Change in specialisation(?) behaviour since 8.0.2 causes 6x slowdown

I was investigating some benchmarks and I noticed some odd results if I duplicated one of my tests.

Reproduction: https://github.com/mpickering/probable-eureka

Looking at the core, it seems that repeating the definition means that one of the key functions doesn't get specialised as expected which leads to a much slower program.

Observe that in the first two benchmarks there is a worker function go :: Int# -> Int -> ReaderT Int (StateT Int Identity Int but in the third benchmark this is specialised to $sgo :: Int# -> Int -> Int -> Int# -> Int# -> Identity (Int, Int). Removing the duplicate benchmark means that specialisation happens properly in the first case as well.

The proper specialisation also happens in 8.0.2.

This causes the first two cases to be 6x slower than the last case.

Trac metadata

Trac field	Value
Version	8.2.1-rc2
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture