introduced umpire host allocator

CMakeLists.txt

@@ -132,6 +132,10 @@ if((ENABLE_GPERFTOOLS OR ENABLE_TCMALLOC_MINIMAL) AND CMAKE_SYSTEM_NAME MATCHES
   set(ENABLE_LIBUNWIND ON)
   add_feature_info(Libunwind ENABLE_LIBUNWIND "Libunwind provides stack unwinding")
 endif()
+
+option(TA_TENSOR_MEM_PROFILE "Turn on instrumented profiling of TA::Tensor memory use" OFF)
+add_feature_info(TENSOR_MEM_PROFILE TA_TENSOR_MEM_PROFILE "instrumented profiling of TA::Tensor memory use")
+
 option(TA_BUILD_UNITTEST "Causes building TiledArray unit tests" ON)
 option(TA_EXPERT "TiledArray Expert mode: disables automatically downloading or building dependencies" OFF)
 
@@ -264,7 +268,7 @@ check_type_size("long double" TILEDARRAY_HAS_LONG_DOUBLE LANGUAGE CXX)
 check_type_size("long long" TILEDARRAY_HAS_LONG_LONG LANGUAGE CXX)
 
 # TA_ASSERT
-set (TA_ASSERT_POLICY TA_ASSERT_THROW CACHE STRING "")
+set (TA_ASSERT_POLICY TA_ASSERT_THROW CACHE STRING "Controls the behavior of TA_ASSERT; TA_ASSERT_THROW causes TA_ASSERT to throw, TA_ASSERT_ABORT causes TA_ASSERT to abort, TA_ASSERT_IGNORE makes TA_ASSERT a no-op")
 set_property(
   CACHE TA_ASSERT_POLICY PROPERTY
   STRINGS TA_ASSERT_THROW TA_ASSERT_ABORT TA_ASSERT_IGNORE)
@@ -294,6 +298,7 @@ endif()
 include(external/madness.cmake)
 detect_MADNESS_configuration()
 include(external/eigen.cmake)
+include(external/umpire.cmake)
 
 ###### discover linear algebra
 

INSTALL.md

@@ -42,7 +42,7 @@ Both methods are supported. However, for most users we _strongly_ recommend to b
   - Boost.Range: header-only, *only used for unit testing*
 - [BTAS](https://github.com/ValeevGroup/BTAS), tag d7794799e4510cf66844081dd8f1f5b648112d33 . If usable BTAS installation is not found, TiledArray will download and compile
   BTAS from source. *This is the recommended way to compile BTAS for all users*.
-- [MADNESS](https://github.com/m-a-d-n-e-s-s/madness), tag 7ce06234c23aa8e0ab3d9e9b87eff9cd85390d80 .
+- [MADNESS](https://github.com/m-a-d-n-e-s-s/madness), tag 12bd24c6cb984a639be863fc0e1364226713f7ff .
   Only the MADworld runtime and BLAS/LAPACK C API component of MADNESS is used by TiledArray.
   If usable MADNESS installation is not found, TiledArray will download and compile
   MADNESS from source. *This is the recommended way to compile MADNESS for all users*.
@@ -65,7 +65,7 @@ Compiling BTAS requires the following prerequisites:
 Optional prerequisites:
 - [CUDA compiler and runtime](https://developer.nvidia.com/cuda-zone) -- for execution on CUDA-enabled accelerators. CUDA 11 or later is required. Support for CUDA also requires the following additional prerequisites, both of which will be built and installed automatically if missing:
   - [cuTT](github.com/ValeevGroup/cutt) -- CUDA transpose library; note that our fork of the [original cuTT repo](github.com/ap-hynninen/cutt) is required to provide thread-safety (tag 0e8685bf82910bc7435835f846e88f1b39f47f09).
-  - [Umpire](github.com/LLNL/Umpire) -- portable memory manager for heterogeneous platforms (tag f04abd1dd038c84262915a493d8f78576bb80fd0).
+  - [Umpire](github.com/LLNL/Umpire) -- portable memory manager for heterogeneous platforms (tag 5201a47a35e3844160dcbecd0916f8c96aa7dd07).
 - [Doxygen](http://www.doxygen.nl/) -- for building documentation (version 1.8.12 or later).
 - [ScaLAPACK](http://www.netlib.org/scalapack/) -- a distributed-memory linear algebra package. If detected, the following C++ components will also be sought and downloaded, if missing:
   - [blacspp](https://github.com/wavefunction91/blacspp.git) -- a modern C++ (C++17) wrapper for BLACS (tag 88076f1706be083ead882f6ce0bfc6884a72fc03)
@@ -405,6 +405,7 @@ support may be added.
 * `TA_ENABLE_RANGEV3` -- Set to `ON` to find or fetch the Range-V3 library and enable additional tests of TA components with constructs anticipated to be supported in the future. [Default=OFF].
 * `TA_SIGNED_1INDEX_TYPE` -- Set to `OFF` to use unsigned 1-index coordinate type (default for TiledArray 1.0.0-alpha.2 and older). The default is `ON`, which enables the use of negative indices in coordinates.
 * `TA_MAX_SOO_RANK_METADATA` -- Specifies the maximum rank for which to use Small Object Optimization (hence, avoid the use of the heap) for metadata. The default is `8`.
+* `TA_TENSOR_MEM_PROFILE` -- Set to `ON` to profile memory allocations in TA::Tensor.
 
