David Gross | 9197d61 | 2016-02-03 10:42:16 -0800 | [diff] [blame] | 1 | ; Check that the #rs_export_reduce node is recognized. |
David Gross | 79e1a05 | 2016-01-11 14:42:51 -0800 | [diff] [blame] | 2 | |
| 3 | ; RUN: llvm-rs-as %s -o %t |
| 4 | ; RUN: bcinfo %t | FileCheck %s |
| 5 | |
| 6 | ; CHECK: exportReduceNewCount: 8 |
| 7 | ; CHECK: exportReduceNewList[0]: addint - 0x00000001 - 1 - 4 |
| 8 | ; CHECK: accumulator(aiAccum) |
| 9 | ; CHECK: exportReduceNewList[1]: mpyint - 0x00000001 - 1 - 4 |
| 10 | ; CHECK: initializer(mpyInit) |
| 11 | ; CHECK: accumulator(mpyAccum) |
| 12 | ; CHECK: exportReduceNewList[2]: dp - 0x00000001 - 2 - 4 |
| 13 | ; CHECK: accumulator(dpAccum) |
| 14 | ; CHECK: combiner(dpSum) |
| 15 | ; CHECK: exportReduceNewList[3]: findMinAndMax - 0x00000009 - 1 - 16 |
| 16 | ; CHECK: initializer(fMMInit) |
| 17 | ; CHECK: accumulator(fMMAccumulator) |
| 18 | ; CHECK: combiner(fMMCombiner) |
| 19 | ; CHECK: outconverter(fMMOutConverter) |
| 20 | ; CHECK: exportReduceNewList[4]: fz - 0x00000009 - 1 - 4 |
| 21 | ; CHECK: initializer(fzInit) |
| 22 | ; CHECK: accumulator(fzAccum) |
| 23 | ; CHECK: combiner(fzCombine) |
| 24 | ; CHECK: halter(fzFound) |
| 25 | ; CHECK: exportReduceNewList[5]: fz2 - 0x00000019 - 1 - 8 |
| 26 | ; CHECK: initializer(fz2Init) |
| 27 | ; CHECK: accumulator(fz2Accum) |
| 28 | ; CHECK: combiner(fz2Combine) |
| 29 | ; CHECK: halter(fz2Found) |
| 30 | ; CHECK: exportReduceNewList[6]: histogram - 0x00000001 - 1 - 1024 |
| 31 | ; CHECK: accumulator(hsgAccum) |
| 32 | ; CHECK: combiner(hsgCombine) |
| 33 | ; CHECK: exportReduceNewList[7]: mode - 0x00000001 - 1 - 1024 |
| 34 | ; CHECK: accumulator(hsgAccum) |
| 35 | ; CHECK: combiner(hsgCombine) |
| 36 | ; CHECK: outconverter(modeOutConvert) |
| 37 | |
| 38 | ; ModuleID = 'reduce_general_examples.bc' |
| 39 | target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128" |
| 40 | target triple = "aarch64-none-linux-gnueabi" |
| 41 | |
| 42 | %struct.MinAndMax.0 = type { %struct.IndexedVal.1, %struct.IndexedVal.1 } |
| 43 | %struct.IndexedVal.1 = type { float, i32 } |
| 44 | |
| 45 | @fMMInit.r = internal unnamed_addr constant %struct.MinAndMax.0 { %struct.IndexedVal.1 { float 0.000000e+00, i32 -1 }, %struct.IndexedVal.1 { float -0.000000e+00, i32 -1 } }, align 4 |
| 46 | @llvm.used = appending global [20 x i8*] [i8* bitcast (void (<2 x i32>*)* @fz2Init to i8*), i8* bitcast (void ([256 x i32]*, [256 x i32]*)* @hsgCombine to i8*), i8* bitcast (i1 (<2 x i32>*)* @fz2Found to i8*), i8* bitcast (void (i32*, i32)* @mpyAccum to i8*), i8* bitcast (void (%struct.MinAndMax.0*)* @fMMInit to i8*), i8* bitcast (void (float*, float, float)* @dpAccum to i8*), i8* bitcast (void (<2 x i32>*, [256 x i32]*)* @modeOutConvert to i8*), i8* bitcast (void ([256 x i32]*, i8)* @hsgAccum to i8*), i8* bitcast (void (i32*)* @mpyInit to i8*), i8* bitcast (void (%struct.MinAndMax.0*, float, i32)* @fMMAccumulator to i8*), i8* bitcast (void (float*, float*)* @dpSum to i8*), i8* bitcast (void (%struct.MinAndMax.0*, %struct.MinAndMax.0*)* @fMMCombiner to i8*), i8* bitcast (void (i32*, i32*)* @fzCombine to i8*), i8* bitcast (void (i32*, i32)* @aiAccum to i8*), i8* bitcast (void (i32*)* @fzInit to i8*), i8* bitcast (void (i32*, i32, i32)* @fzAccum to i8*), i8* bitcast (i1 (i32*)* @fzFound to i8*), i8* bitcast (void (<2 x i32>*, i32, i32, i32)* @fz2Accum to i8*), i8* bitcast (void (<2 x i32>*, %struct.MinAndMax.0*)* @fMMOutConverter to i8*), i8* bitcast (void (<2 x i32>*, <2 x i32>*)* @fz2Combine to i8*)], section "llvm.metadata" |
| 47 | |
| 48 | ; Function Attrs: nounwind |
| 49 | define internal void @aiAccum(i32* nocapture %accum, i32 %val) #0 { |
| 50 | %1 = load i32, i32* %accum, align 4, !tbaa !18 |
| 51 | %2 = add nsw i32 %1, %val |
| 52 | store i32 %2, i32* %accum, align 4, !tbaa !18 |
| 53 | ret void |
| 54 | } |
| 55 | |
| 56 | ; Function Attrs: nounwind |
| 57 | define internal void @mpyInit(i32* nocapture %accum) #0 { |
| 58 | store i32 1, i32* %accum, align 4, !tbaa !18 |
| 59 | ret void |
| 60 | } |
| 61 | |
| 62 | ; Function Attrs: nounwind |
| 63 | define internal void @mpyAccum(i32* nocapture %accum, i32 %val) #0 { |
| 64 | %1 = load i32, i32* %accum, align 4, !tbaa !18 |
| 65 | %2 = mul nsw i32 %1, %val |
| 66 | store i32 %2, i32* %accum, align 4, !tbaa !18 |
| 67 | ret void |
| 68 | } |
| 69 | |
| 70 | ; Function Attrs: nounwind |
| 71 | define internal void @dpAccum(float* nocapture %accum, float %in1, float %in2) #0 { |
| 72 | %1 = fmul float %in1, %in2 |
| 73 | %2 = load float, float* %accum, align 4, !tbaa !22 |
| 74 | %3 = fadd float %1, %2 |
| 75 | store float %3, float* %accum, align 4, !tbaa !22 |
| 76 | ret void |
| 77 | } |
| 78 | |
| 79 | ; Function Attrs: nounwind |
| 80 | define internal void @dpSum(float* nocapture %accum, float* nocapture %val) #0 { |
| 81 | %1 = load float, float* %val, align 4, !tbaa !22 |
| 82 | %2 = load float, float* %accum, align 4, !tbaa !22 |
| 83 | %3 = fadd float %1, %2 |
| 84 | store float %3, float* %accum, align 4, !tbaa !22 |
| 85 | ret void |
| 86 | } |
| 87 | |
| 88 | ; Function Attrs: nounwind |
| 89 | define internal void @fMMInit(%struct.MinAndMax.0* nocapture %accum) #0 { |
| 90 | %1 = bitcast %struct.MinAndMax.0* %accum to i8* |
| 91 | tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %1, i8* bitcast (%struct.MinAndMax.0* @fMMInit.r to i8*), i64 16, i32 4, i1 false), !tbaa.struct !24 |
| 92 | ret void |
| 93 | } |
| 94 | |
| 95 | ; Function Attrs: nounwind |
| 96 | declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1) #0 |
| 97 | |
| 98 | ; Function Attrs: nounwind |
| 99 | define internal void @fMMAccumulator(%struct.MinAndMax.0* nocapture %accum, float %in, i32 %x) #0 { |
| 100 | %1 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %accum, i64 0, i32 0, i32 0 |
| 101 | %2 = load float, float* %1, align 4, !tbaa !22 |
| 102 | %3 = fcmp ogt float %2, %in |
| 103 | br i1 %3, label %4, label %6 |
| 104 | |
| 105 | ; <label>:4 ; preds = %0 |
| 106 | store float %in, float* %1, align 4 |
