1 ; RUN: llc -O3 -march=hexagon < %s | FileCheck %s
3 ; CHECK: loop0(.[[BLOCK:LBB0_[0-9]+]]
5 ; CHECK: = vmemu({{r[0-9]+}}++#1)
6 ; CHECK: = vmemu({{r[0-9]+}}++#1)
7 ; CHECK: = vmemu({{r[0-9]+}}++#1)
8 ; CHECK: = vmemu({{r[0-9]+}}++#1)
11 target triple = "hexagon-unknown--elf"
13 %0 = type { i8*, i32, i32, i32, i32, %1*, %1*, %1* }
16 %3 = type { i8*, i32, i32, i32, i32, i32, i32, i8*, i32, i32* }
17 %4 = type { i64, i8*, [4 x i32], [4 x i32], [4 x i32], i32, i8, i8, [6 x i8] }
19 @g0 = private unnamed_addr constant [5 x i8] c"Load\00", align 1
20 @g1 = private unnamed_addr constant [6 x i8] c"Store\00", align 1
21 @g2 = private unnamed_addr constant [18 x i8] c"Begin realization\00", align 1
22 @g3 = private unnamed_addr constant [16 x i8] c"End realization\00", align 1
23 @g4 = private unnamed_addr constant [8 x i8] c"Produce\00", align 1
24 @g5 = private unnamed_addr constant [7 x i8] c"Update\00", align 1
25 @g6 = private unnamed_addr constant [8 x i8] c"Consume\00", align 1
26 @g7 = private unnamed_addr constant [12 x i8] c"End consume\00", align 1
27 @g8 = private constant [6 x i8] c"input\00", align 32
28 @g9 = private constant [10 x i8] c"dilate3x3\00", align 32
29 @g10 = private constant [2 x %0] [%0 { i8* getelementptr inbounds ([6 x i8], [6 x i8]* @g8, i32 0, i32 0), i32 1, i32 2, i32 1, i32 8, %1* null, %1* null, %1* null }, %0 { i8* getelementptr inbounds ([10 x i8], [10 x i8]* @g9, i32 0, i32 0), i32 2, i32 2, i32 1, i32 8, %1* null, %1* null, %1* null }]
30 @g11 = private constant [64 x i8] c"...............................................................\00", align 32
32 ; Function Attrs: nounwind
33 declare i8* @f0(i8*, i32) #0
35 ; Function Attrs: nounwind
36 declare void @f1(i8*, i8*) #0
38 ; Function Attrs: nounwind
39 declare void @f2(i8*, i8*) #0
41 ; Function Attrs: nounwind
42 declare i32 @f3(i8*, %3*) #0
44 ; Function Attrs: nounwind
47 ; Function Attrs: nounwind
50 ; Function Attrs: nounwind
51 define i32 @f6(%4* noalias nocapture readonly %a0, %4* noalias nocapture readonly %a1) #0 {
53 %v0 = getelementptr inbounds %4, %4* %a0, i32 0, i32 1
54 %v1 = load i8*, i8** %v0, align 4
55 %v2 = getelementptr inbounds %4, %4* %a0, i32 0, i32 3, i32 1
56 %v3 = load i32, i32* %v2, align 4
57 %v4 = getelementptr inbounds %4, %4* %a0, i32 0, i32 4, i32 0
58 %v5 = load i32, i32* %v4, align 4
59 %v6 = getelementptr inbounds %4, %4* %a0, i32 0, i32 4, i32 1
60 %v7 = load i32, i32* %v6, align 4
61 %v8 = getelementptr inbounds %4, %4* %a1, i32 0, i32 1
62 %v9 = load i8*, i8** %v8, align 4
63 %v10 = getelementptr inbounds %4, %4* %a1, i32 0, i32 2, i32 0
64 %v11 = load i32, i32* %v10, align 4
65 %v12 = getelementptr inbounds %4, %4* %a1, i32 0, i32 3, i32 1
66 %v13 = load i32, i32* %v12, align 4
67 %v14 = getelementptr inbounds %4, %4* %a1, i32 0, i32 4, i32 0
68 %v15 = load i32, i32* %v14, align 4
69 %v16 = getelementptr inbounds %4, %4* %a1, i32 0, i32 4, i32 1
70 %v17 = load i32, i32* %v16, align 4
71 %v18 = getelementptr inbounds %4, %4* %a1, i32 0, i32 2, i32 1
