To my knowledge, there are two ways to proceed to make "test-suite" newly-able to pass even when the current compiler configuration has FP contraction enabled:
#1: Test for FP contraction at run-time,
disable tests accordingly.
#2: Assume that FP contraction might be in effect,
require the "test-suite" user to specify that it isn`t when
it isn`t and the user wants the additional tests to be done.
The main advantages of strategy #1:
* it should be much more reliable than a user-specified flag
* it should enable more tests to be active when FP contraction
is _not_ in effect with less [no] user intervention
The main disadvantage of strategy #1 is that it seems likely to be difficult to implement, given the way the CMake+Lit system works, and given the fact that cross-building the tests is supported.
The main advantages of strategy #2:
* easy to implement
* already implemented and tested for the CMake+Lit system
* initial implementation for the old Makefile-based system exists [although not yet tested]
The main disadvantage of strategy #2 is that user intervention will be required in order to activate the sensitive tests that are disabled by default.
Feedback is welcome.