1; RUN: llc -march=hexagon -O2 -mcpu=hexagonv60  < %s | FileCheck %s
2
3; CHECK: vasr(v{{[0-9]+}}.h,v{{[0-9]+}}.h,r{{[0-7]+}}):sat
4
5target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a:0-n16:32"
6target triple = "hexagon-unknown--elf"
7
8%struct.buffer_t = type { i64, i8*, [4 x i32], [4 x i32], [4 x i32], i32, i8, i8, [6 x i8] }
9
10; Function Attrs: norecurse nounwind
11define i32 @__test_vasr(%struct.buffer_t* noalias nocapture %f.buffer, %struct.buffer_t* noalias nocapture %g.buffer, %struct.buffer_t* noalias nocapture %res.buffer) #0 {
12entry:
13  %buf_host = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 1
14  %f.host = load i8*, i8** %buf_host, align 4
15  %buf_dev = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 0
16  %f.dev = load i64, i64* %buf_dev, align 8
17  %0 = icmp eq i8* %f.host, null
18  %1 = icmp eq i64 %f.dev, 0
19  %f.host_and_dev_are_null = and i1 %0, %1
20  %buf_min = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 4, i32 0
21  %f.min.0 = load i32, i32* %buf_min, align 4
22  %buf_host10 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 1
23  %g.host = load i8*, i8** %buf_host10, align 4
24  %buf_dev11 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 0
25  %g.dev = load i64, i64* %buf_dev11, align 8
26  %2 = icmp eq i8* %g.host, null
27  %3 = icmp eq i64 %g.dev, 0
28  %g.host_and_dev_are_null = and i1 %2, %3
29  %buf_min22 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 4, i32 0
30  %g.min.0 = load i32, i32* %buf_min22, align 4
31  %buf_host27 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 1
32  %res.host = load i8*, i8** %buf_host27, align 4
33  %buf_dev28 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 0
34  %res.dev = load i64, i64* %buf_dev28, align 8
35  %4 = icmp eq i8* %res.host, null
36  %5 = icmp eq i64 %res.dev, 0
37  %res.host_and_dev_are_null = and i1 %4, %5
38  %buf_extent31 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 2, i32 0
39  %res.extent.0 = load i32, i32* %buf_extent31, align 4
40  %buf_min39 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 4, i32 0
41  %res.min.0 = load i32, i32* %buf_min39, align 4
42  %6 = add nsw i32 %res.extent.0, -1
43  %7 = and i32 %6, -64
44  %8 = add i32 %res.min.0, 63
45  %9 = add i32 %8, %7
46  %10 = add nsw i32 %res.min.0, %res.extent.0
47  %11 = add nsw i32 %10, -1
48  %12 = icmp slt i32 %9, %11
49  %13 = select i1 %12, i32 %9, i32 %11
50  %14 = add nsw i32 %10, -64
51  %15 = icmp slt i32 %res.min.0, %14
52  %16 = select i1 %15, i32 %res.min.0, i32 %14
53  %f.extent.0.required.s = sub nsw i32 %13, %16
54  br i1 %f.host_and_dev_are_null, label %true_bb, label %after_bb
55
56true_bb:                                          ; preds = %entry
57  %buf_elem_size44 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 5
58  store i32 1, i32* %buf_elem_size44, align 4
59  store i32 %16, i32* %buf_min, align 4
60  %17 = add nsw i32 %f.extent.0.required.s, 1
61  %buf_extent46 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 2, i32 0
62  store i32 %17, i32* %buf_extent46, align 4
63  %buf_stride47 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 3, i32 0
64  store i32 1, i32* %buf_stride47, align 4
65  %buf_min48 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 4, i32 1
66  store i32 0, i32* %buf_min48, align 4
67  %buf_extent49 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 2, i32 1
68  store i32 0, i32* %buf_extent49, align 4
69  %buf_stride50 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 3, i32 1
70  store i32 0, i32* %buf_stride50, align 4
71  %buf_min51 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 4, i32 2
72  store i32 0, i32* %buf_min51, align 4
73  %buf_extent52 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 2, i32 2
74  store i32 0, i32* %buf_extent52, align 4
75  %buf_stride53 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 3, i32 2
76  store i32 0, i32* %buf_stride53, align 4
77  %buf_min54 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 4, i32 3
78  store i32 0, i32* %buf_min54, align 4
79  %buf_extent55 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 2, i32 3
80  store i32 0, i32* %buf_extent55, align 4
81  %buf_stride56 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %f.buffer, i32 0, i32 3, i32 3