72 %v19 = load i32, i32* %v18, align 4
73 %v20 = add nsw i32 %v19, %v17
74 %v21 = icmp sgt i32 %v19, 0
75 br i1 %v21, label %b1, label %b11, !prof !3
78 %v22 = ashr i32 %v11, 7
79 %v23 = icmp slt i32 %v22, 0
80 %v24 = select i1 %v23, i32 0, i32 %v22
81 %v25 = icmp sgt i32 %v24, 0
82 br i1 %v25, label %b5, label %b7, !prof !3
84 b2: ; preds = %b5, %b2
85 %v26 = phi i32 [ %v90, %b2 ], [ 0, %b5 ]
86 %v27 = mul nsw i32 %v7, %v3
87 %v28 = add nsw i32 %v27, %v5
88 %v29 = shl nsw i32 %v26, 7
89 %v30 = add nsw i32 %v29, %v15
90 %v31 = add nsw i32 %v150, -1
91 %v32 = mul nsw i32 %v31, %v3
92 %v33 = mul nsw i32 %v150, %v3
93 %v34 = add nsw i32 %v150, 1
94 %v35 = mul nsw i32 %v34, %v3
95 %v36 = sub i32 %v32, %v28
96 %v37 = add i32 %v36, %v30
97 %v38 = add nsw i32 %v37, -1
98 %v39 = getelementptr inbounds i8, i8* %v1, i32 %v38
99 %v40 = bitcast i8* %v39 to <32 x i32>*
100 %v41 = load <32 x i32>, <32 x i32>* %v40, align 1, !tbaa !4
101 %v42 = getelementptr inbounds i8, i8* %v1, i32 %v37
102 %v43 = bitcast i8* %v42 to <32 x i32>*
103 %v44 = load <32 x i32>, <32 x i32>* %v43, align 1, !tbaa !4
104 %v45 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v41, <32 x i32> %v44)
105 %v46 = add nsw i32 %v37, 1
106 %v47 = getelementptr inbounds i8, i8* %v1, i32 %v46
107 %v48 = bitcast i8* %v47 to <32 x i32>*
108 %v49 = load <32 x i32>, <32 x i32>* %v48, align 1, !tbaa !4
109 %v50 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v45, <32 x i32> %v49)
110 %v51 = sub i32 %v33, %v28
111 %v52 = add i32 %v51, %v30
112 %v53 = add nsw i32 %v52, -1
113 %v54 = getelementptr inbounds i8, i8* %v1, i32 %v53
114 %v55 = bitcast i8* %v54 to <32 x i32>*
115 %v56 = load <32 x i32>, <32 x i32>* %v55, align 1, !tbaa !4
116 %v57 = getelementptr inbounds i8, i8* %v1, i32 %v52
117 %v58 = bitcast i8* %v57 to <32 x i32>*
118 %v59 = load <32 x i32>, <32 x i32>* %v58, align 1, !tbaa !4
119 %v60 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v56, <32 x i32> %v59)
120 %v61 = add nsw i32 %v52, 1
121 %v62 = getelementptr inbounds i8, i8* %v1, i32 %v61
122 %v63 = bitcast i8* %v62 to <32 x i32>*
123 %v64 = load <32 x i32>, <32 x i32>* %v63, align 1, !tbaa !4
124 %v65 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v60, <32 x i32> %v64)
125 %v66 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v50, <32 x i32> %v65)
126 %v67 = sub i32 %v35, %v28
127 %v68 = add i32 %v67, %v30
128 %v69 = add nsw i32 %v68, -1
129 %v70 = getelementptr inbounds i8, i8* %v1, i32 %v69
130 %v71 = bitcast i8* %v70 to <32 x i32>*
131 %v72 = load <32 x i32>, <32 x i32>* %v71, align 1, !tbaa !4
132 %v73 = getelementptr inbounds i8, i8* %v1, i32 %v68
133 %v74 = bitcast i8* %v73 to <32 x i32>*
134 %v75 = load <32 x i32>, <32 x i32>* %v74, align 1, !tbaa !4
135 %v76 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v72, <32 x i32> %v75)
136 %v77 = add nsw i32 %v68, 1
137 %v78 = getelementptr inbounds i8, i8* %v1, i32 %v77
138 %v79 = bitcast i8* %v78 to <32 x i32>*
139 %v80 = load <32 x i32>, <32 x i32>* %v79, align 1, !tbaa !4
