xref: /llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.tbuffer.load.ll (revision 9e9907f1cfa424366fba58d9520f9305b537cec9) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2;RUN: llc < %s -mtriple=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck -check-prefixes=PREGFX10 %s
3;RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck -check-prefixes=PREGFX10 %s
4;RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck -check-prefixes=GFX10 %s
5;RUN: llc < %s -mtriple=amdgcn -mcpu=gfx1100 -verify-machineinstrs | FileCheck -check-prefixes=GFX11 %s
6
7define amdgpu_vs {<4 x float>, <4 x float>, <4 x float>, <4 x float>} @tbuffer_load(ptr addrspace(8) inreg) {
8; PREGFX10-LABEL: tbuffer_load:
9; PREGFX10:       ; %bb.0: ; %main_body
10; PREGFX10-NEXT:    v_mov_b32_e32 v12, 0
11; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v12, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen
12; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[4:7], v12, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen glc
13; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[8:11], v12, s[0:3], 0 format:[BUF_DATA_FORMAT_10_11_11,BUF_NUM_FORMAT_SNORM] idxen slc
14; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[12:15], v12, s[0:3], 0 format:[BUF_DATA_FORMAT_10_11_11,BUF_NUM_FORMAT_SNORM] idxen glc
15; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
16; PREGFX10-NEXT:    ; return to shader part epilog
17;
18; GFX10-LABEL: tbuffer_load:
19; GFX10:       ; %bb.0: ; %main_body
20; GFX10-NEXT:    v_mov_b32_e32 v16, 0
21; GFX10-NEXT:    s_clause 0x3
22; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v16, s[0:3], 0 format:78 idxen
23; GFX10-NEXT:    tbuffer_load_format_xyzw v[4:7], v16, s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen glc
24; GFX10-NEXT:    tbuffer_load_format_xyzw v[8:11], v16, s[0:3], 0 format:[BUF_FMT_32_FLOAT] idxen slc
25; GFX10-NEXT:    tbuffer_load_format_xyzw v[12:15], v16, s[0:3], 0 format:[BUF_FMT_32_FLOAT] idxen glc dlc
26; GFX10-NEXT:    s_waitcnt vmcnt(0)
27; GFX10-NEXT:    ; return to shader part epilog
28;
29; GFX11-LABEL: tbuffer_load:
30; GFX11:       ; %bb.0: ; %main_body
31; GFX11-NEXT:    v_mov_b32_e32 v12, 0
32; GFX11-NEXT:    s_clause 0x3
33; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v12, s[0:3], 0 format:78 idxen
34; GFX11-NEXT:    tbuffer_load_format_xyzw v[4:7], v12, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen glc
35; GFX11-NEXT:    tbuffer_load_format_xyzw v[8:11], v12, s[0:3], 0 format:[BUF_FMT_32_FLOAT] idxen slc
36; GFX11-NEXT:    tbuffer_load_format_xyzw v[12:15], v12, s[0:3], 0 format:[BUF_FMT_32_FLOAT] idxen glc dlc
37; GFX11-NEXT:    s_waitcnt vmcnt(0)
38; GFX11-NEXT:    ; return to shader part epilog
39main_body:
40    %vdata     = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 0, i32 0, i32 78, i32 0)
41    %vdata_glc = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 0, i32 0, i32 63, i32 1)
42    %vdata_slc = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 0, i32 0, i32 22, i32 2)
43    %vdata_f32 = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 0, i32 0, i32 22, i32 5)
44    %vdata.f     = bitcast <4 x i32> %vdata to <4 x float>
45    %vdata_glc.f = bitcast <4 x i32> %vdata_glc to <4 x float>
46    %vdata_slc.f = bitcast <4 x i32> %vdata_slc to <4 x float>
47    %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %vdata.f, 0
48    %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %vdata_glc.f, 1
49    %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %vdata_slc.f, 2
50    %r3 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r2, <4 x float> %vdata_f32, 3
51    ret {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r3
52}
53
54define amdgpu_vs <4 x float> @tbuffer_load_immoffs(ptr addrspace(8) inreg) {
55; PREGFX10-LABEL: tbuffer_load_immoffs:
