gmx::CpuInfo Class Reference

#include <gromacs/hardware/cpuinfo.h>

Description

Detect CPU capabilities and basic logical processor info.

This class provides a lot of information about x86 CPUs, and some very limited information about other hardware. The logical processor information is only available on x86, and is used as a fallback implementation in the HardwareTopology class. If you actually need information about the hardware topology, use the much more general implementation in the HardwareTopology class instead, since that will both be more portable and contain more information.

Classes
struct	LogicalProcessor
	Entry with basic information for a single logical processor. More...

Public Types
enum	SupportLevel { SupportLevel::None, SupportLevel::Name, SupportLevel::Features, SupportLevel::LogicalProcessorInfo }
	Amount of cpu information present (incremental) More...
enum	Vendor {   Vendor::Unknown, Vendor::Intel, Vendor::Amd, Vendor::Fujitsu,   Vendor::Ibm, Vendor::Arm, Vendor::Oracle }
	Processor/system vendors. More...
enum	Feature {   Feature::X86_Aes, Feature::X86_Amd, Feature::X86_Apic, Feature::X86_Avx,   Feature::X86_Avx2, Feature::X86_Avx512F, Feature::X86_Avx512PF, Feature::X86_Avx512ER,   Feature::X86_Avx512CD, Feature::X86_Avx512BW, Feature::X86_Avx512VL, Feature::X86_Clfsh,   Feature::X86_Cmov, Feature::X86_Cx8, Feature::X86_Cx16, Feature::X86_F16C,   Feature::X86_Fma, Feature::X86_Fma4, Feature::X86_Hle, Feature::X86_Htt,   Feature::X86_Intel, Feature::X86_Lahf, Feature::X86_MisalignSse, Feature::X86_Mmx,   Feature::X86_Msr, Feature::X86_NonstopTsc, Feature::X86_Pcid, Feature::X86_Pclmuldq,   Feature::X86_Pdcm, Feature::X86_PDPE1GB, Feature::X86_Popcnt, Feature::X86_Pse,   Feature::X86_Rdrnd, Feature::X86_Rdtscp, Feature::X86_Rtm, Feature::X86_Sha,   Feature::X86_Sse2, Feature::X86_Sse3, Feature::X86_Sse4A, Feature::X86_Sse4_1,   Feature::X86_Sse4_2, Feature::X86_Ssse3, Feature::X86_Tdt, Feature::X86_X2Apic,   Feature::X86_Xop, Feature::Arm_Neon, Feature::Arm_NeonAsimd, Feature::Ibm_Qpx,   Feature::Ibm_Vmx, Feature::Ibm_Vsx, Feature::Fujitsu_HpcAce }
	List of CPU features. More...

Public Member Functions
SupportLevel	supportLevel () const
	Check what cpu information is available. More...
Vendor	vendor () const
	Enumerated value for vendor.
const std::string &	vendorString () const
	String description of vendor: More...
const std::string &	brandString () const
	String description of processor.
int	family () const
	Major version/generation of the processor.
int	model () const
	Middle version of the processor.
int	stepping () const
	Minor version of the processor.
bool	feature (Feature f) const
	Check for availability of specific feature. More...
const std::set< Feature > &	featureSet () const
	Set of all supported features on this processor. More...
const std::vector < LogicalProcessor > &	logicalProcessors () const
	Reference to processing unit topology. More...

Static Public Member Functions
static CpuInfo	detect ()
	Perform detection and construct a CpuInfo class from the results. More...
static const std::string &	featureString (Feature f)
	String description of a specific feature. More...

Member Enumeration Documentation

enum gmx::CpuInfo::Feature

strong

List of CPU features.

These values can be used as arguments to the feature() method to check whether a specific feature was found on the CPU we are running on.

