Runtime Environment¶

Runtime CPU detection and environment configuration.

C¶

• function clock_elapsed(uint64_t) Returns the elapsed time in seconds since start_time, based on clock_now .
• function clock_frequency() Returns the number of ticks per second for the clock used in clock_now().
• function clock_now() Returns the current value of the OS's highest-resolution monotonic clock. On modern hardware, these typically map to hardware instructions (RDTSC/CNTVCT_EL0) via user-mode shared pages, avoiding a heavy syscall context switch.
• function timestamps<max_count, fence>::compute_rdtsc_overhead() Measure and return the RDTSC read overhead in cycles.
• variable timestamps<max_count, fence>::count Number of checkpoints currently stored in dur / msg .
• struct cpu_features Detected x86 CPU capabilities and identifying strings.

D¶

• variable timestamps<max_count, fence>::dur Per-checkpoint elapsed cycle counts (overhead-subtracted), indexed by record order.

E¶

• function stopwatch::elapsed() Return the elapsed time since construction cast to the requested Duration type.
• function stopwatch::elapsed_ms() Convenience wrapper for elapsed returning milliseconds.
• function stopwatch::elapsed_ns() Convenience wrapper for elapsed returning nanoseconds.
• function stopwatch::elapsed_s() Return the elapsed wall-clock time in seconds since construction.
• function stopwatch::elapsed_us() Convenience wrapper for elapsed returning microseconds.
• variable cpu_features::exmax Highest supported extended CPUID leaf (leaf 0x80000000 EAX).

F¶

• variable stopwatch::frequency Tick frequency of the underlying clock, captured at construction.

G¶

• function get_cpu() Returns the CPU instruction set selected for use at runtime.

H¶

• variable cpu_features::has3DNOW AMD 3DNow! instructions.
• variable cpu_features::has3DNOWEXT AMD 3DNow! extension instructions.
• variable cpu_features::hasABM AMD Advanced Bit Manipulation (LZCNT + POPCNT on AMD).
• variable cpu_features::hasADX ADX instruction set (arbitrary-precision add with carry).
• variable cpu_features::hasAES AES instruction set.
• variable cpu_features::hasAVX AVX instruction set (CPUID-reported, see hasAVXOSSUPPORT).
• variable cpu_features::hasAVX2 AVX2 instruction set.
• variable cpu_features::hasAVX512BW AVX-512 Byte and Word.
• variable cpu_features::hasAVX512CD AVX-512 Conflict Detection.
• variable cpu_features::hasAVX512DQ AVX-512 Doubleword and Quadword.
• variable cpu_features::hasAVX512ER AVX-512 Exponential and Reciprocal instructions.
• variable cpu_features::hasAVX512F AVX-512 Foundation.
• variable cpu_features::hasAVX512OSSUPPORT AVX-512 state enabled by the OS (XCR0[7:5] == 0b111).
• variable cpu_features::hasAVX512PF AVX-512 Prefetch.
• variable cpu_features::hasAVX512VL AVX-512 Vector Length extensions.
• variable cpu_features::hasAVXOSSUPPORT AVX state enabled by the OS (XCR0[2:1] == 0b11).
• variable cpu_features::hasBMI1 Bit Manipulation Instructions 1.
• variable cpu_features::hasBMI2 Bit Manipulation Instructions 2.
• variable cpu_features::hasCLFSH CLFLUSH instruction.
• variable cpu_features::hasCMOV CMOVcc conditional move.
• variable cpu_features::hasCMPXCHG16B CMPXCHG16B (128-bit compare-exchange).
• variable cpu_features::hasCX8 CMPXCHG8B (64-bit compare-exchange).
• variable cpu_features::hasERMS Enhanced REP MOVSB/STOSB.
• variable cpu_features::hasF16C F16C half-precision conversion.
• variable cpu_features::hasFMA Fused Multiply-Add (FMA3).
• variable cpu_features::hasFSGSBASE RDFSBASE/WRFSBASE instructions.
• variable cpu_features::hasFXSR FXSAVE/FXRSTOR.
• variable cpu_features::hasHLE Hardware Lock Elision (Intel only).
• variable cpu_features::hasINVPCID INVPCID instruction.
• variable cpu_features::hasLAHF LAHF/SAHF in 64-bit mode.
• variable cpu_features::hasLZCNT LZCNT instruction (Intel only).
• variable cpu_features::hasMMX MMX.
• variable cpu_features::hasMMXEXT AMD MMX extensions.
• variable cpu_features::hasMONITOR MONITOR/MWAIT.
• variable cpu_features::hasMOVBE MOVBE (move byte-swap).
• variable cpu_features::hasMSR Model-Specific Registers (RDMSR/WRMSR).
• variable cpu_features::hasOSXSAVE OS enables XSAVE/XRSTOR (CPUID.1:ECX[27]).
• variable cpu_features::hasPCLMULQDQ PCLMULQDQ carry-less multiplication.
• variable cpu_features::hasPOPCNT POPCNT instruction.
• variable cpu_features::hasPREFETCHWT1 PREFETCHWT1 (Intel Xeon Phi).
• variable cpu_features::hasRDRAND RDRAND random number.
• variable cpu_features::hasRDSEED RDSEED random seed.
• variable cpu_features::hasRDTSCP RDTSCP instruction (Intel only).
• variable cpu_features::hasRTM Restricted Transactional Memory (Intel only).
• variable cpu_features::hasSEP SYSENTER/SYSEXIT.
• variable cpu_features::hasSHA SHA instruction set.
• variable cpu_features::hasSSE SSE.
• variable cpu_features::hasSSE2 SSE2.
• variable cpu_features::hasSSE3 SSE3.
• variable cpu_features::hasSSE41 SSE4.1.
• variable cpu_features::hasSSE42 SSE4.2.
• variable cpu_features::hasSSE4a AMD SSE4a.
• variable cpu_features::hasSSSE3 SSSE3.
• variable cpu_features::hasSYSCALL SYSCALL/SYSRET (Intel only).
• variable cpu_features::hasTBM AMD Trailing Bit Manipulation.
• variable cpu_features::hasXOP AMD XOP.
• variable cpu_features::hasXSAVE XSAVE/XRSTOR.

