mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-12-14 11:32:34 +00:00
dd3383fd09
- Fix operand order for atomic sub, where the minuend is the value loaded from memory and the subtrahend is the parameter specified. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167718 91177308-0d34-0410-b5e6-96231b3b80d8
23 lines
585 B
LLVM
23 lines
585 B
LLVM
; RUN: llc < %s -mtriple=i386-apple-darwin -mcpu=corei7 | FileCheck %s
|
|
; rdar://r7512579
|
|
|
|
; PHI defs in the atomic loop should be used by the add / adc
|
|
; instructions. They should not be dead.
|
|
|
|
define void @t(i64* nocapture %p) nounwind ssp {
|
|
entry:
|
|
; CHECK: t:
|
|
; CHECK: movl ([[REG:%[a-z]+]]), %eax
|
|
; CHECK: movl 4([[REG]]), %edx
|
|
; CHECK: LBB0_1:
|
|
; CHECK: movl %eax, %ebx
|
|
; CHECK: addl {{%[a-z]+}}, %ebx
|
|
; CHECK: movl %edx, %ecx
|
|
; CHECK: adcl {{%[a-z]+}}, %ecx
|
|
; CHECK: lock
|
|
; CHECK-NEXT: cmpxchg8b ([[REG]])
|
|
; CHECK-NEXT: jne
|
|
%0 = atomicrmw add i64* %p, i64 1 seq_cst
|
|
ret void
|
|
}
|