1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN %s
4 define amdgpu_kernel void @zext_i16_to_i32_uniform(ptr addrspace(1) %out, i16 %a, i32 %b) {
5 ; GCN-LABEL: zext_i16_to_i32_uniform:
7 ; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
8 ; GCN-NEXT: s_mov_b32 s7, 0xf000
9 ; GCN-NEXT: s_mov_b32 s6, -1
10 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
11 ; GCN-NEXT: s_mov_b32 s4, s0
12 ; GCN-NEXT: s_and_b32 s0, s2, 0xffff
13 ; GCN-NEXT: s_add_i32 s0, s3, s0
14 ; GCN-NEXT: s_mov_b32 s5, s1
15 ; GCN-NEXT: v_mov_b32_e32 v0, s0
16 ; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0
18 %zext = zext i16 %a to i32
19 %res = add i32 %b, %zext
20 store i32 %res, ptr addrspace(1) %out
25 define amdgpu_kernel void @zext_i16_to_i64_uniform(ptr addrspace(1) %out, i16 %a, i64 %b) {
26 ; GCN-LABEL: zext_i16_to_i64_uniform:
28 ; GCN-NEXT: s_load_dword s6, s[0:1], 0xb
29 ; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0xd
30 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
31 ; GCN-NEXT: s_mov_b32 s3, 0xf000
32 ; GCN-NEXT: s_mov_b32 s2, -1
33 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
34 ; GCN-NEXT: s_and_b32 s6, s6, 0xffff
35 ; GCN-NEXT: s_add_u32 s4, s4, s6
36 ; GCN-NEXT: s_addc_u32 s5, s5, 0
37 ; GCN-NEXT: v_mov_b32_e32 v0, s4
38 ; GCN-NEXT: v_mov_b32_e32 v1, s5
39 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
41 %zext = zext i16 %a to i64
42 %res = add i64 %b, %zext
43 store i64 %res, ptr addrspace(1) %out
47 define amdgpu_kernel void @zext_i16_to_i32_divergent(ptr addrspace(1) %out, i16 %a, i32 %b) {
48 ; GCN-LABEL: zext_i16_to_i32_divergent:
50 ; GCN-NEXT: s_load_dword s4, s[0:1], 0xb
51 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
52 ; GCN-NEXT: s_mov_b32 s3, 0xf000
53 ; GCN-NEXT: s_mov_b32 s2, -1
54 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
55 ; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0
56 ; GCN-NEXT: v_and_b32_e32 v0, 0xffff, v0
57 ; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0
59 %tid = call i32 @llvm.amdgcn.workitem.id.x()
60 %tid.truncated = trunc i32 %tid to i16
61 %divergent.a = add i16 %a, %tid.truncated
62 %zext = zext i16 %divergent.a to i32
63 store i32 %zext, ptr addrspace(1) %out
68 define amdgpu_kernel void @zext_i16_to_i64_divergent(ptr addrspace(1) %out, i16 %a, i64 %b) {
69 ; GCN-LABEL: zext_i16_to_i64_divergent:
71 ; GCN-NEXT: s_load_dword s4, s[0:1], 0xb
72 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
73 ; GCN-NEXT: s_mov_b32 s3, 0xf000
74 ; GCN-NEXT: s_mov_b32 s2, -1
75 ; GCN-NEXT: v_mov_b32_e32 v1, 0
76 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
77 ; GCN-NEXT: v_add_i32_e32 v0, vcc, s4, v0
78 ; GCN-NEXT: v_and_b32_e32 v0, 0xffff, v0
79 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
81 %tid = call i32 @llvm.amdgcn.workitem.id.x()
82 %tid.truncated = trunc i32 %tid to i16
83 %divergent.a = add i16 %a, %tid.truncated
84 %zext = zext i16 %divergent.a to i64
85 store i64 %zext, ptr addrspace(1) %out
89 declare i32 @llvm.amdgcn.workitem.id.x() #1
91 attributes #0 = { nounwind }
92 attributes #1 = { nounwind readnone speculatable }