56; PREGFX10:       ; %bb.0: ; %main_body
57; PREGFX10-NEXT:    v_mov_b32_e32 v0, 0
58; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen offset:42
59; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
60; PREGFX10-NEXT:    ; return to shader part epilog
61;
62; GFX10-LABEL: tbuffer_load_immoffs:
63; GFX10:       ; %bb.0: ; %main_body
64; GFX10-NEXT:    v_mov_b32_e32 v0, 0
65; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:78 idxen offset:42
66; GFX10-NEXT:    s_waitcnt vmcnt(0)
67; GFX10-NEXT:    ; return to shader part epilog
68;
69; GFX11-LABEL: tbuffer_load_immoffs:
70; GFX11:       ; %bb.0: ; %main_body
71; GFX11-NEXT:    v_mov_b32_e32 v0, 0
72; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:78 idxen offset:42
73; GFX11-NEXT:    s_waitcnt vmcnt(0)
74; GFX11-NEXT:    ; return to shader part epilog
75main_body:
76    %vdata   = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 42, i32 0, i32 78, i32 0)
77    %vdata.f = bitcast <4 x i32> %vdata to <4 x float>
78    ret <4 x float> %vdata.f
79}
80
81define amdgpu_vs {<4 x float>, <4 x float>, <4 x float>} @tbuffer_load_immoffs_large(ptr addrspace(8) inreg, i32 inreg %soffs) {
82; PREGFX10-LABEL: tbuffer_load_immoffs_large:
83; PREGFX10:       ; %bb.0:
84; PREGFX10-NEXT:    v_mov_b32_e32 v8, 0
85; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v8, s[0:3], 61 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_USCALED] idxen offset:4095
86; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[4:7], v8, s[0:3], s4 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_SSCALED] idxen offset:73
87; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[8:11], v8, s[0:3], s4 format:[BUF_DATA_FORMAT_32_32_32,BUF_NUM_FORMAT_UINT] idxen offset:1
88; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
89; PREGFX10-NEXT:    ; return to shader part epilog
90;
91; GFX10-LABEL: tbuffer_load_immoffs_large:
92; GFX10:       ; %bb.0:
93; GFX10-NEXT:    v_mov_b32_e32 v12, 0
94; GFX10-NEXT:    s_clause 0x2
95; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v12, s[0:3], 61 format:[BUF_FMT_10_10_10_2_SSCALED] idxen offset:4095
96; GFX10-NEXT:    tbuffer_load_format_xyzw v[4:7], v12, s[0:3], s4 format:[BUF_FMT_32_32_UINT] idxen offset:73
97; GFX10-NEXT:    tbuffer_load_format_xyzw v[8:11], v12, s[0:3], s4 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offset:1
98; GFX10-NEXT:    s_waitcnt vmcnt(0)
99; GFX10-NEXT:    ; return to shader part epilog
100;
101; GFX11-LABEL: tbuffer_load_immoffs_large:
102; GFX11:       ; %bb.0:
103; GFX11-NEXT:    v_mov_b32_e32 v8, 0
104; GFX11-NEXT:    s_clause 0x2
105; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v8, s[0:3], 61 format:[BUF_FMT_8_8_8_8_SINT] idxen offset:4095
106; GFX11-NEXT:    tbuffer_load_format_xyzw v[4:7], v8, s[0:3], s4 format:[BUF_FMT_32_32_32_32_SINT] idxen offset:73
107; GFX11-NEXT:    tbuffer_load_format_xyzw v[8:11], v8, s[0:3], s4 format:77 idxen offset:1
108; GFX11-NEXT:    s_waitcnt vmcnt(0)
109; GFX11-NEXT:    ; return to shader part epilog
110    %vdata     = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 4095, i32 61, i32 47, i32 0)
111    %vdata_glc = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 73, i32 %soffs, i32 62, i32 0)
112    %vdata_slc = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 1, i32 %soffs, i32 77, i32 0)
113    %vdata.f     = bitcast <4 x i32> %vdata to <4 x float>
114    %vdata_glc.f = bitcast <4 x i32> %vdata_glc to <4 x float>
115    %vdata_slc.f = bitcast <4 x i32> %vdata_slc to <4 x float>
116    %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %vdata.f, 0
117    %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %vdata_glc.f, 1
118    %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %vdata_slc.f, 2
119    ret {<4 x float>, <4 x float>, <4 x float>} %r2
