Replace thread_manager::wait with pika::wait

CMakeLists.txt

@@ -171,7 +171,7 @@ else()
 endif()
 
 # ----- pika
-find_package(pika 0.15.1 REQUIRED)
+find_package(pika 0.17.0 REQUIRED)
 find_package(pika-algorithms 0.1.1 REQUIRED)
 
 # ----- BLASPP/LAPACKPP

ci/.gitlab-ci.yml

@@ -17,7 +17,7 @@ stages:
   timeout: 6 hours
   variables:
     GIT_SUBMODULE_STRATEGY: recursive
-    SPACK_SHA: b313b28e64c15761be0d45a16c922c25b2786f76
+    SPACK_SHA: 566754440f9dfed9accd25db7f1a67b0cd074fcd
     SPACK_DLAF_REPO: ./spack
   before_script:
     - !reference [.fetch-registry-tokens, script]

include/dlaf/eigensolver/reduction_to_band/impl.h

@@ -15,6 +15,7 @@
 
 #include <pika/barrier.hpp>
 #include <pika/execution.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/blas/tile.h>
 #include <dlaf/common/assert.h>
@@ -1088,7 +1089,7 @@ Matrix<T, Device::CPU> ReductionToBand<B, D, T>::call(comm::CommunicatorGrid gri
   // Note:
   // This is a temporary workaround.
   // See issue https://github.com/eth-cscs/DLA-Future/issues/729
-  pika::threads::get_thread_manager().wait();
+  pika::wait();
 
   common::Pipeline<comm::Communicator> mpi_col_chain_panel(grid.colCommunicator().clone());
   common::Pipeline<comm::Communicator> mpi_row_chain(grid.rowCommunicator().clone());

spack/packages/dla-future/package.py

@@ -56,6 +56,7 @@ class DlaFuture(CMakePackage, CudaPackage, ROCmPackage):
 
     depends_on("[email protected]:", when="@0.1")
     depends_on("[email protected]:", when="@0.2.0:")
+    depends_on("[email protected]:", when="@master")
     depends_on("[email protected]:")
     depends_on("pika +mpi")
     depends_on("pika +cuda", when="+cuda")

test/include/dlaf_test/eigensolver/test_eigensolver_correctness.h

@@ -15,6 +15,8 @@
 
 #include <blas.hh>
 
+#include <pika/init.hpp>
+
 #include <dlaf/common/single_threaded_blas.h>
 #include <dlaf/communication/communicator_grid.h>
 #include <dlaf/matrix/matrix.h>
@@ -42,7 +44,7 @@ void testEigensolverCorrectness(const blas::Uplo uplo, Matrix<const T, Device::C
   // calls that might lead to deadlocks.
   constexpr bool isDistributed = (sizeof...(grid) == 1);
   if constexpr (isDistributed)
-    pika::threads::get_thread_manager().wait();
+    pika::wait();
 
   const SizeType m = reference.size().rows();
 

test/src/gtest_mpipika_main.cpp

@@ -43,7 +43,6 @@
 
 #include <pika/init.hpp>
 #include <pika/program_options.hpp>
-#include <pika/runtime.hpp>
 
 #include <dlaf/init.h>
 
@@ -89,7 +88,7 @@ GTEST_API_ int main(int argc, char** argv) {
 
   protected:
     virtual void OnTestEnd(const ::testing::TestInfo& test_info) override {
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
       MPIListener::OnTestEnd(test_info);
     }
   };

test/unit/eigensolver/test_bt_band_to_tridiag.cpp

@@ -8,6 +8,8 @@
 // SPDX-License-Identifier: BSD-3-Clause
 //
 
+#include <pika/init.hpp>
+
 #include <dlaf/common/single_threaded_blas.h>
 #include <dlaf/communication/communicator_grid.h>
 #include <dlaf/eigensolver/band_to_tridiag.h>  // for nrSweeps/nrStepsForSweep
@@ -267,7 +269,7 @@ TYPED_TEST(BacktransformationBandToTridiagTestMC, CorrectnessDistributed) {
     for (const auto& [m, n, mb, nb, group_size, b] : configs) {
       getTuneParameters().bt_band_to_tridiag_hh_apply_group_size = group_size;
       testBacktransformation<Backend::MC, Device::CPU, TypeParam>(comm_grid, m, n, mb, nb, b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -285,7 +287,7 @@ TYPED_TEST(BacktransformationBandToTridiagTestGPU, CorrectnessDistributed) {
     for (const auto& [m, n, mb, nb, group_size, b] : configs) {
       getTuneParameters().bt_band_to_tridiag_hh_apply_group_size = group_size;
       testBacktransformation<Backend::GPU, Device::GPU, TypeParam>(comm_grid, m, n, mb, nb, b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -308,7 +310,7 @@ TYPED_TEST(BacktransformationBandToTridiagTestMC, CorrectnessDistributedSubBand)
     for (const auto& [m, n, mb, nb, group_size, b] : configs_subband) {
       getTuneParameters().bt_band_to_tridiag_hh_apply_group_size = group_size;
       testBacktransformation<Backend::MC, Device::CPU, TypeParam>(comm_grid, m, n, mb, nb, b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -326,7 +328,7 @@ TYPED_TEST(BacktransformationBandToTridiagTestGPU, CorrectnessDistributedSubBand
     for (const auto& [m, n, mb, nb, group_size, b] : configs_subband) {
       getTuneParameters().bt_band_to_tridiag_hh_apply_group_size = group_size;
       testBacktransformation<Backend::GPU, Device::GPU, TypeParam>(comm_grid, m, n, mb, nb, b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }

test/unit/eigensolver/test_bt_reduction_to_band.cpp

@@ -14,7 +14,7 @@
 #include <tuple>
 #include <vector>
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/common/assert.h>
 #include <dlaf/common/index2d.h>
@@ -253,7 +253,7 @@ TYPED_TEST(BackTransformationReductionToBandEigenSolverTestMC, CorrectnessDistri
     for (const auto& [m, n, mb, nb, b] : sizes) {
       testBackTransformationReductionToBand<TypeParam, Backend::MC, Device::CPU>(comm_grid, m, n, mb, nb,
                                                                                  b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -264,7 +264,7 @@ TYPED_TEST(BackTransformationReductionToBandEigenSolverTestGPU, CorrectnessDistr
     for (const auto& [m, n, mb, nb, b] : sizes) {
       testBackTransformationReductionToBand<TypeParam, Backend::GPU, Device::GPU>(comm_grid, m, n, mb,
                                                                                   nb, b);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }

test/unit/eigensolver/test_gen_eigensolver.cpp

@@ -11,6 +11,8 @@
 #include <functional>
 #include <tuple>
 
+#include <pika/init.hpp>
+
 #include <dlaf/common/single_threaded_blas.h>
 #include <dlaf/eigensolver/gen_eigensolver.h>
 #include <dlaf/matrix/copy.h>
@@ -73,7 +75,7 @@ void testGenEigensolverCorrectness(const blas::Uplo uplo, Matrix<const T, Device
   // calls that might lead to deadlocks.
   constexpr bool isDistributed = (sizeof...(grid) == 1);
   if constexpr (isDistributed)
-    pika::threads::get_thread_manager().wait();
+    pika::wait();
 
   const SizeType m = reference_a.size().rows();
 

test/unit/eigensolver/test_gen_to_std.cpp

@@ -11,7 +11,7 @@
 #include <functional>
 #include <tuple>
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/communication/communicator_grid.h>
 #include <dlaf/eigensolver/gen_to_std.h>
@@ -124,7 +124,7 @@ TYPED_TEST(EigensolverGenToStdTestMC, CorrectnessDistributed) {
     for (const auto uplo : blas_uplos) {
       for (const auto& [m, mb] : sizes) {
         testGenToStdEigensolver<TypeParam, Backend::MC, Device::CPU>(comm_grid, uplo, m, mb);
-        pika::threads::get_thread_manager().wait();
+        pika::wait();
       }
     }
   }
@@ -144,7 +144,7 @@ TYPED_TEST(EigensolverGenToStdTestGPU, CorrectnessDistributed) {
     for (const auto uplo : blas_uplos) {
       for (const auto& [m, mb] : sizes) {
         testGenToStdEigensolver<TypeParam, Backend::GPU, Device::GPU>(comm_grid, uplo, m, mb);
-        pika::threads::get_thread_manager().wait();
+        pika::wait();
       }
     }
   }

