| blob | a4d5fe4ffa5a758b2ddc013826630e272f3b4629 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -mem-intrinsic-expand-size=3 %s -o - | FileCheck -check-prefix=LOOP %s
3 ; RUN: llc -global-isel -mtriple=amdgcn-- -verify-machineinstrs -mem-intrinsic-expand-size=5 %s -o - | FileCheck -check-prefix=UNROLL %s
5 declare void @llvm.memmove.p1.p1.i32(ptr addrspace(1), ptr addrspace(1), i32, i1)
7 define amdgpu_cs void @memmove_p1i8(ptr addrspace(1) %dst, ptr 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[6:7], 0
16 ; LOOP-NEXT: s_mov_b32 s2, 0
17 ; LOOP-NEXT: s_mov_b32 s3, 0xf000
18 ; LOOP-NEXT: s_mov_b64 s[0:1], 0
19 ; LOOP-NEXT: v_mov_b32_e32 v4, s6
20 ; LOOP-NEXT: v_mov_b32_e32 v5, s7
21 ; LOOP-NEXT: .LBB0_2: ; %copy_forward_loop
22 ; LOOP-NEXT: ; =>This Inner Loop Header: Depth=1
23 ; LOOP-NEXT: v_add_i32_e32 v6, vcc, v2, v4
24 ; LOOP-NEXT: v_addc_u32_e32 v7, vcc, v3, v5, vcc
25 ; LOOP-NEXT: s_waitcnt expcnt(0)
26 ; LOOP-NEXT: buffer_load_ubyte v8, v[6:7], s[0:3], 0 addr64
27 ; LOOP-NEXT: v_add_i32_e32 v6, vcc, v0, v4
28 ; LOOP-NEXT: v_addc_u32_e32 v7, vcc, v1, v5, vcc
29 ; LOOP-NEXT: v_add_i32_e32 v4, vcc, 1, v4
30 ; LOOP-NEXT: v_addc_u32_e32 v5, vcc, 0, v5, vcc
31 ; LOOP-NEXT: v_cmp_ne_u32_e32 vcc, 4, v4
32 ; LOOP-NEXT: s_waitcnt vmcnt(0)
33 ; LOOP-NEXT: buffer_store_byte v8, v[6:7], s[0:3], 0 addr64
34 ; LOOP-NEXT: s_cbranch_vccnz .LBB0_2
35 ; LOOP-NEXT: .LBB0_3: ; %Flow17
36 ; LOOP-NEXT: s_andn2_saveexec_b64 s[0:1], s[4:5]
37 ; LOOP-NEXT: s_cbranch_execz .LBB0_6
38 ; LOOP-NEXT: ; %bb.4: ; %copy_backwards
39 ; LOOP-NEXT: s_mov_b32 s0, -4
40 ; LOOP-NEXT: v_add_i32_e32 v0, vcc, 3, v0
41 ; LOOP-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
42 ; LOOP-NEXT: v_add_i32_e32 v2, vcc, 3, v2
43 ; LOOP-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
44 ; LOOP-NEXT: s_mov_b32 s6, 0
45 ; LOOP-NEXT: s_mov_b32 s7, 0xf000
46 ; LOOP-NEXT: s_mov_b64 s[4:5], 0
47 ; LOOP-NEXT: v_mov_b32_e32 v4, s0
48 ; LOOP-NEXT: .LBB0_5: ; %copy_backwards_loop
49 ; LOOP-NEXT: ; =>This Inner Loop Header: Depth=1
50 ; LOOP-NEXT: s_waitcnt expcnt(0)
51 ; LOOP-NEXT: buffer_load_ubyte v5, v[2:3], s[4:7], 0 addr64
52 ; LOOP-NEXT: v_add_i32_e32 v4, vcc, 1, v4
53 ; LOOP-NEXT: s_and_b64 vcc, vcc, exec
54 ; LOOP-NEXT: s_waitcnt vmcnt(0)
55 ; LOOP-NEXT: buffer_store_byte v5, v[0:1], s[4:7], 0 addr64
56 ; LOOP-NEXT: v_add_i32_e64 v0, s[0:1], -1, v0
57 ; LOOP-NEXT: v_addc_u32_e64 v1, s[0:1], -1, v1, s[0:1]
58 ; LOOP-NEXT: v_add_i32_e64 v2, s[0:1], -1, v2
59 ; LOOP-NEXT: v_addc_u32_e64 v3, s[0:1], -1, v3, s[0:1]
60 ; LOOP-NEXT: s_cbranch_vccz .LBB0_5
61 ; LOOP-NEXT: .LBB0_6: ; %memmove_done
62 ; LOOP-NEXT: s_endpgm
63 ;
64 ; UNROLL-LABEL: memmove_p1i8:
65 ; UNROLL: ; %bb.0:
66 ; UNROLL-NEXT: s_mov_b32 s2, 0
67 ; UNROLL-NEXT: s_mov_b32 s3, 0xf000
68 ; UNROLL-NEXT: s_mov_b64 s[0:1], 0
69 ; UNROLL-NEXT: buffer_load_ubyte v4, v[2:3], s[0:3], 0 addr64
70 ; UNROLL-NEXT: buffer_load_ubyte v5, v[2:3], s[0:3], 0 addr64 offset:1
71 ; UNROLL-NEXT: buffer_load_ubyte v6, v[2:3], s[0:3], 0 addr64 offset:2
72 ; UNROLL-NEXT: buffer_load_ubyte v2, v[2:3], s[0:3], 0 addr64 offset:3
73 ; UNROLL-NEXT: s_waitcnt vmcnt(3)
74 ; UNROLL-NEXT: buffer_store_byte v4, v[0:1], s[0:3], 0 addr64
75 ; UNROLL-NEXT: s_waitcnt vmcnt(3)
76 ; UNROLL-NEXT: buffer_store_byte v5, v[0:1], s[0:3], 0 addr64 offset:1
77 ; UNROLL-NEXT: s_waitcnt vmcnt(3)
78 ; UNROLL-NEXT: buffer_store_byte v6, v[0:1], s[0:3], 0 addr64 offset:2
79 ; UNROLL-NEXT: s_waitcnt vmcnt(3)
80 ; UNROLL-NEXT: buffer_store_byte v2, v[0:1], s[0:3], 0 addr64 offset:3
81 ; UNROLL-NEXT: s_endpgm
82 call void @llvm.memmove.p1.p1.i32(ptr addrspace(1) %dst, ptr addrspace(1) %src, i32 4, i1 false)
83 ret void
84 }