Fixed fftw guards and temp allocation (#660)

Author: aayushg55

include/matx/core/operator_utils.h

@@ -120,7 +120,7 @@ namespace matx {
     template <typename TensorType, typename Executor, typename ShapeType> 
     __MATX_HOST__ __MATX_INLINE__ void AllocateTempTensor(TensorType &tensor, Executor &&ex, ShapeType &&shape, typename TensorType::scalar_type **ptr) {
       const auto ttl_size = std::accumulate(shape.begin(), shape.end(), static_cast<index_t>(1),
-                                  std::multiplies<index_t>()) * sizeof(*ptr);      
+                                  std::multiplies<index_t>()) * sizeof(typename TensorType::scalar_type);      
       if constexpr (is_cuda_executor_v<Executor>) {
         matxAlloc((void**)ptr, ttl_size, MATX_ASYNC_DEVICE_MEMORY, ex.getStream());
         make_tensor(tensor, *ptr, shape);

include/matx/transforms/fft/fft_fftw.h

@@ -32,6 +32,7 @@
 
 #pragma once
 
+#if MATX_EN_CPU_FFT
 #include "matx/core/cache.h"
 #include "matx/core/error.h"
 #include "matx/core/make_tensor.h"
@@ -560,7 +561,6 @@ template<typename OutTensorType, typename InTensorType> class matxFFTWPlan_t {
                                 [[maybe_unused]] const FftFFTWParams_t &params, 
                                 [[maybe_unused]] detail::FFTDirection dir,
                                 [[maybe_unused]] const HostExecutor<MODE> &exec) {
-  #if MATX_EN_CPU_FFT
     using cache_val_type = detail::matxFFTWPlan_t<OutputTensor, InputTensor>;
     detail::GetCache().LookupAndExec<detail::fft_fftw_cache_t>(
       detail::GetCacheIdFromType<detail::fft_fftw_cache_t>(),
@@ -572,7 +572,6 @@ template<typename OutTensorType, typename InTensorType> class matxFFTWPlan_t {
         ctype->Exec(o, i);
       }
     );
-  #endif
   }
 
   template <typename OutputTensor, typename InputTensor, ThreadsMode MODE>
@@ -760,3 +759,4 @@ template<typename OutTensorType, typename InTensorType> class matxFFTWPlan_t {
 
 
 }; // end namespace matx
+#endif