Bug ID: JDK-8309130 x86_64 AVX512 intrinsics for Arrays.sort methods (int, long, float and double arrays)

Type: Enhancement
Component: hotspot
Sub-Component: compiler
Affected Version: 21,22

Priority: P4
Status: Resolved
Resolution: Fixed
OS: linux
CPU: x86

Submitted: 2023-05-30
Updated: 2024-11-23
Resolved: 2023-10-06

JDK 22
22 b19Fixed

The goal is to develop faster sort routines for x68_64 CPUs by taking advantage of AVX512 instructions. This enhancement provides an order of magnitude speedup for Arrays.sort() using int, long, float and double arrays.

A pull request was submitted for review. Branch: master URL: https://git.openjdk.org/jdk/pull/14227 Date: 2023-05-30 18:54:50 +0000
23-11-2024
Hello Matthias, This is a bug in GCC 12 (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105593) and this issue seems to be resolved in GCC 12.3.1 as I was able to successfully build OpenJDK using GCC 12.3.1. Thanks, Vamsi
09-10-2023
Adding DISABLED_WARNINGS_gcc := uninitialized, \ to BUILD_LIB_SIMD_SORT makes the build pass on the problematic Alpine 3.17 environment. The error can be observed as well on Alpine 3.18.2 .
09-10-2023
In the Linux musl (Alpine x86_64) build, we run now into a ton of warnings as errors like this : /usr/lib/gcc/x86_64-alpine-linux-musl/12.2.1/include/avx512fintrin.h:4459:50: error: '__Y' is used uninitialized [-Werror=uninitialized] Should we disable this warning ? Or is it a problematic warning? We use 12.2.1 on our Alpine box running Alpine 3.17.4 . details : In function '__m512i _mm512_shuffle_epi32(__m512i, _MM_PERM_ENUM)', inlined from 'static zmm_vector<int>::zmm_t zmm_vector<int>::shuffle(zmm_t) [with unsigned char mask = 177]' at /openjdk/linuxmuslx86_64/jdk-dev/src/java.base/linux/native/libsimdsort/avx512-32bit-qsort.hpp:96:36, inlined from 'zmm_t sort_zmm_32bit(zmm_t) [with vtype = zmm_vector<int>; zmm_t = __vector(8) long long int]' at /openjdk/linuxmuslx86_64/jdk-dev/src/java.base/linux/native/libsimdsort/avx512-32bit-qsort.hpp:181:27: /usr/lib/gcc/x86_64-alpine-linux-musl/12.2.1/include/avx512fintrin.h:4459:50: error: '__Y' is used uninitialized [-Werror=uninitialized] 4459 \| return (__m512i) __builtin_ia32_pshufd512_mask ((__v16si) __A, \| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~ 4460 \| __mask, \| ~~~~~~~ 4461 \| (__v16si) \| ~~~~~~~~~ 4462 \| _mm512_undefined_epi32 (), \| ~~~~~~~~~~~~~~~~~~~~~~~~~~ 4463 \| (__mmask16) -1); \| ~~~~~~~~~~~~~~~ /usr/lib/gcc/x86_64-alpine-linux-musl/12.2.1/include/avx512fintrin.h: In function 'zmm_t sort_zmm_32bit(zmm_t) [with vtype = zmm_vector<int>; zmm_t = __vector(8) long long int]': /usr/lib/gcc/x86_64-alpine-linux-musl/12.2.1/include/avx512fintrin.h:206:11: note: '__Y' was declared here 206 \| __m512i __Y = __Y; \| ^~~
09-10-2023
Changeset: a4e9168b Author: vamsi-parasa <srinivas.vamsi.parasa@intel.com> Committer: Sandhya Viswanathan <sviswanathan@openjdk.org> Date: 2023-10-06 20:15:30 +0000 URL: https://git.openjdk.org/jdk/commit/a4e9168bab1c2872ce2dbc7971a45c259270271f
06-10-2023

Relates :	JDK-8335221 - Some C2 intrinsics incorrectly assume that type argument is compile-time constant
Relates :	JDK-8317763 - Follow-up to AVX512 intrinsics for Arrays.sort() PR
Relates :	JDK-8319577 - x86_64 AVX2 intrinsics for Arrays.sort methods (int, float arrays)
Relates :	JDK-8318005 - Consider using radix sort in DualPivotQuicksort for int[], or even double[]
Relates :	JDK-8318306 - java/util/Arrays/Sorting.java fails with "Array is not sorted at 8228-th position: 8251.0 and 8153.0"
Relates :	JDK-8317976 - Optimize SIMD sort for AMD Zen 4
Relates :	JDK-8315382 - Add Windows support for x86_64 AVX512 intrinsics for Arrays.sort() and Arrays.parallelSort() using assembly stubs