[llvm-bugs] [Bug 50780] New: Spill in LL/SC loop when using fast register allocator

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=50780

            Bug ID: 50780
           Summary: Spill in LL/SC loop when using fast register allocator
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: PowerPC
          Assignee: unassignedbugs at nondot.org
          Reporter: lkail at cn.ibm.com
                CC: llvm-bugs at lists.llvm.org, nemanja.i.ibm at gmail.com

For llvm IR from pr25526
```
define {i64, i1} @foo(i64* %addr, i64 %desired, i64 %new) {
  %old = cmpxchg i64* %addr, i64 %desired, i64 %new seq_cst seq_cst
  ret {i64, i1} %old
}
```
Run with `llc -verify-machineinstrs -mtriple=powerpc64-ibm-aix-xcoff
-regalloc=fast < bug.ll |less`
We got spills in LL/SC loop
```
# %bb.0:
        std 5, -24(1)                           # 8-byte Folded Spill
        std 4, -16(1)                           # 8-byte Folded Spill
        std 3, -8(1)                            # 8-byte Folded Spill
        sync
L..BB0_1:                               # =>This Inner Loop Header: Depth=1
        ld 4, -8(1)                             # 8-byte Folded Reload
        ld 3, -16(1)                            # 8-byte Folded Reload
        ldarx 4, 0, 4
        cmpd    3, 4
        std 4, -32(1)                           # 8-byte Folded Spill
        bne     0, L..BB0_3
# %bb.2:                                #   in Loop: Header=BB0_1 Depth=1
        ld 3, -24(1)                            # 8-byte Folded Reload
        ld 4, -8(1)                             # 8-byte Folded Reload
        stdcx. 3, 0, 4
        bne     0, L..BB0_1
        b L..BB0_4
L..BB0_3:
        ld 3, -32(1)                            # 8-byte Folded Reload
        ld 4, -8(1)                             # 8-byte Folded Reload
        stdcx. 3, 0, 4
```
Though it hasn't caused infinite loop yet, the size of granule hasn't specified
in POWER ISA, we should avoid spilling considering implementation of POWER ISA
might vary.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210621/36340fd9/attachment.html>