[libc++] <experimental/simd> Add copy functions for class simd/simd_mask #78935

joy2myself · 2024-01-22T04:59:15Z

No description provided.

joy2myself · 2024-01-22T05:07:37Z

@philnik777 @ldionne Hi! I think we may need to discuss how to further reduce the compilation time of SIMD tests. With the current test methods and coverage, the compilation time of this patch may be very long, and the compilation time of subsequent patches such as SIMD binary operators may be even longer. Do you have any suggestions?

philnik777 · 2024-01-26T10:52:52Z

Do you know why the test is taking so long?

joy2myself · 2024-01-29T06:52:34Z

Do you know why the test is taking so long?

@philnik777 Louis has mentioned that the main impact on compile time is the instantiation of classes.

In previous tests, we generally traversed all vectorizable data types and all simd ABIs to instantiate various simd/simd_mask classes for testing. In this case, there would be a total of 20 (data types) x 21 (ABIs) times instantiation of a class. However, some interfaces are template functions (such as the copy function in the current PR), which require additional traversal of the types of the template parameter U, that is, a total of 20 (T types) x 20 (U types) x 21 (ABIs) instantiations. In addition, there are 3 different alignment flags involved in the copy function, so this test has a total of 20 (T types) x 20 (U types) x 21 (ABIs) x 3 (alignment flags) instantiations.

Subsequent patches such as SIMD binary operators involve a total of about 10 operators, and the number of instantiations will also be mutiplied.

How should we simplify this?

philnik777 · 2024-02-10T11:03:18Z

@joy2myself IIRC we already reduced that number of ABIs that we test with. I think it would be good enough to test 4-5 U types. That should reduce the time enough to not be problematic anymore.

joy2myself · 2024-03-04T09:51:09Z

@joy2myself IIRC we already reduced that number of ABIs that we test with. I think it would be good enough to test 4-5 U types. That should reduce the time enough to not be problematic anymore.

@philnik777 Yes, 21 is the number of ABIs after we reduced. I committed a patch to modify the U types to test with #83781. Please review it first. Thank you!

llvmbot · 2024-03-26T09:24:20Z

@llvm/pr-subscribers-libcxx

Author: ZhangYin (joy2myself)

Changes

Full diff: https://github.com/llvm/llvm-project/pull/78935.diff

7 Files Affected:

(modified) libcxx/docs/Status/ParallelismProjects.csv (+2)
(modified) libcxx/include/experimental/__simd/scalar.h (+7)
(modified) libcxx/include/experimental/__simd/simd.h (+13)
(modified) libcxx/include/experimental/__simd/simd_mask.h (+11)
(modified) libcxx/include/experimental/__simd/vec_ext.h (+11)
(added) libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp (+175)
(added) libcxx/test/std/experimental/simd/simd.mask.class/simd_mask_copy.pass.cpp (+127)

