1;RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=SICI
2;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=VI
3
4;CHECK-LABEL: {{^}}buffer_load:
5;CHECK: buffer_load_dwordx4 v[0:3], {{v[0-9]+}}, s[0:3], 0 idxen
6;CHECK: buffer_load_dwordx4 v[4:7], {{v[0-9]+}}, s[0:3], 0 idxen glc
7;CHECK: buffer_load_dwordx4 v[8:11], {{v[0-9]+}}, s[0:3], 0 idxen slc
8define amdgpu_ps {<4 x float>, <4 x float>, <4 x float>} @buffer_load(<4 x i32> inreg) {
9main_body:
10  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 0)
11  %data_glc = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 1)
12  %data_slc = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 2)
13  %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %data, 0
14  %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %data_glc, 1
15  %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %data_slc, 2
16  ret {<4 x float>, <4 x float>, <4 x float>} %r2
17}
18
19;CHECK-LABEL: {{^}}buffer_load_immoffs:
20;CHECK: buffer_load_dwordx4 v[0:3], {{v[0-9]+}}, s[0:3], 0 idxen offset:40
21define amdgpu_ps <4 x float> @buffer_load_immoffs(<4 x i32> inreg) {
22main_body:
23  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 40, i32 0, i32 0)
24  ret <4 x float> %data
25}
26
27;CHECK-LABEL: {{^}}buffer_load_immoffs_large:
28;CHECK: s_movk_i32 [[OFFSET:s[0-9]+]], 0x1ffc
29;CHECK: buffer_load_dwordx4 v[0:3], {{v[0-9]+}}, s[0:3], [[OFFSET]] idxen offset:4
30define amdgpu_ps <4 x float> @buffer_load_immoffs_large(<4 x i32> inreg) {
31main_body:
32  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 4, i32 8188, i32 0)
33  ret <4 x float> %data
34}
35
36;CHECK-LABEL: {{^}}buffer_load_idx:
37;CHECK: buffer_load_dwordx4 v[0:3], v0, s[0:3], 0 idxen
38define amdgpu_ps <4 x float> @buffer_load_idx(<4 x i32> inreg, i32) {
39main_body:
40  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 %1, i32 0, i32 0, i32 0)
41  ret <4 x float> %data
42}
43
44;CHECK-LABEL: {{^}}buffer_load_ofs:
45;CHECK: buffer_load_dwordx4 v[0:3], v[0:1], s[0:3], 0 idxen offen
46define amdgpu_ps <4 x float> @buffer_load_ofs(<4 x i32> inreg, i32) {
47main_body:
48  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 %1, i32 0, i32 0)
49  ret <4 x float> %data
50}
51
52;CHECK-LABEL: {{^}}buffer_load_ofs_imm:
53;CHECK: buffer_load_dwordx4 v[0:3], v[0:1], s[0:3], 0 idxen offen offset:60
54define amdgpu_ps <4 x float> @buffer_load_ofs_imm(<4 x i32> inreg, i32) {
55main_body:
56  %ofs = add i32 %1, 60
57  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 %ofs, i32 0, i32 0)
58  ret <4 x float> %data
59}
60
61;CHECK-LABEL: {{^}}buffer_load_both:
62;CHECK: buffer_load_dwordx4 v[0:3], v[0:1], s[0:3], 0 idxen offen
63define amdgpu_ps <4 x float> @buffer_load_both(<4 x i32> inreg, i32, i32) {
64main_body:
65  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 %1, i32 %2, i32 0, i32 0)
66  ret <4 x float> %data
67}
68
69;CHECK-LABEL: {{^}}buffer_load_both_reversed:
70;CHECK: v_mov_b32_e32 v2, v0
71;CHECK: buffer_load_dwordx4 v[0:3], v[1:2], s[0:3], 0 idxen offen
72define amdgpu_ps <4 x float> @buffer_load_both_reversed(<4 x i32> inreg, i32, i32) {
73main_body:
74  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 %2, i32 %1, i32 0, i32 0)
75  ret <4 x float> %data
76}
77
78;CHECK-LABEL: {{^}}buffer_load_x1:
79;CHECK: buffer_load_dword v0, v[0:1], s[0:3], 0 idxen offen
80define amdgpu_ps float @buffer_load_x1(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
