mir/libreoffice
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
libreoffice/include/tools/simdsupport.hxx

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

84 lines
2.7 KiB
C++
Raw Permalink Normal View History

CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
/*
* This file is part of the LibreOffice project.
*
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*
*/
remove AVX and AVX512 code from Calc It's been a source of numerous problems since the beginning. Poor separation of C++ code causing the compiler to emit some generic code as CPU-specific, compiler optimizations moving CPU-specific code out of #ifdef to unguarded static initialization, etc. And it doesn't seem to even particularly improve performance, on my Ryzen2500U for one full column (1m cells) sumArray() takes about 1.6ms with AVX, 1.9ms with SSE2 and 4.6ms with generic code. So SSE2 code is perhaps worth it, especially given that SSE2 is our baseline requirement on x86_64 everywhere and x86 on Windows, but AVX+ is nowhere near worth the trouble. So this code removes all AVX+ code from Calc, and makes SSE2 a hardcoded option on where it's guaranteed. If we raise the baseline to AVX, the SSE2 code may be replaced by the one removed by this commit. Generic code is there for other platforms, if other platforms add CPU-specific code, they should preferably follow the same rules. This does not necessarily mean that CPU-specific code cannot be used at all. Some externals use them, for example. It just needs to be working, maintained, and worth the trouble. Change-Id: I5ab919930df9d0223db68a94bf84947984d313ac Reviewed-on: https://gerrit.libreoffice.org/c/core/+/129733 Tested-by: Jenkins Reviewed-by: Eike Rathke <erack@redhat.com>
2022-02-09 16:23:33 +01:00
// IMPORTANT: Having CPU-specific routines turned out to be a maintenance
// problem, because of various problems such as compilers moving CPU-specific
// code out of #ifdef code into static initialization or our code using C++
// features that caused the compiler to emit code that used CPU-specific
// instructions (even cpuid.hxx isn't safe, see the comment there).
// The only safe usage is using CPU-specific code that's always available,
// such as SSE2-specific code for x86_64. Do not use for anything else
// unless you really know what you are doing (and you check git history
// to learn from past problems).
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
// Determine the compiler support for SIMD compiler intrinsics.
// This changes from one compiled unit to the other, depending if
// the support has been detected and if the compiled unit contains
// code using intrinsics or not. So we have to (re)set them again
// every time this file has been included.
Add test as an example how to add CPU intrinsics support Also makes sure that if the CPU dataction or compiler detection doesn't work correctly, the test could potentially crash. Change-Id: If0862c0a736c8e1f5ba1117b248918e4c1fb2f4d Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96779 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2020-06-20 16:59:59 +02:00
// In other words... DO NOT ADD "#pragma once" here
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#undef LO_SSE2_AVAILABLE
#undef LO_SSSE3_AVAILABLE
#undef LO_AVX_AVAILABLE
#undef LO_AVX2_AVAILABLE
tdf#142307 - Upgrade SSE2 sum to AVX512 sum with Neumaier 1 This part focuses on allowing it on replacing arrayfunctor By thefault it will try AVX512F (1,17%) If not available will use AVX (94,77%) Use of AVX2 (82,28%) has been avoided even if the code could been more compact Source of hardware statistics: https://store.steampowered.com/hwsurvey Change-Id: Iae737a565379e82c5f84f3fdee6321ac74f59d40 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115675 Tested-by: Jenkins Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
2021-05-16 16:04:20 +02:00
#undef LO_AVX512F_AVAILABLE
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#if defined(_MSC_VER) // VISUAL STUDIO COMPILER
MSVC -arch is independent from the ability to use CPU intrinsics It's possible to write AVX512 intrinsics in code compile only with -arch:AVX . So do not require -arch for being able to do so, especially since there is no -arch option for only AVX512F without other AVX512 subsets (the option enables also CD, BW, DQ and VL https://docs.microsoft.com/en-us/cpp/build/reference/arch-x64). https://crashreport.libreoffice.org/stats/crash_details/55ef825d-c323-4df9-95e2-76672c674e60 is presumably caused by this, I can see use of registers XMM0-15 in arraysumAVX512.cxx built with -arch:AVX2 but when built with -arch:AVX512 registers XMM16-31 are used too (I'm not sure if that's AVX512DQ or something else, I can't find info on it). Change-Id: I74473333a17e618327d43b920b8929d1b0e733b8 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/129724 Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com> Reviewed-by: Luboš Luňák <l.lunak@collabora.com> Tested-by: Jenkins
2022-01-26 13:34:08 +01:00
// With MSVC using -arch is in fact not necessary for being able
// to use CPU intrinsics, code using AVX512F intrinsics will compile
// even if compiled with -arch:AVX, the -arch option really only affects
// instructions generated for C/C++ code.
