xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.load.3d.ll (revision ba52f06f9d92c7ca04b440f618f8d352ea121fcc) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GFX6 %s
3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
4; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
5; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12 %s
6
7define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw(<8 x i32> inreg %rsrc, i32 %s, i32 %t, i32 %r) {
8; GFX6-LABEL: load_3d_v4f32_xyzw:
9; GFX6:       ; %bb.0:
10; GFX6-NEXT:    s_mov_b32 s0, s2
11; GFX6-NEXT:    s_mov_b32 s1, s3
12; GFX6-NEXT:    s_mov_b32 s2, s4
13; GFX6-NEXT:    s_mov_b32 s3, s5
14; GFX6-NEXT:    s_mov_b32 s4, s6
15; GFX6-NEXT:    s_mov_b32 s5, s7
16; GFX6-NEXT:    s_mov_b32 s6, s8
17; GFX6-NEXT:    s_mov_b32 s7, s9
18; GFX6-NEXT:    image_load v[0:3], v[0:2], s[0:7] dmask:0xf unorm
19; GFX6-NEXT:    s_waitcnt vmcnt(0)
20; GFX6-NEXT:    ; return to shader part epilog
21;
22; GFX10PLUS-LABEL: load_3d_v4f32_xyzw:
23; GFX10PLUS:       ; %bb.0:
24; GFX10PLUS-NEXT:    s_mov_b32 s0, s2
25; GFX10PLUS-NEXT:    s_mov_b32 s1, s3
26; GFX10PLUS-NEXT:    s_mov_b32 s2, s4
27; GFX10PLUS-NEXT:    s_mov_b32 s3, s5
28; GFX10PLUS-NEXT:    s_mov_b32 s4, s6
29; GFX10PLUS-NEXT:    s_mov_b32 s5, s7
30; GFX10PLUS-NEXT:    s_mov_b32 s6, s8
31; GFX10PLUS-NEXT:    s_mov_b32 s7, s9
32; GFX10PLUS-NEXT:    image_load v[0:3], v[0:2], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm
33; GFX10PLUS-NEXT:    s_waitcnt vmcnt(0)
34; GFX10PLUS-NEXT:    ; return to shader part epilog
35;
36; GFX12-LABEL: load_3d_v4f32_xyzw:
37; GFX12:       ; %bb.0:
38; GFX12-NEXT:    s_mov_b32 s0, s2
39; GFX12-NEXT:    s_mov_b32 s1, s3
40; GFX12-NEXT:    s_mov_b32 s2, s4
41; GFX12-NEXT:    s_mov_b32 s3, s5
42; GFX12-NEXT:    s_mov_b32 s4, s6
43; GFX12-NEXT:    s_mov_b32 s5, s7
44; GFX12-NEXT:    s_mov_b32 s6, s8
45; GFX12-NEXT:    s_mov_b32 s7, s9
46; GFX12-NEXT:    image_load v[0:3], [v0, v1, v2], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D
47; GFX12-NEXT:    s_wait_loadcnt 0x0
48; GFX12-NEXT:    ; return to shader part epilog
49  %v = call <4 x float> @llvm.amdgcn.image.load.3d.v4f32.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 0, i32 0)
50  ret <4 x float> %v
51}
52
53define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw_tfe(<8 x i32> inreg %rsrc, ptr addrspace(1) inreg %out, i32 %s, i32 %t, i32 %r) {
54; GFX6-LABEL: load_3d_v4f32_xyzw_tfe:
55; GFX6:       ; %bb.0:
56; GFX6-NEXT:    v_mov_b32_e32 v5, v0
57; GFX6-NEXT:    v_mov_b32_e32 v0, 0
58; GFX6-NEXT:    s_mov_b32 s0, s2
59; GFX6-NEXT:    s_mov_b32 s1, s3
60; GFX6-NEXT:    s_mov_b32 s2, s4
61; GFX6-NEXT:    s_mov_b32 s3, s5
62; GFX6-NEXT:    s_mov_b32 s4, s6
63; GFX6-NEXT:    s_mov_b32 s5, s7
64; GFX6-NEXT:    s_mov_b32 s6, s8
65; GFX6-NEXT:    s_mov_b32 s7, s9
66; GFX6-NEXT:    v_mov_b32_e32 v6, v1
67; GFX6-NEXT:    v_mov_b32_e32 v7, v2
68; GFX6-NEXT:    v_mov_b32_e32 v1, v0
69; GFX6-NEXT:    v_mov_b32_e32 v2, v0
70; GFX6-NEXT:    v_mov_b32_e32 v3, v0
71; GFX6-NEXT:    v_mov_b32_e32 v4, v0
72; GFX6-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf unorm tfe
