Skip to content

test: make the MGE bitwise sigma-ladder guards portable across CPUs #551

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Jammy2211 merged 1 commit into from
Aug 5, 2026
Merged

test: make the MGE bitwise sigma-ladder guards portable across CPUs #551

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 24 additions & 6 deletions test_autogalaxy/analysis/test_model_util.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,6 +157,14 @@ def test__mge_model_from__default_sigma_list_is_bitwise_unchanged():
`pytest.approx(rel=1e-8)` is deliberately NOT used: it only fails once the ladder
has moved by a relative ~1e-7, which is already past the point where the
identifier changes.

PORTABILITY TRAP -- the expected ladder is built element by element, exactly as
`mge_model_from` builds it (`10 ** log10_sigma_list[i]`), and must stay that way.
A vectorised `10 ** np.linspace(...)` is a DIFFERENT numpy code path: numpy does
not guarantee its scalar and SIMD power loops agree bit for bit, and on AVX-512
hardware they differ by 1 ULP, so writing the expectation vectorised makes this
test pass on GitHub's runners and fail on an AVX-512 developer machine. Comparing
like for like keeps the assertion exact without measuring the host's CPU.
"""
for mask_radius, total_gaussians in [(3.0, 20), (3.5, 30), (7.5, 10), (1.0, 5)]:
model = ag.model_util.mge_model_from(
Expand All @@ -166,9 +174,11 @@ def test__mge_model_from__default_sigma_list_is_bitwise_unchanged():
instance = model.instance_from_prior_medians()
sigma_list = [profile.sigma for profile in instance.profile_list]

assert sigma_list == list(
10 ** np.linspace(-4, np.log10(mask_radius), total_gaussians)
)
log10_sigma_list = np.linspace(-4, np.log10(mask_radius), total_gaussians)

assert sigma_list == [
10 ** log10_sigma_list[i] for i in range(total_gaussians)
]


def test__mge_point_model_from__returns_basis_model_with_correct_gaussians():
Expand Down Expand Up @@ -207,6 +217,12 @@ def test__mge_point_model_from__default_sigma_list_is_bitwise_unchanged():
As for `mge_model_from`, the default `sigma_min=0.01` must reproduce the
hardcoded `min_log10_sigma = -2.0` ladder that predates the argument EXACTLY,
so the identifier of an existing point-source fit does not change.

The same portability trap applies here: build the expected ladder element by
element, the way `mge_point_model_from` does, so both sides take numpy's scalar
power path. A vectorised `10 ** np.linspace(...)` disagrees with it by 1 ULP on
AVX-512 hardware -- see `test__mge_model_from__default_sigma_list_is_bitwise_unchanged`
for the full reasoning, including why `pytest.approx(rel=1e-8)` is not the fix.
"""
for pixel_scales, total_gaussians in [(0.1, 10), (0.05, 5), (0.2, 3), (0.001, 4)]:
model = ag.model_util.mge_point_model_from(
Expand All @@ -217,9 +233,11 @@ def test__mge_point_model_from__default_sigma_list_is_bitwise_unchanged():

max_sigma = max(2.0 * pixel_scales, 10**-2.0)

assert sigma_list == list(
10 ** np.linspace(-2.0, np.log10(max_sigma), total_gaussians)
)
log10_sigma_list = np.linspace(-2.0, np.log10(max_sigma), total_gaussians)

assert sigma_list == [
10 ** log10_sigma_list[i] for i in range(total_gaussians)
]


def test__mge_point_model_from__sigma_min_input_sets_smallest_gaussian():
Expand Down
Loading