diff --git a/libcxx/test/std/numerics/numeric.ops/numeric.ops.midpoint/midpoint.float.pass.cpp b/libcxx/test/std/numerics/numeric.ops/numeric.ops.midpoint/midpoint.float.pass.cpp
--- a/libcxx/test/std/numerics/numeric.ops/numeric.ops.midpoint/midpoint.float.pass.cpp
+++ b/libcxx/test/std/numerics/numeric.ops/numeric.ops.midpoint/midpoint.float.pass.cpp
@@ -92,7 +92,7 @@
     assert(d0 < d1);  // sanity checking
     assert(d1 < d2);  // sanity checking
 
-#if defined(__PPC__) && __LONG_DOUBLE_128__ && !__LONG_DOUBLE_IEEE128__
+#if defined(__PPC__) && __LONG_DOUBLE_128__ && (!defined(__LONG_DOUBLE_IEEE128__) || !__LONG_DOUBLE_IEEE128__)
 //	For 128 bit long double implemented as 2 doubles on PowerPC,
 //	nextafterl() of libm gives imprecise results which fails the
 //	midpoint() tests below. So skip the test for this case.