diff --git a/CMakeLists.txt b/CMakeLists.txt
index a2e4ee32c83..c75b52555f4 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -215,8 +215,7 @@ set(CMAKE_CUDA_FLAGS_RELEASE "${CMAKE_CUDA_FLAGS_RELEASE} -DDALI_DEBUG=0")
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -O2 -g3")
 set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "${CMAKE_CUDA_FLAGS_RELWITHDEBINFO} -g -lineinfo")
 
-# Use -std=c++14 (and not gnu++14)
-set(CMAKE_CXX_STANDARD 14)
+set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 set(CMAKE_C_STANDARD 11)
diff --git a/dali/kernels/test/tuple_test.cc b/dali/core/tuple_test.cc
similarity index 93%
rename from dali/kernels/test/tuple_test.cc
rename to dali/core/tuple_test.cc
index adce67fa537..963600167be 100644
--- a/dali/kernels/test/tuple_test.cc
+++ b/dali/core/tuple_test.cc
@@ -1,4 +1,4 @@
-// Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2018-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -13,9 +13,8 @@
 // limitations under the License.
 
 #include <gtest/gtest.h>
-#include "dali/kernels/kernel.h"
-#include "dali/kernels/type_tag.h"
-#include "dali/core/static_switch.h"
+#include <tuple>
+#include "dali/core/tuple_helpers.h"
 
 namespace dali {
 namespace kernels {
diff --git a/dali/kernels/test/util_test.cc b/dali/core/util_test.cc
similarity index 100%
rename from dali/kernels/test/util_test.cc
rename to dali/core/util_test.cc
diff --git a/dali/core/utils_test.cu b/dali/core/utils_test.cu
index afbc2022c45..9da68cb6c4d 100644
--- a/dali/core/utils_test.cu
+++ b/dali/core/utils_test.cu
@@ -33,14 +33,14 @@ DEVICE_TEST(CoreUtilsDev, Volume, 1, 1) {
 
 DEVICE_TEST(CoreUtilsDev, Size, 1, 1) {
   int a0[] = { 42 };
-  DEV_EXPECT_EQ(size(a0), 1u);
+  DEV_EXPECT_EQ(dali::size(a0), 1u);
   int a1[] = { 2, 3, 4 };
-  DEV_EXPECT_EQ(size(a1), 3u);
+  DEV_EXPECT_EQ(dali::size(a1), 3u);
 
   SmallVector<int, 5> v;
   v.resize(10);
   DEV_EXPECT_EQ(v.size(), 10u);
-  DEV_EXPECT_EQ(size(v), 10u);
+  DEV_EXPECT_EQ(dali::size(v), 10u);
 }
 
 DEFINE_TEST_KERNEL(CoreUtilsDev, Span, span<float> data) {
diff --git a/dali/kernels/context.h b/dali/kernels/context.h
index c39c37b1104..1ad57649049 100644
--- a/dali/kernels/context.h
+++ b/dali/kernels/context.h
@@ -135,15 +135,16 @@ class Scratchpad {
   template <typename Collection, typename T = std::remove_const_t<element_t<Collection>>>
   if_array_like<Collection, T*>
   ToGPU(cudaStream_t stream, const Collection &c) {
-    T *ptr = AllocateGPU<T>(size(c));
-    CUDA_CALL(cudaMemcpyAsync(ptr, &c[0], size(c) * sizeof(T), cudaMemcpyHostToDevice, stream));
+    auto n = dali::size(c);
+    T *ptr = AllocateGPU<T>(n);
+    CUDA_CALL(cudaMemcpyAsync(ptr, &c[0], n * sizeof(T), cudaMemcpyHostToDevice, stream));
     return ptr;
   }
 
