Code is eliminated in @pstats

[giordano]

julia> using LinuxPerf

julia> function picalc(numsteps)
           slice = 1.0/numsteps
           sum = 0.0
           @simd for i = 1:numsteps
               x = (i - 0.5) * slice
               sum = sum + (4.0/(1.0 + x^2))
           end
           return sum * slice
       end
picalc (generic function with 1 method)

julia> @time picalc(10000000000)
  5.002960 seconds
3.1415926535897762

julia> @pstats picalc(10000000000)
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_frontend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_backend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╶ cpu-cycles               1.44e+02  100.0%  #  0.0 cycles per ns
┌ instructions             1.70e+01  100.0%  #  0.1 insns per cycle
│ branch-instructions      5.00e+00  100.0%  # 29.4% of insns
└ branch-misses            2.00e+00  100.0%  # 40.0% of branch insns
┌ task-clock               2.59e+05  100.0%  # 258.8 μs
│ context-switches         0.00e+00  100.0%
│ cpu-migrations           0.00e+00  100.0%
└ page-faults              0.00e+00  100.0%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

julia> @pstats picalc(10000000000000)
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_frontend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_backend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╶ cpu-cycles               1.57e+02  100.0%  #  0.0 cycles per ns
┌ instructions             1.70e+01  100.0%  #  0.1 insns per cycle
│ branch-instructions      5.00e+00  100.0%  # 29.4% of insns
└ branch-misses            2.00e+00  100.0%  # 40.0% of branch insns
┌ task-clock               3.09e+05  100.0%  # 309.2 μs
│ context-switches         0.00e+00  100.0%
│ cpu-migrations           0.00e+00  100.0%
└ page-faults              0.00e+00  100.0%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

julia> @pstats nothing
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_frontend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
┌ Warning: LinuxPerf.EventTypeExt(hw:stalled_cycles_backend, false, 0x0000000000000006) not supported, skipping
└ @ LinuxPerf ~/.julia/packages/LinuxPerf/396gD/src/LinuxPerf.jl:301
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
╶ cpu-cycles               2.17e+02  100.0%  #  0.0 cycles per ns
┌ instructions             1.70e+01  100.0%  #  0.1 insns per cycle
│ branch-instructions      5.00e+00  100.0%  # 29.4% of insns
└ branch-misses            2.00e+00  100.0%  # 40.0% of branch insns
┌ task-clock               3.47e+05  100.0%  # 347.5 μs
│ context-switches         0.00e+00  100.0%
│ cpu-migrations           0.00e+00  100.0%
└ page-faults              0.00e+00  100.0%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

julia> versioninfo()
Julia Version 1.11.2
Commit 5e9a32e7af2 (2024-12-01 20:02 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 22 × Intel(R) Core(TM) Ultra 7 155H
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, alderlake)
Threads: 1 default, 0 interactive, 1 GC (on 22 virtual cores)

(tmp) pkg> st
Status `/tmp/Project.toml`
  [b4c46c6c] LinuxPerf v0.4.0

picalc(10000000000000) should take 1.4 hours, but it takes ~300 μs inside @pstats, which is about the same time reported for nothing. I see there's some effort to try and avoid the code being eliminated

LinuxPerf.jl/src/LinuxPerf.jl

Lines 1144 to 1145 in eae2705

	# trick the compiler not to eliminate the code
	stats = rand() < 0 ? val : Stats(bench)

but it doesn't seem to be very effective anymore.

[Zentrik]

Looks like llvm is able to figure out that rand() >= 0!

[topolarity]

I think the ::Stats assertion on the next line is probably working against us, because it implies rand() >= 0

[Zentrik]

I would guess LLVM can just see this gives non-negative results https://github.com/JuliaLang/julia/blob/8f5b7ca12ad48c6d740e058312fc8cf2bbe67848/stdlib/Random/src/Xoshiro.jl#L305-L306. Looks like the early CSE pass figured it out https://godbolt.org/z/qnqxrqM9a.