdst == 0 ||