LinuxPerf.jl
LinuxPerf.jl copied to clipboard
JuliaPerf
Can we add an option for normalizing the results with a constant that the user passes? To eliminate noise, we often end up running the experiment multiple times, e.g., run...
I'd love to be able to use annotations to find where in my source code the cache misses happen. Is that currently possible? If not, would you consider it as...
The `SystemError` ``` ERROR: SystemError: ioctl: Bad file descriptor Stacktrace: [1] systemerror(p::Symbol, errno::Int32; extrainfo::Nothing) @ Base ./error.jl:168 [2] #systemerror#62 @ ./error.jl:167 [inlined] [3] systemerror @ ./error.jl:167 [inlined] [4] ioctl @...
It looks to me that syscall numbers differ on different platforms, so this code is only going to work on x86_64 as https://github.com/torvalds/linux/blob/v4.17/arch/x86/entry/syscalls/syscall_32.tbl#L17 says `perf_event_open` is 336 on i686 https://github.com/JuliaPerf/LinuxPerf.jl/blob/b7653aa80ce8e53c101157b592b80c6d2100c87e/src/LinuxPerf.jl#L158-L168...
prctl will enable or disable all benches, so you need to close them after you're done with them otherwise you'll quickly have too many benches and you won't get any...
`fast_prctl` compiles down to just a single llvm call on x86_64 for me.
It's not uncommon for me to get a reading like this: ```julia julia> @pstats "cpu-cycles,instructions,branch-instructions,branch-misses,cache-misses,cache-references" rand(1000,1000) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ╶ cpu-cycles 8.60e+04 62.9% # 0.0 cycles per ns ╶ instructions 7.89e+06 100.0%...
```julia julia> bench = LinuxPerf.make_bench(); julia> enable!(bench); sleep(1.0); disable!(bench); julia> stats = LinuxPerf.ThreadStats(bench); julia> stats.groups[1][1] LinuxPerf.Counter(hw:cycles, 0x0000000001bdc0f6, 0x0000000000a08a80, 0x00000000007a6ed7) julia> reset!(bench) julia> stats = LinuxPerf.ThreadStats(bench); julia> stats.groups[1][1] LinuxPerf.Counter(hw:cycles, 0x0000000000000000, 0x0000000003dad501,...