This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Reloading and Spilling problem
- From: umar janjua <umarj at enabtech dot com>
- To: gcc-help at gcc dot gnu dot org, "gcc.gnu.org" <,gcc at gccc dot gnu dot org,cross-gcc>
- Date: Thu, 20 Jun 2002 11:38:42 +0500
- Subject: Reloading and Spilling problem
In an effort to port gcc to a dummy architecture with sixteen data
registers, we come across a case where gcc allocates all the data registers
and cannot find further data registers. At this point, gcc should spill one
of the data register to stack and proceed further. However , in our case the
gcc is unable to find the data register to spill. I have already defined
#define CLASS_LIKELY_SPILLED_P(class) (( class == DATA_REGS )? 1 :0
)
But it seems to be not working.
I have also defined reload_in and reload_out patterns that work fine
(define_expand "reload_inqi"
[(parallel[(match_operand:QI 0 "register_operand" "=a")
(match_operand:QI 1 "memory_operand" "m")
(match_operand:QI 2 "general_operand" "=&d")])]
""
"{
emit_insn(gen_movqi(operands[2],operands[1]));
emit_insn(gen_movqi(operands[0],operands[2]));
DONE;
}")
(define_expand "reload_outqi"
[(parallel[(match_operand:QI 0 "memory_operand" "=m")
(match_operand:QI 1 "register_operand" "a")
(match_operand:QI 2 "general_operand" "=&d")])]
""
"{
emit_insn(gen_movqi(operands[2],operands[1]));
emit_insn(gen_movqi(operands[0],operands[2]));
DONE;
}")
How do i tell gcc that it should spill data register on to the stack ?