R600/SI: Use scratch memory for large private arrays

[oota-llvm.git] / test / CodeGen / R600 / array-ptr-calc-i32.ll

diff --git a/test/CodeGen/R600/array-ptr-calc-i32.ll b/test/CodeGen/R600/array-ptr-calc-i32.ll
index 2ddc61f3972f6eeb389fe2109e41481dd111c0a1..a2b6978235196aaa7a601dad59220337687791a5 100644 (file)
--- a/test/CodeGen/R600/array-ptr-calc-i32.ll
+++ b/test/CodeGen/R600/array-ptr-calc-i32.ll
@@ -11,15 +11,18 @@ declare void @llvm.AMDGPU.barrier.local() nounwind noduplicate
 
 ; SI-LABEL: @test_private_array_ptr_calc:
 
-; SI: V_ADD_I32_e32 [[PTRREG:v[0-9]+]]
-
-; SI-ALLOCA: V_MOVRELD_B32_e32 {{v[0-9]+}}, [[PTRREG]]
+; FIXME: We end up with zero argument for ADD, because
+; SIRegisterInfo::eliminateFrameIndex() blindly replaces the frame index
+; with the appropriate offset.  We should fold this into the store.
+; SI-ALLOCA: V_ADD_I32_e32 [[PTRREG:v[0-9]+]], 0, v{{[0-9]+}}
+; SI-ALLOCA: BUFFER_STORE_DWORD {{v[0-9]+}}, s[{{[0-9]+:[0-9]+}}], [[PTRREG]]
 ;
 ; FIXME: The AMDGPUPromoteAlloca pass should be able to convert this
 ; alloca to a vector.  It currently fails because it does not know how
 ; to interpret:
 ; getelementptr [4 x i32]* %alloca, i32 1, i32 %b
 
+; SI-PROMOTE: V_ADD_I32_e32 [[PTRREG:v[0-9]+]]
 ; SI-PROMOTE: DS_WRITE_B32 {{v[0-9]+}}, [[PTRREG]]
 define void @test_private_array_ptr_calc(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %inA, i32 addrspace(1)* noalias %inB) {
   %alloca = alloca [4 x i32], i32 4, align 16