Simd.h File Reference

#include "arcane/utils/SimdCommon.h"
#include "arcane/utils/Real3.h"
#include "arcane/utils/Real2.h"
#include "arcane/utils/Real3x3.h"
#include "arcane/utils/Real2x2.h"
#include "arcane/utils/ArrayView.h"
#include "arcane/utils/SimdEMUL.h"

Include dependency graph for Simd.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
class	Arcane::SimdReal3
	Represents a vectorized Real3. More...
class	Arcane::SimdReal2
	Represents a vectorized Real2. More...
class	Arcane::SimdReal3x3
	Represents a vectorized Real3x3. More...
class	Arcane::SimdReal2x2
	Represents a vectorized Real2x2. More...
class	Arcane::SimdTypeTraits< DataType >
	Characteristics of vector types. More...
class	Arcane::SimdTypeTraits< Real >
class	Arcane::SimdTypeTraits< Real2 >
class	Arcane::SimdTypeTraits< Real2x2 >
class	Arcane::SimdTypeTraits< Real3 >
class	Arcane::SimdTypeTraits< Real3x3 >
class	Arcane::SimdEnumeratorBase
	Base class for vector enumerators with indirection. More...

Namespaces
namespace	Arcane
	-- tab-width: 2; indent-tabs-mode: nil; coding: utf-8-with-signature --

Macros
#define	ENUMERATE_SIMD_REAL(_iter)
	Macro to iterate over the indices of a SIMD real or derived vector (Real2, Real3, ...).

Typedefs
typedef EMULSimdInfo	Arcane::SimdInfo
typedef SimdInfo::SimdReal	Arcane::SimdReal
	SIMD vector of real numbers.

Variables
const int	Arcane::SimdSize = SimdReal::Length

Detailed Description

This file contains the declarations of types used to manage vectorization. Since several possible mechanisms exist, a choice must be made based on the machine architecture and compilation options. Each mechanism uses vectors of different sizes. In our case, since vectorization is primarily used for calculations on doubles, the size of a vector will be equal to the number of doubles that can fit into a vector.

Currently, we support the following modes in order of priority. If a mode is supported, the others are not used.

• AVX512 for Intel Knight Landing (KNL) or Xeon Skylake type architectures. The vector size in this mode is 8.
• AVX. For this mode, Arcane must be compiled with the option '–with-avx'. There are two modes: classic AVX and AVX2. For now, only the former is used, due to a lack of machines to test the latter. The vector size in this mode is 4.
• SSE. This mode is available by default because it exists on all x64 platforms. There are also several versions, and we limit ourselves to version 2. The vector size in this mode is 2.
• no mode. In this case, there is no specific vectorization. Nevertheless, to test the code, we allow emulation with vectors of size 2.

Definition in file Simd.h.