1 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
4 @local_memory.local_mem = internal addrspace(3) unnamed_addr global [16 x i32] zeroinitializer, align 4
8 ; Check that the LDS size emitted correctly
10 ; CHECK-NEXT: .long 16
14 ; GROUP_BARRIER must be the last instruction in a clause
15 ; CHECK: GROUP_BARRIER
16 ; CHECK-NEXT: ALU clause
20 define void @local_memory(i32 addrspace(1)* %out) {
22 %y.i = call i32 @llvm.r600.read.tidig.x() #0
23 %arrayidx = getelementptr inbounds [16 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %y.i
24 store i32 %y.i, i32 addrspace(3)* %arrayidx, align 4
25 %add = add nsw i32 %y.i, 1
26 %cmp = icmp eq i32 %add, 16
27 %.add = select i1 %cmp, i32 0, i32 %add
28 call void @llvm.AMDGPU.barrier.local()
29 %arrayidx1 = getelementptr inbounds [16 x i32] addrspace(3)* @local_memory.local_mem, i32 0, i32 %.add
30 %0 = load i32 addrspace(3)* %arrayidx1, align 4
31 %arrayidx2 = getelementptr inbounds i32 addrspace(1)* %out, i32 %y.i
32 store i32 %0, i32 addrspace(1)* %arrayidx2, align 4
36 @local_memory_two_objects.local_mem0 = internal addrspace(3) unnamed_addr global [4 x i32] zeroinitializer, align 4
37 @local_memory_two_objects.local_mem1 = internal addrspace(3) unnamed_addr global [4 x i32] zeroinitializer, align 4
39 ; CHECK: @local_memory_two_objects
41 ; Check that the LDS size emitted correctly
45 ; Make sure the lds writes are using different addresses.
46 ; CHECK: LDS_WRITE {{[*]*}} {{PV|T}}[[ADDRW:[0-9]*\.[XYZW]]]
47 ; CHECK-NOT: LDS_WRITE {{[*]*}} T[[ADDRW]]
49 ; GROUP_BARRIER must be the last instruction in a clause
50 ; CHECK: GROUP_BARRIER
51 ; CHECK-NEXT: ALU clause
53 ; Make sure the lds reads are using different addresses.
54 ; CHECK: LDS_READ_RET {{[*]*}} OQAP, {{PV|T}}[[ADDRR:[0-9]*\.[XYZW]]]
55 ; CHECK-NOT: LDS_READ_RET {{[*]*}} OQAP, T[[ADDRR]]
57 define void @local_memory_two_objects(i32 addrspace(1)* %out) {
59 %x.i = call i32 @llvm.r600.read.tidig.x() #0
60 %arrayidx = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem0, i32 0, i32 %x.i
61 store i32 %x.i, i32 addrspace(3)* %arrayidx, align 4
62 %mul = shl nsw i32 %x.i, 1
63 %arrayidx1 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem1, i32 0, i32 %x.i
64 store i32 %mul, i32 addrspace(3)* %arrayidx1, align 4
65 %sub = sub nsw i32 3, %x.i
66 call void @llvm.AMDGPU.barrier.local()
67 %arrayidx2 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem0, i32 0, i32 %sub
68 %0 = load i32 addrspace(3)* %arrayidx2, align 4
69 %arrayidx3 = getelementptr inbounds i32 addrspace(1)* %out, i32 %x.i
70 store i32 %0, i32 addrspace(1)* %arrayidx3, align 4
71 %arrayidx4 = getelementptr inbounds [4 x i32] addrspace(3)* @local_memory_two_objects.local_mem1, i32 0, i32 %sub
72 %1 = load i32 addrspace(3)* %arrayidx4, align 4
73 %add = add nsw i32 %x.i, 4
74 %arrayidx5 = getelementptr inbounds i32 addrspace(1)* %out, i32 %add
75 store i32 %1, i32 addrspace(1)* %arrayidx5, align 4
79 declare i32 @llvm.r600.read.tidig.x() #0
80 declare void @llvm.AMDGPU.barrier.local()
82 attributes #0 = { readnone }