| 107 | %5 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %accum, i64 0, i32 0, i32 1 |
| 108 | store i32 %x, i32* %5, align 4 |
| 109 | br label %6 |
| 110 | |
| 111 | ; <label>:6 ; preds = %4, %0 |
| 112 | %7 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %accum, i64 0, i32 1, i32 0 |
| 113 | %8 = load float, float* %7, align 4, !tbaa !22 |
| 114 | %9 = fcmp olt float %8, %in |
| 115 | br i1 %9, label %10, label %12 |
| 116 | |
| 117 | ; <label>:10 ; preds = %6 |
| 118 | store float %in, float* %7, align 4 |
| 119 | %11 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %accum, i64 0, i32 1, i32 1 |
| 120 | store i32 %x, i32* %11, align 4 |
| 121 | br label %12 |
| 122 | |
| 123 | ; <label>:12 ; preds = %10, %6 |
| 124 | ret void |
| 125 | } |
| 126 | |
| 127 | ; Function Attrs: nounwind |
| 128 | define internal void @fMMCombiner(%struct.MinAndMax.0* nocapture %accum, %struct.MinAndMax.0* nocapture %val) #0 { |
| 129 | %1 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 0, i32 0 |
| 130 | %2 = load float, float* %1, align 4, !tbaa !22 |
| 131 | %3 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 0, i32 1 |
| 132 | %4 = load i32, i32* %3, align 4, !tbaa !18 |
| 133 | tail call void @fMMAccumulator(%struct.MinAndMax.0* %accum, float %2, i32 %4) |
| 134 | %5 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 1, i32 0 |
| 135 | %6 = load float, float* %5, align 4, !tbaa !22 |
| 136 | %7 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 1, i32 1 |
| 137 | %8 = load i32, i32* %7, align 4, !tbaa !18 |
| 138 | tail call void @fMMAccumulator(%struct.MinAndMax.0* %accum, float %6, i32 %8) |
| 139 | ret void |
| 140 | } |
| 141 | |
| 142 | ; Function Attrs: nounwind |
| 143 | define internal void @fMMOutConverter(<2 x i32>* nocapture %result, %struct.MinAndMax.0* nocapture %val) #0 { |
| 144 | %1 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 0, i32 1 |
| 145 | %2 = load i32, i32* %1, align 4, !tbaa !18 |
| 146 | %3 = load <2 x i32>, <2 x i32>* %result, align 8 |
| 147 | %4 = insertelement <2 x i32> %3, i32 %2, i64 0 |
| 148 | store <2 x i32> %4, <2 x i32>* %result, align 8 |
| 149 | %5 = getelementptr inbounds %struct.MinAndMax.0, %struct.MinAndMax.0* %val, i64 0, i32 1, i32 1 |
| 150 | %6 = load i32, i32* %5, align 4, !tbaa !18 |
| 151 | %7 = insertelement <2 x i32> %4, i32 %6, i64 1 |
| 152 | store <2 x i32> %7, <2 x i32>* %result, align 8 |
| 153 | ret void |
| 154 | } |
| 155 | |
| 156 | ; Function Attrs: nounwind |
| 157 | define internal void @fzInit(i32* nocapture %accumIdx) #0 { |
| 158 | store i32 -1, i32* %accumIdx, align 4, !tbaa !18 |
| 159 | ret void |
| 160 | } |
| 161 | |
| 162 | ; Function Attrs: nounwind |
| 163 | define internal void @fzAccum(i32* nocapture %accumIdx, i32 %inVal, i32 %x) #0 { |
| 164 | %1 = icmp eq i32 %inVal, 0 |
| 165 | br i1 %1, label %2, label %3 |
| 166 | |
| 167 | ; <label>:2 ; preds = %0 |
| 168 | store i32 %x, i32* %accumIdx, align 4, !tbaa !18 |
| 169 | br label %3 |
| 170 | |
