xref: /llvm-project/llvm/test/CodeGen/AMDGPU/sgpr-copy.ll (revision 9e9907f1cfa424366fba58d9520f9305b537cec9)
1; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck %s
2; RUN:  llc -amdgpu-scalarize-global-loads=false  -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck %s
3
4; CHECK-LABEL: {{^}}phi1:
5; CHECK: s_buffer_load_dword [[DST:s[0-9]]], {{s\[[0-9]+:[0-9]+\]}}, 0x0
6; CHECK: ; %bb.1: ; %ELSE
7; CHECK: s_xor_b32 s{{[0-9]}}, [[DST]]
8define amdgpu_ps void @phi1(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19) #0 {
9main_body:
10  %tmp20 = load <4 x i32>, ptr addrspace(4) %arg, !tbaa !0
11  %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 0, i32 0)
12  %tmp22 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 16, i32 0)
13  %tmp23 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 32, i32 0)
14  %tmp24 = fptosi float %tmp22 to i32
15  %tmp25 = icmp ne i32 %tmp24, 0
16  br i1 %tmp25, label %ENDIF, label %ELSE
17
18ELSE:                                             ; preds = %main_body
19  %tmp26 = fsub float -0.000000e+00, %tmp21
20  br label %ENDIF
21
22ENDIF:                                            ; preds = %ELSE, %main_body
23  %temp.0 = phi float [ %tmp26, %ELSE ], [ %tmp21, %main_body ]
24  %tmp27 = fadd float %temp.0, %tmp23
25  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp27, float %tmp27, float 0.000000e+00, float 1.000000e+00, i1 true, i1 true) #0
26  ret void
27}
28
29; Make sure this program doesn't crash
30; CHECK-LABEL: {{^}}phi2:
31define amdgpu_ps void @phi2(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19) #1 {
32main_body:
33  %tmp20 = load <4 x i32>, ptr addrspace(4) %arg, !tbaa !0
34  %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 16, i32 0)
35  %tmp22 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 32, i32 0)
36  %tmp23 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 36, i32 0)
37  %tmp24 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 40, i32 0)
38  %tmp25 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 48, i32 0)
39  %tmp26 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 52, i32 0)
40  %tmp27 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 56, i32 0)
41  %tmp28 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 64, i32 0)
42  %tmp29 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 68, i32 0)
43  %tmp30 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 72, i32 0)
44  %tmp31 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 76, i32 0)
45  %tmp32 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 80, i32 0)
46  %tmp33 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 84, i32 0)
47  %tmp34 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 88, i32 0)
48  %tmp35 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 92, i32 0)
49  %tmp37 = load <8 x i32>, ptr addrspace(4) %arg2, !tbaa !0
50  %tmp39 = load <4 x i32>, ptr addrspace(4) %arg1, !tbaa !0
51  %i.i = extractelement <2 x i32> %arg5, i32 0
52  %j.i = extractelement <2 x i32> %arg5, i32 1
53  %i.f.i = bitcast i32 %i.i to float
54  %j.f.i = bitcast i32 %j.i to float
55  %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 0, i32 0, i32 %arg3) #1
56  %p2.i = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 0, i32 0, i32 %arg3) #1
57  %i.i19 = extractelement <2 x i32> %arg5, i32 0
58  %j.i20 = extractelement <2 x i32> %arg5, i32 1
59  %i.f.i21 = bitcast i32 %i.i19 to float
60  %j.f.i22 = bitcast i32 %j.i20 to float
61  %p1.i23 = call float @llvm.amdgcn.interp.p1(float %i.f.i21, i32 1, i32 0, i32 %arg3) #1
62  %p2.i24 = call float @llvm.amdgcn.interp.p2(float %p1.i23, float %j.f.i22, i32 1, i32 0, i32 %arg3) #1
