1 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI -mattr=-promote-alloca < %s | FileCheck -check-prefix=SI-ALLOCA -check-prefix=SI %s 2 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI -mattr=+promote-alloca < %s | FileCheck -check-prefix=SI-PROMOTE -check-prefix=SI %s 3 4 declare i32 @llvm.SI.tid() nounwind readnone 5 declare void @llvm.AMDGPU.barrier.local() nounwind convergent 6 7 ; The required pointer calculations for the alloca'd actually requires 8 ; an add and won't be folded into the addressing, which fails with a 9 ; 64-bit pointer add. This should work since private pointers should 10 ; be 32-bits. 11 12 ; SI-LABEL: {{^}}test_private_array_ptr_calc: 13 14 ; FIXME: We end up with zero argument for ADD, because 15 ; SIRegisterInfo::eliminateFrameIndex() blindly replaces the frame index 16 ; with the appropriate offset. We should fold this into the store. 17 ; SI-ALLOCA: v_add_i32_e32 [[PTRREG:v[0-9]+]], vcc, 0, v{{[0-9]+}} 18 ; SI-ALLOCA: buffer_store_dword {{v[0-9]+}}, [[PTRREG]], s[{{[0-9]+:[0-9]+}}] 19 ; 20 ; FIXME: The AMDGPUPromoteAlloca pass should be able to convert this 21 ; alloca to a vector. It currently fails because it does not know how 22 ; to interpret: 23 ; getelementptr [4 x i32], [4 x i32]* %alloca, i32 1, i32 %b 24 25 ; SI-PROMOTE: v_add_i32_e32 [[PTRREG:v[0-9]+]], vcc, 16 26 ; SI-PROMOTE: ds_write_b32 [[PTRREG]] 27 define void @test_private_array_ptr_calc(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %inA, i32 addrspace(1)* noalias %inB) { 28 %alloca = alloca [4 x i32], i32 4, align 16 29 %tid = call i32 @llvm.SI.tid() readnone 30 %a_ptr = getelementptr i32, i32 addrspace(1)* %inA, i32 %tid 31 %b_ptr = getelementptr i32, i32 addrspace(1)* %inB, i32 %tid 32 %a = load i32, i32 addrspace(1)* %a_ptr 33 %b = load i32, i32 addrspace(1)* %b_ptr 34 %result = add i32 %a, %b 35 %alloca_ptr = getelementptr [4 x i32], [4 x i32]* %alloca, i32 1, i32 %b 36 store i32 %result, i32* %alloca_ptr, align 4 37 ; Dummy call 38 call void @llvm.AMDGPU.barrier.local() nounwind convergent 39 %reload = load i32, i32* %alloca_ptr, align 4 40 %out_ptr = getelementptr i32, i32 addrspace(1)* %out, i32 %tid 41 store i32 %reload, i32 addrspace(1)* %out_ptr, align 4 42 ret void 43 } 44 45
