[Mirror] [5.1.1] 0 simd: Fix compile failures when ranges are not available

CHANGELOG.md

@@ -1,4 +1,60 @@
 # CHANGELOG
+## 5.1.0
+
+[Full Changelog](https://github.com/kokkos/kokkos/compare/5.0.2...5.1.0)
+
+### Features:
+* Export Kokkos type traits as C++20 concepts [\#8494](https://github.com/kokkos/kokkos/pull/8494)
+
+### Backend and Architecture Enhancements:
+
+#### CUDA:
+* Added `Kokkos_ARCH_BLACKWELL103` configure option for NVIDIA B300 GPUs [\#8791](https://github.com/kokkos/kokkos/pull/8791)
+* Fix compiling with Clang+Cuda+OpenMP with Kokkos_ENABLE_COMPILE_AS_CMAKE_LANGUAGE=ON [\#8810](https://github.com/kokkos/kokkos/pull/8810)
+* `nvcc_wrapper`: Add support for `-Ofc` and `--fdevice-time-trace` flags [\#8865](https://github.com/kokkos/kokkos/pull/8865)
+
+#### HIP:
+* Search the CMake variable `ROCM_PATH` for dependencies [\#8669](https://github.com/kokkos/kokkos/pull/8669)
+* Added support for brain floating-point (`bhalf_t`) [\#8705](https://github.com/kokkos/kokkos/pull/8705)
+* Implemented true reduced-precision mathematical functions (instead of falling back to `float`) [\#8705](https://github.com/kokkos/kokkos/pull/8705)
+* Add support for AMD MI355 and MI350 (`AMD_GFX950`) [\#8839](https://github.com/kokkos/kokkos/pull/8839)
+* Fix race conditions in HIP `parallel_scan` when running on MI300A  [\#8648](https://github.com/kokkos/kokkos/pull/8648)
+
+### General Enhancements
+* Enable ScatterView to contribute into a View that is an rvalue [\#8594](https://github.com/kokkos/kokkos/pull/8594)
+* Add bitwise operators to simd vectors and simd masks [\#8565](https://github.com/kokkos/kokkos/pull/8565)
+* Use Array::size_type for subscript operators [\#8692](https://github.com/kokkos/kokkos/pull/8692)
+* Add missing numeric trait `denorm_min` for `Kokkos::Experimental::half_t` and `Kokkos::Experimental::bhalf_t` [\#8769](https://github.com/kokkos/kokkos/pull/8769)
+* Use StaticBatchSize in ViewFill [\#8795](https://github.com/kokkos/kokkos/pull/8795)
+* Enforce failure when exceeding team_size_max and scratch_size_max checks [\#7445](https://github.com/kokkos/kokkos/pull/7445)
+* Enable MPI detection with PALS [\#8895](https://github.com/kokkos/kokkos/pull/8895)
+* Add simd memory permute functions [\#8775](https://github.com/kokkos/kokkos/pull/8775)
+* Performance improvements using `MDRangePolicy` with `CUDA`, `HIP` and `SYCL` [\#8638](https://github.com/kokkos/kokkos/pull/8638),  [\#8731](https://github.com/kokkos/kokkos/pull/8731)
+* Add `Kokkos::norm`for `Kokkos::complex`- similar to `std::norm` [\#8627](https://github.com/kokkos/kokkos/pull/8927)
+* Use neon and sve SIMD instructions if `nvcc` supports them [\#8667](https://github.com/kokkos/kokkos/pull/8667)
+* Expand math support: complete the implementation of all remaining math functions and increase half-type support [\#8595](https://github.com/kokkos/kokkos/pull/8789) [\#8858](https://github.com/kokkos/kokkos/pull/8858) [\#8873](https://github.com/kokkos/kokkos/pull/8873) [\#8712](https://github.com/kokkos/kokkos/pull/8712) [\#8827](https://github.com/kokkos/kokkos/pull/8827) [\#8819](https://github.com/kokkos/kokkos/pull/8819) [\#8719](https://github.com/kokkos/kokkos/pull/8719) [\#8863](https://github.com/kokkos/kokkos/pull/8863) [\#8862](https://github.com/kokkos/kokkos/pull/8862) [\#8778](https://github.com/kokkos/kokkos/pull/8778) [\#8891](https://github.com/kokkos/kokkos/pull/8891)
+* Improve performance of `deep_copy` from scalar in view fill using StaticBatchSize [\#8795](https://github.com/kokkos/kokkos/pull/8795) [\#8829](https://github.com/kokkos/kokkos/pull/8829)
+
+### Build System Changes
+* Warn about multiple device architectures enabled by `find_package(HIP)` [\#8938](https://github.com/kokkos/kokkos/pull/8938)
+
+### Incompatibilities (i.e. breaking changes)
+* Execution spaces can only be constructed after `Kokkos::initialize()` has been called and must be destructed before `Kokkos::finalize()` [\#8546](https://github.com/kokkos/kokkos/pull/8546) [\#8677](https://github.com/kokkos/kokkos/pull/8677)
+* ScatterValue isn't move constructible/assignable anymore [\#8761](https://github.com/kokkos/kokkos/pull/8761)
+* Enforce TeamPolicy constructor preconditions (includes vector length must be a power of two) [\#8904](https://github.com/kokkos/kokkos/pull/8904) [\#8907](https://github.com/kokkos/kokkos/pull/8907)
+* OpenMP: Warn on exec space instance created within omp region [\#8919](https://github.com/kokkos/kokkos/pull/8919)
+* Remove the deprecated OpenMPTarget backend [\#8701](https://github.com/kokkos/kokkos/pull/8701) [\#8717](https://github.com/kokkos/kokkos/pull/8717) [\#8749](https://github.com/kokkos/kokkos/pull/8749) [\#8767](https://github.com/kokkos/kokkos/pull/8767)
+
+### Bug Fixes
+* Fix reduction_identity for BAnd [\#8715](https://github.com/kokkos/kokkos/pull/8715)
+* Restrict lock free host atomics to the actual sizes that are lock free [\#8809](https://github.com/kokkos/kokkos/pull/8809)
+* Use intrinsics when calling min and max on simd vectors of integral types [\#8899](https://github.com/kokkos/kokkos/pull/8899)
+* Adds missing `constexpr` specifiers on `conj()`, and for the `real()` and `imag()` non-member functions taking complex numbers [\#8928](https://github.com/kokkos/kokkos/pull/8928)
+* Ensure that execution space instances fence on finalize [\#8626](https://github.com/kokkos/kokkos/pull/8626)
+* Update `team_fan_{in|out}` member functions of `ThreadsExecTeamMember` not to call host-only fuctions on the device [\#8730](https://github.com/kokkos/kokkos/pull/8730)
+* Make overloads of `isnormal` compliant with std [\#8857](https://github.com/kokkos/kokkos/pull/8857)
+* Fix compiler macros identify GCC and LLVM Clang on OSX [\#8592](https://github.com/kokkos/kokkos/pull/8592) [\#8952](https://github.com/kokkos/kokkos/pull/8952)
+
 ## 5.0.2
 
 [Full Changelog](https://github.com/kokkos/kokkos/compare/5.0.1...5.0.2)

CMakeLists.txt

@@ -140,8 +140,8 @@ elseif(NOT CMAKE_SIZEOF_VOID_P EQUAL 8)
 endif()
 
 set(Kokkos_VERSION_MAJOR 5)
-set(Kokkos_VERSION_MINOR 0)
-set(Kokkos_VERSION_PATCH 99)
+set(Kokkos_VERSION_MINOR 1)
+set(Kokkos_VERSION_PATCH 1)
 set(Kokkos_VERSION "${Kokkos_VERSION_MAJOR}.${Kokkos_VERSION_MINOR}.${Kokkos_VERSION_PATCH}")
 message(STATUS "Kokkos version: ${Kokkos_VERSION}")
 math(EXPR KOKKOS_VERSION "${Kokkos_VERSION_MAJOR} * 10000 + ${Kokkos_VERSION_MINOR} * 100 + ${Kokkos_VERSION_PATCH}")

core/src/Kokkos_Macros.hpp

@@ -18,6 +18,12 @@
  *  KOKKOS_ENABLE_CUDA_UVM            Use CUDA UVM for Cuda memory space.
  */
 
+#ifndef KOKKOS_DONT_INCLUDE_CORE_CONFIG_H
+#include <KokkosCore_config.h>
+#include <impl/Kokkos_DesulAtomicsConfig.hpp>
+#include <impl/Kokkos_NvidiaGpuArchitectures.hpp>
+#endif
+
 #define KOKKOS_VERSION_LESS(MAJOR, MINOR, PATCH) \
   (KOKKOS_VERSION < ((MAJOR)*10000 + (MINOR)*100 + (PATCH)))
 