 # Build TiledArray
 

external/cutt.cmake

@@ -23,10 +23,10 @@ else()
     enable_language(C)
 
     # set source and build path for cuTT in the TiledArray project
-    set(EXTERNAL_SOURCE_DIR   ${PROJECT_BINARY_DIR}/external/source/cutt)
+    set(EXTERNAL_SOURCE_DIR   ${CMAKE_BINARY_DIR}/_deps/cutt-src)
     # cutt only supports in source build
-    set(EXTERNAL_BUILD_DIR  ${PROJECT_BINARY_DIR}/external/build/cutt)
-    set(EXTERNAL_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/external/cutt)
+    set(EXTERNAL_BUILD_DIR  ${CMAKE_BINARY_DIR}/_deps/cutt-build)
+    set(EXTERNAL_INSTALL_DIR ${CMAKE_INSTALL_PREFIX})
 
     if (NOT CUTT_URL)
         set(CUTT_URL https://github.com/ValeevGroup/cutt.git)
@@ -72,6 +72,9 @@ else()
         -DCMAKE_CUDA_HOST_COMPILER=${CMAKE_CUDA_HOST_COMPILER}
         -DCUDA_TOOLKIT_ROOT_DIR=${CUDAToolkit_ROOT}
         )
+    if (DEFINED CMAKE_CUDA_ARCHITECTURES)
+        list(APPEND CUTT_CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES})
+    endif(DEFINED CMAKE_CUDA_ARCHITECTURES)
     if (CMAKE_TOOLCHAIN_FILE)
         set(CUTT_CMAKE_ARGS "${CUTT_CMAKE_ARGS}"
             "-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TOOLCHAIN_FILE}")
@@ -89,8 +92,8 @@ else()
 
     ExternalProject_Add(cutt
             PREFIX ${CMAKE_INSTALL_PREFIX}
-            STAMP_DIR ${PROJECT_BINARY_DIR}/external/cutt-stamp
-            TMP_DIR ${PROJECT_BINARY_DIR}/external/tmp
+            STAMP_DIR ${CMAKE_BINARY_DIR}/_deps/cutt-ep-artifacts
+            TMP_DIR ${CMAKE_BINARY_DIR}/_deps/cutt-ep-artifacts  # needed in case CMAKE_INSTALL_PREFIX is not writable
             #--Download step--------------
             DOWNLOAD_DIR ${EXTERNAL_SOURCE_DIR}
             GIT_REPOSITORY ${CUTT_URL}

external/eigen.cmake

@@ -93,15 +93,13 @@ else()
   include(ExternalProject)
 
