JIT/jit64/opt/rngchk/RngchkStress2_o/RngchkStress2_o.dll is timing out

[kunalspathak]

export DOTNET_TieredCompilation=0
export DOTNET_JitStress=f5

Observed in osx_x64 as well as linux_arm64

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[amanasifkhalid]

cc @dotnet/jit-contrib

[AndyAyersMS]

Looks like this get stuck in 3-opt:

****** START compiling SimpleArray_01.Test:Test1() (MethodHash=31819848)
Generating code for Unix arm64
OPTIONS: compCodeOpt = BLENDED_CODE
OPTIONS: compDbgCode = false
OPTIONS: compDbgInfo = true
...
Swapping partitions [BB965, BB965] and [BB966, BB964] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB966, BB966] and [BB967, BB965] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB967, BB964] and [BB965, BB966] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB965, BB965] and [BB966, BB964] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB966, BB966] and [BB967, BB965] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB967, BB964] and [BB965, BB966] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB965, BB965] and [BB966, BB964] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)
Swapping partitions [BB966, BB966] and [BB967, BB965] (cost change = -12259964326927110866866776217202473468949912977468817408.000000)

[AndyAyersMS]

@amanasifkhalid want to take a look?

[AndyAyersMS]

Guessing STRESS_BB_PROFILE may be the culprit?

andya@AndyBR3:~/bugs/r111988$ grep STRESS jit.d
*** JitStress: STRESS_GENERIC_VARN ***
*** JitStress: STRESS_BB_PROFILE ***
*** JitStress: STRESS_CLONE_EXPR ***
*** JitStress: STRESS_FOLD ***
*** JitStress: STRESS_REVERSE_FLAG ***
*** JitStress: STRESS_DO_WHILE_LOOPS ***
*** JitStress: STRESS_MAKE_CSE ***
*** JitStress: STRESS_DOWNWARDS_COUNTED_LOOPS ***
*** JitStress: STRESS_SPLIT_TREES_RANDOMLY ***

[amanasifkhalid]

@amanasifkhalid want to take a look?

Sure, I'll take this.

[amanasifkhalid]

Guessing STRESS_BB_PROFILE may be the culprit?

This does seem to be the problem, in this case. The block weights are so unwieldy that I wouldn't expect 3-opt to converge to a local minimum due to floating-point imprecision, and if it isn't actually oscillating in this case, there are still over a thousand basic blocks in the method to reason about. Capping the number of swaps allowed to something large like 1000 doesn't incur any diffs locally, and I wouldn't expect it to kick in under non-stress scenarios unless the profile is hopelessly corrupted.