81main_body:
82  %data = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
83  ret float %data
84}
85
86;CHECK-LABEL: {{^}}buffer_load_x2:
87;CHECK: buffer_load_dwordx2 v[0:1], v[0:1], s[0:3], 0 idxen offen
88define amdgpu_ps <2 x float> @buffer_load_x2(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
89main_body:
90  %data = call <2 x float> @llvm.amdgcn.struct.buffer.load.v2f32(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
91  ret <2 x float> %data
92}
93
94;CHECK-LABEL: {{^}}buffer_load_negative_offset:
95;CHECK: v_add_{{[iu]}}32_e32 {{v[0-9]+}}, vcc, -16, v0
96;CHECK: buffer_load_dwordx4 v[0:3], {{v\[[0-9]+:[0-9]+\]}}, s[0:3], 0 idxen offen
97define amdgpu_ps <4 x float> @buffer_load_negative_offset(<4 x i32> inreg, i32 %ofs) {
98main_body:
99  %ofs.1 = add i32 %ofs, -16
100  %data = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 %ofs.1, i32 0, i32 0)
101  ret <4 x float> %data
102}
103
104; SI won't merge ds memory operations, because of the signed offset bug, so
105; we only have check lines for VI.
106; CHECK-LABEL: buffer_load_mmo:
107; VI: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0
108; VI: ds_write2_b32 v{{[0-9]+}}, [[ZERO]], [[ZERO]] offset1:4
109define amdgpu_ps float @buffer_load_mmo(<4 x i32> inreg %rsrc, float addrspace(3)* %lds) {
110entry:
111  store float 0.0, float addrspace(3)* %lds
112  %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
113  %tmp2 = getelementptr float, float addrspace(3)* %lds, i32 4
114  store float 0.0, float addrspace(3)* %tmp2
115  ret float %val
116}
117
118;CHECK-LABEL: {{^}}buffer_load_int:
119;CHECK: buffer_load_dwordx4 v[0:3], {{v[0-9]+}}, s[0:3], 0 idxen
120;CHECK: buffer_load_dwordx2 v[4:5], {{v[0-9]+}}, s[0:3], 0 idxen glc
121;CHECK: buffer_load_dword v6, {{v[0-9]+}}, s[0:3], 0 idxen slc
122define amdgpu_ps {<4 x float>, <2 x float>, float} @buffer_load_int(<4 x i32> inreg) {
123main_body:
124  %data = call <4 x i32> @llvm.amdgcn.struct.buffer.load.v4i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 0)
125  %data_glc = call <2 x i32> @llvm.amdgcn.struct.buffer.load.v2i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 1)
126  %data_slc = call i32 @llvm.amdgcn.struct.buffer.load.i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 2)
127  %fdata = bitcast <4 x i32> %data to <4 x float>
128  %fdata_glc = bitcast <2 x i32> %data_glc to <2 x float>
129  %fdata_slc = bitcast i32 %data_slc to float
130  %r0 = insertvalue {<4 x float>, <2 x float>, float} undef, <4 x float> %fdata, 0
131  %r1 = insertvalue {<4 x float>, <2 x float>, float} %r0, <2 x float> %fdata_glc, 1
132  %r2 = insertvalue {<4 x float>, <2 x float>, float} %r1, float %fdata_slc, 2
133  ret {<4 x float>, <2 x float>, float} %r2
134}
135
136;CHECK-LABEL: {{^}}struct_buffer_load_ubyte:
137;CHECK-NEXT: %bb.
138;CHECK-NEXT: buffer_load_ubyte v{{[0-9]}}, v[0:1], s[0:3], 0 idxen offen
139;CHECK: s_waitcnt vmcnt(0)
140;CHECK-NEXT: v_cvt_f32_ubyte0_e32 v0, v0
141;CHECK-NEXT: ; return to shader part epilog
142define amdgpu_ps float @struct_buffer_load_ubyte(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
143main_body:
144  %tmp = call i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
145  %tmp2 = zext i8 %tmp to i32
146  %val = uitofp i32 %tmp2 to float
147  ret float %val
148}
149
150;CHECK-LABEL: {{^}}struct_buffer_load_ushort:
151;CHECK-NEXT: %bb.