#if defined(_M_X64) || defined(_M_X86)
// As such, if we're building for X86 or X64, support for these is always available
// with MSVC2019+.
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#define LO_SSE2_AVAILABLE
#define LO_SSSE3_AVAILABLE
#define LO_AVX_AVAILABLE
#define LO_AVX2_AVAILABLE
tdf#142307 - Upgrade SSE2 sum to AVX512 sum with Neumaier 1 This part focuses on allowing it on replacing arrayfunctor By thefault it will try AVX512F (1,17%) If not available will use AVX (94,77%) Use of AVX2 (82,28%) has been avoided even if the code could been more compact Source of hardware statistics: https://store.steampowered.com/hwsurvey Change-Id: Iae737a565379e82c5f84f3fdee6321ac74f59d40 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115675 Tested-by: Jenkins Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
2021-05-16 16:04:20 +02:00
#define LO_AVX512F_AVAILABLE
MSVC -arch is independent from the ability to use CPU intrinsics It's possible to write AVX512 intrinsics in code compile only with -arch:AVX . So do not require -arch for being able to do so, especially since there is no -arch option for only AVX512F without other AVX512 subsets (the option enables also CD, BW, DQ and VL https://docs.microsoft.com/en-us/cpp/build/reference/arch-x64). https://crashreport.libreoffice.org/stats/crash_details/55ef825d-c323-4df9-95e2-76672c674e60 is presumably caused by this, I can see use of registers XMM0-15 in arraysumAVX512.cxx built with -arch:AVX2 but when built with -arch:AVX512 registers XMM16-31 are used too (I'm not sure if that's AVX512DQ or something else, I can't find info on it). Change-Id: I74473333a17e618327d43b920b8929d1b0e733b8 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/129724 Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com> Reviewed-by: Luboš Luňák <l.lunak@collabora.com> Tested-by: Jenkins
2022-01-26 13:34:08 +01:00
#include <intrin.h>
tdf#142307 - Upgrade SSE2 sum to AVX512 sum with Neumaier 1 This part focuses on allowing it on replacing arrayfunctor By thefault it will try AVX512F (1,17%) If not available will use AVX (94,77%) Use of AVX2 (82,28%) has been avoided even if the code could been more compact Source of hardware statistics: https://store.steampowered.com/hwsurvey Change-Id: Iae737a565379e82c5f84f3fdee6321ac74f59d40 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115675 Tested-by: Jenkins Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
2021-05-16 16:04:20 +02:00
#include <immintrin.h>
MSVC -arch is independent from the ability to use CPU intrinsics It's possible to write AVX512 intrinsics in code compile only with -arch:AVX . So do not require -arch for being able to do so, especially since there is no -arch option for only AVX512F without other AVX512 subsets (the option enables also CD, BW, DQ and VL https://docs.microsoft.com/en-us/cpp/build/reference/arch-x64). https://crashreport.libreoffice.org/stats/crash_details/55ef825d-c323-4df9-95e2-76672c674e60 is presumably caused by this, I can see use of registers XMM0-15 in arraysumAVX512.cxx built with -arch:AVX2 but when built with -arch:AVX512 registers XMM16-31 are used too (I'm not sure if that's AVX512DQ or something else, I can't find info on it). Change-Id: I74473333a17e618327d43b920b8929d1b0e733b8 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/129724 Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com> Reviewed-by: Luboš Luňák <l.lunak@collabora.com> Tested-by: Jenkins
2022-01-26 13:34:08 +01:00
#endif
tdf#142307 - Upgrade SSE2 sum to AVX512 sum with Neumaier 1 This part focuses on allowing it on replacing arrayfunctor By thefault it will try AVX512F (1,17%) If not available will use AVX (94,77%) Use of AVX2 (82,28%) has been avoided even if the code could been more compact Source of hardware statistics: https://store.steampowered.com/hwsurvey Change-Id: Iae737a565379e82c5f84f3fdee6321ac74f59d40 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115675 Tested-by: Jenkins Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