63  %i.i13 = extractelement <2 x i32> %arg5, i32 0
64  %j.i14 = extractelement <2 x i32> %arg5, i32 1
65  %i.f.i15 = bitcast i32 %i.i13 to float
66  %j.f.i16 = bitcast i32 %j.i14 to float
67  %p1.i17 = call float @llvm.amdgcn.interp.p1(float %i.f.i15, i32 0, i32 1, i32 %arg3) #1
68  %p2.i18 = call float @llvm.amdgcn.interp.p2(float %p1.i17, float %j.f.i16, i32 0, i32 1, i32 %arg3) #1
69  %i.i7 = extractelement <2 x i32> %arg5, i32 0
70  %j.i8 = extractelement <2 x i32> %arg5, i32 1
71  %i.f.i9 = bitcast i32 %i.i7 to float
72  %j.f.i10 = bitcast i32 %j.i8 to float
73  %p1.i11 = call float @llvm.amdgcn.interp.p1(float %i.f.i9, i32 1, i32 1, i32 %arg3) #1
74  %p2.i12 = call float @llvm.amdgcn.interp.p2(float %p1.i11, float %j.f.i10, i32 1, i32 1, i32 %arg3) #1
75  %i.i1 = extractelement <2 x i32> %arg5, i32 0
76  %j.i2 = extractelement <2 x i32> %arg5, i32 1
77  %i.f.i3 = bitcast i32 %i.i1 to float
78  %j.f.i4 = bitcast i32 %j.i2 to float
79  %p1.i5 = call float @llvm.amdgcn.interp.p1(float %i.f.i3, i32 2, i32 1, i32 %arg3) #1
80  %p2.i6 = call float @llvm.amdgcn.interp.p2(float %p1.i5, float %j.f.i4, i32 2, i32 1, i32 %arg3) #1
81  %tmp39.bc = bitcast <4 x i32> %tmp39 to <4 x i32>
82  %tmp1 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %p2.i, float %p2.i24, <8 x i32> %tmp37, <4 x i32> %tmp39.bc, i1 0, i32 0, i32 0)
83  %tmp50 = extractelement <4 x float> %tmp1, i32 2
84  %tmp51 = call float @llvm.fabs.f32(float %tmp50)
85  %tmp52 = fmul float %p2.i18, %p2.i18
86  %tmp53 = fmul float %p2.i12, %p2.i12
87  %tmp54 = fadd float %tmp53, %tmp52
88  %tmp55 = fmul float %p2.i6, %p2.i6
89  %tmp56 = fadd float %tmp54, %tmp55
90  %tmp57 = call float @llvm.amdgcn.rsq.f32(float %tmp56)
91  %tmp58 = fmul float %p2.i18, %tmp57
92  %tmp59 = fmul float %p2.i12, %tmp57
93  %tmp60 = fmul float %p2.i6, %tmp57
94  %tmp61 = fmul float %tmp58, %tmp22
95  %tmp62 = fmul float %tmp59, %tmp23
96  %tmp63 = fadd float %tmp62, %tmp61
97  %tmp64 = fmul float %tmp60, %tmp24
98  %tmp65 = fadd float %tmp63, %tmp64
99  %tmp66 = fsub float -0.000000e+00, %tmp25
100  %tmp67 = fmul float %tmp65, %tmp51
101  %tmp68 = fadd float %tmp67, %tmp66
102  %tmp69 = fmul float %tmp26, %tmp68
103  %tmp70 = fmul float %tmp27, %tmp68
104  %tmp71 = call float @llvm.fabs.f32(float %tmp69)
105  %tmp72 = fcmp olt float 0x3EE4F8B580000000, %tmp71
106  %tmp73 = sext i1 %tmp72 to i32
107  %tmp74 = bitcast i32 %tmp73 to float
108  %tmp75 = bitcast float %tmp74 to i32
109  %tmp76 = icmp ne i32 %tmp75, 0
110  br i1 %tmp76, label %IF, label %ENDIF
111
112IF:                                               ; preds = %main_body
113  %tmp77 = fsub float -0.000000e+00, %tmp69
114  %tmp78 = call float @llvm.exp2.f32(float %tmp77)
115  %tmp79 = fsub float -0.000000e+00, %tmp78
116  %tmp80 = fadd float 1.000000e+00, %tmp79
117  %tmp81 = fdiv float 1.000000e+00, %tmp69
118  %tmp82 = fmul float %tmp80, %tmp81
119  %tmp83 = fmul float %tmp31, %tmp82
120  br label %ENDIF
121
122ENDIF:                                            ; preds = %IF, %main_body
123  %temp4.0 = phi float [ %tmp83, %IF ], [ %tmp31, %main_body ]
124  %tmp84 = call float @llvm.fabs.f32(float %tmp70)
125  %tmp85 = fcmp olt float 0x3EE4F8B580000000, %tmp84
126  %tmp86 = sext i1 %tmp85 to i32
127  %tmp87 = bitcast i32 %tmp86 to float
128  %tmp88 = bitcast float %tmp87 to i32
129  %tmp89 = icmp ne i32 %tmp88, 0
130  br i1 %tmp89, label %IF25, label %ENDIF24
131
132IF25:                                             ; preds = %ENDIF
133  %tmp90 = fsub float -0.000000e+00, %tmp70
134  %tmp91 = call float @llvm.exp2.f32(float %tmp90)
