Merge branch 'master' into powellsr/fix/debug-attach-handling

src/TiledArray/conversions/vector_of_arrays.h

@@ -5,6 +5,8 @@
 #ifndef TILEDARRAY_CONVERSIONS_VECTOR_OF_ARRAYS_H_
 #define TILEDARRAY_CONVERSIONS_VECTOR_OF_ARRAYS_H_
 
+#include <TiledArray/dist_array.h>
+
 namespace TiledArray {
 
 namespace detail {

src/TiledArray/device/device_task_fn.h

@@ -121,7 +121,8 @@ struct deviceTaskFn : public TaskInterface {
       } else {
         // TODO should we use device callback or device events??
         // insert device callback
-        TiledArray::device::launchHostFunc(*stream_, device_callback, task_);
+        DeviceSafeCall(TiledArray::device::launchHostFunc(
+            *stream_, device_callback, task_));
         // processed sync, clear state
         stream_ = {};
       }

src/TiledArray/device/thrust.h

@@ -32,6 +32,15 @@
 #include <cuda_runtime_api.h>
 #endif
 
+// rocthrust headers rely on THRUST_DEVICE_SYSTEM being defined, which is only
+// defined by the HIP-specific compilers to be usable with host compiler define
+// it here explicitly
+#ifdef TILEDARRAY_HAS_HIP
+#ifndef THRUST_DEVICE_SYSTEM
+#define THRUST_DEVICE_SYSTEM THRUST_DEVICE_SYSTEM_HIP
+#endif
+#endif
+
 #include <thrust/device_vector.h>
 #include <thrust/host_vector.h>
 

src/TiledArray/external/device.h

@@ -503,8 +503,7 @@ class Env {
   /// \param page_size memory added to the pools supporting `this->um_allocator()`, `this->device_allocator()`, and `this->pinned_allocator()` in chunks of at least
   ///                  this size (bytes) [default=2^25]
   /// \param pinned_alloc_limit the maximum total amount of memory (in bytes) that
-  ///        allocator returned by `this->pinned_allocator()` can allocate;
-  ///        this allocator is not used by default [default=0]
+  ///        allocator returned by `this->pinned_allocator()` can allocate [default=2^40]
   // clang-format on
   static void initialize(World& world = TiledArray::get_default_world(),
                          const std::uint64_t page_size = (1ul << 25),
@@ -563,8 +562,9 @@ class Env {
       // allocate all currently-free memory for UM pool
       auto um_dynamic_pool =
           rm.makeAllocator<umpire::strategy::QuickPool, introspect>(
-              "UMDynamicPool", rm.getAllocator("UM"), mem_total_free.second,
-              pinned_alloc_limit);
+              "UMDynamicPool", rm.getAllocator("UM"),
+              /* first_minimum_pool_allocation_size = */ 0,
+              /* next_minimum_pool_allocation_size = */ page_size);
 
       // allocate zero memory for device pool
       auto dev_size_limited_alloc =
@@ -573,8 +573,9 @@ class Env {
               mem_total_free.first);
       auto dev_dynamic_pool =
           rm.makeAllocator<umpire::strategy::QuickPool, introspect>(
-              "DEVICEDynamicPool", dev_size_limited_alloc, 0,
-              pinned_alloc_limit);
+              "DEVICEDynamicPool", dev_size_limited_alloc,
+              /* first_minimum_pool_allocation_size = */ 0,
+              /* next_minimum_pool_allocation_size = */ page_size);
 
       // allocate pinned_alloc_limit in pinned memory
       auto pinned_size_limited_alloc =
@@ -584,7 +585,9 @@ class Env {
       auto pinned_dynamic_pool =
           rm.makeAllocator<umpire::strategy::QuickPool, introspect>(
               "QuickPool_SizeLimited_PINNED", pinned_size_limited_alloc,
-              page_size, page_size, /* alignment */ TILEDARRAY_ALIGN_SIZE);
+              /* first_minimum_pool_allocation_size = */ 0,
+              /* next_minimum_pool_allocation_size = */ page_size,
+              /* alignment */ TILEDARRAY_ALIGN_SIZE);
 
       auto env = std::unique_ptr<Env>(new Env(
           world, num_visible_devices, compute_devices, num_streams_per_device,

src/TiledArray/tile.h

@@ -215,11 +215,23 @@ class Tile {
 
   // Dimension information accessors -----------------------------------------
 
-  /// Size accessors
+  /// Size accessor
 
   /// \return The number of elements in the tensor
   decltype(auto) size() const { return tensor().size(); }
 
+  /// Total size accessor
+
+  /// \return The number of elements in the tensor, tallied across batches (if
+  /// any)
+  decltype(auto) total_size() const {
+    if constexpr (detail::has_member_function_total_size_anyreturn_v<
+                      tensor_type>) {
+      return tensor().total_size();
+    } else
+      return size();
+  }
+
   /// Range accessor
 
   /// \return An object describes the upper and lower bounds of  the tensor data

src/TiledArray/type_traits.h

@@ -322,6 +322,8 @@ GENERATE_HAS_MEMBER_TYPE(mapped_type)
 
 GENERATE_HAS_MEMBER_FUNCTION_ANYRETURN(size)
 GENERATE_HAS_MEMBER_FUNCTION(size)
+GENERATE_HAS_MEMBER_FUNCTION_ANYRETURN(total_size)
+GENERATE_HAS_MEMBER_FUNCTION(total_size)
 GENERATE_HAS_MEMBER_FUNCTION_ANYRETURN(data)
 GENERATE_HAS_MEMBER_FUNCTION(data)
 GENERATE_HAS_MEMBER_FUNCTION_ANYRETURN(empty)

tests/conversions.cpp

@@ -23,13 +23,21 @@
  *
  */
 
-#include "range_fixture.h"
-#include "tiledarray.h"
 #include "unit_test_config.h"
 
-#include "TiledArray/conversions/concat.h"
 #include "TiledArray/conversions/vector_of_arrays.h"
 
+#include "TiledArray/conversions/concat.h"
+
+#include "TiledArray/conversions/dense_to_sparse.h"
+#include "TiledArray/conversions/make_array.h"
+#include "TiledArray/conversions/sparse_to_dense.h"
+#include "TiledArray/conversions/to_new_tile_type.h"
+
+#include "TiledArray/expressions/tsr_expr.h"
+
+#include "range_fixture.h"
+
 using namespace TiledArray;
 
 struct ConversionsFixture : public TiledRangeFixture {