I¶

• function timestamps<max_count, fence>::init() Reset the profiler, capturing the current cycle count as time zero.
• variable cpu_features::isAMD Vendor string is "AuthenticAMD".
• variable cpu_features::isIntel Vendor string is "GenuineIntel".

L¶

• variable timestamps<max_count, fence>::last Cycle count captured by init or the most recent record ; used as the reference point for the next duration measurement.

M¶

• variable cpu_features::max Highest supported standard CPUID leaf (leaf 0 EAX).
• function measure_rdtsc_cycle_time() Measures and returns the estimated cycle time in nanoseconds per tick for rdtsc.
• function timestamps<max_count, fence>::min(const timestamps<max_count, fence> &) Return a new timestamps object containing the per-slot minimum durations.
• variable cpu_features::model NUL-terminated CPUID model/brand string (leaves 0x80000002–4).
• variable timestamps<max_count, fence>::msg Per-checkpoint labels, parallel to dur .

O¶

• function override_cpu(cpu_t) Overrides the CPU instruction set used by KFR at runtime.

P¶

• variable cpu_features::padding1
• variable cpu_features::padding2

R¶

• function rdtsc() Reads the processor's cycle counter.
• variable timestamps<max_count, fence>::rdtsc_overhead
• function timestamps<max_count, fence>::record(const char *) Record a timestamp checkpoint with an optional label.

S¶

• variable stopwatch::start_time Tick count captured at construction.
• function steady_time() Returns the current time in nanoseconds since epoch using steady_clock.
• struct stopwatch RAII-style stopwatch measuring wall-clock elapsed time via clock_now .
• constructor stopwatch::stopwatch()

T¶

• class timestamps<max_count, fence> High-resolution scoped profiler using RDTSC (x86) or CNTVCT_EL0 (AArch64).

V¶

• variable cpu_features::vendor NUL-terminated CPUID vendor string (leaf 0).