https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119131
--- Comment #10 from Wilco <wilco at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #6) > I am trying to understand why there were checks for DECIMAL_FLOAT_MODE_P in > the first place. Removing them allows the testcase to pass. That was because the original code had an expander using GPF_HF which does not include decimal float. But now that is gone, the special cases for decimal float should be removed indeed to avoid that asymmetry. Then the only odd special case left is zero - interestingly it is needed at all in any pattern that already supports all movi immediates.
