1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -global-isel -march=amdgcn -verify-machineinstrs -amdgpu-mem-intrinsic-expand-size=3 %s -o - | FileCheck -check-prefix=LOOP %s
3; RUN: llc -global-isel -march=amdgcn -verify-machineinstrs -amdgpu-mem-intrinsic-expand-size=5 %s -o - | FileCheck -check-prefix=UNROLL %s
4
5declare void @llvm.memmove.p1i8.p1i8.i32(i8 addrspace(1)*, i8 addrspace(1)*, i32, i1)
6
7define amdgpu_cs void @memmove_p1i8(i8 addrspace(1)* %dst, i8 addrspace(1)* %src) {
8; LOOP-LABEL: memmove_p1i8:
9; LOOP:       ; %bb.0:
10; LOOP-NEXT:    v_cmp_ge_u64_e32 vcc, v[2:3], v[0:1]
11; LOOP-NEXT:    s_and_saveexec_b64 s[0:1], vcc
12; LOOP-NEXT:    s_xor_b64 s[4:5], exec, s[0:1]
13; LOOP-NEXT:    s_cbranch_execz .LBB0_3
14; LOOP-NEXT:  ; %bb.1: ; %copy_forward
15; LOOP-NEXT:    s_mov_b64 s[0:1], 0
16; LOOP-NEXT:    s_mov_b32 s2, 0
17; LOOP-NEXT:    s_mov_b32 s3, 0xf000
18; LOOP-NEXT:    v_mov_b32_e32 v5, s1
19; LOOP-NEXT:    v_mov_b32_e32 v4, s0
20; LOOP-NEXT:  .LBB0_2: ; %copy_forward_loop
21; LOOP-NEXT:    ; =>This Inner Loop Header: Depth=1
22; LOOP-NEXT:    v_add_i32_e32 v6, vcc, v2, v4
23; LOOP-NEXT:    v_addc_u32_e32 v7, vcc, v3, v5, vcc
24; LOOP-NEXT:    s_waitcnt expcnt(0)
25; LOOP-NEXT:    buffer_load_ubyte v8, v[6:7], s[0:3], 0 addr64
26; LOOP-NEXT:    v_add_i32_e32 v6, vcc, v0, v4
27; LOOP-NEXT:    v_addc_u32_e32 v7, vcc, v1, v5, vcc
28; LOOP-NEXT:    v_add_i32_e32 v4, vcc, 1, v4
29; LOOP-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
30; LOOP-NEXT:    v_cmp_ne_u32_e32 vcc, 4, v4
31; LOOP-NEXT:    s_waitcnt vmcnt(0)
32; LOOP-NEXT:    buffer_store_byte v8, v[6:7], s[0:3], 0 addr64
33; LOOP-NEXT:    s_cbranch_vccnz .LBB0_2
34; LOOP-NEXT:  .LBB0_3: ; %Flow14
35; LOOP-NEXT:    s_andn2_saveexec_b64 s[0:1], s[4:5]
36; LOOP-NEXT:    s_cbranch_execz .LBB0_6
37; LOOP-NEXT:  ; %bb.4: ; %copy_backwards
38; LOOP-NEXT:    s_mov_b64 s[4:5], 3
39; LOOP-NEXT:    s_mov_b32 s2, 0
40; LOOP-NEXT:    s_mov_b32 s3, 0xf000
41; LOOP-NEXT:    s_mov_b64 s[0:1], 0
42; LOOP-NEXT:    v_mov_b32_e32 v4, s4
43; LOOP-NEXT:    v_mov_b32_e32 v5, s5
44; LOOP-NEXT:  .LBB0_5: ; %copy_backwards_loop
45; LOOP-NEXT:    ; =>This Inner Loop Header: Depth=1
46; LOOP-NEXT:    v_add_i32_e32 v6, vcc, v2, v4
47; LOOP-NEXT:    v_addc_u32_e32 v7, vcc, v3, v5, vcc
48; LOOP-NEXT:    s_waitcnt expcnt(0)
49; LOOP-NEXT:    buffer_load_ubyte v8, v[6:7], s[0:3], 0 addr64
50; LOOP-NEXT:    v_add_i32_e32 v6, vcc, v0, v4
51; LOOP-NEXT:    v_addc_u32_e32 v7, vcc, v1, v5, vcc
52; LOOP-NEXT:    v_add_i32_e32 v4, vcc, -1, v4
53; LOOP-NEXT:    v_addc_u32_e32 v5, vcc, -1, v5, vcc
54; LOOP-NEXT:    v_cmp_eq_u32_e32 vcc, -1, v4
55; LOOP-NEXT:    s_waitcnt vmcnt(0)
56; LOOP-NEXT:    buffer_store_byte v8, v[6:7], s[0:3], 0 addr64
57; LOOP-NEXT:    s_cbranch_vccz .LBB0_5
58; LOOP-NEXT:  .LBB0_6: ; %memmove_done
59; LOOP-NEXT:    s_endpgm
60;
61; UNROLL-LABEL: memmove_p1i8:
62; UNROLL:       ; %bb.0:
63; UNROLL-NEXT:    s_mov_b32 s2, 0
64; UNROLL-NEXT:    s_mov_b32 s3, 0xf000
65; UNROLL-NEXT:    s_mov_b64 s[0:1], 0
66; UNROLL-NEXT:    buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64
67; UNROLL-NEXT:    buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:1
68; UNROLL-NEXT:    buffer_load_ubyte v6, v[2:3], s[0:3], 0 addr64 offset:2
69; UNROLL-NEXT:    buffer_load_ubyte v2, v[2:3], s[0:3], 0 addr64 offset:3
70; UNROLL-NEXT:    s_waitcnt vmcnt(3)
71; UNROLL-NEXT:    buffer_store_byte v4, v[0:1], s[0:3], 0 addr64
72; UNROLL-NEXT:    s_waitcnt vmcnt(3)
73; UNROLL-NEXT:    buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:1
74; UNROLL-NEXT:    s_waitcnt vmcnt(3)
75; UNROLL-NEXT:    buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:2
76; UNROLL-NEXT:    s_waitcnt vmcnt(3)
77; UNROLL-NEXT:    buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:3
78; UNROLL-NEXT:    s_endpgm
79  call void @llvm.memmove.p1i8.p1i8.i32(i8 addrspace(1)* %dst, i8 addrspace(1)* %src, i32 4, i1 false)
80  ret void
81}
82