120}
121
122define amdgpu_vs <4 x float> @tbuffer_load_idx(ptr addrspace(8) inreg, i32 %vindex) {
123; PREGFX10-LABEL: tbuffer_load_idx:
124; PREGFX10:       ; %bb.0: ; %main_body
125; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen
126; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
127; PREGFX10-NEXT:    ; return to shader part epilog
128;
129; GFX10-LABEL: tbuffer_load_idx:
130; GFX10:       ; %bb.0: ; %main_body
131; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:78 idxen
132; GFX10-NEXT:    s_waitcnt vmcnt(0)
133; GFX10-NEXT:    ; return to shader part epilog
134;
135; GFX11-LABEL: tbuffer_load_idx:
136; GFX11:       ; %bb.0: ; %main_body
137; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:78 idxen
138; GFX11-NEXT:    s_waitcnt vmcnt(0)
139; GFX11-NEXT:    ; return to shader part epilog
140main_body:
141    %vdata   = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 %vindex, i32 0, i32 0, i32 78, i32 0)
142    %vdata.f = bitcast <4 x i32> %vdata to <4 x float>
143    ret <4 x float> %vdata.f
144}
145
146define amdgpu_vs <4 x float> @tbuffer_load_ofs(ptr addrspace(8) inreg, i32 %voffs) {
147; PREGFX10-LABEL: tbuffer_load_ofs:
148; PREGFX10:       ; %bb.0: ; %main_body
149; PREGFX10-NEXT:    s_mov_b32 s4, 0
150; PREGFX10-NEXT:    v_mov_b32_e32 v1, v0
151; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
152; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen offen
153; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
154; PREGFX10-NEXT:    ; return to shader part epilog
155;
156; GFX10-LABEL: tbuffer_load_ofs:
157; GFX10:       ; %bb.0: ; %main_body
158; GFX10-NEXT:    s_mov_b32 s4, 0
159; GFX10-NEXT:    v_mov_b32_e32 v1, v0
160; GFX10-NEXT:    v_mov_b32_e32 v0, s4
161; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen
162; GFX10-NEXT:    s_waitcnt vmcnt(0)
163; GFX10-NEXT:    ; return to shader part epilog
164;
165; GFX11-LABEL: tbuffer_load_ofs:
166; GFX11:       ; %bb.0: ; %main_body
167; GFX11-NEXT:    s_mov_b32 s4, 0
168; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
169; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
170; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen
171; GFX11-NEXT:    s_waitcnt vmcnt(0)
172; GFX11-NEXT:    ; return to shader part epilog
173main_body:
174    %vdata   = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 %voffs, i32 0, i32 78, i32 0)
175    %vdata.f = bitcast <4 x i32> %vdata to <4 x float>
176    ret <4 x float> %vdata.f
177}
178
179define amdgpu_vs <4 x float> @tbuffer_load_ofs_imm(ptr addrspace(8) inreg, i32 %voffs) {
180; PREGFX10-LABEL: tbuffer_load_ofs_imm:
181; PREGFX10:       ; %bb.0: ; %main_body
182; PREGFX10-NEXT:    s_mov_b32 s4, 0
183; PREGFX10-NEXT:    v_mov_b32_e32 v1, v0
184; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
185; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen offen offset:52
186; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
187; PREGFX10-NEXT:    ; return to shader part epilog
188;
189; GFX10-LABEL: tbuffer_load_ofs_imm:
190; GFX10:       ; %bb.0: ; %main_body
191; GFX10-NEXT:    s_mov_b32 s4, 0
192; GFX10-NEXT:    v_mov_b32_e32 v1, v0
193; GFX10-NEXT:    v_mov_b32_e32 v0, s4
194; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen offset:52
195; GFX10-NEXT:    s_waitcnt vmcnt(0)
196; GFX10-NEXT:    ; return to shader part epilog
197;
198; GFX11-LABEL: tbuffer_load_ofs_imm:
199; GFX11:       ; %bb.0: ; %main_body
200; GFX11-NEXT:    s_mov_b32 s4, 0
201; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
202; GFX11-NEXT:    v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s4
203; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen offset:52
204; GFX11-NEXT:    s_waitcnt vmcnt(0)
205; GFX11-NEXT:    ; return to shader part epilog
206main_body:
207    %ofs = add i32 %voffs, 52
