[PowerPC] Exclude frexp(long double) on linux

PowerPC on linux currently don't have support for lowering long double for
frexp().  Removing the tests until implementation is provided.

Reviewed By: #libc, amyk, Mordante

Differential Revision: https://reviews.llvm.org/D158547

libcxx/test/libcxx/numerics/c.math/constexpr-cxx23-clang.pass.cpp

@@ -58,9 +58,15 @@ int main(int, char**) {
 
   ASSERT_NOT_CONSTEXPR_CXX23(std::frexp(0.0f, &DummyInt) == 0.0f);
   ASSERT_NOT_CONSTEXPR_CXX23(std::frexp(0.0, &DummyInt) == 0.0);
+//FIXME: currently linux powerpc does not support this expansion
+// since 0.0L lowers to ppcf128 and special handling is required.
+#if !defined(__LONG_DOUBLE_IBM128__)
   ASSERT_NOT_CONSTEXPR_CXX23(std::frexp(0.0L, &DummyInt) == 0.0L);
+#endif
   ASSERT_NOT_CONSTEXPR_CXX23(std::frexpf(0.0f, &DummyInt) == 0.0f);
+#if !defined(__LONG_DOUBLE_IBM128__)
   ASSERT_NOT_CONSTEXPR_CXX23(std::frexpl(0.0L, &DummyInt) == 0.0L);
+#endif
 
   ASSERT_NOT_CONSTEXPR_CXX23(std::ilogb(1.0f) == 0);
   ASSERT_NOT_CONSTEXPR_CXX23(std::ilogb(1.0) == 0);