The order of floating-point operations led to different results between these debug and release builds on ARM64.
This was probably due to different optimizations done by the compiler.
I changed the floating-point precision of all tests to double, which solved the failed transformation tests.
As a side task, I also had to add a precision argument to all Point cloud save functions to keep IO utests from failing.
Description
The order of floating-point operations led to different results between these
debug
andrelease
builds on ARM64. This was probably due to different optimizations done by the compiler. I changed the floating-point precision of all tests todouble
, which solved the failed transformation tests. As a side task, I also had to add a precision argument to all Point cloud save functions to keep IO utests from failing.Summary:
Fixes https://github.com/norlab-ulaval/libpointmatcher/issues/556
Changes and type of changes (quick overview):
double
Checklist:
Code related
PR creation related
base ref
branch is set to thedevelop
branch (the build-system won't be triggered otherwise)develop
branch (the build-system will reject it otherwise)PR description related
# <issue-id>
if changes are of typefix
feat
fix
docs
style
refactor
perf
test
build
ci
chore
revert
<type>!