Merge branch 'master' into kmp5/experimental/upkeep_btas

INSTALL.md

@@ -40,9 +40,9 @@ Both methods are supported. However, for most users we _strongly_ recommend to b
   - Boost.Container: header-only
   - Boost.Test: header-only or (optionally) as a compiled library, *only used for unit testing*
   - Boost.Range: header-only, *only used for unit testing*
-- [BTAS](http://github.com/ValeevGroup/BTAS), tag 85eea7796651de1bcb4781b0081a352b32bf91d5 . If usable BTAS installation is not found, TiledArray will download and compile
+- [BTAS](http://github.com/ValeevGroup/BTAS), tag 4e8f5233aa7881dccdfcc37ce07128833926d3c2 . 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 39de6cb4d262b3df1f67b3c04a37a935564ca657 .
+- [MADNESS](https://github.com/m-a-d-n-e-s-s/madness), tag 8788aea9758bfe6479cc23d39e6c77b7528009db .
   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*.

external/boost.cmake

@@ -1,32 +1,35 @@
-# Boost can be discovered by every (sub)package but only the top package can build it ...
-# if we are the top package need to include the list of Boost components to be built
-if("${CMAKE_PROJECT_NAME}" STREQUAL "${PROJECT_NAME}")
-    set(required_components
-            headers           # TA, BTAS
-            algorithm         # TA
-            container         # TA, BTAS
-            iterator          # TA, BTAS
-            random            # TA, BTAS
-            tuple             # TA
+# Boost can be discovered by every (sub)package but only the top package can *build* it ...
+# in either case must declare the components used by TA
+set(required_components
+        headers
+        algorithm
+        container
+        iterator
+        random
+        tuple
+)
+if (BUILD_TESTING)
+    list(APPEND required_components
+            test
     )
-    if (DEFINED Boost_REQUIRED_COMPONENTS)
-        list(APPEND Boost_REQUIRED_COMPONENTS
-                    ${required_components})
-        list(REMOVE_DUPLICATES Boost_REQUIRED_COMPONENTS)
-    else()
-        set(Boost_REQUIRED_COMPONENTS "${required_components}" CACHE STRING "Components of Boost to discovered or built")
-    endif()
-    set(optional_components
-            serialization     # BTAS
+endif()
+if (DEFINED Boost_REQUIRED_COMPONENTS)
+    list(APPEND Boost_REQUIRED_COMPONENTS
+            ${required_components})
+    list(REMOVE_DUPLICATES Boost_REQUIRED_COMPONENTS)
+else()
+    set(Boost_REQUIRED_COMPONENTS "${required_components}" CACHE STRING "Components of Boost to discovered or built")
+endif()
+set(optional_components
+        serialization     # BTAS
+)
+if (DEFINED Boost_OPTIONAL_COMPONENTS)
+    list(APPEND Boost_OPTIONAL_COMPONENTS
+            ${optional_components}
     )
-    if (DEFINED Boost_OPTIONAL_COMPONENTS)
-        list(APPEND Boost_OPTIONAL_COMPONENTS
-                ${optional_components}
-        )
-        list(REMOVE_DUPLICATES Boost_OPTIONAL_COMPONENTS)
-    else()
-        set(Boost_OPTIONAL_COMPONENTS "${optional_components}" CACHE STRING "Optional components of Boost to discovered or built")
-    endif()
+    list(REMOVE_DUPLICATES Boost_OPTIONAL_COMPONENTS)
+else()
+    set(Boost_OPTIONAL_COMPONENTS "${optional_components}" CACHE STRING "Optional components of Boost to discovered or built")
 endif()
 
 if (NOT DEFINED Boost_FETCH_IF_MISSING)

external/versions.cmake

@@ -1,7 +1,7 @@
 # for each dependency track both current and previous id (the variable for the latter must contain PREVIOUS)
 # to be able to auto-update them
 
-set(TA_TRACKED_VGCMAKEKIT_TAG 9b541a5f3708a58dce59d00f3c47ac030ef4d8b4)
+set(TA_TRACKED_VGCMAKEKIT_TAG d5c0a6f9ff6dc97cbb5132912733e1eb1cf73f1e)
 
 # N.B. may need to update INSTALL.md manually with the CUDA-specific version
 set(TA_TRACKED_EIGEN_VERSION 3.3.5)
@@ -11,13 +11,13 @@ set(TA_INSTALL_EIGEN_PREVIOUS_VERSION 3.3.7)
 set(TA_INSTALL_EIGEN_URL_HASH SHA256=b4c198460eba6f28d34894e3a5710998818515104d6e74e5cc331ce31e46e626)
 set(TA_INSTALL_EIGEN_PREVIOUS_URL_HASH MD5=b9e98a200d2455f06db9c661c5610496)
 