140 %v81 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v76, <32 x i32> %v80)
141 %v82 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v66, <32 x i32> %v81)
142 %v83 = mul nsw i32 %v150, %v13
143 %v84 = mul nsw i32 %v17, %v13
144 %v85 = add i32 %v84, %v15
145 %v86 = sub i32 %v83, %v85
146 %v87 = add i32 %v86, %v30
147 %v88 = getelementptr inbounds i8, i8* %v9, i32 %v87
148 %v89 = bitcast i8* %v88 to <32 x i32>*
149 store <32 x i32> %v82, <32 x i32>* %v89, align 1, !tbaa !7
150 %v90 = add nuw nsw i32 %v26, 1
151 %v91 = icmp eq i32 %v90, %v24
152 br i1 %v91, label %b6, label %b2
154 b3: ; preds = %b6, %b3
155 %v92 = phi i32 [ %v147, %b3 ], [ %v24, %b6 ]
156 %v93 = add nsw i32 %v15, %v11
157 %v94 = sub i32 %v93, %v28
158 %v95 = add i32 %v94, %v32
159 %v96 = add nsw i32 %v95, -129
160 %v97 = getelementptr inbounds i8, i8* %v1, i32 %v96
161 %v98 = bitcast i8* %v97 to <32 x i32>*
162 %v99 = load <32 x i32>, <32 x i32>* %v98, align 1, !tbaa !4
163 %v100 = add nsw i32 %v95, -128
164 %v101 = getelementptr inbounds i8, i8* %v1, i32 %v100
165 %v102 = bitcast i8* %v101 to <32 x i32>*
166 %v103 = load <32 x i32>, <32 x i32>* %v102, align 1, !tbaa !4
167 %v104 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v99, <32 x i32> %v103)
168 %v105 = add nsw i32 %v95, -127
169 %v106 = getelementptr inbounds i8, i8* %v1, i32 %v105
170 %v107 = bitcast i8* %v106 to <32 x i32>*
171 %v108 = load <32 x i32>, <32 x i32>* %v107, align 1, !tbaa !4
172 %v109 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v104, <32 x i32> %v108)
173 %v110 = add i32 %v94, %v33
174 %v111 = add nsw i32 %v110, -129
175 %v112 = getelementptr inbounds i8, i8* %v1, i32 %v111
176 %v113 = bitcast i8* %v112 to <32 x i32>*
177 %v114 = load <32 x i32>, <32 x i32>* %v113, align 1, !tbaa !4
178 %v115 = add nsw i32 %v110, -128
179 %v116 = getelementptr inbounds i8, i8* %v1, i32 %v115
180 %v117 = bitcast i8* %v116 to <32 x i32>*
181 %v118 = load <32 x i32>, <32 x i32>* %v117, align 1, !tbaa !4
182 %v119 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v114, <32 x i32> %v118)
183 %v120 = add nsw i32 %v110, -127
184 %v121 = getelementptr inbounds i8, i8* %v1, i32 %v120
185 %v122 = bitcast i8* %v121 to <32 x i32>*
186 %v123 = load <32 x i32>, <32 x i32>* %v122, align 1, !tbaa !4
187 %v124 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v119, <32 x i32> %v123)
188 %v125 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v109, <32 x i32> %v124)
189 %v126 = add i32 %v94, %v35
190 %v127 = add nsw i32 %v126, -129
191 %v128 = getelementptr inbounds i8, i8* %v1, i32 %v127
192 %v129 = bitcast i8* %v128 to <32 x i32>*
193 %v130 = load <32 x i32>, <32 x i32>* %v129, align 1, !tbaa !4
194 %v131 = add nsw i32 %v126, -128
195 %v132 = getelementptr inbounds i8, i8* %v1, i32 %v131
196 %v133 = bitcast i8* %v132 to <32 x i32>*
197 %v134 = load <32 x i32>, <32 x i32>* %v133, align 1, !tbaa !4
198 %v135 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v130, <32 x i32> %v134)
