Hi,
The fix for PR48250 overhauled arm_legitimize_reload_address taking
into account Thumb2 as well in the code , we forgot to call it for
TARGET_THUMB2 :) . I noticed this while reviewing an earlier patch and
today got reminded about this while looking at something else in the
backend. While the existing behaviour is safe for ARM state code and
has indeed been living in the compiler for over a year now, it is
possibly sub-optimal in some cases for TARGET_THUMB2 and in general
for `coproc_p' based loads and stores since we do the simple thing
with reload.
I want to queue this up for the next stage1 if testing and
benchmarking succeed. Thoughts / comments ?
cheers
Ramana
2011-12-14 Ramana Radhakrishnan <ramana.radhakrish...@linaro.org>
* config/arm/arm.h (LEGITIMIZE_RELOAD_ADDRESS): Call
arm_legitimize_reload_address for Thumb2 as well. Adjust call
to thumb_legitimize_reload_address.
* config/arm/arm-protos.h (thumb1_legitimize_reload_address): Renamed
from thumb_legitimize_reload_address.