-set(TA_TRACKED_MADNESS_TAG 39de6cb4d262b3df1f67b3c04a37a935564ca657)
-set(TA_TRACKED_MADNESS_PREVIOUS_TAG 51c2728d664c096d0ea39d3b9cbf2895d8d99439)
+set(TA_TRACKED_MADNESS_TAG 8788aea9758bfe6479cc23d39e6c77b7528009db)
+set(TA_TRACKED_MADNESS_PREVIOUS_TAG 87715d98a244bff5cbff0bd2c644a8a00d882989)
 set(TA_TRACKED_MADNESS_VERSION 0.10.1)
 set(TA_TRACKED_MADNESS_PREVIOUS_VERSION 0.10.1)
 
 set(TA_TRACKED_BTAS_TAG ba34ccbb5b181f6ceebf34712dd07714cd80a4ba)
-set(TA_TRACKED_BTAS_PREVIOUS_TAG bf0c376d5cdd6f668174b2a4c67b19634d1c0da7)
+set(TA_TRACKED_BTAS_PREVIOUS_TAG b7b2ea7513b087e35c6f1b26184a3904ac1e6b14)
 
 set(TA_TRACKED_LIBRETT_TAG 6eed30d4dd2a5aa58840fe895dcffd80be7fbece)
 set(TA_TRACKED_LIBRETT_PREVIOUS_TAG 354e0ccee54aeb2f191c3ce2c617ebf437e49d83)

src/TiledArray/block_range.h

@@ -85,7 +85,7 @@ class BlockRange : public Range {
       upper[d] = upper_bound_d;
       // Check input dimensions
       TA_ASSERT(lower[d] >= range.lobound(d));
-      TA_ASSERT(lower[d] < upper[d]);
+      TA_ASSERT(lower[d] <= upper[d]);
       TA_ASSERT(upper[d] <= range.upbound(d));
       extent[d] = upper[d] - lower[d];
       TA_ASSERT(extent[d] ==
@@ -132,7 +132,7 @@ class BlockRange : public Range {
       upper[d] = upper_bound_d;
       // Check input dimensions
       TA_ASSERT(lower[d] >= range.lobound(d));
-      TA_ASSERT(lower[d] < upper[d]);
+      TA_ASSERT(lower[d] <= upper[d]);
       TA_ASSERT(upper[d] <= range.upbound(d));
       extent[d] = upper[d] - lower[d];
       TA_ASSERT(extent[d] ==
@@ -177,9 +177,10 @@ class BlockRange : public Range {
   /// \param range the host Range
   /// \param lower_bound A sequence of lower bounds for each dimension
   /// \param upper_bound A sequence of upper bounds for each dimension
+  /// \note Zero-extent blocks along any mode is possible, i.e. `lower_bound[d] == upper_bound[d]` is supported
   /// \throw TiledArray::Exception When the size of \p lower_bound is not
   /// equal to that of \p upper_bound.
-  /// \throw TiledArray::Exception When `lower_bound[i] >= upper_bound[i]`
+  /// \throw TiledArray::Exception When `lower_bound[i] > upper_bound[i]`
   // clang-format on
   template <typename Index1, typename Index2,
             typename = std::enable_if_t<detail::is_integral_range_v<Index1> &&
@@ -204,9 +205,10 @@ class BlockRange : public Range {
   /// \param range the host Range
   /// \param lower_bound An initializer list of lower bounds for each dimension
   /// \param upper_bound An initializer list of upper bounds for each dimension
+  /// \note Zero-extent blocks along any mode is possible, i.e. `lower_bound[d] == upper_bound[d]` is supported
   /// \throw TiledArray::Exception When the size of \p lower_bound is not
   /// equal to that of \p upper_bound.
-  /// \throw TiledArray::Exception When `lower_bound[i] >= upper_bound[i]`
+  /// \throw TiledArray::Exception When `lower_bound[i] > upper_bound[i]`
   // clang-format on
   template <typename Index1, typename Index2,
             typename = std::enable_if_t<std::is_integral_v<Index1> &&
@@ -247,7 +249,8 @@ class BlockRange : public Range {
   /// \endcode
   /// \tparam PairRange Type representing a range of generalized pairs (see TiledArray::detail::is_gpair_v )
   /// \param bounds A range of {lower,upper} bounds for each dimension
-  /// \throw TiledArray::Exception When `bounds[i].lower>=bounds[i].upper` for any \c i .
+  /// \note Zero-extent blocks along any mode is possible, i.e. `bounds[d].lower == bounds[d].upper` is supported
+  /// \throw TiledArray::Exception When `bounds[i].lower>bounds[i].upper` for any \c i .
   // clang-format on
   template <typename PairRange,
             typename = std::enable_if_t<detail::is_gpair_range_v<PairRange>>>
@@ -264,8 +267,9 @@ class BlockRange : public Range {
   ///   BlockRange br0(r, {std::make_pair(0,4), std::pair{1,6}, std::pair(2,8)});
   /// \endcode
   /// \tparam GPair a generalized pair of integral types
-  /// \param bound A range of {lower,upper} bounds for each dimension
-  /// \throw TiledArray::Exception When `bound[i].lower>=bound[i].upper` for any \c i .
+  /// \param bounds A range of {lower,upper} bounds for each dimension
+  /// \note Zero-extent blocks along any mode is possible, i.e. `bounds[d].lower == bounds[d].upper` is supported
+  /// \throw TiledArray::Exception When `bounds[i].lower>bounds[i].upper` for any \c i .
   // clang-format on
   template <typename GPair>
   BlockRange(const Range& range, const std::initializer_list<GPair>& bounds,
@@ -290,8 +294,9 @@ class BlockRange : public Range {
   ///   BlockRange br0(r, {{0,4}, {1,6}, {2,8}});
   /// \endcode
   /// \tparam Index An integral type
-  /// \param bound A range of {lower,upper} bounds for each dimension
-  /// \throw TiledArray::Exception When `bound[i].lower>=bound[i].upper` for any \c i .
+  /// \param bounds A range of {lower,upper} bounds for each dimension
+  /// \note Zero-extent blocks along any mode is possible, i.e. `bounds[d].lower == bounds[d].upper` is supported
+  /// \throw TiledArray::Exception When `bounds[i].lower>bounds[i].upper` for any \c i .
   // clang-format on
   template <typename Index,
             typename = std::enable_if_t<std::is_integral_v<Index>>>
@@ -354,6 +359,8 @@ class BlockRange : public Range {
   /// \return The ordinal index in the
   /// \throw TiledArray::Exception When \c index is not included in this range
   ordinal_type ordinal(ordinal_type ord) const {
+    // ordinals are useless for zero-volume ranges
+    TA_ASSERT(volume() != 0);
     // Check that ord is contained by this range.
     TA_ASSERT(Range::includes_ordinal(ord));
 