@@ -38,12 +44,6 @@
 #error implementation bug
 #endif
 
-#ifndef KOKKOS_DONT_INCLUDE_CORE_CONFIG_H
-#include <KokkosCore_config.h>
-#include <impl/Kokkos_DesulAtomicsConfig.hpp>
-#include <impl/Kokkos_NvidiaGpuArchitectures.hpp>
-#endif
-
 #if __has_include(<version>)
 #include <version>
 #else
@@ -126,7 +126,8 @@
 // CRAY compiler for host code
 #define KOKKOS_COMPILER_CRAYC _CRAYC
 
-#elif defined(__APPLE_CC__) && defined(__clang__)
+#elif defined(__APPLE_CC__) && defined(__clang__) && \
+    defined(__apple_build_version__)
 #define KOKKOS_COMPILER_APPLECC __APPLE_CC__
 
 #elif defined(__NVCOMPILER)

core/src/View/Kokkos_BasicView.hpp

@@ -597,7 +597,8 @@ class BasicView {
     // Explicit cast is needed because submdspan_mapping may return a different
     // layout type.
     using sub_accessor_t = typename OtherAccessorPolicy::offset_policy;
-    m_ptr = src_view.m_acc.offset(src_view.m_ptr, sub_mapping_result.offset);
+    m_ptr                = static_cast<data_handle_type>(
+        src_view.m_acc.offset(src_view.m_ptr, sub_mapping_result.offset));
     m_map = mapping_type(sub_mapping_result.mapping);
     m_acc = sub_accessor_t(src_view.m_acc);
 

core/unit_test/CMakeLists.txt

@@ -299,7 +299,16 @@ foreach(Tag Threads;Serial;OpenMP;Cuda;HPX;OpenACC;HIP;SYCL)
     endforeach()
 
     set(${Tag}_SOURCES2D)
-    foreach(Name SubView_c10 SubView_c11 SubView_c12 SubView_c13 SubView_c14 SubView_c15)
+    foreach(
+      Name
+      SubView_c10
+      SubView_c11
+      SubView_c12
+      SubView_c13
+      SubView_c14
+      SubView_c15
+      SubView_c16
+    )
       set(file ${dir}/Test${Tag}_${Name}.cpp)
       # Write to a temporary intermediate file and call configure_file to avoid
       # updating timestamps triggering unnecessary rebuilds on subsequent cmake runs.

core/unit_test/TestSubView_c15.hpp

@@ -27,7 +27,7 @@ TEST(TEST_CATEGORY_DEATH, view_subview_constructor_layout_compatibility) {
     (void)Kokkos::View<int*, LS>(a1, Kokkos::ALL, 1);
 
     // FIXME: This doesn't compile for BasicView, but should
-    //(void)Kokkos::View<int*, LR>(a, Kokkos::ALL, 1);
+    // (void)Kokkos::View<int*, LR>(a1, Kokkos::ALL, 1);
   }
   {
     // Using subview dims (1, ALL). For a LayoutLeft,

