1 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
3 ; This test checks that uses and defs of the AR register happen in the same
6 ; CHECK: @mova_same_clause
8 ; CHECK-NOT: ALU clause
11 ; CHECK-NOT: ALU clause
14 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {
16 %stack = alloca [5 x i32], align 4
17 %0 = load i32 addrspace(1)* %in, align 4
18 %arrayidx1 = getelementptr inbounds [5 x i32]* %stack, i32 0, i32 %0
19 store i32 4, i32* %arrayidx1, align 4
20 %arrayidx2 = getelementptr inbounds i32 addrspace(1)* %in, i32 1
21 %1 = load i32 addrspace(1)* %arrayidx2, align 4
22 %arrayidx3 = getelementptr inbounds [5 x i32]* %stack, i32 0, i32 %1
23 store i32 5, i32* %arrayidx3, align 4
24 %arrayidx10 = getelementptr inbounds [5 x i32]* %stack, i32 0, i32 0
25 %2 = load i32* %arrayidx10, align 4
26 store i32 %2, i32 addrspace(1)* %out, align 4
27 %arrayidx12 = getelementptr inbounds [5 x i32]* %stack, i32 0, i32 1
28 %3 = load i32* %arrayidx12
29 %arrayidx13 = getelementptr inbounds i32 addrspace(1)* %out, i32 1
30 store i32 %3, i32 addrspace(1)* %arrayidx13
34 ; This test checks that the stack offset is calculated correctly for structs.
35 ; All register loads/stores should be optimized away, so there shouldn't be
36 ; any MOVA instructions.
38 ; XXX: This generated code has unnecessary MOVs, we should be able to optimize
41 ; CHECK: @multiple_structs
44 %struct.point = type { i32, i32 }
46 define void @multiple_structs(i32 addrspace(1)* %out) {
48 %a = alloca %struct.point
49 %b = alloca %struct.point
50 %a.x.ptr = getelementptr %struct.point* %a, i32 0, i32 0
51 %a.y.ptr = getelementptr %struct.point* %a, i32 0, i32 1
52 %b.x.ptr = getelementptr %struct.point* %b, i32 0, i32 0
53 %b.y.ptr = getelementptr %struct.point* %b, i32 0, i32 1
54 store i32 0, i32* %a.x.ptr
55 store i32 1, i32* %a.y.ptr
56 store i32 2, i32* %b.x.ptr
57 store i32 3, i32* %b.y.ptr
58 %a.indirect.ptr = getelementptr %struct.point* %a, i32 0, i32 0
59 %b.indirect.ptr = getelementptr %struct.point* %b, i32 0, i32 0
60 %a.indirect = load i32* %a.indirect.ptr
61 %b.indirect = load i32* %b.indirect.ptr
62 %0 = add i32 %a.indirect, %b.indirect
63 store i32 %0, i32 addrspace(1)* %out
67 ; Test direct access of a private array inside a loop. The private array
68 ; loads and stores should be lowered to copies, so there shouldn't be any
74 define void @direct_loop(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
76 %prv_array_const = alloca [2 x i32]
77 %prv_array = alloca [2 x i32]
78 %a = load i32 addrspace(1)* %in
79 %b_src_ptr = getelementptr i32 addrspace(1)* %in, i32 1
80 %b = load i32 addrspace(1)* %b_src_ptr
81 %a_dst_ptr = getelementptr [2 x i32]* %prv_array_const, i32 0, i32 0
82 store i32 %a, i32* %a_dst_ptr
83 %b_dst_ptr = getelementptr [2 x i32]* %prv_array_const, i32 0, i32 1
84 store i32 %b, i32* %b_dst_ptr
88 %inc = phi i32 [0, %entry], [%count, %for.body]
89 %x_ptr = getelementptr [2 x i32]* %prv_array_const, i32 0, i32 0
91 %y_ptr = getelementptr [2 x i32]* %prv_array, i32 0, i32 0
94 store i32 %xy, i32* %y_ptr
95 %count = add i32 %inc, 1
96 %done = icmp eq i32 %count, 4095
97 br i1 %done, label %for.end, label %for.body
100 %value_ptr = getelementptr [2 x i32]* %prv_array, i32 0, i32 0
101 %value = load i32* %value_ptr
102 store i32 %value, i32 addrspace(1)* %out