135  %tmp92 = fsub float -0.000000e+00, %tmp91
136  %tmp93 = fadd float 1.000000e+00, %tmp92
137  %tmp94 = fdiv float 1.000000e+00, %tmp70
138  %tmp95 = fmul float %tmp93, %tmp94
139  %tmp96 = fmul float %tmp35, %tmp95
140  br label %ENDIF24
141
142ENDIF24:                                          ; preds = %IF25, %ENDIF
143  %temp8.0 = phi float [ %tmp96, %IF25 ], [ %tmp35, %ENDIF ]
144  %tmp97 = fmul float %tmp28, %temp4.0
145  %tmp98 = fmul float %tmp29, %temp4.0
146  %tmp99 = fmul float %tmp30, %temp4.0
147  %tmp100 = fmul float %tmp32, %temp8.0
148  %tmp101 = fadd float %tmp100, %tmp97
149  %tmp102 = fmul float %tmp33, %temp8.0
150  %tmp103 = fadd float %tmp102, %tmp98
151  %tmp104 = fmul float %tmp34, %temp8.0
152  %tmp105 = fadd float %tmp104, %tmp99
153  %tmp106 = call float @llvm.pow.f32(float %tmp51, float %tmp21)
154  %tmp107 = fsub float -0.000000e+00, %tmp101
155  %tmp108 = fmul float %tmp107, %tmp106
156  %tmp109 = fsub float -0.000000e+00, %tmp103
157  %tmp110 = fmul float %tmp109, %tmp106
158  %tmp111 = fsub float -0.000000e+00, %tmp105
159  %tmp112 = fmul float %tmp111, %tmp106
160  %tmp113 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp108, float %tmp110)
161  %tmp115 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp112, float 1.000000e+00)
162  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp113, <2 x half> %tmp115, i1 true, i1 true) #0
163  ret void
164}
165
166; We just want to make sure the program doesn't crash
167; CHECK-LABEL: {{^}}loop:
168define amdgpu_ps void @loop(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, i32 inreg %arg3, <2 x i32> %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <3 x i32> %arg7, <2 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, float %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19) #0 {
169main_body:
170  %tmp20 = load <4 x i32>, ptr addrspace(4) %arg, !tbaa !0
171  %tmp21 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 0, i32 0)
172  %tmp22 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 4, i32 0)
173  %tmp23 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 8, i32 0)
174  %tmp24 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp20, i32 12, i32 0)
175  %tmp25 = fptosi float %tmp24 to i32
176  %tmp26 = bitcast i32 %tmp25 to float
177  %tmp27 = bitcast float %tmp26 to i32
178  br label %LOOP
179
180LOOP:                                             ; preds = %ENDIF, %main_body
181  %temp4.0 = phi float [ %tmp21, %main_body ], [ %temp5.0, %ENDIF ]
182  %temp5.0 = phi float [ %tmp22, %main_body ], [ %temp6.0, %ENDIF ]
183  %temp6.0 = phi float [ %tmp23, %main_body ], [ %temp4.0, %ENDIF ]
184  %temp8.0 = phi float [ 0.000000e+00, %main_body ], [ %tmp36, %ENDIF ]
185  %tmp28 = bitcast float %temp8.0 to i32
186  %tmp29 = icmp sge i32 %tmp28, %tmp27
187  %tmp30 = sext i1 %tmp29 to i32
188  %tmp31 = bitcast i32 %tmp30 to float
189  %tmp32 = bitcast float %tmp31 to i32
190  %tmp33 = icmp ne i32 %tmp32, 0
191  br i1 %tmp33, label %IF, label %ENDIF
192
193IF:                                               ; preds = %LOOP
194  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %temp4.0, float %temp5.0, float %temp6.0, float 1.000000e+00, i1 true, i1 true) #0
195  ret void
196
197ENDIF:                                            ; preds = %LOOP
198  %tmp34 = bitcast float %temp8.0 to i32
199  %tmp35 = add i32 %tmp34, 1
200  %tmp36 = bitcast i32 %tmp35 to float
201  br label %LOOP
202}
203
204; This checks for a bug in the FixSGPRCopies pass where VReg96
205; registers were being identified as an SGPR regclass which was causing
206; an assertion failure.