82  store i32 0, i32* %buf_stride56, align 4
83  br label %after_bb
84
85after_bb:                                         ; preds = %true_bb, %entry
86  br i1 %g.host_and_dev_are_null, label %true_bb57, label %after_bb59
87
88true_bb57:                                        ; preds = %after_bb
89  %buf_elem_size60 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 5
90  store i32 1, i32* %buf_elem_size60, align 4
91  store i32 %16, i32* %buf_min22, align 4
92  %18 = add nsw i32 %f.extent.0.required.s, 1
93  %buf_extent62 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 2, i32 0
94  store i32 %18, i32* %buf_extent62, align 4
95  %buf_stride63 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 3, i32 0
96  store i32 1, i32* %buf_stride63, align 4
97  %buf_min64 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 4, i32 1
98  store i32 0, i32* %buf_min64, align 4
99  %buf_extent65 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 2, i32 1
100  store i32 0, i32* %buf_extent65, align 4
101  %buf_stride66 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 3, i32 1
102  store i32 0, i32* %buf_stride66, align 4
103  %buf_min67 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 4, i32 2
104  store i32 0, i32* %buf_min67, align 4
105  %buf_extent68 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 2, i32 2
106  store i32 0, i32* %buf_extent68, align 4
107  %buf_stride69 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 3, i32 2
108  store i32 0, i32* %buf_stride69, align 4
109  %buf_min70 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 4, i32 3
110  store i32 0, i32* %buf_min70, align 4
111  %buf_extent71 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 2, i32 3
112  store i32 0, i32* %buf_extent71, align 4
113  %buf_stride72 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %g.buffer, i32 0, i32 3, i32 3
114  store i32 0, i32* %buf_stride72, align 4
115  br label %after_bb59
116
117after_bb59:                                       ; preds = %true_bb57, %after_bb
118  br i1 %res.host_and_dev_are_null, label %after_bb75.thread, label %after_bb75
119
120after_bb75.thread:                                ; preds = %after_bb59
121  %buf_elem_size76 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 5
122  store i32 1, i32* %buf_elem_size76, align 4
123  store i32 %16, i32* %buf_min39, align 4
124  %19 = add nsw i32 %f.extent.0.required.s, 1
125  store i32 %19, i32* %buf_extent31, align 4
126  %buf_stride79 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 3, i32 0
127  store i32 1, i32* %buf_stride79, align 4
128  %buf_min80 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 4, i32 1
129  store i32 0, i32* %buf_min80, align 4
130  %buf_extent81 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 2, i32 1
131  store i32 0, i32* %buf_extent81, align 4
132  %buf_stride82 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 3, i32 1
133  store i32 0, i32* %buf_stride82, align 4
134  %buf_min83 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 4, i32 2
135  store i32 0, i32* %buf_min83, align 4
136  %buf_extent84 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 2, i32 2
137  store i32 0, i32* %buf_extent84, align 4
138  %buf_stride85 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 3, i32 2
139  store i32 0, i32* %buf_stride85, align 4
140  %buf_min86 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 4, i32 3
141  store i32 0, i32* %buf_min86, align 4
142  %buf_extent87 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 2, i32 3
143  store i32 0, i32* %buf_extent87, align 4
144  %buf_stride88 = getelementptr inbounds %struct.buffer_t, %struct.buffer_t* %res.buffer, i32 0, i32 3, i32 3
145  store i32 0, i32* %buf_stride88, align 4
146  br label %destructor_block
147
148after_bb75:                                       ; preds = %after_bb59
149  %20 = or i1 %f.host_and_dev_are_null, %g.host_and_dev_are_null
150  br i1 %20, label %destructor_block, label %"produce res"
151
152"produce res":                                    ; preds = %after_bb75
153  %21 = ashr i32 %res.extent.0, 6
154  %22 = icmp sgt i32 %21, 0
155  br i1 %22, label %"for res.s0.x.x", label %"end for res.s0.x.x", !prof !4
156
