https://llvm.org/bugs/show_bug.cgi?id=28966
Bug ID: 28966
Summary: [AArch64] orr wzr vs. mov
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Backend: AArch64
Assignee: unassignedb...@nondot.org
Reporter: spatel+l...@rotateright.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
Forking this off bug 28965.
I can't imagine this would make any perf difference in practice, but it
suggests that there's some missing analysis/fold/pattern-matching because
functionally equivalent IR is producing different code:
$ cat cmpsel.ll
define i32 @goo(i32 %x) {
%cmp = icmp eq i32 %x, 1
%sel = select i1 %cmp, i32 1, i32 -1
ret i32 %sel
}
define i32 @hoo(i32 %x) {
%cmp = icmp ne i32 %x, 1 <--- inverse condition
%sel = select i1 %cmp, i32 -1, i32 1 <--- swap select operands
ret i32 %sel
}
$ ./llc -o - cmpsel.ll -mtriple=aarch64
goo: // @goo
cmp w0, #1 // =1
orr w8, wzr, #0x1
cneg w0, w8, ne
ret
hoo: // @hoo
cmp w0, #1 // =1
mov w8, #-1
cneg w0, w8, eq
ret
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
