1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -verify-machineinstrs < %s | FileCheck %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck %s
4 ; CHECK-LABEL: {{^}}phi1:
5 ; CHECK: s_buffer_load_dword [[DST:s[0-9]]], {{s\[[0-9]+:[0-9]+\]}}, 0x0
6 ; CHECK: v_mov_b32_e32 v{{[0-9]}}, [[DST]]
7 define amdgpu_ps void @phi1(<4 x i32> addrspace(2)* inreg %arg, <4 x i32> addrspace(2)* inreg %arg1, <8 x i32> addrspace(2)* 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 {
9 %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(2)* %arg, i32 0
10 %tmp20 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp, !tbaa !0
11 %tmp21 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 0)
12 %tmp22 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 16)
13 %tmp23 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 32)
14 %tmp24 = fptosi float %tmp22 to i32
15 %tmp25 = icmp ne i32 %tmp24, 0
16 br i1 %tmp25, label %ENDIF, label %ELSE
18 ELSE: ; preds = %main_body
19 %tmp26 = fsub float -0.000000e+00, %tmp21
22 ENDIF: ; 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
29 ; Make sure this program doesn't crash
30 ; CHECK-LABEL: {{^}}phi2:
31 define amdgpu_ps void @phi2(<4 x i32> addrspace(2)* inreg %arg, <4 x i32> addrspace(2)* inreg %arg1, <8 x i32> addrspace(2)* 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 {
33 %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(2)* %arg, i32 0
34 %tmp20 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp, !tbaa !0
35 %tmp21 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 16)
36 %tmp22 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 32)
37 %tmp23 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 36)
38 %tmp24 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 40)
39 %tmp25 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 48)
40 %tmp26 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 52)
41 %tmp27 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 56)
42 %tmp28 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 64)
43 %tmp29 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 68)
44 %tmp30 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 72)
45 %tmp31 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 76)
46 %tmp32 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 80)
47 %tmp33 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 84)
48 %tmp34 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 88)
49 %tmp35 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 92)
50 %tmp36 = getelementptr <8 x i32>, <8 x i32> addrspace(2)* %arg2, i32 0
51 %tmp37 = load <8 x i32>, <8 x i32> addrspace(2)* %tmp36, !tbaa !0
52 %tmp38 = getelementptr <4 x i32>, <4 x i32> addrspace(2)* %arg1, i32 0
53 %tmp39 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp38, !tbaa !0
54 %i.i = extractelement <2 x i32> %arg5, i32 0
55 %j.i = extractelement <2 x i32> %arg5, i32 1
56 %i.f.i = bitcast i32 %i.i to float
57 %j.f.i = bitcast i32 %j.i to float
58 %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 0, i32 0, i32 %arg3) #1
59 %p2.i = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 0, i32 0, i32 %arg3) #1
60 %i.i19 = extractelement <2 x i32> %arg5, i32 0
61 %j.i20 = extractelement <2 x i32> %arg5, i32 1
62 %i.f.i21 = bitcast i32 %i.i19 to float
63 %j.f.i22 = bitcast i32 %j.i20 to float
64 %p1.i23 = call float @llvm.amdgcn.interp.p1(float %i.f.i21, i32 1, i32 0, i32 %arg3) #1