2021-05-16 16:04:20 +02:00
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#else // compiler Clang and GCC
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#if defined(__SSE2__) || defined(__x86_64__) // SSE2 is required for X64
#define LO_SSE2_AVAILABLE
Add test as an example how to add CPU intrinsics support Also makes sure that if the CPU dataction or compiler detection doesn't work correctly, the test could potentially crash. Change-Id: If0862c0a736c8e1f5ba1117b248918e4c1fb2f4d Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96779 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2020-06-20 16:59:59 +02:00
#include <emmintrin.h>
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#endif // defined(__SSE2__)
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#if defined(__SSSE3__)
#define LO_SSSE3_AVAILABLE
Add test as an example how to add CPU intrinsics support Also makes sure that if the CPU dataction or compiler detection doesn't work correctly, the test could potentially crash. Change-Id: If0862c0a736c8e1f5ba1117b248918e4c1fb2f4d Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96779 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2020-06-20 16:59:59 +02:00
#include <tmmintrin.h>
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#endif // defined(__SSSE3__)
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#if defined(__AVX__)
#define LO_AVX_AVAILABLE
Add test as an example how to add CPU intrinsics support Also makes sure that if the CPU dataction or compiler detection doesn't work correctly, the test could potentially crash. Change-Id: If0862c0a736c8e1f5ba1117b248918e4c1fb2f4d Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96779 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2020-06-20 16:59:59 +02:00
#include <immintrin.h>
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#endif // defined(__AVX__)
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
#if defined(__AVX2__)
#define LO_AVX2_AVAILABLE
Add test as an example how to add CPU intrinsics support Also makes sure that if the CPU dataction or compiler detection doesn't work correctly, the test could potentially crash. Change-Id: If0862c0a736c8e1f5ba1117b248918e4c1fb2f4d Reviewed-on: https://gerrit.libreoffice.org/c/core/+/96779 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2020-06-20 16:59:59 +02:00
#include <immintrin.h>
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#endif // defined(__AVX2__)
tdf#142307 - Upgrade SSE2 sum to AVX512 sum with Neumaier 1 This part focuses on allowing it on replacing arrayfunctor By thefault it will try AVX512F (1,17%) If not available will use AVX (94,77%) Use of AVX2 (82,28%) has been avoided even if the code could been more compact Source of hardware statistics: https://store.steampowered.com/hwsurvey Change-Id: Iae737a565379e82c5f84f3fdee6321ac74f59d40 Reviewed-on: https://gerrit.libreoffice.org/c/core/+/115675 Tested-by: Jenkins Reviewed-by: Mike Kaganski <mike.kaganski@collabora.com>
2021-05-16 16:04:20 +02:00
#if defined(__AVX512F__)
#define LO_AVX512F_AVAILABLE
#include <immintrin.h>
#else
#endif // defined(__AVX512F__)
include supported SIMD intrinsics headers automatically If we include simdsupport.hxx, this now includes the supported intrinsics headers automatically, so this removes the need to do it ourselves. Change-Id: Icc406316a7e27b7b5fd5905f51664f4fe47e5468 Reviewed-on: https://gerrit.libreoffice.org/75783 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2019-07-17 21:27:00 +09:00
#endif // end compiler Clang and GCC
CPU intrinsics detection (SSE, AVX) Adds CPU intrinsics detection in configure pass for compile time detection and "cpuid" runtime detection of which CPU instruction sets are available on the user device. Change-Id: I0ee4d0b22a7c51f72796d43e7383a31d03b437ad Reviewed-on: https://gerrit.libreoffice.org/75175 Tested-by: Jenkins Reviewed-by: Tomaž Vajngerl <quikee@gmail.com>
2017-05-08 10:47:04 +02:00
/* vim:set shiftwidth=4 softtabstop=4 expandtab: */
Reference in New Issue Copy Permalink