Fixes for clang17 errors/warnings (#815)

include/matx/operators/base_operator.h

@@ -73,7 +73,7 @@ namespace matx
             tp->Exec(ex);
           }          
           else if constexpr (is_matx_set_op<T>()) {
-            if constexpr (static_cast<const T *>(this)->IsTransformSet()) {
+            if constexpr (is_matx_transform_op<typename T::op_type>() && is_tensor_view_v<typename T::tensor_type>) {
               tp->TransformExec(tp->Shape(), ex);
             }
             else {

include/matx/operators/concat.h

@@ -79,7 +79,7 @@ namespace matx
       {
         static_assert(RANK > 0, "Cannot concatenate rank-0 tensors");
         static_assert(sizeof...(Ts) > 1, "Must have more than one tensor to concatenate");
-        static_assert((... && (RANK == ts.Rank())), "concatenated ops must have the same rank");
+        static_assert((... && (RANK == Ts::Rank())), "concatenated ops must have the same rank");
 
         for (int32_t i = 0; i < RANK; i++) {
           if(i == axis_) {

include/matx/operators/isclose.h

@@ -55,7 +55,7 @@ namespace matx
         __MATX_INLINE__ IsCloseOp(const Op1 &op1, const Op2 &op2, double rtol, double atol) : 
           op1_(op1), op2_(op2), rtol_(static_cast<inner_type>(rtol)), atol_(static_cast<inner_type>(atol)) 
         {
-          static_assert(op1.Rank() == op2.Rank(), "Operator ranks must match in isclose()");
+          static_assert(Op1::Rank() == Op2::Rank(), "Operator ranks must match in isclose()");
           ASSERT_COMPATIBLE_OP_SIZES(op1); 
           ASSERT_COMPATIBLE_OP_SIZES(op2);
         }

include/matx/operators/stack.h

@@ -77,7 +77,7 @@ namespace matx
       __MATX_INLINE__ StackOp(int axis, const Ts&... ts) : ops_(ts...), axis_(axis)
       {
         static_assert(sizeof...(Ts) > 1, "Must have more than one tensor to stack");
-        static_assert((... && (RANK == ts.Rank())), "stacked ops must have the same rank");
+        static_assert((... && (RANK == Ts::Rank())), "stacked ops must have the same rank");
 
         for (int32_t i = 0; i < RANK; i++) {
           MATX_ASSERT_STR(((ts.Size(i) == pp_get<0>(ts).Size(i)) && ...)

include/matx/transforms/fft/fft_cuda.h

@@ -643,7 +643,7 @@ __MATX_INLINE__ auto getCufft1DSupportedTensor( const Op &in, cudaStream_t strea
 template <typename Op>
 __MATX_INLINE__ auto getCufft2DSupportedTensor( const Op &in, cudaStream_t stream) {
   // This would be better as a templated lambda, but we don't have those in C++17 yet
-  const auto support_func = [&in]() {
+  const auto support_func = [&]() {
     if constexpr (is_tensor_view_v<Op>) {
       if ( in.Stride(Op::Rank()-2) != in.Stride(Op::Rank()-1) * in.Size(Op::Rank()-1)) {
         return false;

include/matx/transforms/fft/fft_fftw.h

@@ -500,7 +500,7 @@ template<typename OutTensorType, typename InTensorType> class matxFFTWPlan_t {
   template <typename Op>
   __MATX_INLINE__ auto getFFTW1DSupportedTensor(const Op &in) {
     // This would be better as a templated lambda, but we don't have those in C++17 yet
-    const auto support_func = [&in]() {
+    const auto support_func = [&]() {
       if constexpr (is_tensor_view_v<Op>) {
         if constexpr (Op::Rank() >= 2) {
           if (in.Stride(Op::Rank() - 2) != in.Stride(Op::Rank() - 1) * in.Size(Op::Rank() - 1)) {
@@ -527,7 +527,7 @@ template<typename OutTensorType, typename InTensorType> class matxFFTWPlan_t {
   template <typename Op>
   __MATX_INLINE__ auto getFFTW2DSupportedTensor( const Op &in) {
     // This would be better as a templated lambda, but we don't have those in C++17 yet
-    const auto support_func = [&in]() {
+    const auto support_func = [&]() {
       if constexpr (is_tensor_view_v<Op>) {
         if ( in.Stride(Op::Rank()-2) != in.Stride(Op::Rank()-1) * in.Size(Op::Rank()-1)) {
           return false;

include/matx/transforms/matmul/matmul_cblas.h

@@ -438,7 +438,7 @@ __MATX_INLINE__ void matmul_dispatch(TensorTypeC &c,
 template <typename Op>
 __MATX_INLINE__ auto getCBLASSupportedTensor( const Op &in) {
   // This would be better as a templated lambda, but we don't have those in C++17 yet
-  const auto support_func = [&in]() {
+  const auto support_func = [&]() {
     if constexpr (is_tensor_view_v<Op>) {
       return !(
         (in.Stride(Op::Rank() - 1) != (index_t)1 && in.Stride(Op::Rank() - 2) != (index_t)1) ||

include/matx/transforms/matmul/matmul_cuda.h

@@ -1102,7 +1102,7 @@ using gemm_cuda_cache_t = std::unordered_map<MatMulCUDAParams_t, std::any,  MatM
 template <typename Op>
 __MATX_INLINE__ auto getCublasSupportedTensor( const Op &in, cudaStream_t stream) {
   // This would be better as a templated lambda, but we don't have those in C++17 yet
-  const auto support_func = [&in]() {
+  const auto support_func = [&]() {
     if constexpr (is_tensor_view_v<Op>) {
       return !(
         (in.Stride(Op::Rank()-1) != (index_t)1 && in.Stride(Op::Rank()-2) != (index_t)1) ||

include/matx/transforms/svd/svd_cuda.h

@@ -551,13 +551,11 @@ static __MATX_INLINE__ SVDMethod GetCUDASVDMethod(const ATensor &a) {
   static constexpr int RANK = ATensor::Rank();  
   index_t m = a.Size(RANK - 2);
   index_t n = a.Size(RANK - 1);
-  SVDMethod method;
 
-  // This assumes the matrix sizes are fairly large, in which case gesvd should win out on speed
-  if (a.Rank() == 2) {
-    method = detail::SVDMethod::GESVD;
-  }
-  else {
+  // gesvd is a good default for non-batched
+  SVDMethod method = detail::SVDMethod::GESVD;
+
+  if (a.Rank() != 2) {
     if (a.Size(RANK-2) <= 32 && 
         a.Size(RANK-1) <= 32) {
       if constexpr (is_tensor_view_v<ATensor>) {