https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92237
Bug ID: 92237
Summary: [x86] Missed optimisation opportunity with bit tests
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: andrew.cooper3 at citrix dot com
Target Milestone: ---
See https://godbolt.org/z/mP-8Y7
An expression such as:
bool foo(uint64_t val)
{
return (val & 0x120) == 0x20;
}
gets assembled to:
<foo>:
0: 81 e7 20 01 00 00 and $0x120,%edi
6: 48 83 ff 20 cmp $0x20,%rdi
a: 0f 94 c0 sete %al
d: c3 retq
Some part of optimisation has noticed that, due to the 32bit constant, the AND
can be performed on %edi, but hasn't spotted that the same is true for the
following CMP.
In this example, the CMP could use %edi as well, and save emitting the REX
prefix into the instruction stream.
