Merge pull request #2197 from ndellingwood/master-release-4.3.01

Release 4.3.01

.github/workflows/docs.yml

@@ -16,9 +16,11 @@ jobs:
       - name: Install Dependencies
         run: |
           brew install doxygen
-          python3 -m pip install sphinx -v "sphinx==6.2.1"
-          python3 -m pip install breathe
-          python3 -m pip install sphinx-rtd-theme
+          python3 -m venv .venv
+          . .venv/bin/activate
+          pip install sphinx -v "sphinx==6.2.1"
+          pip install breathe
+          pip install sphinx-rtd-theme
           sphinx-build --version
           doxygen --version
 
@@ -52,8 +54,10 @@ jobs:
         working-directory: kokkos/build
         run: make -j2 install
 
+      # sphinx needs to be available at configure time for the target to be generated
       - name: configure_kokkos_kernels
         run: |
+          . .venv/bin/activate
           mkdir -p kokkos-kernels/{build,install}
           cd kokkos-kernels/build
           cmake \
@@ -81,5 +85,7 @@ jobs:
          fi
 
       - name: build_kokkos_kernels_sphinx
-        working-directory: kokkos-kernels/build
-        run: make Sphinx
+        run: | 
+          . .venv/bin/activate
+          cd kokkos-kernels/build
+          make Sphinx

.github/workflows/osx.yml

@@ -58,7 +58,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           repository: kokkos/kokkos
-          ref: ${{ github.base_ref }}
+          ref: 4.3.00
           path: kokkos
 
       - name: configure_kokkos

CHANGELOG.md

@@ -1,5 +1,17 @@
 # Change Log
 
+## [4.3.01](https://github.com/kokkos/kokkos-kernels/tree/4.3.01)
+[Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/4.3.00...4.3.01)
+
+### Bug Fixes:
+- sparse: block spiluk fixes [\#2172](https://github.com/kokkos/kokkos-kernels/pull/2172)
+- magma: tpl interaction fixes [\#2176](https://github.com/kokkos/kokkos-kernels/pull/2176), [\#2178](https://github.com/kokkos/kokkos-kernels/pull/2178), [\#2181](https://github.com/kokkos/kokkos-kernels/pull/2181)
+- trsv: Add early return if numRows == 0 in trsv to avoid integer divide-by-zero error [\#2180](https://github.com/kokkos/kokkos-kernels/pull/2180)
+- blas tpl: resolve potential duplicate symbol [\#2183](https://github.com/kokkos/kokkos-kernels/pull/2183)
+- spmv: permformance fix, add back special path for rank-2 x/y with 1 column [\#2164](https://github.com/kokkos/kokkos-kernels/pull/2164), [\#2168](https://github.com/kokkos/kokkos-kernels/pull/2168)
+- BsrMatrix: Fix HostMirror typedef [\#2196](https://github.com/kokkos/kokkos-kernels/pull/2196)
+- GA: Fix macOS docs build [\#2190](https://github.com/kokkos/kokkos-kernels/pull/2190)
+
 ## [4.3.00](https://github.com/kokkos/kokkos-kernels/tree/4.3.00) (2024-03-19)
 [Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/4.2.01...4.3.00)
 
@@ -639,7 +651,7 @@
 ## [3.6.00](https://github.com/kokkos/kokkos-kernels/tree/3.6.00) (2022-02-18)
 [Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/3.5.00...3.6.00)
 
-### Features: 
+### Features:
 
 #### Batched Sparse Linear algebra
 - Kokkos Kernels is adding a new component to the library: batched sparse linear algebra.
@@ -673,7 +685,7 @@
 - SpMV: adding support for rocSPARSE TPL [\#1221](https://github.com/kokkos/kokkos-kernels/pull/1221)
 
 #### Additional new features
-- bhalf: Unit test Batched GEMM [\#1251](https://github.com/kokkos/kokkos-kernels/pull/1251) 
+- bhalf: Unit test Batched GEMM [\#1251](https://github.com/kokkos/kokkos-kernels/pull/1251)
 -   and demostrate GMRES example convergence with bhalf_t (https://github.com/kokkos/kokkos-kernels/pull/1300)
 - Stream interface: adding stream support in GEMV and GEMM [\#1131](https://github.com/kokkos/kokkos-kernels/pull/1131)
 - Improve double buffering batched gemm performance [\#1217](https://github.com/kokkos/kokkos-kernels/pull/1217)
@@ -962,7 +974,7 @@
 ## [3.1.01](https://github.com/kokkos/kokkos-kernels/tree/3.1.01) (2020-05-04)
 [Full Changelog](https://github.com/kokkos/kokkos-kernels/compare/3.1.00...3.1.01)
 
-** Fixed bugs:** 
+** Fixed bugs:**
 
 - KokkosBatched QR PR breaking nightly tests [\#691](https://github.com/kokkos/kokkos-kernels/issues/691)
 

CMakeLists.txt

@@ -11,7 +11,7 @@ SET(KOKKOSKERNELS_TOP_SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR})
 
 SET(KokkosKernels_VERSION_MAJOR 4)
 SET(KokkosKernels_VERSION_MINOR 3)
-SET(KokkosKernels_VERSION_PATCH 0)
+SET(KokkosKernels_VERSION_PATCH 1)
 SET(KokkosKernels_VERSION "${KokkosKernels_VERSION_MAJOR}.${KokkosKernels_VERSION_MINOR}.${KokkosKernels_VERSION_PATCH}")
 
 #Set variables for config file
@@ -127,13 +127,13 @@ ELSE()
   IF (NOT KOKKOSKERNELS_HAS_TRILINOS AND NOT KOKKOSKERNELS_HAS_PARENT)
     # This is a standalone build
     FIND_PACKAGE(Kokkos REQUIRED)
-    IF((${Kokkos_VERSION} VERSION_GREATER_EQUAL "4.1.0") AND (${Kokkos_VERSION} VERSION_LESS_EQUAL "4.3.0"))
+    IF((${Kokkos_VERSION} VERSION_GREATER_EQUAL "4.1.0") AND (${Kokkos_VERSION} VERSION_LESS_EQUAL "4.3.1"))
       MESSAGE(STATUS "Found Kokkos version ${Kokkos_VERSION} at ${Kokkos_DIR}")
       IF((${Kokkos_VERSION} VERSION_GREATER "4.3.99"))
         MESSAGE(WARNING "Configuring with Kokkos ${Kokkos_VERSION} which is newer than the expected develop branch - version check may need update")
       ENDIF()
     ELSE()
-      MESSAGE(FATAL_ERROR "Kokkos Kernels ${KokkosKernels_VERSION} requires Kokkos_VERSION 4.1.0, 4.2.0, 4.2.1 or 4.3.0")
+      MESSAGE(FATAL_ERROR "Kokkos Kernels ${KokkosKernels_VERSION} requires Kokkos_VERSION 4.1.0, 4.2.0, 4.2.1, 4.3.0, or 4.3.1")
     ENDIF()
   ENDIF()
 

batched/dense/impl/KokkosBatched_Trsm_Team_Impl.hpp

@@ -99,6 +99,48 @@ struct TeamTrsm<MemberType, Side::Right, Uplo::Upper, Trans::NoTranspose,
   }
 };
 