   # Set source and build path for Eigen3 in the TiledArray Project
-  set(EXTERNAL_SOURCE_DIR   ${PROJECT_BINARY_DIR}/external/source/eigen)
-  set(EXTERNAL_BUILD_DIR  ${PROJECT_BINARY_DIR}/external/build/eigen)
+  set(EXTERNAL_SOURCE_DIR   ${CMAKE_BINARY_DIR}/_deps/eigen-src)
+  set(EXTERNAL_BUILD_DIR  ${CMAKE_BINARY_DIR}/_deps/eigen-build)
 
   message("** Will build Eigen from ${EIGEN3_URL}")
 
   ExternalProject_Add(eigen3
     PREFIX ${CMAKE_INSTALL_PREFIX}
-    STAMP_DIR ${EXTERNAL_BUILD_DIR}/stamp
-    TMP_DIR ${EXTERNAL_BUILD_DIR}/tmp
    #--Download step--------------
     DOWNLOAD_DIR ${EXTERNAL_SOURCE_DIR}
     URL ${EIGEN3_URL}

external/madness.cmake

@@ -129,9 +129,9 @@ else()
 
   # Create a cache entry for MADNESS build variables.
   # Note: This will not overwrite user specified values.
-  set(MADNESS_SOURCE_DIR "${PROJECT_BINARY_DIR}/external/madness-src" CACHE PATH
+  set(MADNESS_SOURCE_DIR "${CMAKE_BINARY_DIR}/_deps/madness-src" CACHE PATH
         "Path to the MADNESS source directory")
-  set(MADNESS_BINARY_DIR "${PROJECT_BINARY_DIR}/external/madness-build" CACHE PATH
+  set(MADNESS_BINARY_DIR "${CMAKE_BINARY_DIR}/_deps/madness-build" CACHE PATH
         "Path to the MADNESS build directory")
   set(MADNESS_URL "https://github.com/m-a-d-n-e-s-s/madness.git" CACHE STRING 
         "Path to the MADNESS repository")
@@ -152,16 +152,16 @@ else()
   # If the MADNESS source directory is the default location and does not exist,
   # MADNESS will be downloaded from git.
   message(STATUS "Checking MADNESS source directory: ${MADNESS_SOURCE_DIR}")
-  if("${MADNESS_SOURCE_DIR}" STREQUAL "${PROJECT_BINARY_DIR}/external/madness-src")
+  if("${MADNESS_SOURCE_DIR}" STREQUAL "${CMAKE_BINARY_DIR}/_deps/madness-src")
 
-    # Create the external source directory
-    if(NOT EXISTS ${PROJECT_BINARY_DIR}/external)
+    # Create the source directory
+    if(NOT EXISTS ${CMAKE_BINARY_DIR}/_deps)
       set(error_code 1)
       execute_process(
-          COMMAND "${CMAKE_COMMAND}" -E make_directory "${PROJECT_BINARY_DIR}/external"
+          COMMAND "${CMAKE_COMMAND}" -E make_directory "${CMAKE_BINARY_DIR}/_deps"
           RESULT_VARIABLE error_code)
       if(error_code)
-        message(FATAL_ERROR "Failed to create directory \"${PROJECT_BINARY_DIR}/external\"")
+        message(FATAL_ERROR "Failed to create directory \"${CMAKE_BINARY_DIR}/_deps\"")
       endif()
     endif()
 