   template <typename Collection, typename T = std::remove_const_t<element_t<Collection>>>
   if_iterable<Collection, T*>
   ToHost(const Collection &c) {
-    T *ptr = AllocateHost<T>(size(c));
+    T *ptr = AllocateHost<T>(dali::size(c));
     std::copy(begin(c), end(c), ptr);
     return ptr;
   }
@@ -151,7 +152,7 @@ class Scratchpad {
   template <typename Collection, typename T = std::remove_const_t<element_t<Collection>>>
   if_iterable<Collection, T*>
   ToPinned(const Collection &c) {
-    T *ptr = AllocatePinned<T>(size(c));
+    T *ptr = AllocatePinned<T>(dali::size(c));
     std::copy(begin(c), end(c), ptr);
     return ptr;
   }
@@ -159,7 +160,7 @@ class Scratchpad {
   template <typename Collection, typename T = std::remove_const_t<element_t<Collection>>>
   if_iterable<Collection, T*>
   ToManaged(const Collection &c) {
-    T *ptr = AllocateManaged<T>(size(c));
+    T *ptr = AllocateManaged<T>(dali::size(c));
     std::copy(begin(c), end(c), ptr);
     return ptr;
   }
diff --git a/dali/kernels/reduce/reduce_drop_dims.h b/dali/kernels/reduce/reduce_drop_dims.h
index 7093b769c04..4eb2be279db 100644
--- a/dali/kernels/reduce/reduce_drop_dims.h
+++ b/dali/kernels/reduce/reduce_drop_dims.h
@@ -136,7 +136,7 @@ struct DropDims {
   template <typename Indices>
   static int simplify(int64_t *out_shape, unsigned &out_mask,
                       const Indices &in_shape, uint64_t axis_mask) {
-    int dims = size(in_shape);
+    int dims = dali::size(in_shape);
     int d = 0;
     out_shape[0] = in_shape[0];
     bool prev = axis_mask & 1;
@@ -169,7 +169,7 @@ struct DropDims {
   template <typename Indices>
   DropDims(const Indices &in_shape, uint64_t reduced_axes) {
     memset(this, 0, sizeof(*this));
-    if (size(in_shape) == 0) {
+    if (dali::size(in_shape) == 0) {
       start = -1;
       return;
     }
diff --git a/dali/kernels/scratch_copy_impl.h b/dali/kernels/scratch_copy_impl.h
index c856c74bb3d..1bbdb0a566d 100644
--- a/dali/kernels/scratch_copy_impl.h
+++ b/dali/kernels/scratch_copy_impl.h
@@ -34,9 +34,9 @@ inline void copy_to_buffer(char *buffer, const size_t *offsets) {}
  */
 template <typename Collection, typename... Collections>
 void copy_to_buffer(char *buffer,
-                  const size_t *offsets,
-                  const Collection &c,
-                  const Collections &... tail) {
+                   const size_t *offsets,
+                   const Collection &c,
+                   const Collections &... tail) {
   using T = std::remove_cv_t<element_t<Collection>>;
   std::copy(dali::begin(c), dali::end(c), reinterpret_cast<T*>(buffer + offsets[0]));
   copy_to_buffer(buffer, offsets+1, tail...);
@@ -54,12 +54,12 @@ inline void GetCollectionOffsets(size_t base, size_t *offsets) { *offsets = base
  */
 template <typename Collection, typename... Collections>
 void GetCollectionOffsets(size_t base, size_t *offsets,
-                              const Collection &c,
-                              const Collections &...tail) {
+                          const Collection &c,
+                          const Collections &...tail) {
   using T = std::remove_cv_t<element_t<Collection>>;
   base = align_up(base, alignof(T));
   *offsets = base;
-  base += size(c) * sizeof(T);
+  base += dali::size(c) * sizeof(T);
   GetCollectionOffsets(base, offsets + 1, tail...);
 }
 
diff --git a/dali/kernels/test/scratch_copy_test.cc b/dali/kernels/test/scratch_copy_test.cc
index b7e428f1169..c621295e8af 100644
--- a/dali/kernels/test/scratch_copy_test.cc
+++ b/dali/kernels/test/scratch_copy_test.cc
@@ -23,7 +23,7 @@ namespace dali {
 namespace kernels {
 
 template <typename C, typename = element_t<C>>
-constexpr auto size_bytes(const C &c) { return size(c) * sizeof(element_t<C>); }
+constexpr auto size_bytes(const C &c) { return dali::size(c) * sizeof(element_t<C>); }
 