65 %p2.i24 = call float @llvm.amdgcn.interp.p2(float %p1.i23, float %j.f.i22, i32 1, i32 0, i32 %arg3) #1
66 %i.i13 = extractelement <2 x i32> %arg5, i32 0
67 %j.i14 = extractelement <2 x i32> %arg5, i32 1
68 %i.f.i15 = bitcast i32 %i.i13 to float
69 %j.f.i16 = bitcast i32 %j.i14 to float
70 %p1.i17 = call float @llvm.amdgcn.interp.p1(float %i.f.i15, i32 0, i32 1, i32 %arg3) #1
71 %p2.i18 = call float @llvm.amdgcn.interp.p2(float %p1.i17, float %j.f.i16, i32 0, i32 1, i32 %arg3) #1
72 %i.i7 = extractelement <2 x i32> %arg5, i32 0
73 %j.i8 = extractelement <2 x i32> %arg5, i32 1
74 %i.f.i9 = bitcast i32 %i.i7 to float
75 %j.f.i10 = bitcast i32 %j.i8 to float
76 %p1.i11 = call float @llvm.amdgcn.interp.p1(float %i.f.i9, i32 1, i32 1, i32 %arg3) #1
77 %p2.i12 = call float @llvm.amdgcn.interp.p2(float %p1.i11, float %j.f.i10, i32 1, i32 1, i32 %arg3) #1
78 %i.i1 = extractelement <2 x i32> %arg5, i32 0
79 %j.i2 = extractelement <2 x i32> %arg5, i32 1
80 %i.f.i3 = bitcast i32 %i.i1 to float
81 %j.f.i4 = bitcast i32 %j.i2 to float
82 %p1.i5 = call float @llvm.amdgcn.interp.p1(float %i.f.i3, i32 2, i32 1, i32 %arg3) #1
83 %p2.i6 = call float @llvm.amdgcn.interp.p2(float %p1.i5, float %j.f.i4, i32 2, i32 1, i32 %arg3) #1
84 %tmp45 = bitcast float %p2.i to i32
85 %tmp46 = bitcast float %p2.i24 to i32
86 %tmp47 = insertelement <2 x i32> undef, i32 %tmp45, i32 0
87 %tmp48 = insertelement <2 x i32> %tmp47, i32 %tmp46, i32 1
88 %tmp39.bc = bitcast <4 x i32> %tmp39 to <4 x i32>
89 %a.bc.i = bitcast <2 x i32> %tmp48 to <2 x float>
90 %tmp1 = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> %a.bc.i, <8 x i32> %tmp37, <4 x i32> %tmp39.bc, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
91 %tmp50 = extractelement <4 x float> %tmp1, i32 2
92 %tmp51 = call float @llvm.fabs.f32(float %tmp50)
93 %tmp52 = fmul float %p2.i18, %p2.i18
94 %tmp53 = fmul float %p2.i12, %p2.i12
95 %tmp54 = fadd float %tmp53, %tmp52
96 %tmp55 = fmul float %p2.i6, %p2.i6
97 %tmp56 = fadd float %tmp54, %tmp55
98 %tmp57 = call float @llvm.amdgcn.rsq.f32(float %tmp56)
99 %tmp58 = fmul float %p2.i18, %tmp57
100 %tmp59 = fmul float %p2.i12, %tmp57
101 %tmp60 = fmul float %p2.i6, %tmp57
102 %tmp61 = fmul float %tmp58, %tmp22
103 %tmp62 = fmul float %tmp59, %tmp23
104 %tmp63 = fadd float %tmp62, %tmp61
105 %tmp64 = fmul float %tmp60, %tmp24
106 %tmp65 = fadd float %tmp63, %tmp64
107 %tmp66 = fsub float -0.000000e+00, %tmp25
108 %tmp67 = fmul float %tmp65, %tmp51
109 %tmp68 = fadd float %tmp67, %tmp66
110 %tmp69 = fmul float %tmp26, %tmp68
111 %tmp70 = fmul float %tmp27, %tmp68
112 %tmp71 = call float @llvm.fabs.f32(float %tmp69)
113 %tmp72 = fcmp olt float 0x3EE4F8B580000000, %tmp71
114 %tmp73 = sext i1 %tmp72 to i32
115 %tmp74 = bitcast i32 %tmp73 to float
116 %tmp75 = bitcast float %tmp74 to i32
117 %tmp76 = icmp ne i32 %tmp75, 0
118 br i1 %tmp76, label %IF, label %ENDIF
120 IF: ; preds = %main_body
121 %tmp77 = fsub float -0.000000e+00, %tmp69
122 %tmp78 = call float @llvm.exp2.f32(float %tmp77)
123 %tmp79 = fsub float -0.000000e+00, %tmp78
124 %tmp80 = fadd float 1.000000e+00, %tmp79
125 %tmp81 = fdiv float 1.000000e+00, %tmp69
126 %tmp82 = fmul float %tmp80, %tmp81
127 %tmp83 = fmul float %tmp31, %tmp82
130 ENDIF: ; preds = %IF, %main_body
131 %temp4.0 = phi float [ %tmp83, %IF ], [ %tmp31, %main_body ]
132 %tmp84 = call float @llvm.fabs.f32(float %tmp70)
133 %tmp85 = fcmp olt float 0x3EE4F8B580000000, %tmp84
