Hi all,
The generic rtx cost table was written with AArch32 cores in mind. It would be a
better idea to use the Cortex-A57 costs for the generic CPU in aarch64. That way
we schedule for the Cortex-A53 and do instruction selection for the Cortex-A57.
Since generic is the default CPU, this patch can change default code generation
but is highly unlikely to affect correctness (unless it exposes latent bugs)
since it only affects instruction selection.
Tested aarch64-none-elf on a model with no regressions.
Is it ok for trunk now or shall we wait for stage1?
Thanks,
Kyrill
2014-02-26 Kyrylo Tkachov <kyrylo.tkac...@arm.com>
* config/aarch64/aarch64.c (generic_tunings):
Use cortexa57_extra_costs.diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index ea90311..b5a2fe1 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -217,7 +217,7 @@ __extension__
#endif
static const struct tune_params generic_tunings =
{
- &generic_extra_costs,
+ &cortexa57_extra_costs,
&generic_addrcost_table,
&generic_regmove_cost,
&generic_vector_cost,