 TEST(Scratchpad, ToContiguous) {
   ScratchpadEstimator se;
diff --git a/dali/operators/geometry/mt_transform_attr.h b/dali/operators/geometry/mt_transform_attr.h
index a64a1e2152f..926849ea559 100644
--- a/dali/operators/geometry/mt_transform_attr.h
+++ b/dali/operators/geometry/mt_transform_attr.h
@@ -1,4 +1,4 @@
-// Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2020-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -106,7 +106,7 @@ class DLL_PUBLIC MTTransformAttr {
   /** @brief Fill the diagonal with a scalar value, put zeros elsewhere */
   template <typename Container>
   void MakeDiagonalMatrix(Container &&mtx, float value) {
-    assert(static_cast<int>(size(mtx)) == input_pt_dim_ * output_pt_dim_);
+    assert(static_cast<int>(dali::size(mtx)) == input_pt_dim_ * output_pt_dim_);
     for (int i = 0, k = 0; i < output_pt_dim_; i++)
       for (int j = 0; j < input_pt_dim_; j++, k++)
           mtx[k] = (i == j ? value : 0);
@@ -114,7 +114,7 @@ class DLL_PUBLIC MTTransformAttr {
 
   template <typename OutRange, typename InRange>
   void Repeat(OutRange &&out, InRange &&in, int times) {
-    ssize_t n = size(in);
+    ssize_t n = dali::size(in);
     resize_if_possible(out, n * times);
     for (ssize_t i = 0, k = 0; i < times; i++)
       for (ssize_t j = 0; j < n; j++, k++)
diff --git a/dali/operators/segmentation/random_object_bbox.cc b/dali/operators/segmentation/random_object_bbox.cc
index 0cf4ba7164a..1a494c26122 100644
--- a/dali/operators/segmentation/random_object_bbox.cc
+++ b/dali/operators/segmentation/random_object_bbox.cc
@@ -241,8 +241,8 @@ void StoreBox(const OutListCPU<int, 1> &out1,
               const OutListCPU<int, 1> &out2,
               RandomObjectBBox::OutputFormat format,
               int sample_idx, Lo &&start, Hi &&end) {
-  assert(size(start) == size(end));
-  int ndim = size(start);
+  assert(dali::size(start) == dali::size(end));
+  int ndim = dali::size(start);
   switch (format) {
     case RandomObjectBBox::Out_Box:
       for (int i = 0; i < ndim; i++) {
diff --git a/dali/pipeline/data/buffer.h b/dali/pipeline/data/buffer.h
index f88482cf148..401ba44f061 100644
--- a/dali/pipeline/data/buffer.h
+++ b/dali/pipeline/data/buffer.h
@@ -324,7 +324,7 @@ class DLL_PUBLIC Buffer {
     return shrink_threshold_;
   }
 
-  static constexpr double kMaxGrowthFactor = 4;
+  DLL_PUBLIC static constexpr double kMaxGrowthFactor = 4;
 
  protected:
   // Helper to resize the underlying allocation
@@ -413,7 +413,7 @@ DLL_PUBLIC double Buffer<Backend>::shrink_threshold_ =
   std::is_same<Backend, CPUBackend>::value ? 0.5 : 0;
 
 template <typename Backend>
-DLL_PUBLIC constexpr double Buffer<Backend>::kMaxGrowthFactor;
+constexpr double Buffer<Backend>::kMaxGrowthFactor;
 
 
 // Macro so we don't have to list these in all
diff --git a/include/dali/core/random.h b/include/dali/core/random.h
index 2a7f63d8204..6acc6d39601 100644
--- a/include/dali/core/random.h
+++ b/include/dali/core/random.h
@@ -1,4 +1,4 @@
-// Copyright (c) 2020, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2020-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -37,7 +37,7 @@ void random_permutation(Collection &out, RNG &rng) {
  */
 template <typename Collection, typename RNG>
 void random_derangement(Collection &out, RNG &rng) {
-  int N = size(out);
+  int N = dali::size(out);
   std::iota(dali::begin(out), dali::end(out), 0);
   for (int i = 0; i < N-1; i++) {
     std::uniform_int_distribution<int> dist(i+1, N-1);
@@ -76,7 +76,7 @@ random_sequence(Collection &out, T lo, T hi, RNG &rng) {
  */
 template <typename Collection, typename RNG>
 void random_sequence_no_fixed_points(Collection &out, int lo, int hi, RNG &rng) {
-  int N = size(out);
+  int N = dali::size(out);
   std::uniform_int_distribution<int> dist1(lo, hi-1);
   int i = 0;
   // when index is below lo, no fixed points possible
diff --git a/include/dali/core/tensor_shape.h b/include/dali/core/tensor_shape.h
index 0b239c526b8..b0edc3a9b48 100644
--- a/include/dali/core/tensor_shape.h
+++ b/include/dali/core/tensor_shape.h
@@ -1363,14 +1363,14 @@ template <int out_ndim, int ndim>
 void collapse_dims(TensorListShape<out_ndim> &result,
                    const TensorListShape<ndim> &shape,
                    std::initializer_list<std::pair<int, int>> dim_groups) {
-  collapse_dims<out_ndim>(result, shape, make_span(dim_groups.begin(), size(dim_groups)));
+  collapse_dims<out_ndim>(result, shape, make_span(dim_groups.begin(), dali::size(dim_groups)));
 }
 
