[libcxx-commits] [libcxx] [libc++] <experimental/simd> Add copy functions for class simd/simd_mask (PR #78935)
via libcxx-commits
libcxx-commits at lists.llvm.org
Sat May 4 22:23:50 PDT 2024
https://github.com/joy2myself updated https://github.com/llvm/llvm-project/pull/78935
>From 5c1abf5f45decff14f3c96fb7389ab6a1a96d7cf Mon Sep 17 00:00:00 2001
From: Yin Zhang <zhangyin2018 at iscas.ac.cn>
Date: Fri, 29 Dec 2023 16:48:51 +0800
Subject: [PATCH] [libc++] <experimental/simd> Add copy functions for class
simd/simd_mask
---
libcxx/docs/Status/ParallelismProjects.csv | 2 +
libcxx/include/experimental/__simd/scalar.h | 7 +
libcxx/include/experimental/__simd/simd.h | 11 ++
.../include/experimental/__simd/simd_mask.h | 11 ++
libcxx/include/experimental/__simd/vec_ext.h | 11 ++
.../simd/simd.class/simd_copy.pass.cpp | 173 ++++++++++++++++++
.../simd.mask.class/simd_mask_copy.pass.cpp | 127 +++++++++++++
7 files changed, 342 insertions(+)
create mode 100644 libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp
create mode 100644 libcxx/test/std/experimental/simd/simd.mask.class/simd_mask_copy.pass.cpp
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..37e334aad6da07 100644
--- a/libcxx/include/experimental/__simd/simd.h
+++ b/libcxx/include/experimental/__simd/simd.h
@@ -70,6 +70,17 @@ class simd {
_Impl::__load(__s_, _Flags::template __apply<simd>(__mem));
}
+ // copy functions
+ template <class _Up, class _Flags, enable_if_t<__is_vectorizable_v<_Up> && is_simd_flag_type_v<_Flags>, int> = 0>
+ _LIBCPP_HIDE_FROM_ABI void copy_from(const _Up* __mem, _Flags) {
+ _Impl::__load(__s_, _Flags::template __apply<simd>(__mem));
+ }
+
+ template <class _Up, class _Flags, enable_if_t<__is_vectorizable_v<_Up> && is_simd_flag_type_v<_Flags>, int> = 0>
+ _LIBCPP_HIDE_FROM_ABI void 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..fd6dee2e28ee91 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, enable_if_t<is_simd_flag_type_v<_Flags>, int> = 0>
+ _LIBCPP_HIDE_FROM_ABI void copy_from(const value_type* __mem, _Flags) {
+ _Impl::__load(__s_, _Flags::template __apply<simd_mask>(__mem));
+ }
+
+ template <class _Flags, enable_if_t<is_simd_flag_type_v<_Flags>, int> = 0>
+ _LIBCPP_HIDE_FROM_ABI void 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..473f58138e5b76
--- /dev/null
+++ b/libcxx/test/std/experimental/simd/simd.class/simd_copy.pass.cpp
@@ -0,0 +1,173 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+
+// FIXME: Fatal error with following targets (remove XFAIL when fixed):
+// Pass-by-value arguments with alignment greater than register width are not supported.
+// XFAIL: target=powerpc.*-ibm-aix7\.2\.5\.7
+
+// <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 {
+ 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 {
+ alignas(ex::memory_alignment_v<ex::simd<T, SimdAbi>, U>) 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 {
+ alignas(bit_ceil(sizeof(U) + 1)) 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<bit_ceil(sizeof(U) + 1)>());
+ 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 {
+ 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 {
+ alignas(ex::memory_alignment_v<ex::simd<T, SimdAbi>, U>) 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 {
+ alignas(bit_ceil(sizeof(U) + 1)) U buffer[array_size];
+ ex::simd<T, SimdAbi> origin_simd([](T i) { return i; });
+ origin_simd.copy_to(buffer, ex::overaligned_tag<bit_ceil(sizeof(U) + 1)>());
+ 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..0c3b4c9ea6d5ac
--- /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(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];
+ 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;
+}
More information about the libcxx-commits
mailing list