134 %tmp86 = sext i1 %tmp85 to i32
135 %tmp87 = bitcast i32 %tmp86 to float
136 %tmp88 = bitcast float %tmp87 to i32
137 %tmp89 = icmp ne i32 %tmp88, 0
138 br i1 %tmp89, label %IF25, label %ENDIF24
140 IF25: ; preds = %ENDIF
141 %tmp90 = fsub float -0.000000e+00, %tmp70
142 %tmp91 = call float @llvm.exp2.f32(float %tmp90)
143 %tmp92 = fsub float -0.000000e+00, %tmp91
144 %tmp93 = fadd float 1.000000e+00, %tmp92
145 %tmp94 = fdiv float 1.000000e+00, %tmp70
146 %tmp95 = fmul float %tmp93, %tmp94
147 %tmp96 = fmul float %tmp35, %tmp95
150 ENDIF24: ; preds = %IF25, %ENDIF
151 %temp8.0 = phi float [ %tmp96, %IF25 ], [ %tmp35, %ENDIF ]
152 %tmp97 = fmul float %tmp28, %temp4.0
153 %tmp98 = fmul float %tmp29, %temp4.0
154 %tmp99 = fmul float %tmp30, %temp4.0
155 %tmp100 = fmul float %tmp32, %temp8.0
156 %tmp101 = fadd float %tmp100, %tmp97
157 %tmp102 = fmul float %tmp33, %temp8.0
158 %tmp103 = fadd float %tmp102, %tmp98
159 %tmp104 = fmul float %tmp34, %temp8.0
160 %tmp105 = fadd float %tmp104, %tmp99
161 %tmp106 = call float @llvm.pow.f32(float %tmp51, float %tmp21)
162 %tmp107 = fsub float -0.000000e+00, %tmp101
163 %tmp108 = fmul float %tmp107, %tmp106
164 %tmp109 = fsub float -0.000000e+00, %tmp103
165 %tmp110 = fmul float %tmp109, %tmp106
166 %tmp111 = fsub float -0.000000e+00, %tmp105
167 %tmp112 = fmul float %tmp111, %tmp106
168 %tmp113 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp108, float %tmp110)
169 %tmp115 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp112, float 1.000000e+00)
170 call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp113, <2 x half> %tmp115, i1 true, i1 true) #0
174 ; We just want ot make sure the program doesn't crash
175 ; CHECK-LABEL: {{^}}loop:
176 define amdgpu_ps void @loop(<4 x i32> addrspace(2)* inreg %arg, <4 x i32> addrspace(2)* inreg %arg1, <8 x i32> addrspace(2)* 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 {
178 %tmp = getelementptr <4 x i32>, <4 x i32> addrspace(2)* %arg, i32 0
179 %tmp20 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp, !tbaa !0
180 %tmp21 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 0)
181 %tmp22 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 4)
182 %tmp23 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 8)
183 %tmp24 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp20, i32 12)
184 %tmp25 = fptosi float %tmp24 to i32
185 %tmp26 = bitcast i32 %tmp25 to float
186 %tmp27 = bitcast float %tmp26 to i32
189 LOOP: ; preds = %ENDIF, %main_body
190 %temp4.0 = phi float [ %tmp21, %main_body ], [ %temp5.0, %ENDIF ]
191 %temp5.0 = phi float [ %tmp22, %main_body ], [ %temp6.0, %ENDIF ]
192 %temp6.0 = phi float [ %tmp23, %main_body ], [ %temp4.0, %ENDIF ]
193 %temp8.0 = phi float [ 0.000000e+00, %main_body ], [ %tmp36, %ENDIF ]
194 %tmp28 = bitcast float %temp8.0 to i32
195 %tmp29 = icmp sge i32 %tmp28, %tmp27
196 %tmp30 = sext i1 %tmp29 to i32
197 %tmp31 = bitcast i32 %tmp30 to float
198 %tmp32 = bitcast float %tmp31 to i32
199 %tmp33 = icmp ne i32 %tmp32, 0
200 br i1 %tmp33, label %IF, label %ENDIF
203 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
206 ENDIF: ; preds = %LOOP
207 %tmp34 = bitcast float %temp8.0 to i32
208 %tmp35 = add i32 %tmp34, 1
209 %tmp36 = bitcast i32 %tmp35 to float
213 ; This checks for a bug in the FixSGPRCopies pass where VReg96
214 ; registers were being identified as an SGPR regclass which was causing
215 ; an assertion failure.
