https://bugs.llvm.org/show_bug.cgi?id=36689
Bug ID: 36689
Summary: Failure to zextload due to unused bits
Product: libraries
Version: trunk
Hardware: PC
OS: Windows NT
Status: NEW
Severity: enhancement
Priority: P
Component: Backend: X86
Assignee: unassignedb...@nondot.org
Reporter: llvm-...@redking.me.uk
CC: llvm-bugs@lists.llvm.org, spatel+l...@rotateright.com
We correctly convert to a zextload for a basic load + mask, but this fails if
we set any extra bits, which I assume is due to SimplifyDemandedBits getting in
the way:
clang -g0 -O3 -march=btver2
int masklo(int *p) {
return (*p & 65535);
}
int masklo_set(int *p) {
return (*p & 65535) | 255;
}
define i32 @_Z6maskloPi(i32*) {
%2 = load i32, i32* %0
%3 = and i32 %2, 65535
ret i32 %3
}
define i32 @_Z10masklo_setPi(i32*) {
%2 = load i32, i32* %0
%3 = and i32 %2, 65280
%4 = or i32 %3, 255
ret i32 %4
}
_Z6maskloPi:
movzwl (%rdi), %eax
retq
_Z10masklo_setPi:
movl $65280, %eax # imm = 0xFF00
andl (%rdi), %eax
orl $255, %eax
retq
Ideally this should have been:
_Z10masklo_setPi:
movzwl (%rdi), %eax
orl $255, %eax
retq
--
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