157"for res.s0.x.x":                                 ; preds = %"for res.s0.x.x", %"produce res"
158  %res.s0.x.x = phi i32 [ %41, %"for res.s0.x.x" ], [ 0, %"produce res" ]
159  %23 = shl nsw i32 %res.s0.x.x, 6
160  %24 = add nsw i32 %23, %res.min.0
161  %25 = sub nsw i32 %24, %f.min.0
162  %26 = getelementptr inbounds i8, i8* %f.host, i32 %25
163  %27 = bitcast i8* %26 to <16 x i32>*
164  %28 = load <16 x i32>, <16 x i32>* %27, align 1, !tbaa !5
165  %29 = tail call <32 x i32> @llvm.hexagon.V6.vzb(<16 x i32> %28)
166  %30 = sub nsw i32 %24, %g.min.0
167  %31 = getelementptr inbounds i8, i8* %g.host, i32 %30
168  %32 = bitcast i8* %31 to <16 x i32>*
169  %33 = load <16 x i32>, <16 x i32>* %32, align 1, !tbaa !8
170  %34 = tail call <32 x i32> @llvm.hexagon.V6.vzb(<16 x i32> %33)
171  %35 = tail call <32 x i32> @llvm.hexagon.V6.vaddh.dv(<32 x i32> %29, <32 x i32> %34)
172  %36 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %35)
173  %37 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %35)
174  %38 = tail call <16 x i32> @llvm.hexagon.V6.vasrhubsat(<16 x i32> %36, <16 x i32> %37, i32 4)
175  %39 = getelementptr inbounds i8, i8* %res.host, i32 %23
176  %40 = bitcast i8* %39 to <16 x i32>*
177  store <16 x i32> %38, <16 x i32>* %40, align 1, !tbaa !10
178  %41 = add nuw nsw i32 %res.s0.x.x, 1
179  %42 = icmp eq i32 %41, %21
180  br i1 %42, label %"end for res.s0.x.x", label %"for res.s0.x.x"
181
182"end for res.s0.x.x":                             ; preds = %"for res.s0.x.x", %"produce res"
183  %43 = add nsw i32 %res.extent.0, 63
184  %44 = ashr i32 %43, 6
185  %45 = icmp sgt i32 %44, %21
186  br i1 %45, label %"for res.s0.x.x92.preheader", label %destructor_block, !prof !4
187
188"for res.s0.x.x92.preheader":                     ; preds = %"end for res.s0.x.x"
189  %46 = sub i32 -64, %f.min.0
190  %47 = add i32 %46, %10
191  %48 = getelementptr inbounds i8, i8* %f.host, i32 %47
192  %49 = bitcast i8* %48 to <16 x i32>*
193  %50 = load <16 x i32>, <16 x i32>* %49, align 1
194  %51 = tail call <32 x i32> @llvm.hexagon.V6.vzb(<16 x i32> %50)
195  %52 = sub i32 -64, %g.min.0
196  %53 = add i32 %52, %10
197  %54 = getelementptr inbounds i8, i8* %g.host, i32 %53
198  %55 = bitcast i8* %54 to <16 x i32>*
199  %56 = load <16 x i32>, <16 x i32>* %55, align 1
200  %57 = tail call <32 x i32> @llvm.hexagon.V6.vzb(<16 x i32> %56)
201  %58 = tail call <32 x i32> @llvm.hexagon.V6.vaddh.dv(<32 x i32> %51, <32 x i32> %57)
202  %59 = tail call <16 x i32> @llvm.hexagon.V6.lo(<32 x i32> %58)
203  %60 = add nsw i32 %res.extent.0, -64
204  %61 = getelementptr inbounds i8, i8* %res.host, i32 %60
205  %62 = tail call <16 x i32> @llvm.hexagon.V6.hi(<32 x i32> %58)
206  %63 = tail call <16 x i32> @llvm.hexagon.V6.vasrhubsat(<16 x i32> %62, <16 x i32> %59, i32 4)
207  %64 = bitcast i8* %61 to <16 x i32>*
208  store <16 x i32> %63, <16 x i32>* %64, align 1, !tbaa !10
209  br label %destructor_block
210
211destructor_block:                                 ; preds = %"for res.s0.x.x92.preheader", %"end for res.s0.x.x", %after_bb75, %after_bb75.thread
212  ret i32 0
213}
214
215; Function Attrs: nounwind readnone
216declare <32 x i32> @llvm.hexagon.V6.vaddh.dv(<32 x i32>, <32 x i32>) #1
217
218; Function Attrs: nounwind readnone
219declare <32 x i32> @llvm.hexagon.V6.vzb(<16 x i32>) #1
220
221; Function Attrs: nounwind readnone
222declare <16 x i32> @llvm.hexagon.V6.hi(<32 x i32>) #1
223
224; Function Attrs: nounwind readnone
225declare <16 x i32> @llvm.hexagon.V6.lo(<32 x i32>) #1
226
227; Function Attrs: nounwind readnone
228declare <16 x i32> @llvm.hexagon.V6.vasrhubsat(<16 x i32>, <16 x i32>, i32) #1
229
230attributes #0 = { norecurse nounwind }
231attributes #1 = { nounwind readnone }
232
233!llvm.ident = !{!0, !0, !0, !0, !0, !0, !0, !0, !0, !0, !0}
234!llvm.module.flags = !{!1, !2, !3}
235
236!0 = !{!"Clang $LLVM_VERSION_MAJOR.$LLVM_VERSION_MINOR (based on LLVM 3.8.0)"}
237!1 = !{i32 2, !"halide_use_soft_float_abi", i32 0}
238!2 = !{i32 2, !"halide_mcpu", !"hexagonv60"}
239!3 = !{i32 2, !"halide_mattrs", !"+hvx"}
240!4 = !{!"branch_weights", i32 1073741824, i32 0}
241!5 = !{!6, !6, i64 0}
242!6 = !{!"f", !7}
243!7 = !{!"Halide buffer"}
244!8 = !{!9, !9, i64 0}
245!9 = !{!"g", !7}
246!10 = !{!11, !11, i64 0}
247!11 = !{!"res", !7}
248