207
208; CHECK-LABEL: {{^}}sample_v3:
209; CHECK: v_mov_b32_e32 v[[SAMPLE_LO:[0-9]+]], 5
210; CHECK: v_mov_b32_e32 v[[SAMPLE_HI:[0-9]+]], 7
211; CHECK: s_cbranch
212
213; CHECK: BB{{[0-9]+_[0-9]+}}:
214; CHECK-DAG: v_mov_b32_e32 v[[SAMPLE_LO:[0-9]+]], 11
215; CHECK-DAG: v_mov_b32_e32 v[[SAMPLE_HI:[0-9]+]], 13
216
217; CHECK: image_sample v{{\[[0-9]+:[0-9]+\]}}, v[[[SAMPLE_LO]]:[[SAMPLE_HI]]]
218; CHECK: exp
219; CHECK: s_endpgm
220define amdgpu_ps void @sample_v3(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, float inreg %arg3, i32 inreg %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <3 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, float %arg20) #0 {
221entry:
222  %tmp21 = load <4 x i32>, ptr addrspace(4) %arg, !tbaa !0
223  %tmp22 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp21, i32 16, i32 0)
224  %tmp24 = load <8 x i32>, ptr addrspace(4) %arg2, !tbaa !0
225  %tmp26 = load <4 x i32>, ptr addrspace(4) %arg1, !tbaa !0
226  %tmp27 = fcmp oeq float %tmp22, 0.000000e+00
227  %tmp26.bc = bitcast <4 x i32> %tmp26 to <4 x i32>
228  br i1 %tmp27, label %if, label %else
229
230if:                                               ; preds = %entry
231  %tmp1 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float 0x36D6000000000000, float 0x36DA000000000000, <8 x i32> %tmp24, <4 x i32> %tmp26.bc, i1 0, i32 0, i32 0)
232  %val.if.0 = extractelement <4 x float> %tmp1, i32 0
233  %val.if.1 = extractelement <4 x float> %tmp1, i32 1
234  %val.if.2 = extractelement <4 x float> %tmp1, i32 2
235  br label %endif
236
237else:                                             ; preds = %entry
238  %tmp2 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float 0x36C4000000000000, float 0x36CC000000000000, <8 x i32> %tmp24, <4 x i32> %tmp26.bc, i1 0, i32 0, i32 0)
239  %val.else.0 = extractelement <4 x float> %tmp2, i32 0
240  %val.else.1 = extractelement <4 x float> %tmp2, i32 1
241  %val.else.2 = extractelement <4 x float> %tmp2, i32 2
242  br label %endif
243
244endif:                                            ; preds = %else, %if
245  %val.0 = phi float [ %val.if.0, %if ], [ %val.else.0, %else ]
246  %val.1 = phi float [ %val.if.1, %if ], [ %val.else.1, %else ]
247  %val.2 = phi float [ %val.if.2, %if ], [ %val.else.2, %else ]
248  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %val.0, float %val.1, float %val.2, float 0.000000e+00, i1 true, i1 true) #0
249  ret void
250}
251
252; CHECK-LABEL: {{^}}copy1:
253; CHECK: buffer_load_dword
254; CHECK: v_add
255; CHECK: s_endpgm
256define amdgpu_kernel void @copy1(ptr addrspace(1) %out, ptr addrspace(1) %in0) {
257entry:
258  %tmp = load float, ptr addrspace(1) %in0
259  %tmp1 = fcmp oeq float %tmp, 0.000000e+00
260  br i1 %tmp1, label %if0, label %endif
261
262if0:                                              ; preds = %entry
263  %tmp2 = bitcast float %tmp to i32
264  %tmp3 = fcmp olt float %tmp, 0.000000e+00
265  br i1 %tmp3, label %if1, label %endif
266
267if1:                                              ; preds = %if0
268  %tmp4 = add i32 %tmp2, 1
269  br label %endif
270
271endif:                                            ; preds = %if1, %if0, %entry
272  %tmp5 = phi i32 [ 0, %entry ], [ %tmp2, %if0 ], [ %tmp4, %if1 ]
273  %tmp6 = bitcast i32 %tmp5 to float
274  store float %tmp6, ptr addrspace(1) %out
275  ret void
276}
277
278; This test is just checking that we don't crash / assertion fail.
