[Python] How can I get LLVM loop vectorization debug output in Numba?

I'm trying to view LLVM debug messages for loop vectorization using Numba and llvmlite. I want to see the loop vectorization "LV:" debug output (e.g., messages like LV: Checking a loop in ...) so I can analyze the vectorization decisions made by LLVM. https://numba.readthedocs.io/en/stable/user/faq.html#does-numba-vectorize-array-computations-simd

I'm using a conda environment with the following YAML file. This makes sure I’m using the llvmlite build from the numba channel (which should have LLVM built with assertions enabled):

name: numbadev

channels:
- defaults
- numba

dependencies:
- python>=3.12.9
- numba::numba
- numba::llvmlite
- intel-cmplr-lib-rt


Running conda list shows:

# Name Version Build Channel
python 3.13.2 hf623796_100_cp313
llvmlite 0.44.0 py313h84b9e52_0 numba
numba 0.61.2 np2.1py3.13hf94e718_g1e70d8ceb_0 numba
...


This is my script debug_loop_vectorization.py:

import llvmlite.binding as llvm
llvm.set_option('', '--debug-only=loop-vectorize')

import numpy as np
from numba import jit

@jit(nopython=True, fastmath=True)
def test_func(a):
result = 0.0
for i in range(a.shape[0]):
result += a * 2.0
return result

# Trigger compilation.
a = np.arange(1000, dtype=np.float64)
test_func(a)

# Print the LLVM IR
print(test_func.inspect_llvm(test_func.signatures[0]))


When I run the script from a Linux terminal I see the LLVM-IR but no lines starting with "LV:...".

How can I get LLVM loop vectorization debug output?

Continue reading...