+///
+/// R/L/NT
+///
+/// B := (alpha*B) inv(tril(A))
+/// A(n x n), B(m x n)
+
+template <typename MemberType, typename ArgDiag>
+struct TeamTrsm<MemberType, Side::Right, Uplo::Lower, Trans::NoTranspose,
+                ArgDiag, Algo::Trsm::Unblocked> {
+  template <typename ScalarType, typename AViewType, typename BViewType>
+  KOKKOS_INLINE_FUNCTION static int invoke(const MemberType &member,
+                                           const ScalarType alpha,
+                                           const AViewType &A,
+                                           const BViewType &B) {
+    return TeamTrsmInternalLeftUpper<Algo::Trsm::Unblocked>::invoke(
+        member, ArgDiag::use_unit_diag, B.extent(1), B.extent(0), alpha,
+        A.data(), A.stride_1(), A.stride_0(), B.data(), B.stride_1(),
+        B.stride_0());
+  }
+};
+
+template <typename MemberType, typename ArgDiag>
+struct TeamTrsm<MemberType, Side::Right, Uplo::Lower, Trans::NoTranspose,
+                ArgDiag, Algo::Trsm::Blocked> {
+  template <typename ScalarType, typename AViewType, typename BViewType>
+  KOKKOS_INLINE_FUNCTION static int invoke(const MemberType &member,
+                                           const ScalarType alpha,
+                                           const AViewType &A,
+                                           const BViewType &B) {
+    return TeamTrsmInternalLeftUpper<Algo::Trsm::Blocked>::invoke(
+        member, ArgDiag::use_unit_diag, B.extent(1), B.extent(0), alpha,
+        A.data(), A.stride_1(), A.stride_0(), B.data(), B.stride_1(),
+        B.stride_0());
+  }
+};
+
+///
+/// R/U/T
+///
+/// B := (alpha*B) inv(triu(A))
+/// A(n x n), B(m x n)
+
 template <typename MemberType, typename ArgDiag>
 struct TeamTrsm<MemberType, Side::Right, Uplo::Upper, Trans::Transpose, ArgDiag,
                 Algo::Trsm::Unblocked> {
@@ -107,7 +149,7 @@ struct TeamTrsm<MemberType, Side::Right, Uplo::Upper, Trans::Transpose, ArgDiag,
                                            const ScalarType alpha,
                                            const AViewType &A,
                                            const BViewType &B) {
-    return TeamTrsmInternalLeftLower<Algo::Trsm::Unblocked>::invoke(
+    return TeamTrsmInternalLeftUpper<Algo::Trsm::Unblocked>::invoke(
         member, ArgDiag::use_unit_diag, B.extent(1), B.extent(0), alpha,
         A.data(), A.stride_0(), A.stride_1(), B.data(), B.stride_1(),
         B.stride_0());
@@ -122,7 +164,7 @@ struct TeamTrsm<MemberType, Side::Right, Uplo::Upper, Trans::Transpose, ArgDiag,
                                            const ScalarType alpha,
                                            const AViewType &A,
                                            const BViewType &B) {
-    return TeamTrsmInternalLeftLower<Algo::Trsm::Blocked>::invoke(
+    return TeamTrsmInternalLeftUpper<Algo::Trsm::Blocked>::invoke(
         member, ArgDiag::use_unit_diag, B.extent(1), B.extent(0), alpha,
         A.data(), A.stride_0(), A.stride_1(), B.data(), B.stride_1(),
         B.stride_0());

batched/dense/unit_test/Test_Batched_BatchedGemm.hpp

@@ -229,7 +229,11 @@ void impl_test_batched_gemm(const int N, const int matAdim1, const int matAdim2,
     ASSERT_EQ(batchedGemmHandleCublas.vecLen, 0);
 #endif
 
-#if defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA)
+    // FIXME temporary workaround to run this magma test only if cublas is not
+    // enabled the design of the BatchedGemmHandle currently does not allow
+    // simultanous testing in this way. See issue #2177
+#if defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA) && \
+    !defined(KOKKOSKERNELS_ENABLE_TPL_CUBLAS)
     magma_queue_t magma_queue;
     BatchedGemmHandle batchedGemmHandleMagma(magma_queue, GemmTplAlgos::MAGMA,
                                              0, 0);

blas/tpls/KokkosBlas2_syr2_tpl_spec_decl_blas.hpp

@@ -163,7 +163,7 @@ namespace Impl {
              ETI_SPEC_AVAIL>::syr2(space, trans, uplo, alpha, X, Y, A);       \
       } else {                                                                \
         if (A_is_ll) {                                                        \
-          HostBlas<std::complex<double>>::zher2(                              \
+          HostBlas<std::complex<double>>::her2(                               \
               uplo[0], N, alpha,                                              \
               reinterpret_cast<const std::complex<double>*>(X.data()), one,   \
               reinterpret_cast<const std::complex<double>*>(Y.data()), one,   \
@@ -220,7 +220,7 @@ namespace Impl {
              ETI_SPEC_AVAIL>::syr2(space, trans, uplo, alpha, X, Y, A);       \
       } else {                                                                \
         if (A_is_ll) {                                                        \
-          HostBlas<std::complex<float>>::cher2(                               \
+          HostBlas<std::complex<float>>::her2(                                \
               uplo[0], N, alpha,                                              \
               reinterpret_cast<const std::complex<float>*>(X.data()), one,    \
               reinterpret_cast<const std::complex<float>*>(Y.data()), one,    \

