On 05/22/2015 09:39 AM, Richard Sandiford wrote:
Eric Botcazou <ebotca...@adacore.com> writes:
Some pieces of code create a temporary REG or MEM and only fill it
in later when they're testing the cost of a particular rtx. This patch
makes sure that even the dummy REG or MEM is valid, rather than force
the gen_* code to handle garbage values.
gcc/
* caller-save.c (init_caller_save): Use word_mode and
FIRST_PSEUDO_REGISTER when creating temporary rtxes.
* expr.c (init_expr_target): Likewise.
* ira.c (setup_prohibited_mode_move_regs): Likewise.
* postreload.c (reload_cse_regs_1): Likewise.
Isn't LAST_VIRTUAL_REGISTER + 1 the canonical regno to be used in this case?
Ah, yeah. Here's patch to fix all instances of that (I hope).
Bootstrapped & regression-tested on x86_64-linux-gnu. OK to install?
Thanks,
Richard
gcc/
* caller-save.c (init_caller_save): Base temporary register numbers
on LAST_VIRTUAL_REGISTER + 1 rather than FIRST_PSEUDO_REGISTER.
* cfgloopanal.c (init_set_costs): Likewise.
* dojump.c (prefer_and_bit_test): Likewise.
* expr.c (init_expr_target): Likewise.
* ira.c (setup_prohibited_mode_move_regs): Likewise.
* lower-subreg.c (init_lower_subreg): Likewise.
* postreload.c (reload_cse_regs_1): Likewise.
OK.
jeff