Enumerator
X86_Aes	x86 advanced encryption standard accel.
X86_Amd	This is an AMD x86 processor.
X86_Apic	APIC support.
X86_Avx	Advanced vector extensions.
X86_Avx2	AVX2 including gather support (not used yet)
X86_Avx512F	Foundation AVX-512 instructions.
X86_Avx512PF	Extended gather/scatter for AVX-512.
X86_Avx512ER	AVX-512 exponential and recpirocal extensions.
X86_Avx512CD	Memory conflict-detection for AVX-512.
X86_Avx512BW	AVX-512 byte and word instructions.
X86_Avx512VL	AVX-512 vector length extensions.
X86_Clfsh	Supports CLFLUSH instruction.
X86_Cmov	Conditional move insn support.
X86_Cx8	Supports CMPXCHG8B (8-byte compare-exchange)
X86_Cx16	Supports CMPXCHG16B (16-byte compare-exchg)
X86_F16C	Supports 16-bit FP conversion instructions.
X86_Fma	Fused-multiply add support (mainly for AVX)
X86_Fma4	4-operand FMA, only on AMD for now
X86_Hle	Hardware lock elision.
X86_Htt	Hyper-Threading enabled (NOTE: might not match the CPUID HTT support flag)
X86_Intel	This is an Intel x86 processor.
X86_Lahf	LAHF/SAHF support in 64 bits.
X86_MisalignSse	Support for misaligned SSE data instructions.
X86_Mmx	MMX registers and instructions.
X86_Msr	Supports Intel model-specific-registers.
X86_NonstopTsc	Invariant TSC (constant rate in ACPI states)
X86_Pcid	Process context identifier support.
X86_Pclmuldq	Carry-less 64-bit multiplication supported.
X86_Pdcm	Perfmon and Debug Capability.
X86_PDPE1GB	Support for 1GB pages.
X86_Popcnt	Supports the POPCNT (population count) insn.
X86_Pse	Supports 4MB-pages (page size extension)
X86_Rdrnd	RDRAND high-quality hardware random numbers.
X86_Rdtscp	Serializing rdtscp instruction available.
X86_Rtm	Restricted transactional memory.
X86_Sha	Intel SHA extensions.
X86_Sse2	SSE 2.
X86_Sse3	SSE 3.
X86_Sse4A	SSE 4A.
X86_Sse4_1	SSE 4.1.
X86_Sse4_2	SSE 4.2.
X86_Ssse3	Supplemental SSE3.
X86_Tdt	TSC deadline timer.
X86_X2Apic	Extended xAPIC Support.
X86_Xop	AMD extended instructions, only AMD for now.
Arm_Neon	32-bit ARM NEON
Arm_NeonAsimd	64-bit ARM AArch64 Advanced SIMD
Ibm_Qpx	IBM QPX SIMD (BlueGene/Q and later)
Ibm_Vmx	IBM VMX SIMD (Altivec on Power6 and later)
Ibm_Vsx	IBM VSX SIMD (Power7 and later)
Fujitsu_HpcAce	Fujitsu Sparc64 HPC-ACE.

enum gmx::CpuInfo::SupportLevel

strong

Amount of cpu information present (incremental)

Enumerator
None	No cpu information whatsoever. Sorry.
Name	Only vendor and/or brand is set.
Features	Some features are set.
LogicalProcessorInfo	Everything includling logical processor information.

enum gmx::CpuInfo::Vendor

strong

Processor/system vendors.

Enumerator
Unknown	Unidentified.
Intel	GenuineIntel.
Amd	AuthenticAMD.
Fujitsu	Only works on Linux (parsed from /proc/cpuinfo)
Ibm	Only works on Linux (parsed from /proc/cpuinfo)
Arm	Only works on Linux (parsed from /proc/cpuinfo)
Oracle	Cannot detect anything else yet (no /proc/cpuinfo available)

Member Function Documentation

CpuInfo gmx::CpuInfo::detect ( )

static

Perform detection and construct a CpuInfo class from the results.

Note

The detection should generally be performed again in different contexts. This might seem like overkill, but there are systems (e.g. Arm) where processors can go completely offline during deep sleep, so at least in theory it is good to have a possibility of forcing re-detection if necessary.

bool gmx::CpuInfo::feature ( Feature  f ) const

inline

Check for availability of specific feature.

Parameters

f	feature to query support for

Returns

True if the feature is available, otherwise false.

const std::set<Feature>& gmx::CpuInfo::featureSet ( ) const

inline

Set of all supported features on this processor.

This is only intended for logfiles, debugging or similar output when we need a full list of all the features available on the CPU.

static const std::string& gmx::CpuInfo::featureString ( Feature  f )

inlinestatic

String description of a specific feature.

Exceptions

std::out_of_range

if the feature is not present in the internal map of feature names. This can only happen if we extend the enum type but forget to add the string with the feature name.

const std::vector<LogicalProcessor>& gmx::CpuInfo::logicalProcessors ( ) const

inline

Reference to processing unit topology.

Only a few systems (x86) provide logical processor information in cpuinfo. This method returns a reference to a vector, whose length will either be zero (if topology information is not available) or the number of enabled processing units, as defined by the operating system. In the latter case, each entry will contain information about the relative rank in the core and socket of this hardware thread.

This is only meant to be use as a fallback implementation for our HardwareTopology class; any user code that needs access to hardware topology information should use that class instead.

Note

For clarity, it is likely better to use the supportLevel() method to check if this information is available rather than relying on the length of the vector.

SupportLevel gmx::CpuInfo::supportLevel ( ) const

inline

Check what cpu information is available.

The amount of cpu information that can be detected depends on the OS, compiler, and CPU, and on non-x86 platforms it can be fragile. Before basing decisions on the output or warning the user about optimizations, you want to check whether it was possible to detect the information you need.

const std::string& gmx::CpuInfo::vendorString ( ) const

inline

String description of vendor:

Exceptions

std::out_of_range

if the vendor is not present in the internal map of vendor names. This can only happen if we extend the enum type but forget to add the string with the vendor name.

---

The documentation for this class was generated from the following files:

• src/gromacs/hardware/cpuinfo.h
• src/gromacs/hardware/cpuinfo.cpp