blas/tpls/KokkosBlas2_syr_tpl_spec_decl_blas.hpp

@@ -139,7 +139,7 @@ namespace Impl {
             space, trans, uplo, alpha, X, A);                                \
       } else {                                                               \
         if (A_is_ll) {                                                       \
-          HostBlas<std::complex<double>>::zher<double>(                      \
+          HostBlas<std::complex<double>>::her<double>(                       \
               uplo[0], N, alpha.real(),                                      \
               reinterpret_cast<const std::complex<double>*>(X.data()), one,  \
               reinterpret_cast<std::complex<double>*>(A.data()), LDA);       \
@@ -188,7 +188,7 @@ namespace Impl {
             space, trans, uplo, alpha, X, A);                                \
       } else {                                                               \
         if (A_is_ll && (alpha.imag() == 0.)) {                               \
-          HostBlas<std::complex<float>>::cher<float>(                        \
+          HostBlas<std::complex<float>>::her<float>(                         \
               uplo[0], N, alpha.real(),                                      \
               reinterpret_cast<const std::complex<float>*>(X.data()), one,   \
               reinterpret_cast<std::complex<float>*>(A.data()), LDA);        \

blas/tpls/KokkosBlas_Cuda_tpl.cpp

@@ -16,3 +16,4 @@
 #include <Kokkos_Core.hpp>
 #include <KokkosKernels_config.h>
 #include <KokkosBlas_Cuda_tpl.hpp>
+#include <KokkosBlas_Magma_tpl.hpp>

blas/tpls/KokkosBlas_Cuda_tpl.hpp

@@ -39,26 +39,4 @@ CudaBlasSingleton& CudaBlasSingleton::singleton() {
 }  // namespace KokkosBlas
 #endif  // defined (KOKKOSKERNELS_ENABLE_TPL_CUBLAS)
 
-#if defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA)
-#include <KokkosBlas_tpl_spec.hpp>
-
-namespace KokkosBlas {
-namespace Impl {
-
-MagmaSingleton::MagmaSingleton() {
-  magma_int_t stat = magma_init();
-  if (stat != MAGMA_SUCCESS) Kokkos::abort("MAGMA initialization failed\n");
-
-  Kokkos::push_finalize_hook([&]() { magma_finalize(); });
-}
-
-MagmaSingleton& MagmaSingleton::singleton() {
-  static MagmaSingleton s;
-  return s;
-}
-
-}  // namespace Impl
-}  // namespace KokkosBlas
-#endif  // defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA)
-
 #endif  // KOKKOSBLAS_CUDA_TPL_HPP_

blas/tpls/KokkosBlas_Host_tpl.cpp

@@ -295,10 +295,10 @@ void F77_BLAS_MANGLE(dsyr, DSYR)(const char*, KK_INT*, const double*,
 
 void F77_BLAS_MANGLE(cher, CHER)(const char*, KK_INT*, const float*,
                                  const std::complex<float>*, KK_INT*,
-                                 std::complex<float>*, KK_INT*);
+                                 /* */ std::complex<float>*, KK_INT*);
 void F77_BLAS_MANGLE(zher, ZHER)(const char*, KK_INT*, const double*,
                                  const std::complex<double>*, KK_INT*,
-                                 std::complex<double>*, KK_INT*);
+                                 /* */ std::complex<double>*, KK_INT*);
 
 ///
 /// Syr2
@@ -322,12 +322,12 @@ void F77_BLAS_MANGLE(cher2, CHER2)(const char*, KK_INT*,
                                    const std::complex<float>*,
                                    const std::complex<float>*, KK_INT*,
                                    const std::complex<float>*, KK_INT*,
-                                   std::complex<float>*, KK_INT*);
+                                   /* */ std::complex<float>*, KK_INT*);
 void F77_BLAS_MANGLE(zher2, ZHER2)(const char*, KK_INT*,
                                    const std::complex<double>*,
                                    const std::complex<double>*, KK_INT*,
                                    const std::complex<double>*, KK_INT*,
-                                   std::complex<double>*, KK_INT*);
+                                   /* */ std::complex<double>*, KK_INT*);
 