| 171 | ; <label>:3 ; preds = %2, %0 |
| 172 | ret void |
| 173 | } |
| 174 | |
| 175 | ; Function Attrs: nounwind |
| 176 | define internal void @fzCombine(i32* nocapture %accumIdx, i32* nocapture %accumIdx2) #0 { |
| 177 | %1 = load i32, i32* %accumIdx2, align 4, !tbaa !18 |
| 178 | %2 = icmp sgt i32 %1, -1 |
| 179 | br i1 %2, label %3, label %4 |
| 180 | |
| 181 | ; <label>:3 ; preds = %0 |
| 182 | store i32 %1, i32* %accumIdx, align 4, !tbaa !18 |
| 183 | br label %4 |
| 184 | |
| 185 | ; <label>:4 ; preds = %3, %0 |
| 186 | ret void |
| 187 | } |
| 188 | |
| 189 | ; Function Attrs: nounwind readonly |
| 190 | define internal i1 @fzFound(i32* nocapture %accumIdx) #1 { |
| 191 | %1 = load i32, i32* %accumIdx, align 4, !tbaa !18 |
| 192 | %2 = icmp sgt i32 %1, -1 |
| 193 | ret i1 %2 |
| 194 | } |
| 195 | |
| 196 | ; Function Attrs: nounwind |
| 197 | define internal void @fz2Init(<2 x i32>* nocapture %accum) #0 { |
| 198 | store <2 x i32> <i32 -1, i32 -1>, <2 x i32>* %accum, align 8 |
| 199 | ret void |
| 200 | } |
| 201 | |
| 202 | ; Function Attrs: nounwind |
| 203 | define internal void @fz2Accum(<2 x i32>* nocapture %accum, i32 %inVal, i32 %x, i32 %y) #0 { |
| 204 | %1 = icmp eq i32 %inVal, 0 |
| 205 | br i1 %1, label %2, label %5 |
| 206 | |
| 207 | ; <label>:2 ; preds = %0 |
| 208 | %3 = insertelement <2 x i32> undef, i32 %x, i64 0 |
| 209 | %4 = insertelement <2 x i32> %3, i32 %y, i64 1 |
| 210 | store <2 x i32> %4, <2 x i32>* %accum, align 8 |
| 211 | br label %5 |
| 212 | |
| 213 | ; <label>:5 ; preds = %2, %0 |
| 214 | ret void |
| 215 | } |
| 216 | |
| 217 | ; Function Attrs: nounwind |
| 218 | define internal void @fz2Combine(<2 x i32>* nocapture %accum, <2 x i32>* nocapture %accum2) #0 { |
| 219 | %1 = load <2 x i32>, <2 x i32>* %accum2, align 8 |
| 220 | %2 = extractelement <2 x i32> %1, i64 0 |
| 221 | %3 = icmp sgt i32 %2, -1 |
| 222 | br i1 %3, label %4, label %5 |
| 223 | |
| 224 | ; <label>:4 ; preds = %0 |
| 225 | store <2 x i32> %1, <2 x i32>* %accum, align 8, !tbaa !25 |
| 226 | br label %5 |
| 227 | |
| 228 | ; <label>:5 ; preds = %4, %0 |
| 229 | ret void |
| 230 | } |
| 231 | |
| 232 | ; Function Attrs: nounwind readonly |
| 233 | define internal i1 @fz2Found(<2 x i32>* nocapture %accum) #1 { |
| 234 | %1 = load <2 x i32>, <2 x i32>* %accum, align 8 |
| 235 | %2 = extractelement <2 x i32> %1, i64 0 |
| 236 | %3 = icmp sgt i32 %2, -1 |
| 237 | ret i1 %3 |
| 238 | } |
| 239 | |
| 240 | ; Function Attrs: nounwind |
| 241 | define internal void @hsgAccum([256 x i32]* nocapture %h, i8 %in) #0 { |
| 242 | %1 = zext i8 %in to i64 |
| 243 | %2 = getelementptr inbounds [256 x i32], [256 x i32]* %h, i64 0, i64 %1 |
| 244 | %3 = load i32, i32* %2, align 4, !tbaa !18 |
| 245 | %4 = add i32 %3, 1 |
| 246 | store i32 %4, i32* %2, align 4, !tbaa !18 |
| 247 | ret void |
| 248 | } |
| 249 | |
| 250 | ; Function Attrs: nounwind |
| 251 | define internal void @hsgCombine([256 x i32]* nocapture %accum, [256 x i32]* nocapture %addend) #0 { |
| 252 | br label %2 |
| 253 | |