diff --git a/libcxx/docs/Status/ParallelismProjects.csv b/libcxx/docs/Status/ParallelismProjects.csv
index 06da008ac5fe90..2ddac1e52f0264 100644
--- a/libcxx/docs/Status/ParallelismProjects.csv
+++ b/libcxx/docs/Status/ParallelismProjects.csv
@@ -24,6 +24,7 @@ Section,Description,Dependencies,Assignee,Complete
 | `[parallel.simd.class] <https://wg21.link/N4808>`_, "`simd generate constructor <https://reviews.llvm.org/D159442>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.class] <https://wg21.link/N4808>`_, "`simd load constructor <https://github.com/llvm/llvm-project/pull/76610>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.class] <https://wg21.link/N4808>`_, "`simd subscript operators <https://github.com/llvm/llvm-project/pull/68960>`_", None, Yin Zhang, |Complete|
+| `[parallel.simd.class] <https://wg21.link/N4808>`_, "`simd copy functions <https://github.com/llvm/llvm-project/pull/78935>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.class] <https://wg21.link/N4808>`_, "Class template simd implementation", None, Yin Zhang, |In Progress|
 | `[parallel.simd.nonmembers] <https://wg21.link/N4808>`_, "simd non-member operations", None, Yin Zhang, |In Progress|
 | `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "`Class template simd_mask declaration and alias <https://reviews.llvm.org/D144362>`_", [parallel.simd.abi], Yin Zhang, |Complete|
@@ -33,5 +34,6 @@ Section,Description,Dependencies,Assignee,Complete
 | `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "`simd_mask implicit type conversion constructor <https://github.com/llvm/llvm-project/pull/71132>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "`simd_mask load constructor <https://github.com/llvm/llvm-project/pull/76610>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "`simd_mask subscript operators <https://github.com/llvm/llvm-project/pull/68960>`_", None, Yin Zhang, |Complete|
+| `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "`simd_mask copy functions <https://github.com/llvm/llvm-project/pull/78935>`_", None, Yin Zhang, |Complete|
 | `[parallel.simd.mask.class] <https://wg21.link/N4808>`_, "Class template simd_mask implementation", None, Yin Zhang, |In Progress|
 | `[parallel.simd.mask.nonmembers] <https://wg21.link/N4808>`_, "simd_mask non-member operations", None, Yin Zhang, |In Progress|
diff --git a/libcxx/include/experimental/__simd/scalar.h b/libcxx/include/experimental/__simd/scalar.h
index aff2cd11cfcfaa..1add4653209ace 100644
--- a/libcxx/include/experimental/__simd/scalar.h
+++ b/libcxx/include/experimental/__simd/scalar.h
@@ -62,6 +62,11 @@ struct __simd_operations<_Tp, simd_abi::__scalar> {
   static _LIBCPP_HIDE_FROM_ABI void __load(_SimdStorage& __s, const _Up* __mem) noexcept {
     __s.__data = static_cast<_Tp>(__mem[0]);
   }
+
+  template <class _Up>
+  static _LIBCPP_HIDE_FROM_ABI void __store(_SimdStorage __s, _Up* __mem) noexcept {
+    *__mem = static_cast<_Up>(__s.__data);
+  }
 };
 
 template <class _Tp>
@@ -71,6 +76,8 @@ struct __mask_operations<_Tp, simd_abi::__scalar> {
   static _LIBCPP_HIDE_FROM_ABI _MaskStorage __broadcast(bool __v) noexcept { return {__v}; }
 
   static _LIBCPP_HIDE_FROM_ABI void __load(_MaskStorage& __s, const bool* __mem) noexcept { __s.__data = __mem[0]; }
+
+  static _LIBCPP_HIDE_FROM_ABI void __store(_MaskStorage __s, bool* __mem) noexcept { __mem[0] = __s.__data; }
 };
 
 } // namespace parallelism_v2
diff --git a/libcxx/include/experimental/__simd/simd.h b/libcxx/include/experimental/__simd/simd.h
index db4ebb8e4a381b..c55e3e42bc7921 100644
--- a/libcxx/include/experimental/__simd/simd.h
+++ b/libcxx/include/experimental/__simd/simd.h
@@ -70,6 +70,19 @@ class simd {
     _Impl::__load(__s_, _Flags::template __apply<simd>(__mem));
   }
 
+  // copy functions
+  template <class _Up, class _Flags>
+  _LIBCPP_HIDE_FROM_ABI enable_if_t<__is_vectorizable_v<_Up> && is_simd_flag_type_v<_Flags>>
+  copy_from(const _Up* __mem, _Flags) {
+    _Impl::__load(__s_, _Flags::template __apply<simd>(__mem));
+  }
+
+  template <class _Up, class _Flags>
+  _LIBCPP_HIDE_FROM_ABI enable_if_t<__is_vectorizable_v<_Up> && is_simd_flag_type_v<_Flags>>
+  copy_to(_Up* __mem, _Flags) const {
+    _Impl::__store(__s_, _Flags::template __apply<simd>(__mem));
+  }
+
   // scalar access [simd.subscr]
   _LIBCPP_HIDE_FROM_ABI reference operator[](size_t __i) noexcept { return reference(__s_, __i); }
   _LIBCPP_HIDE_FROM_ABI value_type operator[](size_t __i) const noexcept { return __s_.__get(__i); }
diff --git a/libcxx/include/experimental/__simd/simd_mask.h b/libcxx/include/experimental/__simd/simd_mask.h
index 754db7992683b1..9b95dc5b9bacbc 100644
--- a/libcxx/include/experimental/__simd/simd_mask.h
+++ b/libcxx/include/experimental/__simd/simd_mask.h
@@ -58,6 +58,17 @@ class simd_mask {
     _Impl::__load(__s_, _Flags::template __apply<simd_mask>(__mem));
   }
 
