https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99083
--- Comment #3 from Uroš Bizjak <ubizjak at gmail dot com> ---
It looks to me another one is in reload1.c, find_reg:
if (this_cost < best_cost
/* Among registers with equal cost, prefer caller-saved ones, or
use REG_ALLOC_ORDER if it is defined. */
|| (this_cost == best_cost
#ifdef REG_ALLOC_ORDER
&& (inv_reg_alloc_order[regno]
< inv_reg_alloc_order[best_reg])
#else
&& crtl->abi->clobbers_full_reg_p (regno)
&& !crtl->abi->clobbers_full_reg_p (best_reg)
#endif
))
{
best_reg = regno;
best_cost = this_cost;
}
According to the comment, REG_ALLOC_ORDER has to be defined to use preferences.
As mentioned by Richard in Comment #2, x86 defines ADJUST_REG_ALLOC_ORDER,
where the real allocation order is computed. But the documentation doesn't
mention that REG_ALLOC_ORDER also needs to be defined. It explicitly says even:
The macro body should not assume anything about the contents of
'reg_alloc_order' before execution of the macro.
But, we want to use the order from reg_alloc_order, so x86 should define
HONOR_REG_ALLOC_ORDER:
-- Macro: HONOR_REG_ALLOC_ORDER
Normally, IRA tries to estimate the costs for saving a register in
the prologue and restoring it in the epilogue. This discourages it
from using call-saved registers. If a machine wants to ensure that
IRA allocates registers in the order given by REG_ALLOC_ORDER even
if some call-saved registers appear earlier than call-used ones,
then define this macro as a C expression to nonzero. Default is 0.
But...
x86_order_regs_for_local_alloc lists general call_used_or_fixed_regs first, so
it should not matter anyway as far as call_used regs are concerned.