217 ; CHECK-LABEL: {{^}}sample_v3:
218 ; CHECK: v_mov_b32_e32 v[[SAMPLE_LO:[0-9]+]], 11
219 ; CHECK: v_mov_b32_e32 v[[SAMPLE_HI:[0-9]+]], 13
222 ; CHECK-DAG: v_mov_b32_e32 v[[SAMPLE_LO:[0-9]+]], 5
223 ; CHECK-DAG: v_mov_b32_e32 v[[SAMPLE_HI:[0-9]+]], 7
225 ; CHECK: BB{{[0-9]+_[0-9]+}}:
226 ; CHECK: image_sample v{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[SAMPLE_LO]]:[[SAMPLE_HI]]{{\]}}
229 define amdgpu_ps void @sample_v3([17 x <4 x i32>] addrspace(2)* byval %arg, [32 x <4 x i32>] addrspace(2)* byval %arg1, [16 x <8 x i32>] addrspace(2)* byval %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 {
231 %tmp = getelementptr [17 x <4 x i32>], [17 x <4 x i32>] addrspace(2)* %arg, i64 0, i32 0
232 %tmp21 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp, !tbaa !0
233 %tmp22 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp21, i32 16)
234 %tmp23 = getelementptr [16 x <8 x i32>], [16 x <8 x i32>] addrspace(2)* %arg2, i64 0, i32 0
235 %tmp24 = load <8 x i32>, <8 x i32> addrspace(2)* %tmp23, !tbaa !0
236 %tmp25 = getelementptr [32 x <4 x i32>], [32 x <4 x i32>] addrspace(2)* %arg1, i64 0, i32 0
237 %tmp26 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp25, !tbaa !0
238 %tmp27 = fcmp oeq float %tmp22, 0.000000e+00
239 %tmp26.bc = bitcast <4 x i32> %tmp26 to <4 x i32>
240 br i1 %tmp27, label %if, label %else
243 %tmp1 = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> <float 0x36D6000000000000, float 0x36DA000000000000>, <8 x i32> %tmp24, <4 x i32> %tmp26.bc, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
244 %val.if.0 = extractelement <4 x float> %tmp1, i32 0
245 %val.if.1 = extractelement <4 x float> %tmp1, i32 1
246 %val.if.2 = extractelement <4 x float> %tmp1, i32 2
249 else: ; preds = %entry
250 %tmp2 = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> <float 0x36C4000000000000, float 0x36CC000000000000>, <8 x i32> %tmp24, <4 x i32> %tmp26.bc, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
251 %val.else.0 = extractelement <4 x float> %tmp2, i32 0
252 %val.else.1 = extractelement <4 x float> %tmp2, i32 1
253 %val.else.2 = extractelement <4 x float> %tmp2, i32 2
256 endif: ; preds = %else, %if
257 %val.0 = phi float [ %val.if.0, %if ], [ %val.else.0, %else ]
258 %val.1 = phi float [ %val.if.1, %if ], [ %val.else.1, %else ]
259 %val.2 = phi float [ %val.if.2, %if ], [ %val.else.2, %else ]
260 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
264 ; CHECK-LABEL: {{^}}copy1:
265 ; CHECK: buffer_load_dword
268 define amdgpu_kernel void @copy1(float addrspace(1)* %out, float addrspace(1)* %in0) {
270 %tmp = load float, float addrspace(1)* %in0
271 %tmp1 = fcmp oeq float %tmp, 0.000000e+00
272 br i1 %tmp1, label %if0, label %endif
274 if0: ; preds = %entry
275 %tmp2 = bitcast float %tmp to i32
276 %tmp3 = fcmp olt float %tmp, 0.000000e+00
277 br i1 %tmp3, label %if1, label %endif
280 %tmp4 = add i32 %tmp2, 1
283 endif: ; preds = %if1, %if0, %entry
284 %tmp5 = phi i32 [ 0, %entry ], [ %tmp2, %if0 ], [ %tmp4, %if1 ]
285 %tmp6 = bitcast i32 %tmp5 to float
286 store float %tmp6, float addrspace(1)* %out
290 ; This test is just checking that we don't crash / assertion fail.
