https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115752
--- Comment #3 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
I found some "interesting" thing in alpha.md:
;; Subregs suck for register allocation. Pretend we can move TFmode
;; data between general registers until after reload.
;; ??? Is this still true now that we have the lower-subreg pass?
(define_expand "movtf"
[(set (match_operand:TF 0 "nonimmediate_operand")
(match_operand:TF 1 "general_operand"))]
""
{
if (MEM_P (operands[0])
&& ! reg_or_0_operand (operands[1], TFmode))
operands[1] = force_reg (TFmode, operands[1]);
})
Do we need this? But it really sucks anyway.
