1;RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefixes=SICI,PREGFX10
2;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefixes=VI,PREGFX10
3;RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=GFX10
4
5;CHECK-LABEL: {{^}}buffer_load:
6;CHECK: buffer_load_dwordx4 v[0:3], off, s[0:3], 0{{$}}
7;CHECK: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 glc{{$}}
8;CHECK: buffer_load_dwordx4 v[8:11], off, s[0:3], 0 slc{{$}}
9;CHECK: s_waitcnt
10define amdgpu_ps {<4 x float>, <4 x float>, <4 x float>} @buffer_load(<4 x i32> inreg) {
11main_body:
12  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 0)
13  %data_glc = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 1)
14  %data_slc = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 2)
15  %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %data, 0
16  %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %data_glc, 1
17  %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %data_slc, 2
18  ret {<4 x float>, <4 x float>, <4 x float>} %r2
19}
20
21;CHECK-LABEL: {{^}}buffer_load_dlc:
22;PREGFX10: buffer_load_dwordx4 v[0:3], off, s[0:3], 0{{$}}
23;PREGFX10: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 glc{{$}}
24;PREGFX10: buffer_load_dwordx4 v[8:11], off, s[0:3], 0 slc{{$}}
25;GFX10: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 dlc{{$}}
26;GFX10: buffer_load_dwordx4 v[4:7], off, s[0:3], 0 glc dlc{{$}}
27;GFX10: buffer_load_dwordx4 v[8:11], off, s[0:3], 0 slc dlc{{$}}
28;CHECK: s_waitcnt
29define amdgpu_ps {<4 x float>, <4 x float>, <4 x float>} @buffer_load_dlc(<4 x i32> inreg) {
30main_body:
31  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 4)
32  %data_glc = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 5)
33  %data_slc = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 6)
34  %r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %data, 0
35  %r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %data_glc, 1
36  %r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %data_slc, 2
37  ret {<4 x float>, <4 x float>, <4 x float>} %r2
38}
39
40;CHECK-LABEL: {{^}}buffer_load_immoffs:
41;CHECK: buffer_load_dwordx4 v[0:3], off, s[0:3], 0 offset:40
42;CHECK: s_waitcnt
43define amdgpu_ps <4 x float> @buffer_load_immoffs(<4 x i32> inreg) {
44main_body:
45  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 40, i32 0, i32 0)
46  ret <4 x float> %data
47}
48
49;CHECK-LABEL: {{^}}buffer_load_immoffs_large:
50;CHECK: s_movk_i32 [[OFFSET:s[0-9]+]], 0x1ffc
51;CHECK: buffer_load_dwordx4 v[0:3], off, s[0:3], [[OFFSET]] offset:4
52;CHECK: s_waitcnt
53define amdgpu_ps <4 x float> @buffer_load_immoffs_large(<4 x i32> inreg) {
54main_body:
55  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 4, i32 8188, i32 0)
56  ret <4 x float> %data
57}
58
59;CHECK-LABEL: {{^}}buffer_load_ofs:
60;CHECK: buffer_load_dwordx4 v[0:3], v0, s[0:3], 0 offen
61;CHECK: s_waitcnt
62define amdgpu_ps <4 x float> @buffer_load_ofs(<4 x i32> inreg, i32) {
63main_body:
64  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 %1, i32 0, i32 0)
65  ret <4 x float> %data
66}
67
68;CHECK-LABEL: {{^}}buffer_load_ofs_imm:
69;CHECK: buffer_load_dwordx4 v[0:3], v0, s[0:3], 0 offen offset:60
70;CHECK: s_waitcnt
71define amdgpu_ps <4 x float> @buffer_load_ofs_imm(<4 x i32> inreg, i32) {
72main_body:
73  %ofs = add i32 %1, 60
74  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 %ofs, i32 0, i32 0)
75  ret <4 x float> %data
76}
77
78;CHECK-LABEL: {{^}}buffer_load_x1:
79;CHECK: buffer_load_dword v0, v0, s[0:3], 0 offen
80;CHECK: s_waitcnt
81define amdgpu_ps float @buffer_load_x1(<4 x i32> inreg %rsrc, i32 %ofs) {
82main_body:
83  %data = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %ofs, i32 0, i32 0)
84  ret float %data
85}
86
87;CHECK-LABEL: {{^}}buffer_load_x2:
88;CHECK: buffer_load_dwordx2 v[0:1], v0, s[0:3], 0 offen
89;CHECK: s_waitcnt
90define amdgpu_ps <2 x float> @buffer_load_x2(<4 x i32> inreg %rsrc, i32 %ofs) {
91main_body:
92  %data = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %ofs, i32 0, i32 0)
93  ret <2 x float> %data
94}
95
96;CHECK-LABEL: {{^}}buffer_load_negative_offset:
97;PREGFX10: v_add_{{[iu]}}32_e32 [[VOFS:v[0-9]+]], vcc, -16, v0
98;GFX10: v_add_nc_{{[iu]}}32_e32 [[VOFS:v[0-9]+]], -16, v0
99;CHECK: buffer_load_dwordx4 v[0:3], [[VOFS]], s[0:3], 0 offen
100define amdgpu_ps <4 x float> @buffer_load_negative_offset(<4 x i32> inreg, i32 %ofs) {
101main_body:
102  %ofs.1 = add i32 %ofs, -16
103  %data = call <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32> %0, i32 %ofs.1, i32 0, i32 0)
104  ret <4 x float> %data
105}
106
107; SI won't merge ds memory operations, because of the signed offset bug, so
108; we only have check lines for VI.
109; CHECK-LABEL: buffer_load_mmo:
110; VI: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0
111; VI: ds_write2_b32 v{{[0-9]+}}, [[ZERO]], [[ZERO]] offset1:4
112define amdgpu_ps float @buffer_load_mmo(<4 x i32> inreg %rsrc, float addrspace(3)* %lds) {
113entry:
114  store float 0.0, float addrspace(3)* %lds
115  %val = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
116  %tmp2 = getelementptr float, float addrspace(3)* %lds, i32 4
117  store float 0.0, float addrspace(3)* %tmp2
118  ret float %val
119}
120
121;CHECK-LABEL: {{^}}buffer_load_x1_offen_merged_and:
122;CHECK-NEXT: %bb.
123;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:4
124;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:28
125;CHECK: s_waitcnt
126define amdgpu_ps void @buffer_load_x1_offen_merged_and(<4 x i32> inreg %rsrc, i32 %a) {
127main_body:
128  %a1 = add i32 %a, 4
129  %a2 = add i32 %a, 8
130  %a3 = add i32 %a, 12
131  %a4 = add i32 %a, 16
132  %a5 = add i32 %a, 28
133  %a6 = add i32 %a, 32
134  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a1, i32 0, i32 0)
135  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a2, i32 0, i32 0)
136  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a3, i32 0, i32 0)
137  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a4, i32 0, i32 0)
138  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a5, i32 0, i32 0)
139  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a6, i32 0, i32 0)
140  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
141  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
142  ret void
143}
144
145;CHECK-LABEL: {{^}}buffer_load_x1_offen_merged_or:
146;CHECK-NEXT: %bb.
147;CHECK-NEXT: v_lshlrev_b32_e32 v{{[0-9]}}, 6, v0
148;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], v{{[0-9]}}, s[0:3], 0 offen offset:4
149;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], v{{[0-9]}}, s[0:3], 0 offen offset:28
150;CHECK: s_waitcnt
151define amdgpu_ps void @buffer_load_x1_offen_merged_or(<4 x i32> inreg %rsrc, i32 %inp) {
152main_body:
153  %a = shl i32 %inp, 6
154  %a1 = or i32 %a, 4
155  %a2 = or i32 %a, 8
156  %a3 = or i32 %a, 12
157  %a4 = or i32 %a, 16
158  %a5 = or i32 %a, 28
159  %a6 = or i32 %a, 32
160  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a1, i32 0, i32 0)
161  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a2, i32 0, i32 0)
162  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a3, i32 0, i32 0)
163  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a4, i32 0, i32 0)
164  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a5, i32 0, i32 0)
165  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a6, i32 0, i32 0)
166  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
167  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
168  ret void
169}
170
171;CHECK-LABEL: {{^}}buffer_load_x1_offen_merged_glc_slc:
172;CHECK-NEXT: %bb.
173;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:4{{$}}
174;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:12 glc{{$}}
175;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:28 glc slc{{$}}
176;CHECK: s_waitcnt
177define amdgpu_ps void @buffer_load_x1_offen_merged_glc_slc(<4 x i32> inreg %rsrc, i32 %a) {
178main_body:
179  %a1 = add i32 %a, 4
180  %a2 = add i32 %a, 8
181  %a3 = add i32 %a, 12
182  %a4 = add i32 %a, 16
183  %a5 = add i32 %a, 28
184  %a6 = add i32 %a, 32
185  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a1, i32 0, i32 0)
186  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a2, i32 0, i32 0)
187  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a3, i32 0, i32 1)
188  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a4, i32 0, i32 1)
189  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a5, i32 0, i32 3)
190  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 %a6, i32 0, i32 3)
191  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
192  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
193  ret void
194}
195
196;CHECK-LABEL: {{^}}buffer_load_x2_offen_merged_and:
197;CHECK-NEXT: %bb.
198;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], v0, s[0:3], 0 offen offset:4
199;CHECK: s_waitcnt
200define amdgpu_ps void @buffer_load_x2_offen_merged_and(<4 x i32> inreg %rsrc, i32 %a) {
201main_body:
202  %a1 = add i32 %a, 4
203  %a2 = add i32 %a, 12
204  %vr1 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %a1, i32 0, i32 0)
205  %vr2 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %a2, i32 0, i32 0)
206  %r1 = extractelement <2 x float> %vr1, i32 0
207  %r2 = extractelement <2 x float> %vr1, i32 1
208  %r3 = extractelement <2 x float> %vr2, i32 0
209  %r4 = extractelement <2 x float> %vr2, i32 1
210  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
211  ret void
212}
213
214;CHECK-LABEL: {{^}}buffer_load_x2_offen_merged_or:
215;CHECK-NEXT: %bb.
216;CHECK-NEXT: v_lshlrev_b32_e32 v{{[0-9]}}, 4, v0
217;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], v{{[0-9]}}, s[0:3], 0 offen offset:4
218;CHECK: s_waitcnt
219define amdgpu_ps void @buffer_load_x2_offen_merged_or(<4 x i32> inreg %rsrc, i32 %inp) {
220main_body:
221  %a = shl i32 %inp, 4
222  %a1 = add i32 %a, 4
223  %a2 = add i32 %a, 12
224  %vr1 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %a1, i32 0, i32 0)
225  %vr2 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 %a2, i32 0, i32 0)
226  %r1 = extractelement <2 x float> %vr1, i32 0
227  %r2 = extractelement <2 x float> %vr1, i32 1
228  %r3 = extractelement <2 x float> %vr2, i32 0
229  %r4 = extractelement <2 x float> %vr2, i32 1
230  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
231  ret void
232}
233
234;CHECK-LABEL: {{^}}buffer_load_x1_offset_merged:
235;CHECK-NEXT: %bb.
236;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], off, s[0:3], 0 offset:4
237;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], off, s[0:3], 0 offset:28
238;CHECK: s_waitcnt
239define amdgpu_ps void @buffer_load_x1_offset_merged(<4 x i32> inreg %rsrc) {
240main_body:
241  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 4, i32 0, i32 0)
242  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 8, i32 0, i32 0)
243  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 12, i32 0, i32 0)
244  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 16, i32 0, i32 0)
245  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 28, i32 0, i32 0)
246  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 32, i32 0, i32 0)
247  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
248  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
249  ret void
250}
251
252;CHECK-LABEL: {{^}}buffer_load_x2_offset_merged:
253;CHECK-NEXT: %bb.
254;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], off, s[0:3], 0 offset:4
255;CHECK: s_waitcnt
256define amdgpu_ps void @buffer_load_x2_offset_merged(<4 x i32> inreg %rsrc) {
257main_body:
258  %vr1 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 4, i32 0, i32 0)
259  %vr2 = call <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32> %rsrc, i32 12, i32 0, i32 0)
260  %r1 = extractelement <2 x float> %vr1, i32 0
261  %r2 = extractelement <2 x float> %vr1, i32 1
262  %r3 = extractelement <2 x float> %vr2, i32 0
263  %r4 = extractelement <2 x float> %vr2, i32 1
264  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
265  ret void
266}
267
268;CHECK-LABEL: {{^}}buffer_load_int:
269;CHECK: buffer_load_dwordx4 v[0:3], off, s[0:3], 0
270;CHECK: buffer_load_dwordx2 v[4:5], off, s[0:3], 0 glc
271;CHECK: buffer_load_dword v6, off, s[0:3], 0 slc
272;CHECK: s_waitcnt
273define amdgpu_ps {<4 x float>, <2 x float>, float} @buffer_load_int(<4 x i32> inreg) {
274main_body:
275  %data = call <4 x i32> @llvm.amdgcn.raw.buffer.load.v4i32(<4 x i32> %0, i32 0, i32 0, i32 0)
276  %data_glc = call <2 x i32> @llvm.amdgcn.raw.buffer.load.v2i32(<4 x i32> %0, i32 0, i32 0, i32 1)
277  %data_slc = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %0, i32 0, i32 0, i32 2)
278  %fdata = bitcast <4 x i32> %data to <4 x float>
279  %fdata_glc = bitcast <2 x i32> %data_glc to <2 x float>
280  %fdata_slc = bitcast i32 %data_slc to float
281  %r0 = insertvalue {<4 x float>, <2 x float>, float} undef, <4 x float> %fdata, 0
282  %r1 = insertvalue {<4 x float>, <2 x float>, float} %r0, <2 x float> %fdata_glc, 1
283  %r2 = insertvalue {<4 x float>, <2 x float>, float} %r1, float %fdata_slc, 2
284  ret {<4 x float>, <2 x float>, float} %r2
285}
286
287;CHECK-LABEL: {{^}}raw_buffer_load_ubyte:
288;CHECK-NEXT: %bb.
289;CHECK-NEXT: buffer_load_ubyte v{{[0-9]}}, off, s[0:3], 0
290;CHECK: s_waitcnt vmcnt(0)
291;CHECK-NEXT: v_cvt_f32_ubyte0_e32 v0, v0
292;CHECK-NEXT: ; return to shader part epilog
293define amdgpu_ps float @raw_buffer_load_ubyte(<4 x i32> inreg %rsrc) {
294main_body:
295  %tmp = call i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
296  %tmp2 = zext i8 %tmp to i32
297  %val = uitofp i32 %tmp2 to float
298  ret float %val
299}
300
301;CHECK-LABEL: {{^}}raw_buffer_load_i16:
302;CHECK-NEXT: %bb.
303;CHECK-NEXT: buffer_load_ushort v{{[0-9]}}, off, s[0:3], 0
304;CHECK: s_waitcnt vmcnt(0)
305;CHECK-NEXT: v_cvt_f32_u32_e32 v0, v0
306;CHECK-NEXT: ; return to shader part epilog
307define amdgpu_ps float @raw_buffer_load_i16(<4 x i32> inreg %rsrc) {
308main_body:
309  %tmp = call i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
310  %tmp2 = zext i16 %tmp to i32
311  %val = uitofp i32 %tmp2 to float
312  ret float %val
313}
314
315;CHECK-LABEL: {{^}}raw_buffer_load_sbyte:
316;CHECK-NEXT: %bb.
317;CHECK-NEXT: buffer_load_sbyte v{{[0-9]}}, off, s[0:3], 0
318;CHECK: s_waitcnt vmcnt(0)
319;CHECK-NEXT: v_cvt_f32_i32_e32 v0, v0
320;CHECK-NEXT: ; return to shader part epilog
321define amdgpu_ps float @raw_buffer_load_sbyte(<4 x i32> inreg %rsrc) {
322main_body:
323  %tmp = call i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
324  %tmp2 = sext i8 %tmp to i32
325  %val = sitofp i32 %tmp2 to float
326  ret float %val
327}
328
329;CHECK-LABEL: {{^}}raw_buffer_load_sshort:
330;CHECK-NEXT: %bb.
331;CHECK-NEXT: buffer_load_sshort v{{[0-9]}}, off, s[0:3], 0
332;CHECK: s_waitcnt vmcnt(0)
333;CHECK-NEXT: v_cvt_f32_i32_e32 v0, v0
334;CHECK-NEXT: ; return to shader part epilog
335define amdgpu_ps float @raw_buffer_load_sshort(<4 x i32> inreg %rsrc) {
336main_body:
337  %tmp = call i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
338  %tmp2 = sext i16 %tmp to i32
339  %val = sitofp i32 %tmp2 to float
340  ret float %val
341}
342
343;CHECK-LABEL: {{^}}raw_buffer_load_f16:
344;CHECK-NEXT: %bb.
345;CHECK-NEXT: buffer_load_ushort [[VAL:v[0-9]+]], off, s[0:3], 0
346;CHECK: s_waitcnt vmcnt(0)
347;CHECK: ds_write_b16 v0, [[VAL]]
348define amdgpu_ps void @raw_buffer_load_f16(<4 x i32> inreg %rsrc, half addrspace(3)* %ptr) {
349main_body:
350  %val = call half @llvm.amdgcn.raw.buffer.load.f16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
351  store half %val, half addrspace(3)* %ptr
352  ret void
353}
354
355;CHECK-LABEL: {{^}}raw_buffer_load_v2f16:
356;CHECK-NEXT: %bb.
357;CHECK-NEXT: buffer_load_dword [[VAL:v[0-9]+]], off, s[0:3], 0
358;CHECK: s_waitcnt vmcnt(0)
359;CHECK: ds_write_b32 v0, [[VAL]]
360define amdgpu_ps void @raw_buffer_load_v2f16(<4 x i32> inreg %rsrc, <2 x half> addrspace(3)* %ptr) {
361main_body:
362  %val = call <2 x half> @llvm.amdgcn.raw.buffer.load.v2f16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
363  store <2 x half> %val, <2 x half> addrspace(3)* %ptr
364  ret void
365}
366
367;CHECK-LABEL: {{^}}raw_buffer_load_v4f16:
368;CHECK-NEXT: %bb.
369;CHECK-NEXT: buffer_load_dwordx2 [[VAL:v\[[0-9]+:[0-9]+\]]], off, s[0:3], 0
370;CHECK: s_waitcnt vmcnt(0)
371;CHECK: ds_write_b64 v0, [[VAL]]
372define amdgpu_ps void @raw_buffer_load_v4f16(<4 x i32> inreg %rsrc, <4 x half> addrspace(3)* %ptr) {
373main_body:
374  %val = call <4 x half> @llvm.amdgcn.raw.buffer.load.v4f16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
375  store <4 x half> %val, <4 x half> addrspace(3)* %ptr
376  ret void
377}
378
379;CHECK-LABEL: {{^}}raw_buffer_load_v2i16:
380;CHECK-NEXT: %bb.
381;CHECK-NEXT: buffer_load_dword [[VAL:v[0-9]+]], off, s[0:3], 0
382;CHECK: s_waitcnt vmcnt(0)
383;CHECK: ds_write_b32 v0, [[VAL]]
384define amdgpu_ps void @raw_buffer_load_v2i16(<4 x i32> inreg %rsrc, <2 x i16> addrspace(3)* %ptr) {
385main_body:
386  %val = call <2 x i16> @llvm.amdgcn.raw.buffer.load.v2i16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
387  store <2 x i16> %val, <2 x i16> addrspace(3)* %ptr
388  ret void
389}
390
391;CHECK-LABEL: {{^}}raw_buffer_load_v4i16:
392;CHECK-NEXT: %bb.
393;CHECK-NEXT: buffer_load_dwordx2 [[VAL:v\[[0-9]+:[0-9]+\]]], off, s[0:3], 0
394;CHECK: s_waitcnt vmcnt(0)
395;CHECK: ds_write_b64 v0, [[VAL]]
396define amdgpu_ps void @raw_buffer_load_v4i16(<4 x i32> inreg %rsrc, <4 x i16> addrspace(3)* %ptr) {
397main_body:
398  %val = call <4 x i16> @llvm.amdgcn.raw.buffer.load.v4i16(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
399  store <4 x i16> %val, <4 x i16> addrspace(3)* %ptr
400  ret void
401}
402
403;CHECK-LABEL: {{^}}raw_buffer_load_x1_offset_merged:
404;CHECK-NEXT: %bb.
405;CHECK-NEXT: buffer_load_dwordx4 v[{{[0-9]}}:{{[0-9]}}], off, s[0:3], 0 offset:4
406;CHECK-NEXT: buffer_load_dwordx2 v[{{[0-9]}}:{{[0-9]}}], off, s[0:3], 0 offset:28
407;CHECK: s_waitcnt
408define amdgpu_ps void @raw_buffer_load_x1_offset_merged(<4 x i32> inreg %rsrc) {
409main_body:
410  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 4, i32 0, i32 0)
411  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 8, i32 0, i32 0)
412  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 12, i32 0, i32 0)
413  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 16, i32 0, i32 0)
414  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 28, i32 0, i32 0)
415  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 32, i32 0, i32 0)
416  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
417  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
418  ret void
419}
420
421;CHECK-LABEL: {{^}}raw_buffer_load_x1_offset_swizzled_not_merged:
422;CHECK-NEXT: %bb.
423;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:4
424;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:8
425;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:12
426;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:16
427;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:28
428;CHECK-NEXT: buffer_load_dword v{{[0-9]}}, off, s[0:3], 0 offset:32
429;CHECK: s_waitcnt
430define amdgpu_ps void @raw_buffer_load_x1_offset_swizzled_not_merged(<4 x i32> inreg %rsrc) {
431main_body:
432  %r1 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 4, i32 0, i32 8)
433  %r2 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 8, i32 0, i32 8)
434  %r3 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 12, i32 0, i32 8)
435  %r4 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 16, i32 0, i32 8)
436  %r5 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 28, i32 0, i32 8)
437  %r6 = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %rsrc, i32 32, i32 0, i32 8)
438  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r1, float %r2, float %r3, float %r4, i1 true, i1 true)
439  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %r5, float %r6, float undef, float undef, i1 true, i1 true)
440  ret void
441}
442
443declare float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32>, i32, i32, i32) #0
444declare <2 x float> @llvm.amdgcn.raw.buffer.load.v2f32(<4 x i32>, i32, i32, i32) #0
445declare <4 x float> @llvm.amdgcn.raw.buffer.load.v4f32(<4 x i32>, i32, i32, i32) #0
446declare i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32>, i32, i32, i32) #0
447declare <2 x i32> @llvm.amdgcn.raw.buffer.load.v2i32(<4 x i32>, i32, i32, i32) #0
448declare <4 x i32> @llvm.amdgcn.raw.buffer.load.v4i32(<4 x i32>, i32, i32, i32) #0
449declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
450declare i8 @llvm.amdgcn.raw.buffer.load.i8(<4 x i32>, i32, i32, i32) #0
451declare i16 @llvm.amdgcn.raw.buffer.load.i16(<4 x i32>, i32, i32, i32) #0
452declare <2 x i16> @llvm.amdgcn.raw.buffer.load.v2i16(<4 x i32>, i32, i32, i32) #0
453declare <4 x i16> @llvm.amdgcn.raw.buffer.load.v4i16(<4 x i32>, i32, i32, i32) #0
454declare half @llvm.amdgcn.raw.buffer.load.f16(<4 x i32>, i32, i32, i32) #0
455declare <2 x half> @llvm.amdgcn.raw.buffer.load.v2f16(<4 x i32>, i32, i32, i32) #0
456declare <4 x half> @llvm.amdgcn.raw.buffer.load.v4f16(<4 x i32>, i32, i32, i32) #0
457
458attributes #0 = { nounwind readonly }
459