@@ -414,7 +421,7 @@ class BlockRange : public Range {
   template <typename Archive>
   void serialize(Archive& ar) const {
     Range::serialize(ar);
-    ar& block_offset_;
+    ar & block_offset_;
   }
 };  // BlockRange
 

src/TiledArray/device/btas.h

@@ -77,10 +77,12 @@ ::btas::Tensor<T, Range, Storage> gemm(
   gemm_helper.compute_matrix_sizes(m, n, k, left.range(), right.range());
 
   // Get the leading dimension for left and right matrices.
-  const integer lda =
-      (gemm_helper.left_op() == TiledArray::math::blas::Op::NoTrans ? k : m);
-  const integer ldb =
-      (gemm_helper.right_op() == TiledArray::math::blas::Op::NoTrans ? n : k);
+  const integer lda = std::max(
+      integer{1},
+      (gemm_helper.left_op() == TiledArray::math::blas::Op::NoTrans ? k : m));
+  const integer ldb = std::max(
+      integer{1},
+      (gemm_helper.right_op() == TiledArray::math::blas::Op::NoTrans ? n : k));
 
   T factor_t = T(factor);
   T zero(0);
@@ -112,10 +114,11 @@ ::btas::Tensor<T, Range, Storage> gemm(
 
     static_assert(::btas::boxrange_iteration_order<Range>::value ==
                   ::btas::boxrange_iteration_order<Range>::row_major);
+    const integer ldc = std::max(integer{1}, n);
     blas::gemm(blas::Layout::ColMajor, gemm_helper.right_op(),
                gemm_helper.left_op(), n, m, k, factor_t,
                device_data(right.storage()), ldb, device_data(left.storage()),
-               lda, zero, device_data(result.storage()), n, queue);
+               lda, zero, device_data(result.storage()), ldc, queue);
 
     device::sync_madness_task_with(stream);
   }
@@ -185,10 +188,12 @@ void gemm(::btas::Tensor<T, Range, Storage> &result,
   gemm_helper.compute_matrix_sizes(m, n, k, left.range(), right.range());
 
   // Get the leading dimension for left and right matrices.
-  const integer lda =
-      (gemm_helper.left_op() == TiledArray::math::blas::Op::NoTrans ? k : m);
-  const integer ldb =
-      (gemm_helper.right_op() == TiledArray::math::blas::Op::NoTrans ? n : k);
+  const integer lda = std::max(
+      integer{1},
+      (gemm_helper.left_op() == TiledArray::math::blas::Op::NoTrans ? k : m));
+  const integer ldb = std::max(
+      integer{1},
+      (gemm_helper.right_op() == TiledArray::math::blas::Op::NoTrans ? n : k));
 
   auto &queue = blasqueue_for(result.range());
   const auto stream = device::Stream(queue.device(), queue.stream());
@@ -207,10 +212,11 @@ void gemm(::btas::Tensor<T, Range, Storage> &result,
 
     static_assert(::btas::boxrange_iteration_order<Range>::value ==
                   ::btas::boxrange_iteration_order<Range>::row_major);
+    const integer ldc = std::max(integer{1}, n);
     blas::gemm(blas::Layout::ColMajor, gemm_helper.right_op(),
                gemm_helper.left_op(), n, m, k, factor_t,
                device_data(right.storage()), ldb, device_data(left.storage()),
-               lda, one, device_data(result.storage()), n, queue);
+               lda, one, device_data(result.storage()), ldc, queue);
     device::sync_madness_task_with(stream);
   }
 }