 ///
 /// Trsv
@@ -901,14 +901,14 @@ void HostBlas<std::complex<float> >::gerc(
 }
 template <>
 template <>
-void HostBlas<std::complex<float> >::cher<float>(
+void HostBlas<std::complex<float> >::her<float>(
     const char uplo, KK_INT n, const float alpha, const std::complex<float>* x,
     KK_INT incx, std::complex<float>* a, KK_INT lda) {
   F77_FUNC_CHER(&uplo, &n, &alpha, (const std::complex<float>*)x, &incx,
                 (std::complex<float>*)a, &lda);
 }
 template <>
-void HostBlas<std::complex<float> >::cher2(
+void HostBlas<std::complex<float> >::her2(
     const char uplo, KK_INT n, const std::complex<float> alpha,
     const std::complex<float>* x, KK_INT incx, const std::complex<float>* y,
     KK_INT incy, std::complex<float>* a, KK_INT lda) {
@@ -1069,15 +1069,17 @@ void HostBlas<std::complex<double> >::gerc(
 }
 template <>
 template <>
-void HostBlas<std::complex<double> >::zher<double>(
-    const char uplo, KK_INT n, const double alpha,
-    const std::complex<double>* x, KK_INT incx, std::complex<double>* a,
-    KK_INT lda) {
+void HostBlas<std::complex<double> >::her<double>(const char uplo, KK_INT n,
+                                                  const double alpha,
+                                                  const std::complex<double>* x,
+                                                  KK_INT incx,
+                                                  std::complex<double>* a,
+                                                  KK_INT lda) {
   F77_FUNC_ZHER(&uplo, &n, &alpha, (const std::complex<double>*)x, &incx,
                 (std::complex<double>*)a, &lda);
 }
 template <>
-void HostBlas<std::complex<double> >::zher2(
+void HostBlas<std::complex<double> >::her2(
     const char uplo, KK_INT n, const std::complex<double> alpha,
     const std::complex<double>* x, KK_INT incx, const std::complex<double>* y,
     KK_INT incy, std::complex<double>* a, KK_INT lda) {

blas/tpls/KokkosBlas_Host_tpl.hpp

@@ -90,18 +90,11 @@ struct HostBlas {
                    KK_INT incx, const T *y, KK_INT incy, T *a, KK_INT lda);
 
   template <typename tAlpha>
-  static void cher(const char uplo, KK_INT n, const tAlpha alpha, const T *x,
-                   KK_INT incx, T *a, KK_INT lda);
-
-  template <typename tAlpha>
-  static void zher(const char uplo, KK_INT n, const tAlpha alpha, const T *x,
-                   KK_INT incx, T *a, KK_INT lda);
-
-  static void cher2(const char uplo, KK_INT n, const T alpha, const T *x,
-                    KK_INT incx, const T *y, KK_INT incy, T *a, KK_INT lda);
+  static void her(const char uplo, KK_INT n, const tAlpha alpha, const T *x,
+                  KK_INT incx, T *a, KK_INT lda);
 
-  static void zher2(const char uplo, KK_INT n, const T alpha, const T *x,
-                    KK_INT incx, const T *y, KK_INT incy, T *a, KK_INT lda);
+  static void her2(const char uplo, KK_INT n, const T alpha, const T *x,
+                   KK_INT incx, const T *y, KK_INT incy, T *a, KK_INT lda);
 
   static void trsv(const char uplo, const char transa, const char diag,
                    KK_INT m, const T *a, KK_INT lda,

blas/tpls/KokkosBlas_Magma_tpl.hpp

@@ -0,0 +1,41 @@
+//@HEADER
+// ************************************************************************
+//
+//                        Kokkos v. 4.0
+//       Copyright (2022) National Technology & Engineering
+//               Solutions of Sandia, LLC (NTESS).
+//
+// Under the terms of Contract DE-NA0003525 with NTESS,
+// the U.S. Government retains certain rights in this software.
+//
+// Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions.
+// See https://kokkos.org/LICENSE for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//@HEADER
+#ifndef KOKKOSBLAS_MAGMA_TPL_HPP_
+#define KOKKOSBLAS_MAGMA_TPL_HPP_
+
+#if defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA)
+#include <KokkosBlas_magma.hpp>
+
+namespace KokkosBlas {
+namespace Impl {
+
+MagmaSingleton::MagmaSingleton() {
+  magma_int_t stat = magma_init();
+  if (stat != MAGMA_SUCCESS) Kokkos::abort("MAGMA initialization failed\n");
+
+  Kokkos::push_finalize_hook([&]() { magma_finalize(); });
+}
+
+MagmaSingleton& MagmaSingleton::singleton() {
+  static MagmaSingleton s;
+  return s;
+}
+
+}  // namespace Impl
+}  // namespace KokkosBlas
+#endif  // defined(KOKKOSKERNELS_ENABLE_TPL_MAGMA)
+
+#endif  // KOKKOSBLAS_MAGMA_TPL_HPP_