1 ; RUN: llc -O3 -march=hexagon < %s | FileCheck %s
2 ; CHECK: v{{[0-9]+}}.w = vadd
4 target triple = "hexagon"
6 ; Function Attrs: nounwind readnone
7 declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #0
9 ; Function Attrs: nounwind readnone
10 declare <16 x i32> @llvm.hexagon.V6.vshuffb(<16 x i32>) #0
12 ; Function Attrs: nounwind readnone
13 declare <32 x i32> @llvm.hexagon.V6.vmpyubv(<16 x i32>, <16 x i32>) #0
15 ; Function Attrs: nounwind readnone
16 declare <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32>, <16 x i32>, i32) #0
18 ; Function Attrs: nounwind readnone
19 declare <16 x i32> @llvm.hexagon.V6.vaddh(<16 x i32>, <16 x i32>) #0
21 ; Function Attrs: nounwind
22 define void @f0(i16* noalias nocapture %a0, i32* noalias nocapture readonly %a1, i32 %a2, i8* noalias nocapture readonly %a3, i1 %cond) #1 {
24 %v0 = add nsw i32 %a2, 63
26 %v2 = bitcast i16* %a0 to <16 x i32>*
27 %v3 = bitcast i8* %a3 to <16 x i32>*
28 %v4 = getelementptr inbounds i32, i32* %a1, i32 32
29 %v5 = bitcast i32* %v4 to <16 x i32>*
30 %v6 = load <16 x i32>, <16 x i32>* %v5, align 64, !tbaa !0
31 %v7 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 32768)
32 %v8 = tail call <16 x i32> @llvm.hexagon.V6.lvsplatw(i32 2147450879)
33 %v9 = icmp sgt i32 %v1, 0
34 br i1 %v9, label %b1, label %b4
37 %v10 = bitcast i32* %a1 to <16 x i32>*
38 %v11 = load <16 x i32>, <16 x i32>* %v10, align 64, !tbaa !0
39 %v12 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v6, <16 x i32> %v11, i32 2)
40 %v13 = getelementptr inbounds i32, i32* %a1, i32 48
41 %v14 = tail call <16 x i32> @llvm.hexagon.V6.vaddh(<16 x i32> %v12, <16 x i32> undef)
42 %v15 = bitcast i32* %v13 to <16 x i32>*
43 br i1 %cond, label %b2, label %b3
46 %v16 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 1
47 %v17 = load <16 x i32>, <16 x i32>* %v16, align 64, !tbaa !0
48 %v18 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v17, <16 x i32> %v6, i32 4)
49 %v19 = load <16 x i32>, <16 x i32>* %v15, align 64, !tbaa !0
50 %v20 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 2
51 %v21 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v18, <16 x i32> %v19)
52 %v22 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v21, <16 x i32> %v14, i32 4)
53 %v23 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v21, <16 x i32> %v14, i32 8)
54 %v24 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v21, <16 x i32> %v14, i32 12)
55 %v25 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v14, <16 x i32> %v22)
56 %v26 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v25, <16 x i32> %v23)
57 %v27 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v26, <16 x i32> %v24)
58 %v28 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v19, <16 x i32> undef, i32 16)
59 %v29 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v27, <16 x i32> %v11)
60 %v30 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v27, <16 x i32> %v28)
61 %v31 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v29, i32 53019433)
62 %v32 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v30, i32 53019433)
63 %v33 = load <16 x i32>, <16 x i32>* %v3, align 64, !tbaa !0
64 %v34 = tail call <16 x i32> @llvm.hexagon.V6.vshuffb(<16 x i32> %v33)
65 %v35 = tail call <32 x i32> @llvm.hexagon.V6.vmpyubv(<16 x i32> %v34, <16 x i32> %v34)
66 %v36 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v32, <16 x i32> %v31)
67 store <16 x i32> %v36, <16 x i32>* %v2, align 64, !tbaa !0
68 %v37 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 3
69 %v38 = load <16 x i32>, <16 x i32>* %v37, align 64, !tbaa !0
70 %v39 = load <16 x i32>, <16 x i32>* %v20, align 64, !tbaa !0
71 %v40 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 4
72 %v41 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> undef, <16 x i32> %v39)
73 %v42 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v41, <16 x i32> %v21, i32 4)