208    %vdata   = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 0, i32 %ofs, i32 0, i32 78, i32 0)
209    %vdata.f = bitcast <4 x i32> %vdata to <4 x float>
210    ret <4 x float> %vdata.f
211}
212
213define amdgpu_vs <4 x float> @tbuffer_load_both(ptr addrspace(8) inreg, i32 %vindex, i32 %voffs) {
214; PREGFX10-LABEL: tbuffer_load_both:
215; PREGFX10:       ; %bb.0: ; %main_body
216; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32_32,BUF_NUM_FORMAT_UINT] idxen offen
217; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
218; PREGFX10-NEXT:    ; return to shader part epilog
219;
220; GFX10-LABEL: tbuffer_load_both:
221; GFX10:       ; %bb.0: ; %main_body
222; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen
223; GFX10-NEXT:    s_waitcnt vmcnt(0)
224; GFX10-NEXT:    ; return to shader part epilog
225;
226; GFX11-LABEL: tbuffer_load_both:
227; GFX11:       ; %bb.0: ; %main_body
228; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:78 idxen offen
229; GFX11-NEXT:    s_waitcnt vmcnt(0)
230; GFX11-NEXT:    ; return to shader part epilog
231main_body:
232    %vdata   = call <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8) %0, i32 %vindex, i32 %voffs, i32 0, i32 78, i32 0)
233    %vdata.f = bitcast <4 x i32> %vdata to <4 x float>
234    ret <4 x float> %vdata.f
235}
236
237define amdgpu_vs <2 x float> @buffer_load_xy(ptr addrspace(8) inreg %rsrc) {
238; PREGFX10-LABEL: buffer_load_xy:
239; PREGFX10:       ; %bb.0:
240; PREGFX10-NEXT:    v_mov_b32_e32 v0, 0
241; PREGFX10-NEXT:    tbuffer_load_format_xy v[0:1], v0, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32,BUF_NUM_FORMAT_UINT] idxen
242; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
243; PREGFX10-NEXT:    ; return to shader part epilog
244;
245; GFX10-LABEL: buffer_load_xy:
246; GFX10:       ; %bb.0:
247; GFX10-NEXT:    v_mov_b32_e32 v0, 0
248; GFX10-NEXT:    tbuffer_load_format_xy v[0:1], v0, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen
249; GFX10-NEXT:    s_waitcnt vmcnt(0)
250; GFX10-NEXT:    ; return to shader part epilog
251;
252; GFX11-LABEL: buffer_load_xy:
253; GFX11:       ; %bb.0:
254; GFX11-NEXT:    v_mov_b32_e32 v0, 0
255; GFX11-NEXT:    tbuffer_load_format_xy v[0:1], v0, s[0:3], 0 format:77 idxen
256; GFX11-NEXT:    s_waitcnt vmcnt(0)
257; GFX11-NEXT:    ; return to shader part epilog
258    %vdata = call <2 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v2i32(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 77, i32 0)
259    %vdata.f = bitcast <2 x i32> %vdata to <2 x float>
260    ret <2 x float> %vdata.f
261}
262
263define amdgpu_vs float @buffer_load_x(ptr addrspace(8) inreg %rsrc) {
264; PREGFX10-LABEL: buffer_load_x:
265; PREGFX10:       ; %bb.0:
266; PREGFX10-NEXT:    v_mov_b32_e32 v0, 0
267; PREGFX10-NEXT:    tbuffer_load_format_x v0, v0, s[0:3], 0 format:[BUF_DATA_FORMAT_32_32_32,BUF_NUM_FORMAT_UINT] idxen
268; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
269; PREGFX10-NEXT:    ; return to shader part epilog
270;
271; GFX10-LABEL: buffer_load_x:
272; GFX10:       ; %bb.0:
273; GFX10-NEXT:    v_mov_b32_e32 v0, 0
274; GFX10-NEXT:    tbuffer_load_format_x v0, v0, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen
275; GFX10-NEXT:    s_waitcnt vmcnt(0)
276; GFX10-NEXT:    ; return to shader part epilog
277;
278; GFX11-LABEL: buffer_load_x:
279; GFX11:       ; %bb.0:
280; GFX11-NEXT:    v_mov_b32_e32 v0, 0
281; GFX11-NEXT:    tbuffer_load_format_x v0, v0, s[0:3], 0 format:77 idxen
282; GFX11-NEXT:    s_waitcnt vmcnt(0)
283; GFX11-NEXT:    ; return to shader part epilog
284    %vdata = call i32 @llvm.amdgcn.struct.ptr.tbuffer.load.i32(ptr addrspace(8) %rsrc, i32 0, i32 0, i32 0, i32 77, i32 0)
285    %vdata.f = bitcast i32 %vdata to float
286    ret float %vdata.f
287}
288
289define amdgpu_ps <4 x float> @buffer_load_voffset_large_12bit(ptr addrspace(8) inreg) {