@@ -173,7 +173,7 @@ else()
       while(error_code AND number_of_tries LESS 3)
         execute_process(
             COMMAND ${GIT_EXECUTABLE} clone ${MADNESS_URL} madness-src
-            WORKING_DIRECTORY ${PROJECT_BINARY_DIR}/external
+            WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/_deps
             RESULT_VARIABLE error_code)
         math(EXPR number_of_tries "${number_of_tries} + 1")
       endwhile()

external/umpire.cmake

@@ -2,6 +2,8 @@
 ## find Umpire
 ##
 
+if (NOT TARGET TiledArray_UMPIRE)
+
 find_path(_UMPIRE_INSTALL_DIR NAMES include/umpire/Umpire.hpp HINTS ${UMPIRE_INSTALL_DIR})
 
 # if user provides UMPIRE, use it
@@ -27,9 +29,9 @@ else()
     enable_language(C)
 
     # set source and build path for Umpire in the TiledArray project
-    set(EXTERNAL_SOURCE_DIR   ${PROJECT_BINARY_DIR}/external/source/Umpire)
-    set(EXTERNAL_BUILD_DIR  ${PROJECT_BINARY_DIR}/external/build/Umpire)
-    set(EXTERNAL_INSTALL_DIR ${CMAKE_INSTALL_PREFIX}/external/Umpire)
+    set(EXTERNAL_SOURCE_DIR ${CMAKE_BINARY_DIR}/_deps/umpire-src)
+    set(EXTERNAL_BUILD_DIR ${CMAKE_BINARY_DIR}/_deps/umpire-build)
+    set(EXTERNAL_INSTALL_DIR ${CMAKE_INSTALL_PREFIX})
 
     if (NOT UMPIRE_URL)
         set(UMPIRE_URL https://github.com/LLNL/Umpire.git)
@@ -40,9 +42,6 @@ else()
 
     message("** Will clone Umpire from ${UMPIRE_URL}")
 
-    ## use patched BLT
-    include(FetchBLT)
-
     if (TA_ASSERT_POLICY EQUAL TA_ASSERT_IGNORE)
         set(enable_umpire_asserts OFF)
     else()
@@ -69,21 +68,27 @@ else()
         -DCMAKE_CXX_STANDARD=${CMAKE_CXX_STANDARD}
         -DCMAKE_CXX_EXTENSIONS=${CMAKE_CXX_EXTENSIONS}
         -DCMAKE_AR=${CMAKE_AR}
-        -DBLT_SOURCE_DIR=${BLT_SOURCE_DIR}
-        -DBLT_CXX_STD=c++${CMAKE_CUDA_STANDARD}
-        -DENABLE_CUDA=ON
+        -DBLT_CXX_STD=c++${CMAKE_CXX_STANDARD}
         -DENABLE_BENCHMARKS=OFF
         -DENABLE_OPENMP=OFF
         -DENABLE_TESTS=OFF
         -DENABLE_EXAMPLES=OFF
         -DENABLE_LOGGING=OFF
         -DENABLE_ASSERTS=${enable_umpire_asserts}
-        -DCMAKE_CUDA_COMPILER=${CMAKE_CUDA_COMPILER}
-        -DCMAKE_CUDA_STANDARD=${CMAKE_CUDA_STANDARD}
-        -DCMAKE_CUDA_EXTENSIONS=${CMAKE_CUDA_EXTENSIONS}
-        -DCMAKE_CUDA_HOST_COMPILER=${CMAKE_CUDA_HOST_COMPILER}
-        -DCUDA_TOOLKIT_ROOT_DIR=${CUDAToolkit_ROOT}
         )
+    if (ENABLE_CUDA)
+        list(APPEND UMPIRE_CMAKE_ARGS
+                -DENABLE_CUDA=ON
+                -DCMAKE_CUDA_COMPILER=${CMAKE_CUDA_COMPILER}
+                -DCMAKE_CUDA_STANDARD=${CMAKE_CUDA_STANDARD}
+                -DCMAKE_CUDA_EXTENSIONS=${CMAKE_CUDA_EXTENSIONS}
+                -DCMAKE_CUDA_HOST_COMPILER=${CMAKE_CUDA_HOST_COMPILER}
+                -DCUDA_TOOLKIT_ROOT_DIR=${CUDAToolkit_ROOT}
+                )
+        if (DEFINED CMAKE_CUDA_ARCHITECTURES)
+            list(APPEND UMPIRE_CMAKE_ARGS -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES})
+        endif(DEFINED CMAKE_CUDA_ARCHITECTURES)
+    endif(ENABLE_CUDA)
     if (CMAKE_TOOLCHAIN_FILE)
         set(UMPIRE_CMAKE_ARGS "${UMPIRE_CMAKE_ARGS}"
             "-DCMAKE_TOOLCHAIN_FILE=${CMAKE_TOOLCHAIN_FILE}"
@@ -102,13 +107,12 @@ else()
 
     ExternalProject_Add(Umpire
             PREFIX ${CMAKE_INSTALL_PREFIX}
-            STAMP_DIR ${PROJECT_BINARY_DIR}/external/Umpire-stamp
-            TMP_DIR ${PROJECT_BINARY_DIR}/external/tmp
+            STAMP_DIR ${CMAKE_BINARY_DIR}/_deps/umpire-ep-artifacts
+            TMP_DIR ${CMAKE_BINARY_DIR}/_deps/umpire-ep-artifacts   # needed in case CMAKE_INSTALL_PREFIX is not writable
             #--Download step--------------
             DOWNLOAD_DIR ${EXTERNAL_SOURCE_DIR}
             GIT_REPOSITORY ${UMPIRE_URL}
             GIT_TAG ${UMPIRE_TAG}
-            GIT_SUBMODULES ""  # N.B. do not initialize modules!
             #--Configure step-------------
             SOURCE_DIR ${EXTERNAL_SOURCE_DIR}
             LIST_SEPARATOR ::
@@ -152,11 +156,13 @@ set_target_properties(
         TiledArray_UMPIRE
         PROPERTIES
         INTERFACE_INCLUDE_DIRECTORIES
-        "$<BUILD_INTERFACE:${EXTERNAL_SOURCE_DIR}/src>;$<BUILD_INTERFACE:${EXTERNAL_BUILD_DIR}/include>;$<INSTALL_INTERFACE:${_UMPIRE_INSTALL_DIR}/include>"
+        "$<BUILD_INTERFACE:${EXTERNAL_SOURCE_DIR}/src>;$<BUILD_INTERFACE:${EXTERNAL_SOURCE_DIR}/src/umpire/tpl/camp/include>;$<BUILD_INTERFACE:${EXTERNAL_BUILD_DIR}/include>;$<INSTALL_INTERFACE:${_UMPIRE_INSTALL_DIR}/include>"
         INTERFACE_LINK_LIBRARIES
         "$<BUILD_INTERFACE:${UMPIRE_BUILD_BYPRODUCTS}>;$<INSTALL_INTERFACE:${_UMPIRE_INSTALL_DIR}/lib/libumpire.${UMPIRE_DEFAULT_LIBRARY_SUFFIX}>"
         )
 
 install(TARGETS TiledArray_UMPIRE EXPORT tiledarray COMPONENT tiledarray)
 
 #TODO test Umpire
+
+endif(NOT TARGET TiledArray_UMPIRE)

external/versions.cmake

@@ -19,8 +19,8 @@ set(TA_INSTALL_EIGEN_PREVIOUS_VERSION 3.3.7)
 set(TA_INSTALL_EIGEN_URL_HASH b9e98a200d2455f06db9c661c5610496)
 set(TA_INSTALL_EIGEN_PREVIOUS_URL_HASH b9e98a200d2455f06db9c661c5610496)
 
