xref: /llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane64.ptr.ll (revision 6548b6354d1d990e1c98736f5e7c3de876bedc8e) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
2; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11-SDAG %s
3
4define amdgpu_kernel void @test_p0(ptr addrspace(1) %out, ptr %src0) {
5; GFX11-SDAG-LABEL: test_p0:
6; GFX11-SDAG:       ; %bb.0:
7; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
8; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
9; GFX11-SDAG-NEXT:    v_dual_mov_b32 v3, 0 :: v_dual_mov_b32 v0, s3
10; GFX11-SDAG-NEXT:    v_mov_b32_e32 v2, s2
11; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
12; GFX11-SDAG-NEXT:    v_permlane64_b32 v1, v0
13; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v2
14; GFX11-SDAG-NEXT:    global_store_b64 v3, v[0:1], s[0:1]
15; GFX11-SDAG-NEXT:    s_endpgm
16  %v = call ptr @llvm.amdgcn.permlane64.p0(ptr %src0)
17  store ptr %v, ptr addrspace(1) %out
18  ret void
19}
20
21define amdgpu_kernel void @test_v3p0(ptr addrspace(1) %out, <3 x ptr> %src0) {
22; GFX11-SDAG-LABEL: test_v3p0:
23; GFX11-SDAG:       ; %bb.0:
24; GFX11-SDAG-NEXT:    s_clause 0x2
25; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x44
26; GFX11-SDAG-NEXT:    s_load_b64 s[6:7], s[4:5], 0x54
27; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x24
28; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
29; GFX11-SDAG-NEXT:    v_dual_mov_b32 v6, 0 :: v_dual_mov_b32 v1, s2
30; GFX11-SDAG-NEXT:    v_dual_mov_b32 v4, s1 :: v_dual_mov_b32 v5, s7
31; GFX11-SDAG-NEXT:    v_mov_b32_e32 v8, s6
32; GFX11-SDAG-NEXT:    v_dual_mov_b32 v0, s3 :: v_dual_mov_b32 v7, s0
33; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
34; GFX11-SDAG-NEXT:    v_permlane64_b32 v2, v1
35; GFX11-SDAG-NEXT:    v_permlane64_b32 v1, v4
36; GFX11-SDAG-NEXT:    v_permlane64_b32 v5, v5
37; GFX11-SDAG-NEXT:    v_permlane64_b32 v4, v8
38; GFX11-SDAG-NEXT:    v_permlane64_b32 v3, v0
39; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v7
40; GFX11-SDAG-NEXT:    s_clause 0x1
41; GFX11-SDAG-NEXT:    global_store_b64 v6, v[4:5], s[4:5] offset:16
42; GFX11-SDAG-NEXT:    global_store_b128 v6, v[0:3], s[4:5]
43; GFX11-SDAG-NEXT:    s_endpgm
44  %v = call <3 x ptr> @llvm.amdgcn.permlane64.v3p0(<3 x ptr> %src0)
45  store <3 x ptr> %v, ptr addrspace(1) %out
46  ret void
47}
48
49define amdgpu_kernel void @test_p3(ptr addrspace(1) %out, ptr addrspace(3) %src0) {
50; GFX11-SDAG-LABEL: test_p3:
51; GFX11-SDAG:       ; %bb.0:
52; GFX11-SDAG-NEXT:    s_clause 0x1
53; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
54; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
55; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
56; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
57; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
58; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v0
59; GFX11-SDAG-NEXT:    global_store_b32 v1, v0, s[0:1]
60; GFX11-SDAG-NEXT:    s_endpgm
61  %v = call ptr addrspace(3) @llvm.amdgcn.permlane64.v3p0(ptr addrspace(3) %src0)
62  store ptr addrspace(3) %v, ptr addrspace(1) %out
63  ret void
64}
65
66define amdgpu_kernel void @test_v3p3(ptr addrspace(1) %out, <3 x ptr addrspace(3)> %src0) {
67; GFX11-SDAG-LABEL: test_v3p3:
68; GFX11-SDAG:       ; %bb.0:
69; GFX11-SDAG-NEXT:    s_clause 0x1
70; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
71; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x24
72; GFX11-SDAG-NEXT:    v_mov_b32_e32 v4, 0
73; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
74; GFX11-SDAG-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s1
75; GFX11-SDAG-NEXT:    v_mov_b32_e32 v3, s0
76; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
77; GFX11-SDAG-NEXT:    v_permlane64_b32 v2, v0
78; GFX11-SDAG-NEXT:    v_permlane64_b32 v1, v1
79; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3)
80; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v3
81; GFX11-SDAG-NEXT:    global_store_b96 v4, v[0:2], s[4:5]
