Performance of hvm-c is unstable with some programs

[developedby]

Reproducing the behavior

This seems to be the underlying issue of HigherOrderCO/Bend#525.

For some programs, the C runtime can have wildly different performance on different executions.
This might be due to bad memory allocations, bad distribution of work among threads or something completely different.

I'm able to reproduce this issue with this bend program

def sum(depth, x):
  switch depth:
    case 0:
      return x
    case _:
      fst = sum(depth-1, x*2+0) # adds the fst half
      snd = sum(depth-1, x*2+1) # adds the snd half
      return fst + snd
    
def main:
  return sum(25, 0)

@imaqtkatt also tried this.
It seems that on ARM Macs the performance is a bit more consistent

System Settings

Example:

• OS: Linux (openSUSE Tumbleweed)
• CPU: AMD Ryzen 7 5800H
• GPU: [e.g. RTX 4090]
• HVM: Tested on 2.0.17 and 2.0.18

Additional context

No response