+  // copy functions
+  template <class _Flags>
+  _LIBCPP_HIDE_FROM_ABI enable_if_t<is_simd_flag_type_v<_Flags>> copy_from(const value_type* __mem, _Flags) {
+    _Impl::__load(__s_, _Flags::template __apply<simd_mask>(__mem));
+  }
+
+  template <class _Flags>
+  _LIBCPP_HIDE_FROM_ABI enable_if_t<is_simd_flag_type_v<_Flags>> copy_to(value_type* __mem, _Flags) const {
+    _Impl::__store(__s_, _Flags::template __apply<simd_mask>(__mem));
+  }
+
   // scalar access [simd.mask.subscr]
   _LIBCPP_HIDE_FROM_ABI reference operator[](size_t __i) noexcept { return reference(__s_, __i); }
   _LIBCPP_HIDE_FROM_ABI value_type operator[](size_t __i) const noexcept { return __s_.__get(__i); }
diff --git a/libcxx/include/experimental/__simd/vec_ext.h b/libcxx/include/experimental/__simd/vec_ext.h
index c9423df93cfacc..316866b84873dd 100644
--- a/libcxx/include/experimental/__simd/vec_ext.h
+++ b/libcxx/include/experimental/__simd/vec_ext.h
@@ -80,6 +80,12 @@ struct __simd_operations<_Tp, simd_abi::__vec_ext<_Np>> {
     for (size_t __i = 0; __i < _Np; __i++)
       __s.__data[__i] = static_cast<_Tp>(__mem[__i]);
   }
+
+  template <class _Up>
+  static _LIBCPP_HIDE_FROM_ABI void __store(_SimdStorage __s, _Up* __mem) noexcept {
+    for (size_t __i = 0; __i < _Np; __i++)
+      __mem[__i] = static_cast<_Up>(__s.__data[__i]);
+  }
 };
 
 template <class _Tp, int _Np>
@@ -99,6 +105,11 @@ struct __mask_operations<_Tp, simd_abi::__vec_ext<_Np>> {
     for (size_t __i = 0; __i < _Np; __i++)
       __s.__data[__i] = experimental::__set_all_bits<_Tp>(__mem[__i]);
   }
+
+  static _LIBCPP_HIDE_FROM_ABI void __store(_MaskStorage __s, bool* __mem) noexcept {
+    for (size_t __i = 0; __i < _Np; __i++)
+      __mem[__i] = static_cast<bool>(__s.__data[__i]);
+  }
 };
 
 } // namespace parallelism_v2
diff --git a/libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp b/libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp
new file mode 100644
index 00000000000000..cb2e7c6d46c7c6
--- /dev/null
+++ b/libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp
@@ -0,0 +1,175 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+// UNSUPPORTED: c++03, c++11, c++14
+
+// <experimental/simd>
+//
+// [simd.class]
+// template<class U, class Flags> void copy_from(const U* mem, Flags);
+// template<class U, class Flags> void copy_to(U* mem, Flags) const;
+
+#include "../test_utils.h"
+
+namespace ex = std::experimental::parallelism_v2;
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct ElementAlignedCopyFromHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = alignof(U);
+    alignas(alignas_size) U buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      buffer[i] = static_cast<U>(i);
+    ex::simd<T, SimdAbi> origin_simd;
+    origin_simd.copy_from(buffer, ex::element_aligned_tag());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct VectorAlignedCopyFromHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = ex::memory_alignment_v<ex::simd<T, SimdAbi>, U>;
+    alignas(alignas_size) U buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      buffer[i] = static_cast<U>(i);
+    ex::simd<T, SimdAbi> origin_simd;
+    origin_simd.copy_from(buffer, ex::vector_aligned_tag());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct OveralignedCopyFromHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = bit_ceil(sizeof(U) + 1);
+    alignas(alignas_size) U buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      buffer[i] = static_cast<U>(i);
+    ex::simd<T, SimdAbi> origin_simd;
+    origin_simd.copy_from(buffer, ex::overaligned_tag<alignas_size>());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, std::size_t>
+struct CheckSimdCopyFrom {
+  template <class SimdAbi>
+  void operator()() {
+    constexpr std::size_t array_size = ex::simd_size_v<T, SimdAbi>;
+
+    types::for_each(simd_test_types(), ElementAlignedCopyFromHelper<T, SimdAbi, array_size>());
+    types::for_each(simd_test_types(), VectorAlignedCopyFromHelper<T, SimdAbi, array_size>());
+    types::for_each(simd_test_types(), OveralignedCopyFromHelper<T, SimdAbi, array_size>());
+  }
+};
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct ElementAlignedCopyToHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = alignof(U);
+    alignas(alignas_size) U buffer[array_size];
+    ex::simd<T, SimdAbi> origin_simd([](T i) { return i; });
+    origin_simd.copy_to(buffer, ex::element_aligned_tag());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct VectorAlignedCopyToHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = ex::memory_alignment_v<ex::simd<T, SimdAbi>, U>;
+    alignas(alignas_size) U buffer[array_size];
+    ex::simd<T, SimdAbi> origin_simd([](T i) { return i; });
+    origin_simd.copy_to(buffer, ex::vector_aligned_tag());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, class SimdAbi, std::size_t array_size>
+struct OveralignedCopyToHelper {
+  template <class U>
+  void operator()() const {
+    constexpr std::size_t alignas_size = bit_ceil(sizeof(U) + 1);
+    alignas(alignas_size) U buffer[array_size];
+    ex::simd<T, SimdAbi> origin_simd([](T i) { return i; });
+    origin_simd.copy_to(buffer, ex::overaligned_tag<alignas_size>());
+    assert_simd_values_equal(origin_simd, buffer);
+  }
+};
+
+template <class T, std::size_t>
+struct CheckSimdCopyTo {
+  template <class SimdAbi>
+  void operator()() {
+    constexpr std::size_t array_size = ex::simd_size_v<T, SimdAbi>;
+
+    types::for_each(simd_test_types(), ElementAlignedCopyToHelper<T, SimdAbi, array_size>());
+    types::for_each(simd_test_types(), VectorAlignedCopyToHelper<T, SimdAbi, array_size>());