152;CHECK-NEXT: buffer_load_ushort v{{[0-9]}}, v[0:1], s[0:3], 0 idxen offen
153;CHECK-NEXT: s_waitcnt vmcnt(0)
154;CHECK-NEXT: v_cvt_f32_u32_e32 v0, v0
155;CHECK-NEXT: ; return to shader part epilog
156define amdgpu_ps float @struct_buffer_load_ushort(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
157main_body:
158  %tmp = call i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
159  %tmp2 = zext i16 %tmp to i32
160  %val = uitofp i32 %tmp2 to float
161  ret float %val
162}
163
164;CHECK-LABEL: {{^}}struct_buffer_load_sbyte:
165;CHECK-NEXT: %bb.
166;CHECK-NEXT: buffer_load_sbyte v{{[0-9]}}, v[0:1], s[0:3], 0 idxen offen
167;CHECK-NEXT: s_waitcnt vmcnt(0)
168;CHECK-NEXT: v_cvt_f32_i32_e32 v0, v0
169;CHECK-NEXT: ; return to shader part epilog
170define amdgpu_ps float @struct_buffer_load_sbyte(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
171main_body:
172  %tmp = call i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
173  %tmp2 = sext i8 %tmp to i32
174  %val = sitofp i32 %tmp2 to float
175  ret float %val
176}
177
178;CHECK-LABEL: {{^}}struct_buffer_load_sshort:
179;CHECK-NEXT: %bb.
180;CHECK-NEXT: buffer_load_sshort v{{[0-9]}}, v[0:1], s[0:3], 0 idxen offen
181;CHECK-NEXT: s_waitcnt vmcnt(0)
182;CHECK-NEXT: v_cvt_f32_i32_e32 v0, v0
183;CHECK-NEXT: ; return to shader part epilog
184define amdgpu_ps float @struct_buffer_load_sshort(<4 x i32> inreg %rsrc, i32 %idx, i32 %ofs) {
185main_body:
186  %tmp = call i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32> %rsrc, i32 %idx, i32 %ofs, i32 0, i32 0)
187  %tmp2 = sext i16 %tmp to i32
188  %val = sitofp i32 %tmp2 to float
189  ret float %val
190}
191
192;CHECK-LABEL: {{^}}struct_buffer_load_f16:
193;CHECK-NEXT: %bb.
194;CHECK-NEXT: buffer_load_ushort [[VAL:v[0-9]+]], v1, s[0:3], 0 idxen
195;CHECK: s_waitcnt vmcnt(0)
196;CHECK: ds_write_b16 v0, [[VAL]]
197define amdgpu_ps void @struct_buffer_load_f16(<4 x i32> inreg %rsrc, half addrspace(3)* %ptr, i32 %idx) {
198main_body:
199  %val = call half @llvm.amdgcn.struct.buffer.load.f16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
200  store half %val, half addrspace(3)* %ptr
201  ret void
202}
203
204;CHECK-LABEL: {{^}}struct_buffer_load_v2f16:
205;CHECK-NEXT: %bb.
206;CHECK-NEXT: buffer_load_dword [[VAL:v[0-9]+]], v1, s[0:3], 0 idxen
207;CHECK: s_waitcnt vmcnt(0)
208;CHECK: ds_write_b32 v0, [[VAL]]
209define amdgpu_ps void @struct_buffer_load_v2f16(<4 x i32> inreg %rsrc, <2 x half> addrspace(3)* %ptr, i32 %idx) {
210main_body:
211  %val = call <2 x half> @llvm.amdgcn.struct.buffer.load.v2f16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
212  store <2 x half> %val, <2 x half> addrspace(3)* %ptr
213  ret void
214}
215
216;CHECK-LABEL: {{^}}struct_buffer_load_v4f16:
217;CHECK-NEXT: %bb.