73; GFX6-NEXT:    s_mov_b32 s8, s10
74; GFX6-NEXT:    s_mov_b32 s9, s11
75; GFX6-NEXT:    s_mov_b32 s10, -1
76; GFX6-NEXT:    s_mov_b32 s11, 0xf000
77; GFX6-NEXT:    s_waitcnt vmcnt(0)
78; GFX6-NEXT:    buffer_store_dword v4, off, s[8:11], 0
79; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
80; GFX6-NEXT:    ; return to shader part epilog
81;
82; GFX10-LABEL: load_3d_v4f32_xyzw_tfe:
83; GFX10:       ; %bb.0:
84; GFX10-NEXT:    v_mov_b32_e32 v8, 0
85; GFX10-NEXT:    v_mov_b32_e32 v5, v0
86; GFX10-NEXT:    v_mov_b32_e32 v6, v1
87; GFX10-NEXT:    v_mov_b32_e32 v7, v2
88; GFX10-NEXT:    s_mov_b32 s0, s2
89; GFX10-NEXT:    v_mov_b32_e32 v9, v8
90; GFX10-NEXT:    v_mov_b32_e32 v10, v8
91; GFX10-NEXT:    v_mov_b32_e32 v11, v8
92; GFX10-NEXT:    v_mov_b32_e32 v12, v8
93; GFX10-NEXT:    s_mov_b32 s1, s3
94; GFX10-NEXT:    s_mov_b32 s2, s4
95; GFX10-NEXT:    s_mov_b32 s3, s5
96; GFX10-NEXT:    s_mov_b32 s4, s6
97; GFX10-NEXT:    s_mov_b32 s5, s7
98; GFX10-NEXT:    s_mov_b32 s6, s8
99; GFX10-NEXT:    s_mov_b32 s7, s9
100; GFX10-NEXT:    v_mov_b32_e32 v0, v8
101; GFX10-NEXT:    v_mov_b32_e32 v1, v9
102; GFX10-NEXT:    v_mov_b32_e32 v2, v10
103; GFX10-NEXT:    v_mov_b32_e32 v3, v11
104; GFX10-NEXT:    v_mov_b32_e32 v4, v12
105; GFX10-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe
106; GFX10-NEXT:    s_waitcnt vmcnt(0)
107; GFX10-NEXT:    global_store_dword v8, v4, s[10:11]
108; GFX10-NEXT:    ; return to shader part epilog
109;
110; GFX11-LABEL: load_3d_v4f32_xyzw_tfe:
111; GFX11:       ; %bb.0:
112; GFX11-NEXT:    v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0
113; GFX11-NEXT:    v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1
114; GFX11-NEXT:    s_mov_b32 s0, s2
115; GFX11-NEXT:    s_mov_b32 s1, s3
116; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
117; GFX11-NEXT:    v_mov_b32_e32 v9, v8
118; GFX11-NEXT:    v_mov_b32_e32 v10, v8
119; GFX11-NEXT:    v_mov_b32_e32 v11, v8
120; GFX11-NEXT:    v_mov_b32_e32 v12, v8
121; GFX11-NEXT:    s_mov_b32 s2, s4
122; GFX11-NEXT:    s_mov_b32 s3, s5
123; GFX11-NEXT:    s_mov_b32 s4, s6
124; GFX11-NEXT:    s_mov_b32 s5, s7
125; GFX11-NEXT:    s_mov_b32 s6, s8
126; GFX11-NEXT:    s_mov_b32 s7, s9
127; GFX11-NEXT:    v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9
128; GFX11-NEXT:    v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11
129; GFX11-NEXT:    v_mov_b32_e32 v4, v12
130; GFX11-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe
131; GFX11-NEXT:    s_waitcnt vmcnt(0)
132; GFX11-NEXT:    global_store_b32 v8, v4, s[10:11]
133; GFX11-NEXT:    ; return to shader part epilog
134;
135; GFX12-LABEL: load_3d_v4f32_xyzw_tfe:
136; GFX12:       ; %bb.0:
137; GFX12-NEXT:    v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0
138; GFX12-NEXT:    v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1
139; GFX12-NEXT:    s_mov_b32 s0, s2
140; GFX12-NEXT:    s_mov_b32 s1, s3
141; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
142; GFX12-NEXT:    v_dual_mov_b32 v9, v8 :: v_dual_mov_b32 v10, v8
143; GFX12-NEXT:    v_dual_mov_b32 v11, v8 :: v_dual_mov_b32 v12, v8
