This is an automated email from the ASF dual-hosted git repository.
raulcd pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/arrow.git
The following commit(s) were added to refs/heads/main by this push:
new a10f254cc3 GH-49921: [C++] Bump xsimd to 14.2.0 (#49922)
a10f254cc3 is described below
commit a10f254cc3f3ecea54b0da9743c83894bec46ef1
Author: Antoine Prouvost <[email protected]>
AuthorDate: Wed Jun 3 14:11:48 2026 +0200
GH-49921: [C++] Bump xsimd to 14.2.0 (#49922)
### Rationale for this change
- Fix for the current SVE ODR violation in the latest version
https://github.com/xtensor-stack/xsimd/pull/1311
- Pre-requisit to GH-45331
### What changes are included in this PR?
Bump version and cleanup backports.
### Are these changes tested?
Yes, in CI.
### Are there any user-facing changes?
No
* GitHub Issue: #49921
Authored-by: AntoinePrv <[email protected]>
Signed-off-by: Raúl Cumplido <[email protected]>
---
c_glib/vcpkg.json | 2 +-
ci/conda_env_cpp.txt | 2 +-
cpp/cmake_modules/ThirdpartyToolchain.cmake | 2 +-
cpp/src/arrow/util/bpacking_simd_kernel_internal.h | 92 ++--------------------
cpp/thirdparty/versions.txt | 4 +-
cpp/vcpkg.json | 2 +-
6 files changed, 13 insertions(+), 91 deletions(-)
diff --git a/c_glib/vcpkg.json b/c_glib/vcpkg.json
index 321f47bf63..e7919df65b 100644
--- a/c_glib/vcpkg.json
+++ b/c_glib/vcpkg.json
@@ -7,5 +7,5 @@
"pkgconf"
],
"$comment": "We can update builtin-baseline by 'vcpkg x-update-baseline'",
- "builtin-baseline": "40c89449f0ccce12d21f8a906639f6c2c649b9e7"
+ "builtin-baseline": "9b965a116838c6cdcd36bca60d1b81b030c8ab8d"
}
diff --git a/ci/conda_env_cpp.txt b/ci/conda_env_cpp.txt
index 470db4f8b9..e8a7d55239 100644
--- a/ci/conda_env_cpp.txt
+++ b/ci/conda_env_cpp.txt
@@ -47,6 +47,6 @@ rapidjson
re2
snappy
thrift-cpp>=0.11.0
-xsimd>=14.0
+xsimd>=14.2
zlib
zstd
diff --git a/cpp/cmake_modules/ThirdpartyToolchain.cmake
b/cpp/cmake_modules/ThirdpartyToolchain.cmake
index c08d0c4292..fd1752928e 100644
--- a/cpp/cmake_modules/ThirdpartyToolchain.cmake
+++ b/cpp/cmake_modules/ThirdpartyToolchain.cmake
@@ -2858,7 +2858,7 @@ if(ARROW_USE_XSIMD)
IS_RUNTIME_DEPENDENCY
FALSE
REQUIRED_VERSION
- "14.0.0")
+ "14.2.0")
if(xsimd_SOURCE STREQUAL "BUNDLED")
set(ARROW_XSIMD arrow::xsimd)
diff --git a/cpp/src/arrow/util/bpacking_simd_kernel_internal.h
b/cpp/src/arrow/util/bpacking_simd_kernel_internal.h
index fe879bb5b0..83a969cca6 100644
--- a/cpp/src/arrow/util/bpacking_simd_kernel_internal.h
+++ b/cpp/src/arrow/util/bpacking_simd_kernel_internal.h
@@ -151,73 +151,6 @@ constexpr bool IsSse2 = std::is_base_of_v<xsimd::sse2,
Arch>;
template <typename Arch>
constexpr bool IsAvx2 = std::is_base_of_v<xsimd::avx2, Arch>;
-/// Whether we are compiling for the Neon or above in the arm64 family.
-template <typename Arch>
-constexpr bool IsNeon = std::is_base_of_v<xsimd::neon, Arch>;
-
-/// Wrapper around ``xsimd::bitwise_lshift`` with optimizations for non
implemented sizes.
-///
-/// We replace the variable left shift by a variable multiply with a power of
two.
-///
-/// This trick is borrowed from Daniel Lemire and Leonid Boytsov, Decoding
billions of
-/// integers per second through vectorization, Software Practice & Experience
45 (1),
-/// 2015. http://arxiv.org/abs/1209.2137
-///
-/// TODO(xsimd) Tracking in https://github.com/xtensor-stack/xsimd/pull/1220
-/// When migrating, be sure to use batch_constant overload, and not the batch
one.
-template <typename Arch, typename Int, Int... kShifts>
-ARROW_FORCE_INLINE auto left_shift(const xsimd::batch<Int, Arch>& batch,
- xsimd::batch_constant<Int, Arch,
kShifts...> shifts)
- -> xsimd::batch<Int, Arch> {
- constexpr bool kIsSse2 = IsSse2<Arch>;
- constexpr bool kIsAvx2 = IsAvx2<Arch>;
- static_assert(
- !(kIsSse2 && kIsAvx2),
- "In xsimd, an x86 arch is either part of the SSE family or of the AVX
family,"
- "not both. If this check fails, it means the assumptions made here to
detect SSE "
- "and AVX are out of date.");
-
- constexpr auto kMults = xsimd::make_batch_constant<Int, 1, Arch>() << shifts;
-
- constexpr auto IntSize = sizeof(Int);
-
- // Sizes and architecture for which there is no variable left shift and
there is a
- // multiplication
- if constexpr (
//
- (kIsSse2 && (IntSize == sizeof(uint16_t) || IntSize ==
sizeof(uint32_t))) //
- || (kIsAvx2 && (IntSize == sizeof(uint16_t)))
//
- ) {
- return batch * kMults;
- }
-
- // Architecture for which there is no variable left shift on uint8_t but a
fallback
- // exists for uint16_t.
- if constexpr ((kIsSse2 || kIsAvx2) && (IntSize == sizeof(uint8_t))) {
- const auto batch16 = xsimd::bitwise_cast<uint16_t>(batch);
-
- constexpr auto kShifts0 = select_stride<uint16_t, 0>(shifts);
- const auto shifted0 = left_shift(batch16, kShifts0) & 0x00FF;
-
- constexpr auto kShifts1 = select_stride<uint16_t, 1>(shifts);
- const auto shifted1 = left_shift(batch16 & 0xFF00, kShifts1);
-
- return xsimd::bitwise_cast<Int>(shifted0 | shifted1);
- }
-
- // TODO(xsimd) bug fixed in xsimd 14.1.0
- // https://github.com/xtensor-stack/xsimd/pull/1266
-#if XSIMD_VERSION_MAJOR < 14 || ((XSIMD_VERSION_MAJOR == 14) &&
XSIMD_VERSION_MINOR == 0)
- if constexpr (IsNeon<Arch>) {
- using SInt = std::make_signed_t<Int>;
- constexpr auto signed_shifts =
- xsimd::batch_constant<SInt, Arch, static_cast<SInt>(kShifts)...>();
- return xsimd::kernel::bitwise_lshift(batch, signed_shifts.as_batch(),
Arch{});
- }
-#endif
-
- return batch << shifts;
-}
-
/// Fallback for variable shift right.
///
/// When we know that the relevant bits will not overflow, we can instead
shift left all
@@ -243,9 +176,8 @@ ARROW_FORCE_INLINE auto right_shift_by_excess(
constexpr auto IntSize = sizeof(Int);
- // Architecture for which there is no variable right shift but a larger
fallback exists.
- // TODO(xsimd) Tracking for Avx2 in
https://github.com/xtensor-stack/xsimd/pull/1220
- // When migrating, be sure to use batch_constant overload, and not the batch
one.
+ // Architectures for which there is no variable right shift but a larger
fallback
+ // exists.
if constexpr (kIsAvx2 && (IntSize == sizeof(uint8_t) || IntSize ==
sizeof(uint16_t))) {
using twice_uint = SizedUint<2 * IntSize>;
@@ -262,27 +194,17 @@ ARROW_FORCE_INLINE auto right_shift_by_excess(
return xsimd::bitwise_cast<Int>(shifted0 | shifted1);
}
- // These conditions are the ones matched in `left_shift`, i.e. the ones
where variable
- // shift right will not be available but a left shift (fallback) exists.
+ // Architectures for which there is no variable right shift but a left shift
exists
+ // (possibly using the multiply trick inside of xsimd).
+ // We use a variable left shift and fixed right shift.
if constexpr (kIsSse2 && (IntSize != sizeof(uint64_t))) {
constexpr Int kMaxRShift = max_value(std::array{kShifts...});
constexpr auto kLShifts =
xsimd::make_batch_constant<Int, kMaxRShift, Arch>() - shifts;
- return xsimd::bitwise_rshift<kMaxRShift>(left_shift(batch, kLShifts));
- }
-
- // TODO(xsimd) bug fixed in xsimd 14.1.0
- // https://github.com/xtensor-stack/xsimd/pull/1266
-#if XSIMD_VERSION_MAJOR < 14 || ((XSIMD_VERSION_MAJOR == 14) &&
XSIMD_VERSION_MINOR == 0)
- if constexpr (IsNeon<Arch>) {
- using SInt = std::make_signed_t<Int>;
- constexpr auto signed_shifts =
- xsimd::batch_constant<SInt, Arch, static_cast<SInt>(kShifts)...>();
- return xsimd::kernel::bitwise_rshift(batch, signed_shifts.as_batch(),
Arch{});
+ return xsimd::bitwise_rshift<kMaxRShift>(batch << kLShifts);
}
-#endif
return batch >> shifts;
}
@@ -1040,7 +962,7 @@ struct LargeKernel {
const auto high_swizzled = xsimd::swizzle(bytes, kHighSwizzles);
const auto high_words = xsimd::bitwise_cast<unpacked_type>(high_swizzled);
- const auto high_shifted = left_shift(high_words, kHighLShifts);
+ const auto high_shifted = high_words << kHighLShifts;
// We can have a single mask and apply it after OR because the shifts will
ensure that
// there are zeros where the high/low values are incomplete.
diff --git a/cpp/thirdparty/versions.txt b/cpp/thirdparty/versions.txt
index 2895247e9f..c6f4b01a71 100644
--- a/cpp/thirdparty/versions.txt
+++ b/cpp/thirdparty/versions.txt
@@ -115,8 +115,8 @@
ARROW_UTF8PROC_BUILD_SHA256_CHECKSUM=6f4f1b639daa6dca9f80bc5db1233e9cbaa31a67790
# WIL (Windows Implementation Libraries) is required by Azure SDK on Windows
for WinHTTP transport
ARROW_WIL_BUILD_VERSION=v1.0.250325.1
ARROW_WIL_BUILD_SHA256_CHECKSUM=c9e667d5f86ded43d17b5669d243e95ca7b437e3a167c170805ffd4aa8a9a786
-ARROW_XSIMD_BUILD_VERSION=14.0.0
-ARROW_XSIMD_BUILD_SHA256_CHECKSUM=17de0236954955c10c09d6938d4c5f3a3b92d31be5dadd1d5d09fc1b15490dce
+ARROW_XSIMD_BUILD_VERSION=14.2.0
+ARROW_XSIMD_BUILD_SHA256_CHECKSUM=21e841ab684b05331e81e7f782431753a029ef7b7d9d6d3ddab837e7782a40ee
ARROW_ZLIB_BUILD_VERSION=1.3.1
ARROW_ZLIB_BUILD_SHA256_CHECKSUM=9a93b2b7dfdac77ceba5a558a580e74667dd6fede4585b91eefb60f03b72df23
ARROW_ZSTD_BUILD_VERSION=1.5.7
diff --git a/cpp/vcpkg.json b/cpp/vcpkg.json
index 4b018476d2..1773c216b5 100644
--- a/cpp/vcpkg.json
+++ b/cpp/vcpkg.json
@@ -57,5 +57,5 @@
"zstd"
],
"$comment": "We can update builtin-baseline by 'vcpkg x-update-baseline'",
- "builtin-baseline": "40c89449f0ccce12d21f8a906639f6c2c649b9e7"
+ "builtin-baseline": "9b965a116838c6cdcd36bca60d1b81b030c8ab8d"
}