74 %v43 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v41, <16 x i32> %v21, i32 8)
75 %v44 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v41, <16 x i32> %v21, i32 12)
76 %v45 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v21, <16 x i32> %v42)
77 %v46 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v45, <16 x i32> %v43)
78 %v47 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v46, <16 x i32> %v44)
79 %v48 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v47, <16 x i32> %v6)
80 %v49 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v47, <16 x i32> undef)
81 %v50 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v48, i32 53019433)
82 %v51 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v49, i32 53019433)
83 %v52 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v51, <16 x i32> %v50)
84 %v53 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v52, <16 x i32> undef, i32 56)
85 %v54 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v35)
86 %v55 = tail call <16 x i32> @llvm.hexagon.V6.vsubuhsat(<16 x i32> %v53, <16 x i32> %v54)
87 %v56 = tail call <16 x i32> @llvm.hexagon.V6.vminuh(<16 x i32> %v55, <16 x i32> %v8)
88 %v57 = getelementptr inbounds <16 x i32>, <16 x i32>* %v2, i32 undef
89 store <16 x i32> %v56, <16 x i32>* %v57, align 64, !tbaa !0
90 %v58 = getelementptr <16 x i32>, <16 x i32>* %v2, i32 2
91 %v59 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 5
92 %v60 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> zeroinitializer, <16 x i32> %v38, i32 4)
93 %v61 = load <16 x i32>, <16 x i32>* %v40, align 64, !tbaa !0
94 %v62 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v60, <16 x i32> %v61)
95 %v63 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v62, <16 x i32> %v41, i32 4)
96 %v64 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v62, <16 x i32> %v41, i32 8)
97 %v65 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v62, <16 x i32> %v41, i32 12)
98 %v66 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v41, <16 x i32> %v63)
99 %v67 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v66, <16 x i32> %v64)
100 %v68 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v67, <16 x i32> %v65)
101 %v69 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v61, <16 x i32> %v39, i32 16)
102 %v70 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 1
103 %v71 = load <16 x i32>, <16 x i32>* %v70, align 64, !tbaa !0
104 %v72 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v68, <16 x i32> %v71)
105 %v73 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v68, <16 x i32> %v69)
106 %v74 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v72, i32 53019433)
107 %v75 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v73, i32 53019433)
108 %v76 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v75, <16 x i32> %v74)
109 store <16 x i32> %v76, <16 x i32>* %v58, align 64, !tbaa !0
110 %v77 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 7
111 %v78 = load <16 x i32>, <16 x i32>* %v77, align 64, !tbaa !0
112 %v79 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> undef, <16 x i32> undef)
113 %v80 = getelementptr <16 x i32>, <16 x i32>* %v2, i32 4
114 %v81 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 9
115 %v82 = load <16 x i32>, <16 x i32>* %v81, align 64, !tbaa !0
116 %v83 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v82, <16 x i32> %v78, i32 4)
117 %v84 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 10
118 %v85 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v83, <16 x i32> undef)
119 %v86 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v85, <16 x i32> %v79, i32 4)
120 %v87 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v85, <16 x i32> %v79, i32 8)
121 %v88 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v85, <16 x i32> %v79, i32 12)
122 %v89 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v79, <16 x i32> %v86)