144; GFX12-NEXT:    s_mov_b32 s2, s4
145; GFX12-NEXT:    s_mov_b32 s3, s5
146; GFX12-NEXT:    s_mov_b32 s4, s6
147; GFX12-NEXT:    s_mov_b32 s5, s7
148; GFX12-NEXT:    s_mov_b32 s6, s8
149; GFX12-NEXT:    s_mov_b32 s7, s9
150; GFX12-NEXT:    v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9
151; GFX12-NEXT:    v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11
152; GFX12-NEXT:    v_mov_b32_e32 v4, v12
153; GFX12-NEXT:    image_load v[0:4], [v5, v6, v7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D tfe
154; GFX12-NEXT:    s_wait_loadcnt 0x0
155; GFX12-NEXT:    global_store_b32 v8, v4, s[10:11]
156; GFX12-NEXT:    ; return to shader part epilog
157  %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 1, i32 0)
158  %v.vec = extractvalue { <4 x float>, i32 } %v, 0
159  %v.err = extractvalue { <4 x float>, i32 } %v, 1
160  store i32 %v.err, ptr addrspace(1) %out, align 4
161  ret <4 x float> %v.vec
162}
163
164define amdgpu_ps <4 x float> @load_3d_v4f32_xyzw_tfe_lwe(<8 x i32> inreg %rsrc, ptr addrspace(1) inreg %out, i32 %s, i32 %t, i32 %r) {
165; GFX6-LABEL: load_3d_v4f32_xyzw_tfe_lwe:
166; GFX6:       ; %bb.0:
167; GFX6-NEXT:    v_mov_b32_e32 v5, v0
168; GFX6-NEXT:    v_mov_b32_e32 v0, 0
169; GFX6-NEXT:    s_mov_b32 s0, s2
170; GFX6-NEXT:    s_mov_b32 s1, s3
171; GFX6-NEXT:    s_mov_b32 s2, s4
172; GFX6-NEXT:    s_mov_b32 s3, s5
173; GFX6-NEXT:    s_mov_b32 s4, s6
174; GFX6-NEXT:    s_mov_b32 s5, s7
175; GFX6-NEXT:    s_mov_b32 s6, s8
176; GFX6-NEXT:    s_mov_b32 s7, s9
177; GFX6-NEXT:    v_mov_b32_e32 v6, v1
178; GFX6-NEXT:    v_mov_b32_e32 v7, v2
179; GFX6-NEXT:    v_mov_b32_e32 v1, v0
180; GFX6-NEXT:    v_mov_b32_e32 v2, v0
181; GFX6-NEXT:    v_mov_b32_e32 v3, v0
182; GFX6-NEXT:    v_mov_b32_e32 v4, v0
183; GFX6-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf unorm tfe lwe
184; GFX6-NEXT:    s_mov_b32 s8, s10
185; GFX6-NEXT:    s_mov_b32 s9, s11
186; GFX6-NEXT:    s_mov_b32 s10, -1
187; GFX6-NEXT:    s_mov_b32 s11, 0xf000
188; GFX6-NEXT:    s_waitcnt vmcnt(0)
189; GFX6-NEXT:    buffer_store_dword v4, off, s[8:11], 0
190; GFX6-NEXT:    s_waitcnt vmcnt(0) expcnt(0)
191; GFX6-NEXT:    ; return to shader part epilog
192;
193; GFX10-LABEL: load_3d_v4f32_xyzw_tfe_lwe:
194; GFX10:       ; %bb.0:
195; GFX10-NEXT:    v_mov_b32_e32 v8, 0
196; GFX10-NEXT:    v_mov_b32_e32 v5, v0
197; GFX10-NEXT:    v_mov_b32_e32 v6, v1
198; GFX10-NEXT:    v_mov_b32_e32 v7, v2
199; GFX10-NEXT:    s_mov_b32 s0, s2
200; GFX10-NEXT:    v_mov_b32_e32 v9, v8
201; GFX10-NEXT:    v_mov_b32_e32 v10, v8
202; GFX10-NEXT:    v_mov_b32_e32 v11, v8
203; GFX10-NEXT:    v_mov_b32_e32 v12, v8
204; GFX10-NEXT:    s_mov_b32 s1, s3
205; GFX10-NEXT:    s_mov_b32 s2, s4
206; GFX10-NEXT:    s_mov_b32 s3, s5
207; GFX10-NEXT:    s_mov_b32 s4, s6
208; GFX10-NEXT:    s_mov_b32 s5, s7
209; GFX10-NEXT:    s_mov_b32 s6, s8
210; GFX10-NEXT:    s_mov_b32 s7, s9
211; GFX10-NEXT:    v_mov_b32_e32 v0, v8
212; GFX10-NEXT:    v_mov_b32_e32 v1, v9
213; GFX10-NEXT:    v_mov_b32_e32 v2, v10
214; GFX10-NEXT:    v_mov_b32_e32 v3, v11
215; GFX10-NEXT:    v_mov_b32_e32 v4, v12
216; GFX10-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe lwe
