Home | History | Annotate | Download | only in AMDGPU
      1 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
      2 ; XUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s
      3 
      4 ; FIXME: broken on VI because flat instructions need to be emitted
      5 ; instead of addr64 equivalent of the _OFFSET variants.
      6 
      7 ; Check that moving the pointer out of the resource descriptor to
      8 ; vaddr works for atomics.
      9 
     10 declare i32 @llvm.amdgcn.workitem.id.x() #1
     11 
     12 ; GCN-LABEL: {{^}}atomic_max_i32:
     13 ; GCN: buffer_atomic_smax v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:400 glc{{$}}
     14 define void @atomic_max_i32(i32 addrspace(1)* %out, i32 addrspace(1)* addrspace(1)* %in, i32 addrspace(1)* %x, i32 %y) #0 {
     15   %tid = call i32 @llvm.amdgcn.workitem.id.x()
     16   %tid.gep = getelementptr i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %in, i32 %tid
     17   %ptr = load volatile i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %tid.gep
     18   %xor = xor i32 %tid, 1
     19   %cmp = icmp ne i32 %xor, 0
     20   br i1 %cmp, label %atomic, label %exit
     21 
     22 atomic:
     23   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 100
     24   %ret = atomicrmw max i32 addrspace(1)* %gep, i32 %y seq_cst
     25   store i32 %ret, i32 addrspace(1)* %out
     26   br label %exit
     27 
     28 exit:
     29   ret void
     30 }
     31 
     32 ; GCN-LABEL: {{^}}atomic_max_i32_noret:
     33 ; GCN: buffer_atomic_smax v{{[0-9]+}}, v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:400{{$}}
     34 define void @atomic_max_i32_noret(i32 addrspace(1)* %out, i32 addrspace(1)* addrspace(1)* %in, i32 addrspace(1)* %x, i32 %y) #0 {
     35   %tid = call i32 @llvm.amdgcn.workitem.id.x()
     36   %tid.gep = getelementptr i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %in, i32 %tid
     37   %ptr = load volatile i32 addrspace(1)*, i32 addrspace(1)* addrspace(1)* %tid.gep
     38   %xor = xor i32 %tid, 1
     39   %cmp = icmp ne i32 %xor, 0
     40   br i1 %cmp, label %atomic, label %exit
     41 
     42 atomic:
     43   %gep = getelementptr i32, i32 addrspace(1)* %ptr, i32 100
     44   %ret = atomicrmw max i32 addrspace(1)* %gep, i32 %y seq_cst
     45   br label %exit
     46 
     47 exit:
     48   ret void
     49 }
     50 
     51 attributes #0 = { nounwind }
     52 attributes #1 = { nounwind readnone }
     53