82; GFX11-SDAG-NEXT:    s_endpgm
83  %v = call <3 x ptr addrspace(3)> @llvm.amdgcn.permlane64.v3p3(<3 x ptr addrspace(3)> %src0)
84  store <3 x ptr addrspace(3)> %v, ptr addrspace(1) %out
85  ret void
86}
87
88define amdgpu_kernel void @test_p5(ptr addrspace(1) %out, ptr addrspace(5) %src0) {
89; GFX11-SDAG-LABEL: test_p5:
90; GFX11-SDAG:       ; %bb.0:
91; GFX11-SDAG-NEXT:    s_clause 0x1
92; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
93; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
94; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
95; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
96; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
97; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v0
98; GFX11-SDAG-NEXT:    global_store_b32 v1, v0, s[0:1]
99; GFX11-SDAG-NEXT:    s_endpgm
100  %v = call ptr addrspace(5) @llvm.amdgcn.permlane64.p5(ptr addrspace(5) %src0)
101  store ptr addrspace(5) %v, ptr addrspace(1) %out
102  ret void
103}
104
105define amdgpu_kernel void @test_v3p5(ptr addrspace(1) %out, <3 x ptr addrspace(5)> %src0) {
106; GFX11-SDAG-LABEL: test_v3p5:
107; GFX11-SDAG:       ; %bb.0:
108; GFX11-SDAG-NEXT:    s_clause 0x1
109; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
110; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x24
111; GFX11-SDAG-NEXT:    v_mov_b32_e32 v4, 0
112; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
113; GFX11-SDAG-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s1
114; GFX11-SDAG-NEXT:    v_mov_b32_e32 v3, s0
115; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
116; GFX11-SDAG-NEXT:    v_permlane64_b32 v2, v0
117; GFX11-SDAG-NEXT:    v_permlane64_b32 v1, v1
118; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3)
119; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v3
120; GFX11-SDAG-NEXT:    global_store_b96 v4, v[0:2], s[4:5]
121; GFX11-SDAG-NEXT:    s_endpgm
122  %v = call <3 x ptr addrspace(5)> @llvm.amdgcn.permlane64.v3p5(<3 x ptr addrspace(5)> %src0)
123  store <3 x ptr addrspace(5)> %v, ptr addrspace(1) %out
124  ret void
125}
126
127define amdgpu_kernel void @test_p6(ptr addrspace(1) %out, ptr addrspace(6) %src0) {
128; GFX11-SDAG-LABEL: test_p6:
129; GFX11-SDAG:       ; %bb.0:
130; GFX11-SDAG-NEXT:    s_clause 0x1
131; GFX11-SDAG-NEXT:    s_load_b32 s2, s[4:5], 0x2c
132; GFX11-SDAG-NEXT:    s_load_b64 s[0:1], s[4:5], 0x24
133; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
134; GFX11-SDAG-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
135; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_1)
136; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v0
137; GFX11-SDAG-NEXT:    global_store_b32 v1, v0, s[0:1]
138; GFX11-SDAG-NEXT:    s_endpgm
139  %v = call ptr addrspace(6) @llvm.amdgcn.permlane64.p6(ptr addrspace(6) %src0)
140  store ptr addrspace(6) %v, ptr addrspace(1) %out
141  ret void
142}
143
144define amdgpu_kernel void @test_v3p6(ptr addrspace(1) %out, <3 x ptr addrspace(6)> %src0) {
145; GFX11-SDAG-LABEL: test_v3p6:
146; GFX11-SDAG:       ; %bb.0:
147; GFX11-SDAG-NEXT:    s_clause 0x1
148; GFX11-SDAG-NEXT:    s_load_b128 s[0:3], s[4:5], 0x34
149; GFX11-SDAG-NEXT:    s_load_b64 s[4:5], s[4:5], 0x24
150; GFX11-SDAG-NEXT:    v_mov_b32_e32 v4, 0
151; GFX11-SDAG-NEXT:    s_waitcnt lgkmcnt(0)
152; GFX11-SDAG-NEXT:    v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s1
153; GFX11-SDAG-NEXT:    v_mov_b32_e32 v3, s0
154; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_3)
155; GFX11-SDAG-NEXT:    v_permlane64_b32 v2, v0
156; GFX11-SDAG-NEXT:    v_permlane64_b32 v1, v1
157; GFX11-SDAG-NEXT:    s_delay_alu instid0(VALU_DEP_3)
158; GFX11-SDAG-NEXT:    v_permlane64_b32 v0, v3
159; GFX11-SDAG-NEXT:    global_store_b96 v4, v[0:2], s[4:5]
160; GFX11-SDAG-NEXT:    s_endpgm
161  %v = call <3 x ptr addrspace(6)> @llvm.amdgcn.permlane64.v3p6(<3 x ptr addrspace(6)> %src0)
162  store <3 x ptr addrspace(6)> %v, ptr addrspace(1) %out
163  ret void
164}
165