 template <int out_ndim = DynamicDimensions, int ndim>
 TensorListShape<out_ndim> collapse_dims(const TensorListShape<ndim> &shape,
                                         std::initializer_list<std::pair<int, int>> dim_groups) {
   TensorListShape<out_ndim> result;
-  collapse_dims<out_ndim>(result, shape, make_span(dim_groups.begin(), size(dim_groups)));
+  collapse_dims<out_ndim>(result, shape, make_span(dim_groups.begin(), dali::size(dim_groups)));
   return result;
 }
 
diff --git a/include/dali/core/tensor_view.h b/include/dali/core/tensor_view.h
index ba4ce3cc463..f5785326fbc 100644
--- a/include/dali/core/tensor_view.h
+++ b/include/dali/core/tensor_view.h
@@ -1,4 +1,4 @@
-// Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2017-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -38,8 +38,8 @@ struct check_implicit_conversion {
 
 template <typename Shape, typename Position>
 bool ContainsCoords(const Shape &shape, const Position &pos) {
-  const int shape_dim = size(shape);
-  const int pos_dim = size(pos);
+  const int shape_dim = dali::size(shape);
+  const int pos_dim = dali::size(pos);
   if (pos_dim > shape_dim) {
     return false;
   }
@@ -60,8 +60,8 @@ template <typename Shape, typename Position>
 DALI_HOST_DEV if_array_like<Position, ptrdiff_t> CalcOffset(const Shape &shape,
                                                             const Position &pos) {
   ptrdiff_t ofs = pos[0];
-  const int pos_dim = size(pos);
-  const int shape_dim = size(shape);
+  const int pos_dim = dali::size(pos);
+  const int shape_dim = dali::size(shape);
   int i;
   for (i = 1; i < pos_dim; i++) {
     ofs *= shape[i];
diff --git a/include/dali/core/tuple_helpers.h b/include/dali/core/tuple_helpers.h
index 95fb16b69d5..8c4e1114fab 100644
--- a/include/dali/core/tuple_helpers.h
+++ b/include/dali/core/tuple_helpers.h
@@ -1,4 +1,4 @@
-// Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
+// Copyright (c) 2018-2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
 // you may not use this file except in compliance with the License.
@@ -155,8 +155,8 @@ constexpr auto apply(F &&f, std::tuple<T...> &&args)
 
 template <typename F, typename... Args>
 constexpr auto apply_all(F &&f, Args&&... args)
-  ->decltype(apply(f, std::tuple_cat(as_tuple(args)...))) {
-  return apply(f, std::tuple_cat(as_tuple(args)...));
+  ->decltype(dali::detail::apply(f, std::tuple_cat(as_tuple(args)...))) {
+  return dali::detail::apply(f, std::tuple_cat(as_tuple(args)...));
 }
 
 template <size_t total, typename Type, typename Tuple>
diff --git a/include/dali/core/util.h b/include/dali/core/util.h
index 80cbc0c955a..0d9cf114059 100644
--- a/include/dali/core/util.h
+++ b/include/dali/core/util.h
@@ -217,7 +217,7 @@ using if_indexable = if_istype<decltype(std::declval<Collection>()[0]), T>;
 
 template <typename Collection, typename T = void>
 using if_array_like = if_indexable<Collection,
-                                   if_istype<decltype(size(std::declval<Collection>())), T>>;
+                                  if_istype<decltype(dali::size(std::declval<Collection>())), T>>;
 
 template <typename It>
 using is_integer_iterator = std::is_integral<
diff --git a/third_party/libcudacxx b/third_party/libcudacxx
index adf4d157ada..03e17c25869 160000
--- a/third_party/libcudacxx
+++ b/third_party/libcudacxx
@@ -1 +1 @@
-Subproject commit adf4d157ada537ef500553b695f32bb10a2a11e4
+Subproject commit 03e17c25869a2a2a4203e5cde129c833438e2047