+    types::for_each(simd_test_types(), OveralignedCopyToHelper<T, SimdAbi, array_size>());
+  }
+};
+
+template <class U, class T, class Flags, class SimdAbi = ex::simd_abi::compatible<T>, class = void>
+struct has_copy_from : std::false_type {};
+
+template <class U, class T, class Flags, class SimdAbi>
+struct has_copy_from<U,
+                     T,
+                     Flags,
+                     SimdAbi,
+                     std::void_t<decltype(std::declval<ex::simd<T, SimdAbi>>().copy_from(
+                         std::declval<const U*>(), std::declval<Flags>()))>> : std::true_type {};
+
+template <class U, class T, class Flags, class SimdAbi = ex::simd_abi::compatible<T>, class = void>
+struct has_copy_to : std::false_type {};
+
+template <class U, class T, class Flags, class SimdAbi>
+struct has_copy_to<
+    U,
+    T,
+    Flags,
+    SimdAbi,
+    std::void_t<decltype(std::declval<ex::simd<T, SimdAbi>>().copy_to(std::declval<U*>(), std::declval<Flags>()))>>
+    : std::true_type {};
+
+template <class T, std::size_t>
+struct CheckSimdCopyTraits {
+  template <class SimdAbi>
+  void operator()() {
+    // These functions shall not participate in overload resolution unless
+    // is_simd_flag_type_v<Flags> is true, and
+    // U is a vectorizable type.
+    static_assert(has_copy_from<int, T, ex::element_aligned_tag, SimdAbi>::value);
+    static_assert(has_copy_to<int, T, ex::element_aligned_tag, SimdAbi>::value);
+
+    // is_simd_flag_type_v<Flags> is false
+    static_assert(!has_copy_from<int, T, T, SimdAbi>::value);
+    static_assert(!has_copy_to<int, T, T, SimdAbi>::value);
+    static_assert(!has_copy_from<int, T, SimdAbi, SimdAbi>::value);
+    static_assert(!has_copy_to<int, T, SimdAbi, SimdAbi>::value);
+
+    // U is not a vectorizable type.
+    static_assert(!has_copy_from<SimdAbi, T, ex::element_aligned_tag, SimdAbi>::value);
+    static_assert(!has_copy_to<SimdAbi, T, ex::element_aligned_tag, SimdAbi>::value);
+    static_assert(!has_copy_from<ex::element_aligned_tag, T, ex::element_aligned_tag, SimdAbi>::value);
+    static_assert(!has_copy_to<ex::element_aligned_tag, T, ex::element_aligned_tag, SimdAbi>::value);
+  }
+};
+
+int main(int, char**) {
+  test_all_simd_abi<CheckSimdCopyFrom>();
+  test_all_simd_abi<CheckSimdCopyTo>();
+  test_all_simd_abi<CheckSimdCopyTraits>();
+  return 0;
+}
diff --git a/libcxx/test/std/experimental/simd/simd.mask.class/simd_mask_copy.pass.cpp b/libcxx/test/std/experimental/simd/simd.mask.class/simd_mask_copy.pass.cpp
new file mode 100644
index 00000000000000..1b6b38f53d8aba
--- /dev/null
+++ b/libcxx/test/std/experimental/simd/simd.mask.class/simd_mask_copy.pass.cpp
@@ -0,0 +1,127 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+// UNSUPPORTED: c++03, c++11, c++14
+
+// <experimental/simd>
+//
+// [simd.class]
+// template<class Flags> void copy_from(const value_type* mem, Flags);
+// template<class Flags> void copy_to(value_type* mem, Flags);
+
+#include "../test_utils.h"
+
+namespace ex = std::experimental::parallelism_v2;
+
+template <class T, std::size_t>
+struct CheckSimdMaskCopyFrom {
+  template <class SimdAbi>
+  void operator()() {
+    constexpr std::size_t array_size = ex::simd_size_v<T, SimdAbi>;
+
+    // element aligned tag
+    constexpr std::size_t element_alignas_size = alignof(bool);
+    alignas(element_alignas_size) bool element_buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      element_buffer[i] = static_cast<bool>(i % 2);
+    ex::simd_mask<T, SimdAbi> element_mask;
+    element_mask.copy_from(element_buffer, ex::element_aligned_tag());
+    assert_simd_mask_values_equal(element_mask, element_buffer);
+
+    // vector aligned tag
+    constexpr std::size_t vector_alignas_size = ex::memory_alignment_v<ex::simd_mask<T, SimdAbi>>;