279; CHECK-LABEL: {{^}}copy2:
280; CHECK: s_endpgm
281define amdgpu_ps void @copy2(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, float inreg %arg3, i32 inreg %arg4, <2 x i32> %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <3 x i32> %arg8, <2 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, float %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, float %arg19, float %arg20) #0 {
282entry:
283  br label %LOOP68
284
285LOOP68:                                           ; preds = %ENDIF69, %entry
286  %temp4.7 = phi float [ 0.000000e+00, %entry ], [ %v, %ENDIF69 ]
287  %t = phi i32 [ 20, %entry ], [ %x, %ENDIF69 ]
288  %g = icmp eq i32 0, %t
289  %l = bitcast float %temp4.7 to i32
290  br i1 %g, label %IF70, label %ENDIF69
291
292IF70:                                             ; preds = %LOOP68
293  %q = icmp ne i32 %l, 13
294  %temp.8 = select i1 %q, float 1.000000e+00, float 0.000000e+00
295  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %temp.8, float 0.000000e+00, float 0.000000e+00, float 1.000000e+00, i1 true, i1 true) #0
296  ret void
297
298ENDIF69:                                          ; preds = %LOOP68
299  %u = add i32 %l, %t
300  %v = bitcast i32 %u to float
301  %x = add i32 %t, -1
302  br label %LOOP68
303}
304
305; This test checks that image_sample resource descriptors aren't loaded into
306; vgprs.  The verifier will fail if this happens.
307; CHECK-LABEL:{{^}}sample_rsrc
308
309; CHECK: s_cmp_eq_u32
310; CHECK: s_cbranch_scc1 [[END:.LBB[0-9]+_[0-9]+]]
311
312; CHECK: image_sample v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}
313; CHECK: s_endpgm
314
315; [[END]]:
316; CHECK: v_add_{{[iu]}}32_e32 v[[ADD:[0-9]+]], vcc, 1, v{{[0-9]+}}
317; CHECK: image_sample v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+}}:[[ADD]]]
318; CHECK: s_branch
319define amdgpu_ps void @sample_rsrc(ptr addrspace(4) inreg %arg, ptr addrspace(4) inreg %arg1, ptr addrspace(4) inreg %arg2, ptr addrspace(4) inreg %arg3, float inreg %arg4, i32 inreg %arg5, <2 x i32> %arg6, <2 x i32> %arg7, <2 x i32> %arg8, <3 x i32> %arg9, <2 x i32> %arg10, <2 x i32> %arg11, <2 x i32> %arg12, float %arg13, float %arg14, float %arg15, float %arg16, float %arg17, float %arg18, i32 %arg19, float %arg20, float %arg21) #0 {
320bb:
321  %tmp22 = load <4 x i32>, ptr addrspace(4) %arg1, !tbaa !3
322  %tmp23 = call float @llvm.amdgcn.s.buffer.load.f32(<4 x i32> %tmp22, i32 16, i32 0)
323  %tmp26 = load <8 x i32>, ptr addrspace(4) %arg3, !tbaa !3
324  %tmp28 = load <4 x i32>, ptr addrspace(4) %arg2, !tbaa !3
325  %i.i = extractelement <2 x i32> %arg7, i32 0
326  %j.i = extractelement <2 x i32> %arg7, i32 1
327  %i.f.i = bitcast i32 %i.i to float
328  %j.f.i = bitcast i32 %j.i to float
329  %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 0, i32 0, i32 %arg5) #0
330  %p2.i = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 0, i32 0, i32 %arg5) #0
331  %i.i1 = extractelement <2 x i32> %arg7, i32 0
332  %j.i2 = extractelement <2 x i32> %arg7, i32 1
333  %i.f.i3 = bitcast i32 %i.i1 to float
334  %j.f.i4 = bitcast i32 %j.i2 to float
335  %p1.i5 = call float @llvm.amdgcn.interp.p1(float %i.f.i3, i32 1, i32 0, i32 %arg5) #0
336  %p2.i6 = call float @llvm.amdgcn.interp.p2(float %p1.i5, float %j.f.i4, i32 1, i32 0, i32 %arg5) #0
337  %tmp31 = bitcast float %tmp23 to i32
338  %tmp36 = icmp ne i32 %tmp31, 0
339  br i1 %tmp36, label %bb38, label %bb80
340
341bb38:                                             ; preds = %bb
342  %tmp56 = bitcast <8 x i32> %tmp26 to <8 x i32>
343  %tmp2 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %p2.i, float %p2.i6, <8 x i32> %tmp56, <4 x i32> %tmp28, i1 0, i32 0, i32 0)
