Script stuck when using GPU with bend-cu but not when using CPU with bend or bend-c

[2lian]

Reproducing the behavior

I have this code that can only execute when using bend or bend-c (or their compiled equivalent). When using the GPU with bend-cu (or their compiled equivalent) it get stuck and use half of my GPU memory.

def BASE(): # global variable
  return 10

def pow_binary(number, power): # power of an int
  if power < 1:
    return 1
  else:
    bend x=power:
      when x>1:
        left = (x / 2)
        right = x - left
        mult = fork(left) * fork(right)
      else:
        mult = number
    return mult

def get_digit(number, Nth_digit):
  return (number / pow_binary(BASE(0), Nth_digit - 1)) % BASE(0)

def validate_narc(number, Nd): # Check if the number is a Narcissistic number
  bend Nth_digit = Nd:
    when Nth_digit > 1:
      digit = get_digit(number, Nth_digit)
      power_sum = pow_binary(digit, Nd) + fork(Nth_digit - 1)
    else:
      power_sum = pow_binary(number % BASE(0), Nd)
  return power_sum == number

def main:
  return validate_arm(407, 3)

The following functions work fine: get_digit(407, 1), get_digit(407, 2), get_digit(407, 3), get_digit(407, 4), pow_binary(10, 0), pow_binary(10, 1), ... Only validate_arm creates the bug.

Even more strange is that I have tried replacing number by the value 407 (while using bend-cu) like so:

def validate_narc(number, Nd): # Check if the number is a Narcissistic number
  bend Nth_digit = Nd:
    when Nth_digit > 1:
      digit = get_digit(407, Nth_digit)
      power_sum = pow_binary(digit, Nd) + fork(Nth_digit - 1)
    else:
      power_sum = pow_binary(number % BASE(0), Nd)
  return power_sum == number

And this did not get stuck.

(Other numbers than 407 also do not work)

System Settings

• OS: WSL (Ubuntu 22.04)
• CPU: Intel i5-12600K
• GPU: GTX 1080Ti
• Cuda Version 12.3
• Updated version of bend+HVM 5 minutes ago

Additional context

No response