-set(TA_TRACKED_MADNESS_TAG 6694b3adc9204dc86aba9911444aa6737171c9e3)
-set(TA_TRACKED_MADNESS_PREVIOUS_TAG f47f962213be0b1e149f5b56826992f27278128e)
+set(TA_TRACKED_MADNESS_TAG 12bd24c6cb984a639be863fc0e1364226713f7ff)
+set(TA_TRACKED_MADNESS_PREVIOUS_TAG 6694b3adc9204dc86aba9911444aa6737171c9e3)
 set(TA_TRACKED_MADNESS_VERSION 0.10.1)
 set(TA_TRACKED_MADNESS_PREVIOUS_VERSION 0.10.1)
 
@@ -30,7 +30,7 @@ set(TA_TRACKED_BTAS_PREVIOUS_TAG f4c965b778ff7df74f276f7fcf6224ac419e8ee3)
 set(TA_TRACKED_CUTT_TAG 0e8685bf82910bc7435835f846e88f1b39f47f09)
 set(TA_TRACKED_CUTT_PREVIOUS_TAG 592198b93c93b7ca79e7900b9a9f2e79f9dafec3)
 
-set(TA_TRACKED_UMPIRE_TAG f04abd1dd038c84262915a493d8f78576bb80fd0)
+set(TA_TRACKED_UMPIRE_TAG 5201a47a35e3844160dcbecd0916f8c96aa7dd07)
 set(TA_TRACKED_UMPIRE_PREVIOUS_TAG f04abd1dd038c84262915a493d8f78576bb80fd0)
 
 #set(TA_TRACKED_BLACSPP_TAG 20cfd414c5b719be1c958f4a2d57abef06df83b6 )

src/CMakeLists.txt

@@ -35,7 +35,6 @@ TiledArray/dense_shape.h
 TiledArray/dist_array.h
 TiledArray/distributed_storage.h
 TiledArray/error.h
-TiledArray/external/madness.h
 TiledArray/initialize.h
 TiledArray/perm_index.h
 TiledArray/permutation.h
@@ -117,6 +116,10 @@ TiledArray/expressions/unary_engine.h
 TiledArray/expressions/unary_expr.h
 TiledArray/expressions/index_list.h
 TiledArray/external/btas.h
+TiledArray/external/madness.h
+TiledArray/external/umpire.h
+TiledArray/host/env.h
+TiledArray/host/allocator.h
 TiledArray/math/blas.h
 TiledArray/math/gemm_helper.h
 TiledArray/math/outer.h
@@ -217,7 +220,7 @@ TiledArray/math/linalg/rank-local.cpp
 )
 
 # the list of libraries on which TiledArray depends on, will be cached later
-set(_TILEDARRAY_DEPENDENCIES MADworld TiledArray_Eigen BTAS::BTAS blaspp_headers)
+set(_TILEDARRAY_DEPENDENCIES MADworld TiledArray_Eigen BTAS::BTAS blaspp_headers TiledArray_UMPIRE)
 
 # TODO better ways to handle tiledarray cuda dependency
 if(CUDA_FOUND)
@@ -234,7 +237,7 @@ if(CUDA_FOUND)
           LANGUAGE CUDA)
 
   # the list of libraries on which TiledArray depends on
-  list(APPEND _TILEDARRAY_DEPENDENCIES CUDA::cublas CUDA::nvToolsExt TiledArray_UMPIRE TiledArray_CUTT)
+  list(APPEND _TILEDARRAY_DEPENDENCIES CUDA::cublas CUDA::nvToolsExt TiledArray_CUTT)
 
 endif(CUDA_FOUND)
 

src/TiledArray/config.h.in

@@ -78,6 +78,9 @@
 #cmakedefine TILEDARRAY_HAS_CUDA @TILEDARRAY_HAS_CUDA@
 #cmakedefine TILEDARRAY_CHECK_CUDA_ERROR @TILEDARRAY_CHECK_CUDA_ERROR@
 
+/* Is TA::Tensor memory profiling enabled? */
+#cmakedefine TA_TENSOR_MEM_PROFILE 1
+
 /* Use preprocessor to check if BTAS is available */
 #ifndef TILEDARRAY_HAS_BTAS
 #ifdef __has_include