344  br label %bb71
345
346bb80:                                             ; preds = %bb
347  %tmp81 = bitcast float %p2.i to i32
348  %tmp82 = bitcast float %p2.i6 to i32
349  %tmp82.2 = add i32 %tmp82, 1
350  %tmp83 = bitcast i32 %tmp81 to float
351  %tmp84 = bitcast i32 %tmp82.2 to float
352  %tmp85 = bitcast <8 x i32> %tmp26 to <8 x i32>
353  %tmp3 = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %tmp83, float %tmp84, <8 x i32> %tmp85, <4 x i32> %tmp28, i1 0, i32 0, i32 0)
354  br label %bb71
355
356bb71:                                             ; preds = %bb80, %bb38
357  %tmp72 = phi <4 x float> [ %tmp2, %bb38 ], [ %tmp3, %bb80 ]
358  %tmp88 = extractelement <4 x float> %tmp72, i32 0
359  call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp88, float %tmp88, float %tmp88, float %tmp88, i1 true, i1 true) #0
360  ret void
361}
362
363; Check the resource descriptor is stored in an sgpr.
364; CHECK-LABEL: {{^}}mimg_srsrc_sgpr:
365; CHECK: image_sample v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}] dmask:0x1
366define amdgpu_ps void @mimg_srsrc_sgpr(ptr addrspace(4) inreg %arg) #0 {
367bb:
368  %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
369  %tmp7 = getelementptr [34 x <8 x i32>], ptr addrspace(4) %arg, i32 0, i32 %tid
370  %tmp8 = load <8 x i32>, ptr addrspace(4) %tmp7, align 32, !tbaa !0
371  %tmp = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float 7.500000e-01, float 2.500000e-01, <8 x i32> %tmp8, <4 x i32> undef, i1 0, i32 0, i32 0)
372  %tmp10 = extractelement <4 x float> %tmp, i32 0
373  %tmp12 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float %tmp10)
374  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp12, <2 x half> undef, i1 true, i1 true) #0
375  ret void
376}
377
378; Check the sampler is stored in an sgpr.
379; CHECK-LABEL: {{^}}mimg_ssamp_sgpr:
380; CHECK: image_sample v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}] dmask:0x1
381define amdgpu_ps void @mimg_ssamp_sgpr(ptr addrspace(4) inreg %arg) #0 {
382bb:
383  %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
384  %tmp7 = getelementptr [17 x <4 x i32>], ptr addrspace(4) %arg, i32 0, i32 %tid
385  %tmp8 = load <4 x i32>, ptr addrspace(4) %tmp7, align 16, !tbaa !0
386  %tmp = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float 7.500000e-01, float 2.500000e-01, <8 x i32> undef, <4 x i32> %tmp8, i1 0, i32 0, i32 0)
387  %tmp10 = extractelement <4 x float> %tmp, i32 0
388  %tmp12 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp10, float undef)
389  call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp12, <2 x half> undef, i1 true, i1 true) #0
390  ret void
391}
392
393declare float @llvm.fabs.f32(float) #1
394declare float @llvm.amdgcn.rsq.f32(float) #1
395declare float @llvm.exp2.f32(float) #1
396declare float @llvm.pow.f32(float, float) #1
397declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #1
398declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32) #1
399declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32) #1
400declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
401declare void @llvm.amdgcn.exp.compr.v2f16(i32, i32, <2 x half>, <2 x half>, i1, i1) #0
402declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) #1
403declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #2
404declare float @llvm.amdgcn.s.buffer.load.f32(<4 x i32>, i32, i32) #1
405
406attributes #0 = { nounwind }
407attributes #1 = { nounwind readnone }
408attributes #2 = { nounwind readonly }
409
410!0 = !{!1, !1, i64 0, i32 1}
411!1 = !{!"const", !2}
412!2 = !{!"tbaa root"}
413!3 = !{!1, !1, i64 0}
414