test/unit/eigensolver/test_reduction_to_band.cpp

@@ -11,7 +11,7 @@
 #include <cmath>
 
 #include <pika/execution.hpp>
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/common/index2d.h>
 #include <dlaf/common/pipeline.h>
@@ -405,7 +405,7 @@ void testReductionToBand(comm::CommunicatorGrid grid, const LocalElementSize siz
   checkUpperPartUnchanged(reference, matrix_a_h);
 
   // Wait for all work to finish before doing blocking communication
-  pika::threads::get_thread_manager().wait();
+  pika::wait();
 
   auto mat_v = allGather(blas::Uplo::Lower, matrix_a_h, grid);
   auto mat_b = makeLocal(matrix_a_h);

test/unit/eigensolver/test_tridiag_solver_distributed.cpp

@@ -8,7 +8,7 @@
 // SPDX-License-Identifier: BSD-3-Clause
 //
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/communication/kernels.h>
 #include <dlaf/eigensolver/tridiag_solver.h>
@@ -251,7 +251,7 @@ TYPED_TEST(TridiagSolverDistTestMC, Laplace1D) {
   for (const auto& comm_grid : this->commGrids()) {
     for (auto [n, nb] : tested_problems) {
       solveDistributedLaplace1D<Backend::MC, Device::CPU, TypeParam>(comm_grid, n, nb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -260,7 +260,7 @@ TYPED_TEST(TridiagSolverDistTestMC, Random) {
   for (const auto& comm_grid : this->commGrids()) {
     for (auto [n, nb] : tested_problems) {
       solveRandomTridiagMatrix<Backend::MC, Device::CPU, TypeParam>(comm_grid, n, nb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -270,7 +270,7 @@ TYPED_TEST(TridiagSolverDistTestGPU, Laplace1D) {
   for (const auto& comm_grid : this->commGrids()) {
     for (auto [n, nb] : tested_problems) {
       solveDistributedLaplace1D<Backend::GPU, Device::GPU, TypeParam>(comm_grid, n, nb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -279,7 +279,7 @@ TYPED_TEST(TridiagSolverDistTestGPU, Random) {
   for (const auto& comm_grid : this->commGrids()) {
     for (auto [n, nb] : tested_problems) {
       solveRandomTridiagMatrix<Backend::GPU, Device::GPU, TypeParam>(comm_grid, n, nb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }

test/unit/factorization/test_cholesky.cpp

@@ -11,7 +11,7 @@
 #include <functional>
 #include <tuple>
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/communication/communicator_grid.h>
 #include <dlaf/factorization/cholesky.h>
@@ -111,7 +111,7 @@ TYPED_TEST(CholeskyTestMC, CorrectnessDistributed) {
     for (auto uplo : blas_uplos) {
       for (const auto& [m, mb] : sizes) {
         testCholesky<TypeParam, Backend::MC, Device::CPU>(comm_grid, uplo, m, mb);
-        pika::threads::get_thread_manager().wait();
+        pika::wait();
       }
     }
   }
@@ -131,7 +131,7 @@ TYPED_TEST(CholeskyTestGPU, CorrectnessDistributed) {
     for (auto uplo : blas_uplos) {
       for (const auto& [m, mb] : sizes) {
         testCholesky<TypeParam, Backend::GPU, Device::GPU>(comm_grid, uplo, m, mb);
-        pika::threads::get_thread_manager().wait();
+        pika::wait();
       }
     }
   }

test/unit/matrix/test_matrix.cpp

@@ -13,6 +13,7 @@
 #include <vector>
 
 #include <pika/execution.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/communication/communicator_grid.h>
 #include <dlaf/matrix/copy.h>
@@ -1892,7 +1893,7 @@ void testReshuffling(const TestReshuffling& config, CommunicatorGrid grid) {
 
