https://llvm.org/bugs/show_bug.cgi?id=28569
Bug ID: 28569
Summary: [IPRA] Failed to mark return address register as
clobbered
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Register Allocator
Assignee: unassignedb...@nondot.org
Reporter: zyfw...@gmail.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
=== ipra-call.c ===
__attribute__((noinline))
int foo2(int b)
{
return b + 2;
}
__attribute__((noinline))
int bar2(int b)
{
return b + foo2(b);
}
$ clang -target arm -Oz -emit-llvm -S ipra-call.c -o ipra-call.ll
$ llc -march=arm ipra-call.ll -enable-ipra=true -print-regusage
bar2 Clobbered Registers: SP R0 R1 R2 R3 R12 R12_SP R0_R1 R2_R3
foo2 Clobbered Registers: R0 R1 R0_R1
===== ipra-call.s =====
foo2: @ @foo2
add r0, r0, #2
mov pc, lr
bar2: @ @bar2
push {r11, lr}
mov r11, sp
mov r2, r0
bl foo2
add r0, r0, r2
pop {r11, lr}
mov pc, lr
I think bar2 should mark LR as Clobbered.
Current implementation thinks that a `call foo' instruction only clobbers
registers that clobbers by function `foo', but it's not true: `call foo' will
clobbers both registers clobbers by function `foo' and other physical registers
, in this case, LR.
--
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
