Add detection of Intel x86 AVX-VNNI instructions. (#196)

Tested using Intel SDE:

bash scripts/


OPTIONS+=(-quark); PLATFORMS+=("Quark")
OPTIONS+=(-p4); PLATFORMS+=("Pentium4")
OPTIONS+=(-p4p); PLATFORMS+=("Pentium4 Prescott")
OPTIONS+=(-mrm); PLATFORMS+=("Merom")
OPTIONS+=(-pnr); PLATFORMS+=("Penryn")
OPTIONS+=(-nhm); PLATFORMS+=("Nehalem")
OPTIONS+=(-wsm); PLATFORMS+=("Westmere")
OPTIONS+=(-snb); PLATFORMS+=("Sandy Bridge")
OPTIONS+=(-ivb); PLATFORMS+=("Ivy Bridge")
OPTIONS+=(-hsw); PLATFORMS+=("Haswell")
OPTIONS+=(-bdw); PLATFORMS+=("Broadwell")
OPTIONS+=(-slt); PLATFORMS+=("Saltwell")
OPTIONS+=(-slm); PLATFORMS+=("Silvermont")
OPTIONS+=(-glm); PLATFORMS+=("Goldmont")
OPTIONS+=(-glp); PLATFORMS+=("Goldmont Plus")
OPTIONS+=(-tnt); PLATFORMS+=("Tremont")
OPTIONS+=(-snr); PLATFORMS+=("Snow Ridge")
OPTIONS+=(-skl); PLATFORMS+=("Skylake")
OPTIONS+=(-cnl); PLATFORMS+=("Cannon Lake")
OPTIONS+=(-icl); PLATFORMS+=("Ice Lake")
OPTIONS+=(-skx); PLATFORMS+=("Skylake server")
OPTIONS+=(-clx); PLATFORMS+=("Cascade Lake")
OPTIONS+=(-cpx); PLATFORMS+=("Cooper Lake")
OPTIONS+=(-icx); PLATFORMS+=("Ice Lake server")
OPTIONS+=(-knl); PLATFORMS+=("Knights landing")
OPTIONS+=(-knm); PLATFORMS+=("Knights mill")
OPTIONS+=(-tgl); PLATFORMS+=("Tiger Lake")
OPTIONS+=(-adl); PLATFORMS+=("Alder Lake")
OPTIONS+=(-mtl); PLATFORMS+=("Meteor Lake")
OPTIONS+=(-rpl); PLATFORMS+=("Raptor Lake")
OPTIONS+=(-spr); PLATFORMS+=("Sapphire Rapids")
OPTIONS+=(-gnr); PLATFORMS+=("Granite Rapids")
OPTIONS+=(-srf); PLATFORMS+=("Sierra Forest")
OPTIONS+=(-grr); PLATFORMS+=("Grand Ridge")
OPTIONS+=(-future); PLATFORMS+=("Future chip")


for I in "${!PLATFORMS[@]}"; do
  echo "${PLATFORMS["${I}"]}"
  "${SDE_BIN}" "${OPTIONS[$I]}" -- ./build/local/isa-info | grep "AVXVNNI"


        AVXVNNI: no
Sandy Bridge
        AVXVNNI: no
Ivy Bridge
        AVXVNNI: no
        AVXVNNI: no
        AVXVNNI: no
        AVXVNNI: no
        AVXVNNI: no
Goldmont Plus
        AVXVNNI: no
        AVXVNNI: no
Snow Ridge
        AVXVNNI: no
        AVXVNNI: no
Cannon Lake
        AVXVNNI: no
Ice Lake
        AVXVNNI: no
Skylake server
        AVXVNNI: no
Cascade Lake
        AVXVNNI: no
Cooper Lake
        AVXVNNI: no
Ice Lake server
        AVXVNNI: no
Knights landing
        AVXVNNI: no
Knights mill
        AVXVNNI: no
Tiger Lake
        AVXVNNI: no
Alder Lake
        AVXVNNI: yes
Meteor Lake
        AVXVNNI: yes
Raptor Lake
        AVXVNNI: yes
Sapphire Rapids
        AVXVNNI: yes
Granite Rapids
        AVXVNNI: yes
Sierra Forest
        AVXVNNI: yes
Grand Ridge
        AVXVNNI: yes
Future chip
        AVXVNNI: yes
3 files changed
tree: 007610914ea9a9d54b71ff412cf9b07c6bb078ee
  1. .github/
  2. bench/
  3. cmake/
  4. deps/
  5. include/
  6. jni/
  7. scripts/
  8. src/
  9. test/
  10. tools/
  11. .bazelrc
  12. .bazelversion
  13. .gitignore
  14. .travis.yml
  15. appveyor.yml
  16. BUILD.bazel
  17. CMakeLists.txt
  18. CMakePresets.json
  21. confu.yaml
  26. WORKSPACE.bazel

CPU INFOrmation library

BSD (2 clause) License Linux/Mac build status Windows build status

cpuinfo is a library to detect essential for performance optimization information about host CPU.


