Hi all,
This patch adds an -mverbose-cost-dump option, similar to the one in
aarch64.
It makes the RTX cost dump print the RTX we're costing in the backend,
as well as its cost.
This can be distracting in other cost-related RTL dumps like combine's.
So now we don't dump the backend information by default, but provide the
power-user option -mverbose-cost-dump
to enable the old verbose dumping.
This option is for GCC developers debugging the compiler only, so no
documentation are added.
It's also trivially simple in functionality so no test is added either.
Bootstrapped and tested on arm-none-linux-gnueabihf.
Committed to trunk.
Thanks,
Kyrill
2017-12-14 Kyrylo Tkachov <kyrylo.tkac...@arm.com>
* config/arm/arm.opt (mverbose-cost-dump): New option.
* config/arm/arm.c (arm_rtx_costs): Use it.
diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
index f34b6e0..9b29a0a 100644
--- a/gcc/config/arm/arm.c
+++ b/gcc/config/arm/arm.c
@@ -11068,7 +11068,7 @@ arm_rtx_costs (rtx x, machine_mode mode ATTRIBUTE_UNUSED, int outer_code,
current_tune->insn_extra_cost,
total, speed);
- if (dump_file && (dump_flags & TDF_DETAILS))
+ if (dump_file && arm_verbose_cost)
{
print_rtl_single (dump_file, x);
fprintf (dump_file, "\n%s cost: %d (%s)\n", speed ? "Hot" : "Cold",
diff --git a/gcc/config/arm/arm.opt b/gcc/config/arm/arm.opt
index 6060516..48f64588 100644
--- a/gcc/config/arm/arm.opt
+++ b/gcc/config/arm/arm.opt
@@ -254,6 +254,10 @@ mvectorize-with-neon-double
Target Report RejectNegative Mask(NEON_VECTORIZE_DOUBLE)
Use Neon double-word (rather than quad-word) registers for vectorization.
+mverbose-cost-dump
+Common Undocumented Var(arm_verbose_cost) Init(0)
+Enable more verbose RTX cost dumps during debug. For GCC developers use only.
+
mword-relocations
Target Report Var(target_word_relocations) Init(TARGET_DEFAULT_WORD_RELOCATIONS)
Only generate absolute relocations on word sized values.
