Deltablue crashes register allocator if YKD_OPT=0

[vext01]

Using yk a4488568e0:

===> 1: timeout 30 /home/vext01/research/yklua/src/lua harness.lua DeltaBlue 1 12000
Starting DeltaBlue benchmark ...

thread '<unnamed>' panicked at ykrt/src/compile/jitc_yk/codegen/x64/lsregalloc.rs:1009:22:
called `Option::unwrap()` on a `None` value
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

thread '<unnamed>' panicked at ykrt/src/compile/jitc_yk/codegen/x64/lsregalloc.rs:1009:22:
called `Option::unwrap()` on a `None` value

thread '<unnamed>' panicked at ykrt/src/compile/jitc_yk/codegen/x64/lsregalloc.rs:1009:22:
called `Option::unwrap()` on a `None` value
^[[ADeltaBlue: iterations=1 runtime: 2418808us
DeltaBlue: iterations=1 average: 2418808us total: 2418808us

Total Runtime: 2418808us
fatal runtime error: failed to initiate panic, error 5

[ltratt]

Does this happen deterministically?

[vext01]

Appears to be.

[ltratt]

How did this get through CI? I thought we tried the AWFY benchmarks in release mode?

[vext01]

We don't benchmark with the optimiser off. That's a requirement for this crash.

[vext01]

Here's a reduced test case for this one:

C = {}
do
	function C.A()
		getmetatable({})
	end
end
B = {}
do
	function B.A(A)
		for A = 0, A * 3 do
			C.A()
		end
	end
	function A(A)
		B.A(A)
	end
end
A(40000)

It crashes about 50% of the time.

[ltratt]

This fails 100% of the time for me (as I would expect for a register allocator bug).

[vext01]

It crashes every time if you serialise compilation, but not if you don't. Probably the program can end before the trace is compiled if you don't block on compilation.