1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -mtriple=amdgcn-- -mcpu=pitcairn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
4 define amdgpu_kernel void @zext_shl64_to_32(ptr addrspace(1) nocapture %out, i32 %x) {
5 ; GCN-LABEL: zext_shl64_to_32:
7 ; GCN-NEXT: s_load_dword s4, s[2:3], 0xb
8 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
9 ; GCN-NEXT: s_mov_b32 s3, 0xf000
10 ; GCN-NEXT: s_mov_b32 s2, -1
11 ; GCN-NEXT: v_mov_b32_e32 v1, 0
12 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
13 ; GCN-NEXT: s_lshl_b32 s4, s4, 2
14 ; GCN-NEXT: v_mov_b32_e32 v0, s4
15 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
17 %and = and i32 %x, 1073741823
18 %ext = zext i32 %and to i64
19 %shl = shl i64 %ext, 2
20 store i64 %shl, ptr addrspace(1) %out, align 4
24 define amdgpu_kernel void @sext_shl64_to_32(ptr addrspace(1) nocapture %out, i32 %x) {
25 ; GCN-LABEL: sext_shl64_to_32:
27 ; GCN-NEXT: s_load_dword s4, s[2:3], 0xb
28 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
29 ; GCN-NEXT: s_mov_b32 s3, 0xf000
30 ; GCN-NEXT: s_mov_b32 s2, -1
31 ; GCN-NEXT: v_mov_b32_e32 v1, 0
32 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
33 ; GCN-NEXT: s_and_b32 s4, s4, 0x1fffffff
34 ; GCN-NEXT: s_lshl_b32 s4, s4, 2
35 ; GCN-NEXT: v_mov_b32_e32 v0, s4
36 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
38 %and = and i32 %x, 536870911
39 %ext = sext i32 %and to i64
40 %shl = shl i64 %ext, 2
41 store i64 %shl, ptr addrspace(1) %out, align 4
45 define amdgpu_kernel void @zext_shl64_overflow(ptr addrspace(1) nocapture %out, i32 %x) {
46 ; GCN-LABEL: zext_shl64_overflow:
48 ; GCN-NEXT: s_load_dword s4, s[2:3], 0xb
49 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
50 ; GCN-NEXT: s_mov_b32 s5, 0
51 ; GCN-NEXT: s_mov_b32 s3, 0xf000
52 ; GCN-NEXT: s_mov_b32 s2, -1
53 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
54 ; GCN-NEXT: s_bitset0_b32 s4, 31
55 ; GCN-NEXT: s_lshl_b64 s[4:5], s[4:5], 2
56 ; GCN-NEXT: v_mov_b32_e32 v0, s4
57 ; GCN-NEXT: v_mov_b32_e32 v1, s5
58 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
60 %and = and i32 %x, 2147483647
61 %ext = zext i32 %and to i64
62 %shl = shl i64 %ext, 2
63 store i64 %shl, ptr addrspace(1) %out, align 4
67 define amdgpu_kernel void @sext_shl64_overflow(ptr addrspace(1) nocapture %out, i32 %x) {
68 ; GCN-LABEL: sext_shl64_overflow:
70 ; GCN-NEXT: s_load_dword s4, s[2:3], 0xb
71 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
72 ; GCN-NEXT: s_mov_b32 s5, 0
73 ; GCN-NEXT: s_mov_b32 s3, 0xf000
74 ; GCN-NEXT: s_mov_b32 s2, -1
75 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
76 ; GCN-NEXT: s_bitset0_b32 s4, 31
77 ; GCN-NEXT: s_lshl_b64 s[4:5], s[4:5], 2
78 ; GCN-NEXT: v_mov_b32_e32 v0, s4
79 ; GCN-NEXT: v_mov_b32_e32 v1, s5
80 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
82 %and = and i32 %x, 2147483647
83 %ext = sext i32 %and to i64
84 %shl = shl i64 %ext, 2
85 store i64 %shl, ptr addrspace(1) %out, align 4
89 define amdgpu_kernel void @mulu24_shl64(ptr addrspace(1) nocapture %arg) {
90 ; GCN-LABEL: mulu24_shl64:
92 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[2:3], 0x9
93 ; GCN-NEXT: v_and_b32_e32 v0, 6, v0
94 ; GCN-NEXT: v_mul_u32_u24_e32 v0, 7, v0
95 ; GCN-NEXT: s_mov_b32 s3, 0xf000
96 ; GCN-NEXT: s_mov_b32 s2, 0
97 ; GCN-NEXT: v_lshlrev_b32_e32 v0, 2, v0
98 ; GCN-NEXT: v_mov_b32_e32 v1, 0
99 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
100 ; GCN-NEXT: buffer_store_dword v1, v[0:1], s[0:3], 0 addr64
103 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
104 %tmp1 = and i32 %tmp, 6
105 %mulconv = mul nuw nsw i32 %tmp1, 7
106 %tmp2 = zext i32 %mulconv to i64
107 %tmp3 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %tmp2
108 store i32 0, ptr addrspace(1) %tmp3, align 4
112 define amdgpu_kernel void @muli24_shl64(ptr addrspace(1) nocapture %arg, ptr addrspace(1) nocapture readonly %arg1) {
113 ; GCN-LABEL: muli24_shl64:
114 ; GCN: ; %bb.0: ; %bb
115 ; GCN-NEXT: s_load_dwordx4 s[0:3], s[2:3], 0x9
116 ; GCN-NEXT: s_mov_b32 s7, 0xf000
117 ; GCN-NEXT: s_mov_b32 s6, 0
118 ; GCN-NEXT: v_lshlrev_b32_e32 v1, 2, v0
119 ; GCN-NEXT: v_mov_b32_e32 v2, 0
120 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
121 ; GCN-NEXT: s_mov_b64 s[4:5], s[2:3]
122 ; GCN-NEXT: buffer_load_dword v1, v[1:2], s[4:7], 0 addr64
123 ; GCN-NEXT: v_lshlrev_b32_e32 v3, 3, v0
124 ; GCN-NEXT: s_mov_b64 s[2:3], s[6:7]
125 ; GCN-NEXT: v_mov_b32_e32 v4, v2
126 ; GCN-NEXT: s_waitcnt vmcnt(0)
127 ; GCN-NEXT: v_or_b32_e32 v0, 0x800000, v1
128 ; GCN-NEXT: v_mul_i32_i24_e32 v0, -7, v0
129 ; GCN-NEXT: v_lshlrev_b32_e32 v1, 3, v0
130 ; GCN-NEXT: buffer_store_dwordx2 v[1:2], v[3:4], s[0:3], 0 addr64
133 %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
134 %tmp2 = sext i32 %tmp to i64
135 %tmp3 = getelementptr inbounds i32, ptr addrspace(1) %arg1, i64 %tmp2
136 %tmp4 = load i32, ptr addrspace(1) %tmp3, align 4
137 %tmp5 = or i32 %tmp4, -8388608
138 %tmp6 = mul nsw i32 %tmp5, -7
139 %tmp7 = zext i32 %tmp6 to i64
140 %tmp8 = shl nuw nsw i64 %tmp7, 3
141 %tmp9 = getelementptr inbounds i64, ptr addrspace(1) %arg, i64 %tmp2
142 store i64 %tmp8, ptr addrspace(1) %tmp9, align 8
146 declare i32 @llvm.amdgcn.workitem.id.x()