core/unit_test/TestSubView_c16.hpp

@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+// SPDX-FileCopyrightText: Copyright Contributors to the Kokkos project
+
+#include <gtest/gtest.h>
+#include <Kokkos_Macros.hpp>
+#ifdef KOKKOS_ENABLE_EXPERIMENTAL_CXX20_MODULES
+import kokkos.core;
+#else
+#include <Kokkos_Core.hpp>
+#endif
+
+namespace {
+// Test that the subview constructor works for constructing subviews from
+// Views where one of them has the Kokkos::Unmanaged memory trait
+TEST(TEST_CATEGORY, view_create_unmanaged_subview_from_managed) {
+  int N    = 10;
+  using LR = Kokkos::LayoutRight;
+  using LL = Kokkos::LayoutLeft;
+  using LS = Kokkos::LayoutStride;
+
+  Kokkos::View<int**, LL> a1("A1", N, N);
+  Kokkos::View<int**, LL, Kokkos::MemoryTraits<Kokkos::Unmanaged>> a1_unmanaged(
+      a1.data(), N, N);
+  {
+    // Using subview dims (ALL, 1). For a LayoutLeft,
+    // any subview layout should be appropriate.
+    (void)Kokkos::View<int*, LL, Kokkos::MemoryTraits<Kokkos::Unmanaged>>(
+        a1, Kokkos::ALL, 1);
+    (void)Kokkos::View<int*, LS, Kokkos::MemoryTraits<Kokkos::Unmanaged>>(
+        a1, Kokkos::ALL, 1);
+
+    (void)Kokkos::View<int*, LL>(a1_unmanaged, Kokkos::ALL, 1);
+    (void)Kokkos::View<int*, LS>(a1_unmanaged, Kokkos::ALL, 1);
+    // FIXME: This doesn't compile for BasicView, but should
+    // (void)Kokkos::View<int*, LR>(a1_unmanaged, Kokkos::ALL, 1);
+  }
+
+  Kokkos::View<int**, LL> a2("A2", 1, N);
+  Kokkos::View<int**, LL, Kokkos::MemoryTraits<Kokkos::Unmanaged>> a2_unmanaged(
+      a2.data(), 1, N);
+  {
+    // Using subview dims (0, ALL). Any subview layout should be appropriate.
+    (void)Kokkos::View<int*, LL, Kokkos::MemoryTraits<Kokkos::Unmanaged>>(
+        a2, 0, Kokkos::ALL);
+    (void)Kokkos::View<int*, LS, Kokkos::MemoryTraits<Kokkos::Unmanaged>>(
+        a2, 0, Kokkos::ALL);
+    (void)Kokkos::View<int*, LR, Kokkos::MemoryTraits<Kokkos::Unmanaged>>(
+        a2, 0, Kokkos::ALL);
+
+    (void)Kokkos::View<int*, LL>(a2_unmanaged, 0, Kokkos::ALL);
+    (void)Kokkos::View<int*, LS>(a2_unmanaged, 0, Kokkos::ALL);
+    (void)Kokkos::View<int*, LR>(a2_unmanaged, 0, Kokkos::ALL);
+  }
+}
+}  // namespace

simd/src/Kokkos_SIMD.hpp

@@ -236,29 +236,29 @@ simd_unchecked_load(const T* ptr,
   return simd_unchecked_load<basic_simd<T, simd_abi::scalar>>(ptr, flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::NonScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto unchecked_gather_from(
     R&& in, const I& indices, simd_flags<Flags...> flag = simd_flag_default) {
   return unchecked_gather_from<
       basic_simd<T, simd_abi::Impl::host_fixed_native<T>>>(in, indices, flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::ScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_FORCEINLINE_FUNCTION auto unchecked_gather_from(
     R&& in, const I& indices, simd_flags<Flags...> flag = simd_flag_default) {
   return unchecked_gather_from<basic_simd<T, simd_abi::scalar>>(in, indices,
                                                                 flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::NonScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto unchecked_gather_from(
     R&& in, const typename I::mask_type& mask, const I& indices,
@@ -268,9 +268,9 @@ KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto unchecked_gather_from(
                                                            flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::ScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_FORCEINLINE_FUNCTION auto unchecked_gather_from(
     R&& in, const typename I::mask_type& mask, const I& indices,
@@ -279,29 +279,29 @@ KOKKOS_FORCEINLINE_FUNCTION auto unchecked_gather_from(
                                                                 indices, flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::NonScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto partial_gather_from(
     R&& in, const I& indices, simd_flags<Flags...> flag = simd_flag_default) {
   return partial_gather_from<
       basic_simd<T, simd_abi::Impl::host_fixed_native<T>>>(in, indices, flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::ScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_FORCEINLINE_FUNCTION auto partial_gather_from(
     R&& in, const I& indices, simd_flags<Flags...> flag = simd_flag_default) {
   return partial_gather_from<basic_simd<T, simd_abi::scalar>>(in, indices,
                                                               flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::NonScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto partial_gather_from(
     R&& in, const typename I::mask_type& mask, const I& indices,
@@ -311,9 +311,9 @@ KOKKOS_IMPL_HOST_FORCEINLINE_FUNCTION auto partial_gather_from(
                                                            flag);
 }
 
-template <std::ranges::contiguous_range R, Impl::SimdIntegral I,
-          typename... Flags, typename T = std::ranges::range_value_t<R>>
-  requires std::ranges::sized_range<R> &&
+template <Impl::Ranges::contiguous_range R, Impl::SimdIntegral I,
+          typename... Flags, typename T = Impl::Ranges::range_value_t<R>>
+  requires Impl::Ranges::sized_range<R> &&
            Impl::ScalarAbi<simd_abi::Impl::host_fixed_native<T>>
 KOKKOS_FORCEINLINE_FUNCTION auto partial_gather_from(
     R&& in, const typename I::mask_type& mask, const I& indices,