218;CHECK-NEXT: buffer_load_dwordx2 [[VAL:v\[[0-9]+:[0-9]+\]]], v1, s[0:3], 0 idxen
219;CHECK: s_waitcnt vmcnt(0)
220;CHECK: ds_write_b64 v0, [[VAL]]
221define amdgpu_ps void @struct_buffer_load_v4f16(<4 x i32> inreg %rsrc, <4 x half> addrspace(3)* %ptr, i32 %idx) {
222main_body:
223  %val = call <4 x half> @llvm.amdgcn.struct.buffer.load.v4f16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
224  store <4 x half> %val, <4 x half> addrspace(3)* %ptr
225  ret void
226}
227
228;CHECK-LABEL: {{^}}struct_buffer_load_i16:
229;CHECK-NEXT: %bb.
230;CHECK-NEXT: buffer_load_ushort [[VAL:v[0-9]+]], v1, s[0:3], 0 idxen
231;CHECK: s_waitcnt vmcnt(0)
232;CHECK: ds_write_b16 v0, [[VAL]]
233define amdgpu_ps void @struct_buffer_load_i16(<4 x i32> inreg %rsrc, i16 addrspace(3)* %ptr, i32 %idx) {
234main_body:
235  %val = call i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
236  store i16 %val, i16 addrspace(3)* %ptr
237  ret void
238}
239
240;CHECK-LABEL: {{^}}struct_buffer_load_v2i16:
241;CHECK-NEXT: %bb.
242;CHECK-NEXT: buffer_load_dword [[VAL:v[0-9]+]], v1, s[0:3], 0 idxen
243;CHECK: s_waitcnt vmcnt(0)
244;CHECK: ds_write_b32 v0, [[VAL]]
245define amdgpu_ps void @struct_buffer_load_v2i16(<4 x i32> inreg %rsrc, <2 x i16> addrspace(3)* %ptr, i32 %idx) {
246main_body:
247  %val = call <2 x i16> @llvm.amdgcn.struct.buffer.load.v2i16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
248  store <2 x i16> %val, <2 x i16> addrspace(3)* %ptr
249  ret void
250}
251
252;CHECK-LABEL: {{^}}struct_buffer_load_v4i16:
253;CHECK-NEXT: %bb.
254;CHECK-NEXT: buffer_load_dwordx2 [[VAL:v\[[0-9]+:[0-9]+\]]], v1, s[0:3], 0 idxen
255;CHECK: s_waitcnt vmcnt(0)
256;CHECK: ds_write_b64 v0, [[VAL]]
257define amdgpu_ps void @struct_buffer_load_v4i16(<4 x i32> inreg %rsrc, <4 x i16> addrspace(3)* %ptr, i32 %idx) {
258main_body:
259  %val = call <4 x i16> @llvm.amdgcn.struct.buffer.load.v4i16(<4 x i32> %rsrc, i32 %idx, i32 0, i32 0, i32 0)
260  store <4 x i16> %val, <4 x i16> addrspace(3)* %ptr
261  ret void
262}
263
264declare float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32>, i32, i32, i32, i32) #0
265declare <2 x float> @llvm.amdgcn.struct.buffer.load.v2f32(<4 x i32>, i32, i32, i32, i32) #0
266declare <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32>, i32, i32, i32, i32) #0
267declare i32 @llvm.amdgcn.struct.buffer.load.i32(<4 x i32>, i32, i32, i32, i32) #0
268declare <2 x i32> @llvm.amdgcn.struct.buffer.load.v2i32(<4 x i32>, i32, i32, i32, i32) #0
269declare <4 x i32> @llvm.amdgcn.struct.buffer.load.v4i32(<4 x i32>, i32, i32, i32, i32) #0
270declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
271declare i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32>, i32, i32, i32, i32) #0
272
273declare half @llvm.amdgcn.struct.buffer.load.f16(<4 x i32>, i32, i32, i32, i32) #0
274declare <2 x half> @llvm.amdgcn.struct.buffer.load.v2f16(<4 x i32>, i32, i32, i32, i32) #0
275declare <4 x half> @llvm.amdgcn.struct.buffer.load.v4f16(<4 x i32>, i32, i32, i32, i32) #0
276
277declare i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32>, i32, i32, i32, i32) #0
278declare <2 x i16> @llvm.amdgcn.struct.buffer.load.v2i16(<4 x i32>, i32, i32, i32, i32) #0
279declare <4 x i16> @llvm.amdgcn.struct.buffer.load.v4i16(<4 x i32>, i32, i32, i32, i32) #0
280
281attributes #0 = { nounwind readonly }
282