+    alignas(vector_alignas_size) bool vector_buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      vector_buffer[i] = static_cast<bool>(i % 2);
+    ex::simd_mask<T, SimdAbi> vector_mask;
+    vector_mask.copy_from(element_buffer, ex::vector_aligned_tag());
+    assert_simd_mask_values_equal(vector_mask, vector_buffer);
+
+    // overaligned tag
+    constexpr std::size_t over_alignas_size = bit_ceil(sizeof(bool) + 1);
+    alignas(over_alignas_size) bool overaligned_buffer[array_size];
+    for (size_t i = 0; i < array_size; ++i)
+      overaligned_buffer[i] = static_cast<bool>(i % 2);
+    ex::simd_mask<T, SimdAbi> overaligned_mask;
+    overaligned_mask.copy_from(overaligned_buffer, ex::overaligned_tag<over_alignas_size>());
+    assert_simd_mask_values_equal(overaligned_mask, overaligned_buffer);
+  }
+};
+
+template <class T, std::size_t>
+struct CheckSimdMaskCopyTo {
+  template <class SimdAbi>
+  void operator()() {
+    constexpr std::size_t array_size = ex::simd_size_v<T, SimdAbi>;
+
+    // element aligned tag
+    constexpr std::size_t element_alignas_size = alignof(bool);
+    alignas(element_alignas_size) bool element_buffer[array_size];
+    ex::simd_mask<T, SimdAbi> element_mask(true);
+    element_mask.copy_to(element_buffer, ex::element_aligned_tag());
+    assert_simd_mask_values_equal(element_mask, element_buffer);
+
+    // vector aligned tag
+    constexpr std::size_t vector_alignas_size = ex::memory_alignment_v<ex::simd_mask<T, SimdAbi>>;
+    alignas(vector_alignas_size) bool vector_buffer[array_size];
+    ex::simd_mask<T, SimdAbi> vector_mask(false);
+    vector_mask.copy_to(vector_buffer, ex::vector_aligned_tag());
+    assert_simd_mask_values_equal(vector_mask, vector_buffer);
+
+    // overaligned tag
+    constexpr std::size_t over_alignas_size = bit_ceil(sizeof(bool) + 1);
+    alignas(over_alignas_size) bool overaligned_buffer[array_size];
+    ex::simd_mask<T, SimdAbi> overaligned_mask(true);
+    overaligned_mask.copy_to(overaligned_buffer, ex::overaligned_tag<over_alignas_size>());
+    assert_simd_mask_values_equal(overaligned_mask, overaligned_buffer);
+  }
+};
+
+template <class T, class Flags, class SimdAbi = ex::simd_abi::compatible<T>, class = void>
+struct has_copy_from : std::false_type {};
+
+template <class T, class Flags, class SimdAbi>
+struct has_copy_from<T,
+                     Flags,
+                     SimdAbi,
+                     std::void_t<decltype(std::declval<ex::simd_mask<T, SimdAbi>>().copy_from(
+                         std::declval<const bool*>(), std::declval<Flags>()))>> : std::true_type {};
+
+template <class T, class Flags, class SimdAbi = ex::simd_abi::compatible<T>, class = void>
+struct has_copy_to : std::false_type {};
+
+template <class T, class Flags, class SimdAbi>
+struct has_copy_to<T,
+                   Flags,
+                   SimdAbi,
+                   std::void_t<decltype(std::declval<ex::simd_mask<T, SimdAbi>>().copy_to(
+                       std::declval<bool*>(), std::declval<Flags>()))>> : std::true_type {};
+
+template <class T, std::size_t>
+struct CheckSimdMaskCopyTraits {
+  template <class SimdAbi>
+  void operator()() {
+    // These functions shall not participate in overload resolution unless
+    // is_simd_flag_type_v<Flags> is true
+    static_assert(has_copy_from<T, ex::element_aligned_tag, SimdAbi>::value);
+    static_assert(has_copy_to<T, ex::element_aligned_tag, SimdAbi>::value);
+
+    // is_simd_flag_type_v<Flags> is false
+    static_assert(!has_copy_from<T, T, SimdAbi>::value);
+    static_assert(!has_copy_to<T, T, SimdAbi>::value);
+    static_assert(!has_copy_from<T, SimdAbi, SimdAbi>::value);
+    static_assert(!has_copy_to<T, SimdAbi, SimdAbi>::value);
+  }
+};
+
+int main(int, char**) {
+  test_all_simd_abi<CheckSimdMaskCopyFrom>();
+  test_all_simd_abi<CheckSimdMaskCopyTo>();
+  test_all_simd_abi<CheckSimdMaskCopyTraits>();
+  return 0;
+}

