https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91400
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Target| |x86_64-*-*, i?86-*-*
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-08-09
Component|tree-optimization |target
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. The issue is that the target builds a new(!) VAR_DECL for
__cpu_mode for _each_(!) invocation of __builtin_cpu_supports so the GIMPLE
level optimizers have no chance to CSE the loads. The two VAR_DECLs for
this testcase are
<var_decl 0x7ffff7fefb40 __cpu_model type <record_type 0x7ffff69c8888
__processor_model>
public static external weak preserve BLK (null):0:0
size <integer_cst 0x7ffff69ade10 constant 160>
unit-size <integer_cst 0x7ffff69ade40 constant 20>
align:32 warn_if_not_align:0>
<var_decl 0x7ffff7fefab0 __cpu_model type <record_type 0x7ffff69c8498
__processor_model>
public static external weak preserve BLK (null):0:0
size <integer_cst 0x7ffff69ade10 constant 160>
unit-size <integer_cst 0x7ffff69ade40 constant 20>
align:32 warn_if_not_align:0>
so you can see that we even re-build the RECORD_TYPE from scratch... :/