123 %v90 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v89, <16 x i32> %v87)
124 %v91 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v90, <16 x i32> %v88)
125 %v92 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> undef, <16 x i32> undef, i32 16)
126 %v93 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v91, <16 x i32> zeroinitializer)
127 %v94 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v91, <16 x i32> %v92)
128 %v95 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v93, i32 53019433)
129 %v96 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v94, i32 53019433)
130 %v97 = tail call <32 x i32> @llvm.hexagon.V6.vmpyubv(<16 x i32> undef, <16 x i32> undef)
131 %v98 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v96, <16 x i32> %v95)
132 store <16 x i32> %v98, <16 x i32>* %v80, align 64, !tbaa !0
133 %v99 = getelementptr inbounds <16 x i32>, <16 x i32>* %v15, i32 11
134 %v100 = load <16 x i32>, <16 x i32>* %v99, align 64, !tbaa !0
135 %v101 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v100, <16 x i32> %v82, i32 4)
136 %v102 = load <16 x i32>, <16 x i32>* %v84, align 64, !tbaa !0
137 %v103 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v101, <16 x i32> %v102)
138 %v104 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v103, <16 x i32> %v85, i32 4)
139 %v105 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v103, <16 x i32> %v85, i32 8)
140 %v106 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v85, <16 x i32> %v104)
141 %v107 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v106, <16 x i32> %v105)
142 %v108 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v107, <16 x i32> undef)
143 %v109 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v102, <16 x i32> undef, i32 16)
144 %v110 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v108, <16 x i32> %v78)
145 %v111 = tail call <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32> %v108, <16 x i32> %v109)
146 %v112 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v110, i32 53019433)
147 %v113 = tail call <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32> %v7, <16 x i32> %v111, i32 53019433)
148 %v114 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> %v113, <16 x i32> %v112)
149 %v115 = tail call <16 x i32> @llvm.hexagon.V6.valignb(<16 x i32> %v114, <16 x i32> undef, i32 56)
150 %v116 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %v97)
151 %v117 = tail call <16 x i32> @llvm.hexagon.V6.vsubuhsat(<16 x i32> %v115, <16 x i32> %v116)
152 %v118 = tail call <16 x i32> @llvm.hexagon.V6.vminuh(<16 x i32> %v117, <16 x i32> %v8)
153 %v119 = getelementptr inbounds <16 x i32>, <16 x i32>* %v2, i32 undef
154 store <16 x i32> %v118, <16 x i32>* %v119, align 64, !tbaa !0
155 %v120 = getelementptr <16 x i32>, <16 x i32>* %v2, i32 6
156 %v121 = tail call <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32> undef, <16 x i32> undef)
157 store <16 x i32> %v121, <16 x i32>* %v120, align 64, !tbaa !0
167 ; Function Attrs: nounwind readnone
168 declare <16 x i32> @llvm.hexagon.V6.lvsplatw(i32) #0
170 ; Function Attrs: nounwind readnone
171 declare <16 x i32> @llvm.hexagon.V6.vaddw(<16 x i32>, <16 x i32>) #0
173 ; Function Attrs: nounwind readnone
174 declare <16 x i32> @llvm.hexagon.V6.vmpyiwh.acc(<16 x i32>, <16 x i32>, i32) #0
176 ; Function Attrs: nounwind readnone
177 declare <16 x i32> @llvm.hexagon.V6.vshufoh(<16 x i32>, <16 x i32>) #0
179 ; Function Attrs: nounwind readnone
180 declare <16 x i32> @llvm.hexagon.V6.vsubuhsat(<16 x i32>, <16 x i32>) #0
182 ; Function Attrs: nounwind readnone
183 declare <16 x i32> @llvm.hexagon.V6.vminuh(<16 x i32>, <16 x i32>) #0
185 attributes #0 = { nounwind readnone }
186 attributes #1 = { nounwind "target-cpu"="hexagonv60" "target-features"="+hvxv60,+hvx-length64b" }
188 !0 = !{!1, !1, i64 0}
189 !1 = !{!"omnipotent char", !2, i64 0}
190 !2 = !{!"Simple C/C++ TBAA"}