saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-29 20:14:01 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Add some CPU feature detection for Apple platforms. (#20769)

Browse source 
Add CPUIDInfo::ArmAppleInit() to detect CPU features on Apple platforms. This initial implementation is not comprehensive.
This commit is contained in:
Edward Chen 2024-05-23 15:59:46 -07:00 committed by GitHub
parent b522df0ae4
commit d1af19db9d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 42 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

48
onnxruntime/core/common/cpuid_info.cc
View file

@ -73,7 +73,7 @@ void decodeMIDR(uint32_t midr, uint32_t uarch[1]);
namespace onnxruntime {
#ifdef CPUIDINFO_ARCH_X86
#if defined(CPUIDINFO_ARCH_X86)
static inline void GetCPUID(int function_id, int data[4]) { // NOLINT
#if defined(_MSC_VER)
@ -141,10 +141,11 @@ void CPUIDInfo::X86Init() {
}
}
#endif /* CPUIDINFO_ARCH_X86 */
#endif // defined(CPUIDINFO_ARCH_X86)
#if defined(CPUIDINFO_ARCH_ARM)
#ifdef __linux__
#if defined(__linux__)
void CPUIDInfo::ArmLinuxInit() {
// Assuming no hyper-threading, no NUMA groups
@ -190,11 +191,11 @@ void CPUIDInfo::ArmLinuxInit() {
}
}
#elif defined(_WIN32)
#elif defined(_WIN32) // ^ defined(__linux__)
void CPUIDInfo::ArmWindowsInit() {
// ARM32 certainly doesn't have fp16, so we will skip the logic to avoid using RegGetValueA Windows API
#ifndef _M_ARM
#if !defined(_M_ARM)
#pragma region Application Family or OneCore Family
#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP | WINAPI_PARTITION_SYSTEM)
// Read MIDR from windows registry
@ -243,12 +244,12 @@ void CPUIDInfo::ArmWindowsInit() {
lastUarch = uarch;
}
}
#endif /* Application Family or OneCore Family */
#endif // WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP | WINAPI_PARTITION_SYSTEM)
has_arm_neon_dot_ = (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE) != 0);
#else
#else // ^ !defined(_M_ARM) / v defined(_M_ARM)
has_arm_neon_dot_ = false;
#endif
#endif // defined(_M_ARM)
#if defined(CPUINFO_SUPPORTED)
if (pytorch_cpuinfo_init_) {
@ -266,8 +267,29 @@ void CPUIDInfo::ArmWindowsInit() {
}
}
#endif /* (arm or arm64) and windows */
#endif /* arm or arm64*/
#elif defined(__APPLE__) // ^ defined(_WIN32)
void CPUIDInfo::ArmAppleInit() {
#if defined(CPUINFO_SUPPORTED)
if (pytorch_cpuinfo_init_) {
is_hybrid_ = cpuinfo_get_uarchs_count() > 1;
has_arm_neon_dot_ = cpuinfo_has_arm_neon_dot();
has_fp16_ = cpuinfo_has_arm_neon_fp16_arith();
has_arm_neon_i8mm_ = cpuinfo_has_arm_i8mm();
has_arm_sve_i8mm_ = cpuinfo_has_arm_sve() && cpuinfo_has_arm_i8mm();
has_arm_neon_bf16_ = cpuinfo_has_arm_neon_bf16();
// TODO figure out how to set core_uarchs_ and is_armv8_narrow_ld_
} else
#endif // defined(CPUINFO_SUPPORTED)
{
// No fallback detection attempted now. Add if needed.
}
}
#endif // defined(__APPLE__)
#endif // defined(CPUIDINFO_ARCH_ARM)
uint32_t CPUIDInfo::GetCurrentCoreIdx() const {
#ifdef _WIN32
@ -295,11 +317,13 @@ CPUIDInfo::CPUIDInfo() {
"due to undetected CPU features.";
}
#endif // defined(CPUINFO_SUPPORTED)
#ifdef __linux__
#if defined(__linux__)
ArmLinuxInit();
#elif defined(_WIN32)
ArmWindowsInit();
#endif /* (arm or arm64) and windows */
#elif defined(__APPLE__)
ArmAppleInit();
#endif
#endif // defined(CPUIDINFO_ARCH_ARM)
}
} // namespace onnxruntime

8
onnxruntime/core/common/cpuid_info.h
View file

@ -129,7 +129,7 @@ class CPUIDInfo {
bool pytorch_cpuinfo_init_{false};
#endif // defined(CPUINFO_SUPPORTED)
#ifdef __linux__
#if defined(__linux__)
void ArmLinuxInit();
@ -137,7 +137,11 @@ class CPUIDInfo {
void ArmWindowsInit();
#endif /* (arm or arm64) and windows */
#elif defined(__APPLE__)
void ArmAppleInit();
#endif
#endif // defined(CPUIDINFO_ARCH_ARM)
};