|0.2.1||Jun 20, 2021|
|0.2.0||Jun 20, 2021|
|0.1.1||Feb 4, 2021|
|0.1.0||Apr 30, 2018|
#386 in Unix APIs
40 downloads per month
Used in benchpmc
pmc-rs provides a safe abstraction for interacting with libpmc/hwpmc's
Performance Monitor Counters on FreeBSD.
PMCs are part of the CPU hardware and are typically used to profile CPU micro-architecture events such as L1/L2/L3 cache hits & misses, instructions processed per CPU tick, TLB lookups, branch mispredictions, etc for a particular application or algorithm. Using PMCs an algorithm can be tuned for performance by minimising CPU stalls, optimising CPU cache usage, and ensuring the CPU is always doing useful work.
The events are defined by the CPU manufacturer (here is the Intel 64 and IA-32
Architectures Developer's Manual: vol. 3B where the events can be
found in section
18.104.22.168 "Pre-defined Architectural Performance Events",
Table 18-1 "UMask and Event Select Encodings for Pre-Defined Architectural Performance Events").
The latest release of
pmc-rs generally targets the latest FreeBSD release. Due
to changes in
libpmc between FreeBSD versions, compatibility with older
FreeBSD versions may require pinning
pmc-rs to an older release.
Versions prior to FreeBSD 11 are untested, but may work.
FreeBSD 12 linked C++ into the libpmc stack causing compatibility issues with Rust (see this issue). Fortunately this is not the case with FreeBSD 13.
- Support sampling PMCs.
- Read counters with the
RDPMCinstruction to avoid context switching.