199 %v136 = add nsw i32 %v126, -127
200 %v137 = getelementptr inbounds i8, i8* %v1, i32 %v136
201 %v138 = bitcast i8* %v137 to <32 x i32>*
202 %v139 = load <32 x i32>, <32 x i32>* %v138, align 1, !tbaa !4
203 %v140 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v135, <32 x i32> %v139)
204 %v141 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v125, <32 x i32> %v140)
205 %v142 = add i32 %v11, -128
206 %v143 = sub i32 %v142, %v84
207 %v144 = add i32 %v143, %v83
208 %v145 = getelementptr inbounds i8, i8* %v9, i32 %v144
209 %v146 = bitcast i8* %v145 to <32 x i32>*
210 store <32 x i32> %v141, <32 x i32>* %v146, align 1, !tbaa !7
211 %v147 = add nuw nsw i32 %v92, 1
212 %v148 = icmp eq i32 %v147, %v152
213 br i1 %v148, label %b4, label %b3
215 b4: ; preds = %b6, %b3
216 %v149 = icmp eq i32 %v34, %v20
217 br i1 %v149, label %b11, label %b5
219 b5: ; preds = %b4, %b1
220 %v150 = phi i32 [ %v34, %b4 ], [ %v17, %b1 ]
224 %v151 = add nsw i32 %v11, 127
225 %v152 = ashr i32 %v151, 7
226 %v153 = icmp slt i32 %v24, %v152
227 br i1 %v153, label %b3, label %b4, !prof !3
230 %v154 = add nsw i32 %v11, 127
231 %v155 = ashr i32 %v154, 7
232 %v156 = icmp slt i32 %v24, %v155
233 br i1 %v156, label %b9, label %b11, !prof !3
235 b8: ; preds = %b9, %b8
236 %v157 = phi i32 [ %v221, %b8 ], [ %v24, %b9 ]
237 %v158 = mul nsw i32 %v7, %v3
238 %v159 = add nsw i32 %v158, %v5
239 %v160 = add nsw i32 %v15, %v11
240 %v161 = add nsw i32 %v223, -1
241 %v162 = mul nsw i32 %v161, %v3
242 %v163 = mul nsw i32 %v223, %v3
243 %v164 = add nsw i32 %v223, 1
244 %v165 = mul nsw i32 %v164, %v3
245 %v166 = sub i32 %v160, %v159
246 %v167 = add i32 %v166, %v162
247 %v168 = add nsw i32 %v167, -129
248 %v169 = getelementptr inbounds i8, i8* %v1, i32 %v168
249 %v170 = bitcast i8* %v169 to <32 x i32>*
250 %v171 = load <32 x i32>, <32 x i32>* %v170, align 1, !tbaa !4
251 %v172 = add nsw i32 %v167, -128
252 %v173 = getelementptr inbounds i8, i8* %v1, i32 %v172
253 %v174 = bitcast i8* %v173 to <32 x i32>*
254 %v175 = load <32 x i32>, <32 x i32>* %v174, align 1, !tbaa !4
255 %v176 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v171, <32 x i32> %v175)
256 %v177 = add nsw i32 %v167, -127
257 %v178 = getelementptr inbounds i8, i8* %v1, i32 %v177
258 %v179 = bitcast i8* %v178 to <32 x i32>*
259 %v180 = load <32 x i32>, <32 x i32>* %v179, align 1, !tbaa !4
260 %v181 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v176, <32 x i32> %v180)
261 %v182 = add i32 %v166, %v163
262 %v183 = add nsw i32 %v182, -129
263 %v184 = getelementptr inbounds i8, i8* %v1, i32 %v183
264 %v185 = bitcast i8* %v184 to <32 x i32>*
265 %v186 = load <32 x i32>, <32 x i32>* %v185, align 1, !tbaa !4
266 %v187 = add nsw i32 %v182, -128
267 %v188 = getelementptr inbounds i8, i8* %v1, i32 %v187
268 %v189 = bitcast i8* %v188 to <32 x i32>*
269 %v190 = load <32 x i32>, <32 x i32>* %v189, align 1, !tbaa !4
270 %v191 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v186, <32 x i32> %v190)
271 %v192 = add nsw i32 %v182, -127