290; PREGFX10-LABEL: buffer_load_voffset_large_12bit:
291; PREGFX10:       ; %bb.0: ; %main_body
292; PREGFX10-NEXT:    v_mov_b32_e32 v0, 0
293; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen offset:4092
294; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
295; PREGFX10-NEXT:    ; return to shader part epilog
296;
297; GFX10-LABEL: buffer_load_voffset_large_12bit:
298; GFX10:       ; %bb.0: ; %main_body
299; GFX10-NEXT:    v_mov_b32_e32 v0, 0
300; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen offset:4092
301; GFX10-NEXT:    s_waitcnt vmcnt(0)
302; GFX10-NEXT:    ; return to shader part epilog
303;
304; GFX11-LABEL: buffer_load_voffset_large_12bit:
305; GFX11:       ; %bb.0: ; %main_body
306; GFX11-NEXT:    v_mov_b32_e32 v0, 0
307; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v0, s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offset:4092
308; GFX11-NEXT:    s_waitcnt vmcnt(0)
309; GFX11-NEXT:    ; return to shader part epilog
310main_body:
311  %data = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 4092, i32 0, i32 63, i32 0)
312  ret <4 x float> %data
313}
314
315define amdgpu_ps <4 x float> @tbuffer_load_voffset_large_13bit(ptr addrspace(8) inreg) {
316; PREGFX10-LABEL: tbuffer_load_voffset_large_13bit:
317; PREGFX10:       ; %bb.0: ; %main_body
318; PREGFX10-NEXT:    s_mov_b32 s4, 0
319; PREGFX10-NEXT:    v_mov_b32_e32 v1, 0x1000
320; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
321; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen offen offset:4092
322; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
323; PREGFX10-NEXT:    ; return to shader part epilog
324;
325; GFX10-LABEL: tbuffer_load_voffset_large_13bit:
326; GFX10:       ; %bb.0: ; %main_body
327; GFX10-NEXT:    s_mov_b32 s4, 0
328; GFX10-NEXT:    v_mov_b32_e32 v1, 0x1000
329; GFX10-NEXT:    v_mov_b32_e32 v0, s4
330; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen offen offset:4092
331; GFX10-NEXT:    s_waitcnt vmcnt(0)
332; GFX10-NEXT:    ; return to shader part epilog
333;
334; GFX11-LABEL: tbuffer_load_voffset_large_13bit:
335; GFX11:       ; %bb.0: ; %main_body
336; GFX11-NEXT:    s_mov_b32 s4, 0
337; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
338; GFX11-NEXT:    v_dual_mov_b32 v1, 0x1000 :: v_dual_mov_b32 v0, s4
339; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offen offset:4092
340; GFX11-NEXT:    s_waitcnt vmcnt(0)
341; GFX11-NEXT:    ; return to shader part epilog
342main_body:
343  %data = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 8188, i32 0, i32 63, i32 0)
344  ret <4 x float> %data
345}
346
347define amdgpu_ps <4 x float> @tbuffer_load_voffset_large_16bit(ptr addrspace(8) inreg) {
348; PREGFX10-LABEL: tbuffer_load_voffset_large_16bit:
349; PREGFX10:       ; %bb.0: ; %main_body
350; PREGFX10-NEXT:    s_mov_b32 s4, 0
351; PREGFX10-NEXT:    v_mov_b32_e32 v1, 0xf000
352; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
353; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen offen offset:4092
354; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
355; PREGFX10-NEXT:    ; return to shader part epilog
356;
357; GFX10-LABEL: tbuffer_load_voffset_large_16bit:
358; GFX10:       ; %bb.0: ; %main_body
359; GFX10-NEXT:    s_mov_b32 s4, 0
360; GFX10-NEXT:    v_mov_b32_e32 v1, 0xf000
361; GFX10-NEXT:    v_mov_b32_e32 v0, s4
362; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen offen offset:4092
363; GFX10-NEXT:    s_waitcnt vmcnt(0)
364; GFX10-NEXT:    ; return to shader part epilog
365;
366; GFX11-LABEL: tbuffer_load_voffset_large_16bit:
367; GFX11:       ; %bb.0: ; %main_body
368; GFX11-NEXT:    s_mov_b32 s4, 0
369; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
370; GFX11-NEXT:    v_dual_mov_b32 v1, 0xf000 :: v_dual_mov_b32 v0, s4
371; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offen offset:4092
372; GFX11-NEXT:    s_waitcnt vmcnt(0)
373; GFX11-NEXT:    ; return to shader part epilog
374main_body:
375  %data = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 65532, i32 0, i32 63, i32 0)
376  ret <4 x float> %data
377}
378
379define amdgpu_ps <4 x float> @tbuffer_load_voffset_large_23bit(ptr addrspace(8) inreg) {
380; PREGFX10-LABEL: tbuffer_load_voffset_large_23bit:
381; PREGFX10:       ; %bb.0: ; %main_body
382; PREGFX10-NEXT:    s_mov_b32 s4, 0
383; PREGFX10-NEXT:    v_mov_b32_e32 v1, 0x7ff000
384; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
385; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen offen offset:4092
386; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
387; PREGFX10-NEXT:    ; return to shader part epilog
388;
389; GFX10-LABEL: tbuffer_load_voffset_large_23bit:
390; GFX10:       ; %bb.0: ; %main_body
391; GFX10-NEXT:    s_mov_b32 s4, 0
392; GFX10-NEXT:    v_mov_b32_e32 v1, 0x7ff000
393; GFX10-NEXT:    v_mov_b32_e32 v0, s4
394; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen offen offset:4092
395; GFX10-NEXT:    s_waitcnt vmcnt(0)
396; GFX10-NEXT:    ; return to shader part epilog
397;
398; GFX11-LABEL: tbuffer_load_voffset_large_23bit:
399; GFX11:       ; %bb.0: ; %main_body
400; GFX11-NEXT:    s_mov_b32 s4, 0
401; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
402; GFX11-NEXT:    v_dual_mov_b32 v1, 0x7ff000 :: v_dual_mov_b32 v0, s4
403; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offen offset:4092
404; GFX11-NEXT:    s_waitcnt vmcnt(0)
405; GFX11-NEXT:    ; return to shader part epilog
406main_body:
407  %data = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 8388604, i32 0, i32 63, i32 0)
408  ret <4 x float> %data
409}
410
411define amdgpu_ps <4 x float> @tbuffer_load_voffset_large_24bit(ptr addrspace(8) inreg) {
412; PREGFX10-LABEL: tbuffer_load_voffset_large_24bit:
413; PREGFX10:       ; %bb.0: ; %main_body
414; PREGFX10-NEXT:    s_mov_b32 s4, 0
415; PREGFX10-NEXT:    v_mov_b32_e32 v1, 0xfff000
416; PREGFX10-NEXT:    v_mov_b32_e32 v0, s4
417; PREGFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_DATA_FORMAT_RESERVED_15,BUF_NUM_FORMAT_SSCALED] idxen offen offset:4092
418; PREGFX10-NEXT:    s_waitcnt vmcnt(0)
419; PREGFX10-NEXT:    ; return to shader part epilog
420;
421; GFX10-LABEL: tbuffer_load_voffset_large_24bit:
422; GFX10:       ; %bb.0: ; %main_body
423; GFX10-NEXT:    s_mov_b32 s4, 0
424; GFX10-NEXT:    v_mov_b32_e32 v1, 0xfff000
425; GFX10-NEXT:    v_mov_b32_e32 v0, s4
426; GFX10-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_SINT] idxen offen offset:4092
427; GFX10-NEXT:    s_waitcnt vmcnt(0)
428; GFX10-NEXT:    ; return to shader part epilog
429;
430; GFX11-LABEL: tbuffer_load_voffset_large_24bit:
431; GFX11:       ; %bb.0: ; %main_body
432; GFX11-NEXT:    s_mov_b32 s4, 0
433; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
434; GFX11-NEXT:    v_dual_mov_b32 v1, 0xfff000 :: v_dual_mov_b32 v0, s4
435; GFX11-NEXT:    tbuffer_load_format_xyzw v[0:3], v[0:1], s[0:3], 0 format:[BUF_FMT_32_32_32_32_FLOAT] idxen offen offset:4092
436; GFX11-NEXT:    s_waitcnt vmcnt(0)
437; GFX11-NEXT:    ; return to shader part epilog
438main_body:
439  %data = call <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8) %0, i32 0, i32 16777212, i32 0, i32 63, i32 0)
440  ret <4 x float> %data
441}
442
443declare i32 @llvm.amdgcn.struct.ptr.tbuffer.load.i32(ptr addrspace(8), i32, i32, i32, i32, i32)
444declare <2 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v2i32(ptr addrspace(8), i32, i32, i32, i32, i32)
445declare <4 x i32> @llvm.amdgcn.struct.ptr.tbuffer.load.v4i32(ptr addrspace(8), i32, i32, i32, i32, i32)
446declare <4 x float> @llvm.amdgcn.struct.ptr.tbuffer.load.v4f32(ptr addrspace(8), i32, i32, i32, i32, i32)
447