291 ; CHECK-LABEL: {{^}}copy2:
293 define amdgpu_ps void @copy2([17 x <4 x i32>] addrspace(2)* byval %arg, [32 x <4 x i32>] addrspace(2)* byval %arg1, [16 x <8 x i32>] addrspace(2)* byval %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 {
297 LOOP68: ; preds = %ENDIF69, %entry
298 %temp4.7 = phi float [ 0.000000e+00, %entry ], [ %v, %ENDIF69 ]
299 %t = phi i32 [ 20, %entry ], [ %x, %ENDIF69 ]
300 %g = icmp eq i32 0, %t
301 %l = bitcast float %temp4.7 to i32
302 br i1 %g, label %IF70, label %ENDIF69
304 IF70: ; preds = %LOOP68
305 %q = icmp ne i32 %l, 13
306 %temp.8 = select i1 %q, float 1.000000e+00, float 0.000000e+00
307 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
310 ENDIF69: ; preds = %LOOP68
312 %v = bitcast i32 %u to float
317 ; This test checks that image_sample resource descriptors aren't loaded into
318 ; vgprs. The verifier will fail if this happens.
319 ; CHECK-LABEL:{{^}}sample_rsrc
321 ; CHECK: s_cmp_eq_u32
322 ; CHECK: s_cbranch_scc0 [[END:BB[0-9]+_[0-9]+]]
324 ; CHECK: v_add_{{[iu]}}32_e32 v[[ADD:[0-9]+]], vcc, 1, v{{[0-9]+}}
327 ; CHECK: image_sample v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+}}:[[ADD]]{{\]}}
329 define amdgpu_ps void @sample_rsrc([6 x <4 x i32>] addrspace(2)* byval %arg, [17 x <4 x i32>] addrspace(2)* byval %arg1, [16 x <4 x i32>] addrspace(2)* byval %arg2, [32 x <8 x i32>] addrspace(2)* byval %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 {
331 %tmp = getelementptr [17 x <4 x i32>], [17 x <4 x i32>] addrspace(2)* %arg1, i32 0, i32 0
332 %tmp22 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp, !tbaa !3
333 %tmp23 = call float @llvm.SI.load.const.v4i32(<4 x i32> %tmp22, i32 16)
334 %tmp25 = getelementptr [32 x <8 x i32>], [32 x <8 x i32>] addrspace(2)* %arg3, i32 0, i32 0
335 %tmp26 = load <8 x i32>, <8 x i32> addrspace(2)* %tmp25, !tbaa !3
336 %tmp27 = getelementptr [16 x <4 x i32>], [16 x <4 x i32>] addrspace(2)* %arg2, i32 0, i32 0
337 %tmp28 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp27, !tbaa !3
338 %i.i = extractelement <2 x i32> %arg7, i32 0
339 %j.i = extractelement <2 x i32> %arg7, i32 1
340 %i.f.i = bitcast i32 %i.i to float
341 %j.f.i = bitcast i32 %j.i to float
342 %p1.i = call float @llvm.amdgcn.interp.p1(float %i.f.i, i32 0, i32 0, i32 %arg5) #0
343 %p2.i = call float @llvm.amdgcn.interp.p2(float %p1.i, float %j.f.i, i32 0, i32 0, i32 %arg5) #0
344 %i.i1 = extractelement <2 x i32> %arg7, i32 0
345 %j.i2 = extractelement <2 x i32> %arg7, i32 1
346 %i.f.i3 = bitcast i32 %i.i1 to float
347 %j.f.i4 = bitcast i32 %j.i2 to float
348 %p1.i5 = call float @llvm.amdgcn.interp.p1(float %i.f.i3, i32 1, i32 0, i32 %arg5) #0
349 %p2.i6 = call float @llvm.amdgcn.interp.p2(float %p1.i5, float %j.f.i4, i32 1, i32 0, i32 %arg5) #0
350 %tmp31 = bitcast float %tmp23 to i32
351 %tmp36 = icmp ne i32 %tmp31, 0
352 br i1 %tmp36, label %bb38, label %bb80
355 %tmp52 = bitcast float %p2.i to i32
356 %tmp53 = bitcast float %p2.i6 to i32
357 %tmp54 = insertelement <2 x i32> undef, i32 %tmp52, i32 0
358 %tmp55 = insertelement <2 x i32> %tmp54, i32 %tmp53, i32 1
359 %tmp56 = bitcast <8 x i32> %tmp26 to <8 x i32>
360 %a.bc.i = bitcast <2 x i32> %tmp55 to <2 x float>
361 %tmp2 = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> %a.bc.i, <8 x i32> %tmp56, <4 x i32> %tmp28, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
365 %tmp81 = bitcast float %p2.i to i32
366 %tmp82 = bitcast float %p2.i6 to i32
367 %tmp82.2 = add i32 %tmp82, 1
368 %tmp83 = insertelement <2 x i32> undef, i32 %tmp81, i32 0
369 %tmp84 = insertelement <2 x i32> %tmp83, i32 %tmp82.2, i32 1
370 %tmp85 = bitcast <8 x i32> %tmp26 to <8 x i32>
371 %a.bc.i1 = bitcast <2 x i32> %tmp84 to <2 x float>
372 %tmp3 = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> %a.bc.i1, <8 x i32> %tmp85, <4 x i32> %tmp28, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
375 bb71: ; preds = %bb80, %bb38
376 %tmp72 = phi <4 x float> [ %tmp2, %bb38 ], [ %tmp3, %bb80 ]
377 %tmp88 = extractelement <4 x float> %tmp72, i32 0
378 call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %tmp88, float %tmp88, float %tmp88, float %tmp88, i1 true, i1 true) #0
382 ; Check the the resource descriptor is stored in an sgpr.
