This change contains the fixes for following issues:-

1) When AMD optimized FFTW (using --enable-amd-opt) is configured with --enable-amd-trans, the hybrid OpenMP+MPI tests were failing.
   With this fix, AMD_OPT_TRANS is made undefined in case of hybrid OpenMP+MPI configuration. So, even using --enable-amd-trans, AMD optimized
   transpose would not be used when running for hybrid OpenMP+MPI configuration. For, single-threaded FFTW, AMD_OPT_TRANS will remain enabled.
2) long double and quad precision tests with --enable-amd-opt were failing. The cpy2d routine will use normal C version now in case of
   long double and quad precision. For single and double precision, AMD optimized cpy2d routine will remained enabled.

Change-Id: I30cdb461bd6d24f5563faba9f4c85b17f1c08006

kernel/cpy2d.c

@@ -38,7 +38,7 @@
 #  define WIDE_TYPE double
 #endif
 
-#ifdef AMD_OPT_ALL //AMD optimized routines
+#if defined(AMD_OPT_ALL) && (!defined(FFTW_LDOUBLE) && !defined(FFTW_QUAD)) //AMD optimized routines
 
 #ifdef FFTW_SINGLE//SINGLE PRECISION CPY2d starts
 #ifdef AMD_OPT_IN_PLACE_1D_CPY2D_STABLE_INTRIN//SIMD optimized function

kernel/ifftw.h

@@ -96,6 +96,10 @@ extern "C"
 //--------------------------------
 //In-place Transpose related optimization switches :-
 //The below switches are defined through config.h using configure script run-time feature arg --enable-amd-trans
+//AMD_OPT_TRANS is currently tested and supported only for single-threaded, so undefining when MPI or openMP used
+#if defined(HAVE_MPI) || defined(HAVE_OPENMP)
+#undef AMD_OPT_TRANS
+#endif
 #ifdef AMD_OPT_TRANS
 #define AMD_OPT_AUTO_TUNED_TRANS_BLK_SIZE
 #define AMD_OPT_AUTO_TUNED_RASTER_TILED_TRANS_METHOD