217; GFX10-NEXT:    s_waitcnt vmcnt(0)
218; GFX10-NEXT:    global_store_dword v8, v4, s[10:11]
219; GFX10-NEXT:    ; return to shader part epilog
220;
221; GFX11-LABEL: load_3d_v4f32_xyzw_tfe_lwe:
222; GFX11:       ; %bb.0:
223; GFX11-NEXT:    v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0
224; GFX11-NEXT:    v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1
225; GFX11-NEXT:    s_mov_b32 s0, s2
226; GFX11-NEXT:    s_mov_b32 s1, s3
227; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2)
228; GFX11-NEXT:    v_mov_b32_e32 v9, v8
229; GFX11-NEXT:    v_mov_b32_e32 v10, v8
230; GFX11-NEXT:    v_mov_b32_e32 v11, v8
231; GFX11-NEXT:    v_mov_b32_e32 v12, v8
232; GFX11-NEXT:    s_mov_b32 s2, s4
233; GFX11-NEXT:    s_mov_b32 s3, s5
234; GFX11-NEXT:    s_mov_b32 s4, s6
235; GFX11-NEXT:    s_mov_b32 s5, s7
236; GFX11-NEXT:    s_mov_b32 s6, s8
237; GFX11-NEXT:    s_mov_b32 s7, s9
238; GFX11-NEXT:    v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9
239; GFX11-NEXT:    v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11
240; GFX11-NEXT:    v_mov_b32_e32 v4, v12
241; GFX11-NEXT:    image_load v[0:4], v[5:7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D unorm tfe lwe
242; GFX11-NEXT:    s_waitcnt vmcnt(0)
243; GFX11-NEXT:    global_store_b32 v8, v4, s[10:11]
244; GFX11-NEXT:    ; return to shader part epilog
245;
246; GFX12-LABEL: load_3d_v4f32_xyzw_tfe_lwe:
247; GFX12:       ; %bb.0:
248; GFX12-NEXT:    v_dual_mov_b32 v7, v2 :: v_dual_mov_b32 v8, 0
249; GFX12-NEXT:    v_dual_mov_b32 v5, v0 :: v_dual_mov_b32 v6, v1
250; GFX12-NEXT:    s_mov_b32 s0, s2
251; GFX12-NEXT:    s_mov_b32 s1, s3
252; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2)
253; GFX12-NEXT:    v_dual_mov_b32 v9, v8 :: v_dual_mov_b32 v10, v8
254; GFX12-NEXT:    v_dual_mov_b32 v11, v8 :: v_dual_mov_b32 v12, v8
255; GFX12-NEXT:    s_mov_b32 s2, s4
256; GFX12-NEXT:    s_mov_b32 s3, s5
257; GFX12-NEXT:    s_mov_b32 s4, s6
258; GFX12-NEXT:    s_mov_b32 s5, s7
259; GFX12-NEXT:    s_mov_b32 s6, s8
260; GFX12-NEXT:    s_mov_b32 s7, s9
261; GFX12-NEXT:    v_dual_mov_b32 v0, v8 :: v_dual_mov_b32 v1, v9
262; GFX12-NEXT:    v_dual_mov_b32 v2, v10 :: v_dual_mov_b32 v3, v11
263; GFX12-NEXT:    v_mov_b32_e32 v4, v12
264; GFX12-NEXT:    image_load v[0:4], [v5, v6, v7], s[0:7] dmask:0xf dim:SQ_RSRC_IMG_3D tfe
265; GFX12-NEXT:    s_wait_loadcnt 0x0
266; GFX12-NEXT:    global_store_b32 v8, v4, s[10:11]
267; GFX12-NEXT:    ; return to shader part epilog
268  %v = call { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 15, i32 %s, i32 %t, i32 %r, <8 x i32> %rsrc, i32 3, i32 0)
269  %v.vec = extractvalue { <4 x float>, i32 } %v, 0
270  %v.err = extractvalue { <4 x float>, i32 } %v, 1
271  store i32 %v.err, ptr addrspace(1) %out, align 4
272  ret <4 x float> %v.vec
273}
274
275declare <4 x float> @llvm.amdgcn.image.load.3d.v4f32.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
276declare { <4 x float>, i32 } @llvm.amdgcn.image.load.3d.sl_v4f32i32s.i32(i32 immarg, i32, i32, i32, <8 x i32>, i32 immarg, i32 immarg) #0
277
278attributes #0 = { nounwind readonly }
279