https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82580
Bug ID: 82580
Summary: Optimize comparisons for __int128 on x86-64
Product: gcc
Version: 7.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: morwenn29 at hotmail dot fr
Target Milestone: ---
Given the following simple code:
bool foobar(unsigned __int128 lhs, unsigned __int128 rhs) {
return lhs < rhs;
}
GCC generates branchful code for x86-64 at -O3 optimization level:
foobar(unsigned __int128, unsigned __int128):
cmp rsi, rcx
mov eax, 1
jb .L2
jbe .L6
.L3:
xor eax, eax
.L2:
rep ret
.L6:
cmp rdi, rdx
jnb .L3
rep ret
On the other hand, Clang is able to generate branchless code with just a few
instructions at the same optimization level:
foobar(unsigned __int128, unsigned __int128): # @foobar(unsigned __int128,
unsigned __int128)
cmp rdi, rdx
sbb rsi, rcx
setb al
ret
The codegen results are equivalent for the other comparison instructions. Would
it be possible to optimize these comparison instructions the same way for GCC?