libcxx/include/experimental/__simd/simd.h

libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp

joy2myself · 2024-04-07T07:48:59Z

Do you have any suggestions for solving the AIX errors in buildkite CI?

joy2myself · 2024-05-05T09:17:29Z

I marked XFAIL temporarily for target powerpc-ibm-aix7.2.5.7 and powerpc64-ibm-aix7.2.5.7. The error "Pass-by-value arguments with alignment greater than register width are not supported." seems a target specific limitation, and no related issues have been encountered on other targets. Therefore, this test may require special handling on these two targets. Fix it later.

joy2myself · 2024-05-07T03:45:12Z

@philnik777 Any other comments?

philnik777

Sorry for the long delay. LGTM.

joy2myself force-pushed the split_simd branch from 64bb284 to 1be369c Compare March 26, 2024 09:21

joy2myself marked this pull request as ready for review March 26, 2024 09:23

joy2myself requested a review from a team as a code owner March 26, 2024 09:23

llvmbot added the libc++ libc++ C++ Standard Library. Not GNU libstdc++. Not libc++abi. label Mar 26, 2024

joy2myself force-pushed the split_simd branch 2 times, most recently from fb27f54 to 898a99c Compare April 2, 2024 07:23

philnik777 reviewed Apr 5, 2024

View reviewed changes

libcxx/include/experimental/__simd/simd.h Outdated Show resolved Hide resolved

libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp Outdated Show resolved Hide resolved

joy2myself force-pushed the split_simd branch 2 times, most recently from 1d7f730 to ab6225e Compare April 7, 2024 05:23

ldionne assigned philnik777 Apr 26, 2024

joy2myself force-pushed the split_simd branch from ab6225e to 5c1abf5 Compare May 5, 2024 05:23

[libc++] <experimental/simd> Add copy functions for class simd/simd_mask

71af200

joy2myself force-pushed the split_simd branch from 5c1abf5 to 71af200 Compare May 5, 2024 06:48

joy2myself requested a review from philnik777 May 15, 2024 05:55

philnik777 approved these changes May 22, 2024

View reviewed changes

philnik777 merged commit 058e445 into llvm:main May 22, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[libc++] <experimental/simd> Add copy functions for class simd/simd_mask #78935

[libc++] <experimental/simd> Add copy functions for class simd/simd_mask #78935

Uh oh!

joy2myself commented Jan 22, 2024

Uh oh!

joy2myself commented Jan 22, 2024

Uh oh!

philnik777 commented Jan 26, 2024

Uh oh!

joy2myself commented Jan 29, 2024

Uh oh!

philnik777 commented Feb 10, 2024

Uh oh!

joy2myself commented Mar 4, 2024

Uh oh!

llvmbot commented Mar 26, 2024

Uh oh!

Uh oh!

Uh oh!

joy2myself commented Apr 7, 2024 •

edited

Loading

Uh oh!

joy2myself commented May 5, 2024

Uh oh!

joy2myself commented May 7, 2024

Uh oh!

philnik777 left a comment

Uh oh!

Uh oh!

[libc++] <experimental/simd> Add copy functions for class simd/simd_mask #78935

[libc++] <experimental/simd> Add copy functions for class simd/simd_mask #78935

Uh oh!

Conversation

joy2myself commented Jan 22, 2024

Uh oh!

joy2myself commented Jan 22, 2024

Uh oh!

philnik777 commented Jan 26, 2024

Uh oh!

joy2myself commented Jan 29, 2024

Uh oh!

philnik777 commented Feb 10, 2024

Uh oh!

joy2myself commented Mar 4, 2024

Uh oh!

llvmbot commented Mar 26, 2024

Uh oh!

Uh oh!

Uh oh!

joy2myself commented Apr 7, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joy2myself commented May 5, 2024

Uh oh!

joy2myself commented May 7, 2024

Uh oh!

philnik777 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

joy2myself commented Apr 7, 2024 •

edited

Loading