Debug

Author: cliffburdick

include/matx/transforms/chol/chol_cuda.h

@@ -103,7 +103,7 @@ class matxDnCholCUDAPlan_t : matxDnCUDASolver_t {
     MATX_STATIC_ASSERT_STR(!is_half_v<T1>, matxInvalidType, "Cholesky solver does not support half precision");
     MATX_STATIC_ASSERT_STR((std::is_same_v<T1, typename OutTensor_t::value_type>), matxInavlidType, "Input and Output types must match");
 
-    params = GetCholParams(a, uplo);
+    params = GetCholParams(a, uplo, exec);
 
     this->GetWorkspaceSize();
     this->AllocateWorkspace(params.batch_size, false, exec);

include/matx/transforms/eig/eig_cuda.h

@@ -160,6 +160,7 @@ class matxDnEigCUDAPlan_t : matxDnCUDASolver_t {
     params.jobz = jobz;
     params.uplo = uplo;
     params.exec = exec;    
+    printf("setting stream to %p\n", exec.getStream());
     params.dtype = TypeToInt<T1>();
 
     return params;
@@ -273,6 +274,7 @@ struct DnEigCUDAParamsKeyHash {
 struct DnEigCUDAParamsKeyEq {
   bool operator()(const DnEigCUDAParams_t &l, const DnEigCUDAParams_t &t) const noexcept
   {
+    printf("%ld %ld %zu %zu %d %d %p %p\n", l.n , t.n, l.batch_size , t.batch_size , (int)l.dtype , (int)t.dtype , l.exec.getStream() , t.exec.getStream());
     return l.n == t.n && l.batch_size == t.batch_size && l.dtype == t.dtype && l.exec.getStream() == t.exec.getStream();
   }
 };

include/matx/transforms/svd/svd_cuda.h

@@ -724,8 +724,8 @@ class matxDnSVDCUDAPlan_t : matxDnCUDASolver_t {
 
   static DnSVDCUDAParams_t
   GetSVDParams(UTensor &u, STensor &s,
-               VtTensor &vt, const ATensor &a, const cudaExecutor &exec,
-               const char jobz = 'A')
+               VtTensor &vt, const ATensor &a, 
+               const char jobz, const cudaExecutor &exec)
   {
     DnSVDCUDAParams_t params;
     params.batch_size = GetNumBatches(a);
@@ -997,7 +997,7 @@ void svd_impl(UTensor &&u, STensor &&s,
 
     // Get parameters required by these tensors
     auto params = detail::matxDnSVDCUDAPlan_t<decltype(u_in), decltype(s_new), decltype(vt_in), decltype(at_col_maj)>::
-      GetSVDParams(u_in, s_new, vt_in, at_col_maj, exec, job_cusolver);
+      GetSVDParams(u_in, s_new, vt_in, at_col_maj, job_cusolver, exec);
 
     // Get cache or new SVD plan if it doesn't exist
     using cache_val_type = detail::matxDnSVDCUDAPlan_t<decltype(u_in), decltype(s_new), decltype(vt_in), decltype(at_col_maj)>;
@@ -1034,7 +1034,7 @@ void svd_impl(UTensor &&u, STensor &&s,
 
     // Get parameters required by these tensors
     auto params = detail::matxDnSVDCUDAPlan_t<decltype(u_col_maj), decltype(s_new), decltype(vt_col_maj), decltype(tvt)>::
-        GetSVDParams(u_col_maj, s_new, vt_col_maj, tvt, exec, job_cusolver);
+        GetSVDParams(u_col_maj, s_new, vt_col_maj, tvt, job_cusolver, exec);
 
     // Get cache or new SVD plan if it doesn't exist
     using cache_val_type = detail::matxDnSVDCUDAPlan_t<decltype(u_col_maj), decltype(s_new), decltype(vt_col_maj), decltype(tvt)>;