1 ; RUN: llc -verify-machineinstrs -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI %s
3 declare i32 @llvm.SI.tid() nounwind readnone
4 declare void @llvm.AMDGPU.barrier.local() nounwind noduplicate
6 ; The required pointer calculations for the alloca'd actually requires
7 ; an add and won't be folded into the addressing, which fails with a
8 ; 64-bit pointer add. This should work since private pointers should
11 ; SI-LABEL: @test_private_array_ptr_calc:
12 ; SI: V_ADD_I32_e32 [[PTRREG:v[0-9]+]]
14 ; FIXME: The AMDGPUPromoteAlloca pass should be able to convert this
15 ; alloca to a vector. It currently fails because it does not know how
17 ; getelementptr [4 x i32]* %alloca, i32 1, i32 %b
18 ; SI: DS_WRITE_B32 {{v[0-9]+}}, [[PTRREG]]
19 define void @test_private_array_ptr_calc(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %inA, i32 addrspace(1)* noalias %inB) {
20 %alloca = alloca [4 x i32], i32 4, align 16
21 %tid = call i32 @llvm.SI.tid() readnone
22 %a_ptr = getelementptr i32 addrspace(1)* %inA, i32 %tid
23 %b_ptr = getelementptr i32 addrspace(1)* %inB, i32 %tid
24 %a = load i32 addrspace(1)* %a_ptr
25 %b = load i32 addrspace(1)* %b_ptr
26 %result = add i32 %a, %b
27 %alloca_ptr = getelementptr [4 x i32]* %alloca, i32 1, i32 %b
28 store i32 %result, i32* %alloca_ptr, align 4
30 call void @llvm.AMDGPU.barrier.local() nounwind noduplicate
31 %reload = load i32* %alloca_ptr, align 4
32 %out_ptr = getelementptr i32 addrspace(1)* %out, i32 %tid
33 store i32 %reload, i32 addrspace(1)* %out_ptr, align 4