272 %v193 = getelementptr inbounds i8, i8* %v1, i32 %v192
273 %v194 = bitcast i8* %v193 to <32 x i32>*
274 %v195 = load <32 x i32>, <32 x i32>* %v194, align 1, !tbaa !4
275 %v196 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v191, <32 x i32> %v195)
276 %v197 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v181, <32 x i32> %v196)
277 %v198 = add i32 %v166, %v165
278 %v199 = add nsw i32 %v198, -129
279 %v200 = getelementptr inbounds i8, i8* %v1, i32 %v199
280 %v201 = bitcast i8* %v200 to <32 x i32>*
281 %v202 = load <32 x i32>, <32 x i32>* %v201, align 1, !tbaa !4
282 %v203 = add nsw i32 %v198, -128
283 %v204 = getelementptr inbounds i8, i8* %v1, i32 %v203
284 %v205 = bitcast i8* %v204 to <32 x i32>*
285 %v206 = load <32 x i32>, <32 x i32>* %v205, align 1, !tbaa !4
286 %v207 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v202, <32 x i32> %v206)
287 %v208 = add nsw i32 %v198, -127
288 %v209 = getelementptr inbounds i8, i8* %v1, i32 %v208
289 %v210 = bitcast i8* %v209 to <32 x i32>*
290 %v211 = load <32 x i32>, <32 x i32>* %v210, align 1, !tbaa !4
291 %v212 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v207, <32 x i32> %v211)
292 %v213 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %v197, <32 x i32> %v212)
293 %v214 = mul nsw i32 %v223, %v13
294 %v215 = mul nsw i32 %v17, %v13
295 %v216 = add i32 %v11, -128
296 %v217 = sub i32 %v216, %v215
297 %v218 = add i32 %v217, %v214
298 %v219 = getelementptr inbounds i8, i8* %v9, i32 %v218
299 %v220 = bitcast i8* %v219 to <32 x i32>*
300 store <32 x i32> %v213, <32 x i32>* %v220, align 1, !tbaa !7
301 %v221 = add nuw nsw i32 %v157, 1
302 %v222 = icmp eq i32 %v221, %v155
303 br i1 %v222, label %b10, label %b8
305 b9: ; preds = %b10, %b7
306 %v223 = phi i32 [ %v164, %b10 ], [ %v17, %b7 ]
310 %v224 = icmp eq i32 %v164, %v20
311 br i1 %v224, label %b11, label %b9
313 b11: ; preds = %b10, %b7, %b4, %b0
317 ; Function Attrs: nounwind readnone
318 declare <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32>, <32 x i32>) #1
320 ; Function Attrs: nounwind
321 define i32 @f7(%4* noalias nocapture readonly %a0, %4* noalias nocapture readonly %a1) #0 {
323 %v0 = tail call i32 @f6(%4* %a0, %4* %a1) #0
327 ; Function Attrs: nounwind
328 define i32 @f8(i8** nocapture readonly %a0) #0 {
330 %v0 = bitcast i8** %a0 to %4**
331 %v1 = load %4*, %4** %v0, align 4
332 %v2 = getelementptr i8*, i8** %a0, i32 1
333 %v3 = bitcast i8** %v2 to %4**
334 %v4 = load %4*, %4** %v3, align 4
335 %v5 = tail call i32 @f7(%4* %v1, %4* %v4)
339 attributes #0 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length128b" }
340 attributes #1 = { nounwind readnone }
342 !llvm.module.flags = !{!0, !1, !2}
344 !0 = !{i32 2, !"halide_use_soft_float_abi", i32 0}
345 !1 = !{i32 2, !"halide_mcpu", !"hexagonv60"}
346 !2 = !{i32 2, !"halide_mattrs", !"+hvxv60,+hvx-length64b"}
347 !3 = !{!"branch_weights", i32 1073741824, i32 0}
348 !4 = !{!5, !5, i64 0}
350 !6 = !{!"Halide buffer"}
351 !7 = !{!8, !8, i64 0}
352 !8 = !{!"dilate3x3", !6}