  • Cross-platform availability:
    • Linux, Windows, macOS, Android, and iOS operating systems
    • x86, x86-64, ARM, and ARM64 architectures
  • Modern C/C++ interface
    • Thread-safe
    • No memory allocation after initialization
    • No exceptions thrown
  • Detection of supported instruction sets, up to AVX512 (x86) and ARMv8.3 extensions
  • Detection of SoC and core information:
    • Processor (SoC) name
    • Vendor and microarchitecture for each CPU core
    • ID (MIDR on ARM, CPUID leaf 1 EAX value on x86) for each CPU core
  • Detection of cache information:
    • Cache type (instruction/data/unified), size and line size
    • Cache associativity
    • Cores and logical processors (hyper-threads) sharing the cache
  • Detection of topology information (relative between logical processors, cores, and processor packages)
  • Well-tested production-quality code:
    • 60+ mock tests based on data from real devices
    • Includes work-arounds for common bugs in hardware and OS kernels
    • Supports systems with heterogenous cores, such as big.LITTLE and Max.Med.Min
  • Permissive open-source license (Simplified BSD)


Log processor name:

printf("Running on %s CPU\n", cpuinfo_get_package(0)->name);

Detect if target is a 32-bit or 64-bit ARM system:

    /* 32-bit ARM-specific code here */

Check if the host CPU supports ARM NEON

if (cpuinfo_has_arm_neon()) {

Check if the host CPU supports x86 AVX

if (cpuinfo_has_x86_avx()) {

Check if the thread runs on a Cortex-A53 core

switch (cpuinfo_get_current_core()->uarch) {
    case cpuinfo_uarch_cortex_a53:

Get the size of level 1 data cache on the fastest core in the processor (e.g. big core in big.LITTLE ARM systems):

const size_t l1_size = cpuinfo_get_processor(0)->cache.l1d->size;

Pin thread to cores sharing L2 cache with the current core (Linux or Android)

cpu_set_t cpu_set;
const struct cpuinfo_cache* current_l2 = cpuinfo_get_current_processor()->cache.l2;
for (uint32_t i = 0; i < current_l2->processor_count; i++) {
    CPU_SET(cpuinfo_get_processor(current_l2->processor_start + i)->linux_id, &cpu_set);
pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpu_set);

Use via pkg-config

If you would like to provide your project's build environment with the necessary compiler and linker flags in a portable manner, the library by default when built enables CPUINFO_BUILD_PKG_CONFIG and will generate a pkg-config manifest (libcpuinfo.pc). Here are several examples of how to use it:

Command Line

If you used your distro's package manager to install the library, you can verify that it is available to your build environment like so:

$ pkg-config --cflags --libs libcpuinfo
-I/usr/include/x86_64-linux-gnu/ -L/lib/x86_64-linux-gnu/ -lcpuinfo

If you have installed the library from source into a non-standard prefix, pkg-config may need help finding it:

$ PKG_CONFIG_PATH="/home/me/projects/cpuinfo/prefix/lib/pkgconfig/:$PKG_CONFIG_PATH" pkg-config --cflags --libs libcpuinfo
-I/home/me/projects/cpuinfo/prefix/include -L/home/me/projects/cpuinfo/prefix/lib -lcpuinfo

GNU Autotools

To use with the GNU Autotools include the following snippet in your project's

# CPU INFOrmation library...
    [libcpuinfo], [libcpuinfo], [],
    [AC_MSG_ERROR([libcpuinfo missing...])])


To use with Meson you just need to add dependency('libcpuinfo') as a dependency for your executable.

    meson_version: '>=0.55.0'

    sources: 'main.cpp',
    dependencies: dependency('libcpuinfo')


This project can be built using Bazel.

You can also use this library as a dependency to your Bazel project. Add to the WORKSPACE file:

load("@bazel_tools//tools/build_defs/repo:git.bzl", "git_repository")

    name = "org_pytorch_cpuinfo",
    branch = "master",
    remote = "",

And to your BUILD file:

    name = "cpuinfo_test",
    srcs = [
        # ...
    deps = [


To use with CMake use the FindPkgConfig module. Here is an example:

cmake_minimum_required(VERSION 3.6)

pkg_check_modules(CpuInfo REQUIRED IMPORTED_TARGET libcpuinfo)

add_executable(${PROJECT_NAME} main.cpp)
target_link_libraries(${PROJECT_NAME} PkgConfig::CpuInfo)


To use within a vanilla makefile, you can call pkg-config directly to supply compiler and linker flags using shell substitution.