   // Note: ensure that everything finishes before next call to Communicator::clone() that might block a
   // working thread (and if it is just one, it would deadlock)
-  pika::threads::get_thread_manager().wait();
+  pika::wait();
 }
 
 TYPED_TEST(MatrixTest, CopyReshuffling) {

test/unit/multiplication/test_multiplication_general.cpp

@@ -8,6 +8,8 @@
 // SPDX-License-Identifier: BSD-3-Clause
 //
 
+#include <pika/init.hpp>
+
 #include <dlaf/blas/enum_output.h>
 #include <dlaf/common/assert.h>
 #include <dlaf/communication/communicator_grid.h>
@@ -172,7 +174,7 @@ TYPED_TEST(GeneralSubMultiplicationDistTestMC, CorrectnessDistributed) {
       const TypeParam beta = TypeUtilities<TypeParam>::element(-2.6, .7);
       testGeneralSubMultiplication<TypeParam, Backend::MC, Device::CPU>(comm_grid, a, b, alpha, beta, m,
                                                                         mb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }
@@ -185,7 +187,7 @@ TYPED_TEST(GeneralSubMultiplicationDistTestGPU, CorrectnessDistributed) {
       const TypeParam beta = TypeUtilities<TypeParam>::element(-2.6, .7);
       testGeneralSubMultiplication<TypeParam, Backend::GPU, Device::GPU>(comm_grid, a, b, alpha, beta, m,
                                                                          mb);
-      pika::threads::get_thread_manager().wait();
+      pika::wait();
     }
   }
 }

test/unit/multiplication/test_multiplication_hermitian.cpp

@@ -11,7 +11,7 @@
 #include <functional>
 #include <tuple>
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/blas/tile.h>
 #include <dlaf/communication/communicator_grid.h>
@@ -175,7 +175,7 @@ TYPED_TEST(HermitianMultiplicationTestMC, CorrectnessDistributed) {
 
           testHermitianMultiplication<TypeParam, Backend::MC, Device::CPU>(comm_grid, side, uplo, m, n,
                                                                            mb, nb, alpha, beta);
-          pika::threads::get_thread_manager().wait();
+          pika::wait();
         }
       }
     }
@@ -212,7 +212,7 @@ TYPED_TEST(HermitianMultiplicationTestGPU, CorrectnessDistributed) {
 
           testHermitianMultiplication<TypeParam, Backend::GPU, Device::GPU>(comm_grid, side, uplo, m, n,
                                                                             mb, nb, alpha, beta);
-          pika::threads::get_thread_manager().wait();
+          pika::wait();
         }
       }
     }

test/unit/multiplication/test_multiplication_triangular.cpp

@@ -11,7 +11,7 @@
 #include <functional>
 #include <tuple>
 
-#include <pika/runtime.hpp>
+#include <pika/init.hpp>
 
 #include <dlaf/blas/tile.h>
 #include <dlaf/communication/communicator_grid.h>
@@ -172,7 +172,7 @@ TYPED_TEST(TriangularMultiplicationTestMC, CorrectnessDistributed) {
               testTriangularMultiplication<TypeParam, Backend::MC, Device::CPU>(comm_grid, side, uplo,
                                                                                 op, diag, alpha, m, n,
                                                                                 mb, nb);
-              pika::threads::get_thread_manager().wait();
+              pika::wait();
             }
           }
         }
@@ -213,7 +213,7 @@ TYPED_TEST(TriangularMultiplicationTestGPU, CorrectnessDistributed) {
               testTriangularMultiplication<TypeParam, Backend::GPU, Device::GPU>(comm_grid, side, uplo,
                                                                                  op, diag, alpha, m, n,
                                                                                  mb, nb);
-              pika::threads::get_thread_manager().wait();
+              pika::wait();
             }
           }
         }