383 ; CHECK-LABEL: {{^}}mimg_srsrc_sgpr:
384 ; CHECK: image_sample v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}] dmask:0x1
385 define amdgpu_ps void @mimg_srsrc_sgpr([34 x <8 x i32>] addrspace(2)* byval %arg) #0 {
387 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
388 %tmp7 = getelementptr [34 x <8 x i32>], [34 x <8 x i32>] addrspace(2)* %arg, i32 0, i32 %tid
389 %tmp8 = load <8 x i32>, <8 x i32> addrspace(2)* %tmp7, align 32, !tbaa !0
390 %tmp = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> <float 7.500000e-01, float 2.500000e-01>, <8 x i32> %tmp8, <4 x i32> undef, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
391 %tmp10 = extractelement <4 x float> %tmp, i32 0
392 %tmp12 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float undef, float %tmp10)
393 call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp12, <2 x half> undef, i1 true, i1 true) #0
397 ; Check the the sampler is stored in an sgpr.
398 ; CHECK-LABEL: {{^}}mimg_ssamp_sgpr:
399 ; CHECK: image_sample v{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}] dmask:0x1
400 define amdgpu_ps void @mimg_ssamp_sgpr([17 x <4 x i32>] addrspace(2)* byval %arg) #0 {
402 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
403 %tmp7 = getelementptr [17 x <4 x i32>], [17 x <4 x i32>] addrspace(2)* %arg, i32 0, i32 %tid
404 %tmp8 = load <4 x i32>, <4 x i32> addrspace(2)* %tmp7, align 16, !tbaa !0
405 %tmp = call <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float> <float 7.500000e-01, float 2.500000e-01>, <8 x i32> undef, <4 x i32> %tmp8, i32 15, i1 false, i1 false, i1 false, i1 false, i1 false)
406 %tmp10 = extractelement <4 x float> %tmp, i32 0
407 %tmp12 = call <2 x half> @llvm.amdgcn.cvt.pkrtz(float %tmp10, float undef)
408 call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> %tmp12, <2 x half> undef, i1 true, i1 true) #0
412 declare float @llvm.fabs.f32(float) #1
413 declare float @llvm.amdgcn.rsq.f32(float) #1
414 declare float @llvm.exp2.f32(float) #1
415 declare float @llvm.pow.f32(float, float) #1
416 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #1
417 declare float @llvm.amdgcn.interp.p1(float, i32, i32, i32) #1
418 declare float @llvm.amdgcn.interp.p2(float, float, i32, i32, i32) #1
419 declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
420 declare void @llvm.amdgcn.exp.compr.v2f16(i32, i32, <2 x half>, <2 x half>, i1, i1) #0
421 declare <2 x half> @llvm.amdgcn.cvt.pkrtz(float, float) #1
422 declare <4 x float> @llvm.amdgcn.image.sample.v4f32.v2f32.v8i32(<2 x float>, <8 x i32>, <4 x i32>, i32, i1, i1, i1, i1, i1) #2
423 declare float @llvm.SI.load.const.v4i32(<4 x i32>, i32) #1
425 attributes #0 = { nounwind }
426 attributes #1 = { nounwind readnone }
427 attributes #2 = { nounwind readonly }
429 !0 = !{!1, !1, i64 0, i32 1}
432 !3 = !{!1, !1, i64 0}