CFLAGS=-g3 -Wall -Wextra -Werror ...
LDFLAGS=-lfoo ...
CFLAGS+= $(pkg-config --cflags libcpuinfo)
LDFLAGS+= $(pkg-config --libs libcpuinfo)

Exposed information

  • [x] Processor (SoC) name
  • [x] Microarchitecture
  • [x] Usable instruction sets
  • [ ] CPU frequency
  • [x] Cache
    • [x] Size
    • [x] Associativity
    • [x] Line size
    • [x] Number of partitions
    • [x] Flags (unified, inclusive, complex hash function)
    • [x] Topology (logical processors that share this cache level)
  • [ ] TLB
    • [ ] Number of entries
    • [ ] Associativity
    • [ ] Covered page types (instruction, data)
    • [ ] Covered page sizes
  • [x] Topology information
    • [x] Logical processors
    • [x] Cores
    • [x] Packages (sockets)

Supported environments:

  • [x] Android
    • [x] x86 ABI
    • [x] x86_64 ABI
    • [x] armeabi ABI
    • [x] armeabiv7-a ABI
    • [x] arm64-v8a ABI
    • [ ] mips ABI
    • [ ] mips64 ABI
  • [x] Linux
    • [x] x86
    • [x] x86-64
    • [x] 32-bit ARM (ARMv5T and later)
    • [x] ARM64
    • [ ] PowerPC64
  • [x] iOS
    • [x] x86 (iPhone simulator)
    • [x] x86-64 (iPhone simulator)
    • [x] ARMv7
    • [x] ARM64
  • [x] macOS
    • [x] x86
    • [x] x86-64
    • [x] ARM64 (Apple silicon)
  • [x] Windows
    • [x] x86
    • [x] x86-64
    • [x] arm64


  • Processor (SoC) name detection
    • [x] Using CPUID leaves 0x80000002–0x80000004 on x86/x86-64
    • [x] Using /proc/cpuinfo on ARM
    • [x] Using ro.chipname, ro.board.platform, ro.product.board, ro.mediatek.platform, ro.arch properties (Android)
    • [ ] Using kernel log (dmesg) on ARM Linux
    • [x] Using Windows registry on ARM64 Windows
  • Vendor and microarchitecture detection
    • [x] Intel-designed x86/x86-64 cores (up to Sunny Cove, Goldmont Plus, and Knights Mill)
    • [x] AMD-designed x86/x86-64 cores (up to Puma/Jaguar and Zen 2)
    • [ ] VIA-designed x86/x86-64 cores
    • [ ] Other x86 cores (DM&P, RDC, Transmeta, Cyrix, Rise)
    • [x] ARM-designed ARM cores (up to Cortex-A55, Cortex-A77, and Neoverse E1/V1/N2/V2)
    • [x] Qualcomm-designed ARM cores (Scorpion, Krait, and Kryo)
    • [x] Nvidia-designed ARM cores (Denver and Carmel)
    • [x] Samsung-designed ARM cores (Exynos)
    • [x] Intel-designed ARM cores (XScale up to 3rd-gen)
    • [x] Apple-designed ARM cores (up to Lightning and Thunder)
    • [x] Cavium-designed ARM cores (ThunderX)
    • [x] AppliedMicro-designed ARM cores (X-Gene)
  • Instruction set detection
    • [x] Using CPUID (x86/x86-64)
    • [x] Using /proc/cpuinfo on 32-bit ARM EABI (Linux)
    • [x] Using microarchitecture heuristics on (32-bit ARM)
    • [x] Using FPSID and WCID registers (32-bit ARM)
    • [x] Using getauxval (Linux/ARM)
    • [x] Using /proc/self/auxv (Android/ARM)
    • [ ] Using instruction probing on ARM (Linux)
    • [ ] Using CPUID registers on ARM64 (Linux)
    • [x] Using IsProcessorFeaturePresent on ARM64 Windows
  • Cache detection
    • [x] Using CPUID leaf 0x00000002 (x86/x86-64)
    • [x] Using CPUID leaf 0x00000004 (non-AMD x86/x86-64)
    • [ ] Using CPUID leaves 0x80000005-0x80000006 (AMD x86/x86-64)
    • [x] Using CPUID leaf 0x8000001D (AMD x86/x86-64)
    • [x] Using /proc/cpuinfo (Linux/pre-ARMv7)
    • [x] Using microarchitecture heuristics (ARM)
    • [x] Using chipset name (ARM)
    • [x] Using sysctlbyname (Mach)
    • [x] Using sysfs typology directories (ARM/Linux)
    • [ ] Using sysfs cache directories (Linux)
    • [x] Using GetLogicalProcessorInformationEx on ARM64 Windows
  • TLB detection
    • [x] Using CPUID leaf 0x00000002 (x86/x86-64)
    • [ ] Using CPUID leaves 0x80000005-0x80000006 and 0x80000019 (AMD x86/x86-64)
    • [x] Using microarchitecture heuristics (ARM)
  • Topology detection
    • [x] Using CPUID leaf 0x00000001 on x86/x86-64 (legacy APIC ID)
    • [x] Using CPUID leaf 0x0000000B on x86/x86-64 (Intel APIC ID)
    • [ ] Using CPUID leaf 0x8000001E on x86/x86-64 (AMD APIC ID)
    • [x] Using /proc/cpuinfo (Linux)
    • [x] Using host_info (Mach)
    • [x] Using GetLogicalProcessorInformationEx (Windows)
    • [x] Using sysfs (Linux)
    • [x] Using chipset name (ARM/Linux)