| 254 | ; <label>:1 ; preds = %2 |
| 255 | ret void |
| 256 | |
| 257 | ; <label>:2 ; preds = %2, %0 |
| 258 | %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %2 ] |
| 259 | %3 = getelementptr inbounds [256 x i32], [256 x i32]* %addend, i64 0, i64 %indvars.iv |
| 260 | %4 = load i32, i32* %3, align 4, !tbaa !18 |
| 261 | %5 = getelementptr inbounds [256 x i32], [256 x i32]* %accum, i64 0, i64 %indvars.iv |
| 262 | %6 = load i32, i32* %5, align 4, !tbaa !18 |
| 263 | %7 = add i32 %6, %4 |
| 264 | store i32 %7, i32* %5, align 4, !tbaa !18 |
| 265 | %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 |
| 266 | %exitcond = icmp eq i64 %indvars.iv.next, 256 |
| 267 | br i1 %exitcond, label %1, label %2 |
| 268 | } |
| 269 | |
| 270 | ; Function Attrs: nounwind |
| 271 | define internal void @modeOutConvert(<2 x i32>* nocapture %result, [256 x i32]* nocapture %h) #0 { |
| 272 | br label %8 |
| 273 | |
| 274 | ; <label>:1 ; preds = %8 |
| 275 | %2 = load <2 x i32>, <2 x i32>* %result, align 8 |
| 276 | %3 = insertelement <2 x i32> %2, i32 %i.0.mode.0, i64 0 |
| 277 | store <2 x i32> %3, <2 x i32>* %result, align 8 |
| 278 | %4 = zext i32 %i.0.mode.0 to i64 |
| 279 | %5 = getelementptr inbounds [256 x i32], [256 x i32]* %h, i64 0, i64 %4 |
| 280 | %6 = load i32, i32* %5, align 4, !tbaa !18 |
| 281 | %7 = insertelement <2 x i32> %3, i32 %6, i64 1 |
| 282 | store <2 x i32> %7, <2 x i32>* %result, align 8 |
| 283 | ret void |
| 284 | |
| 285 | ; <label>:8 ; preds = %8, %0 |
| 286 | %indvars.iv = phi i64 [ 1, %0 ], [ %indvars.iv.next, %8 ] |
| 287 | %mode.01 = phi i32 [ 0, %0 ], [ %i.0.mode.0, %8 ] |
| 288 | %9 = getelementptr inbounds [256 x i32], [256 x i32]* %h, i64 0, i64 %indvars.iv |
| 289 | %10 = load i32, i32* %9, align 4, !tbaa !18 |
| 290 | %11 = zext i32 %mode.01 to i64 |
| 291 | %12 = getelementptr inbounds [256 x i32], [256 x i32]* %h, i64 0, i64 %11 |
| 292 | %13 = load i32, i32* %12, align 4, !tbaa !18 |
| 293 | %14 = icmp ugt i32 %10, %13 |
| 294 | %15 = trunc i64 %indvars.iv to i32 |
| 295 | %i.0.mode.0 = select i1 %14, i32 %15, i32 %mode.01 |
| 296 | %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 |
| 297 | %exitcond = icmp eq i64 %indvars.iv.next, 256 |
| 298 | br i1 %exitcond, label %1, label %8 |
| 299 | } |
| 300 | |
| 301 | attributes #0 = { nounwind } |
| 302 | attributes #1 = { nounwind readonly } |
| 303 | |
| 304 | !llvm.ident = !{!0} |
| 305 | !\23pragma = !{!1, !2} |
David Gross | 9197d61 | 2016-02-03 10:42:16 -0800 | [diff] [blame] | 306 | !\23rs_export_reduce = !{!3, !5, !7, !9, !11, !13, !15, !17} |
David Gross | 79e1a05 | 2016-01-11 14:42:51 -0800 | [diff] [blame] | 307 | |
| 308 | !0 = !{!"clang version 3.6 "} |
| 309 | !1 = !{!"version", !"1"} |
| 310 | !2 = !{!"java_package_name", !"examples"} |
| 311 | !3 = !{!"addint", !"4", !4} |
| 312 | !4 = !{!"aiAccum", !"1"} |
| 313 | !5 = !{!"mpyint", !"4", !6, !"mpyInit"} |
| 314 | !6 = !{!"mpyAccum", !"1"} |
| 315 | !7 = !{!"dp", !"4", !8, null, !"dpSum"} |
| 316 | !8 = !{!"dpAccum", !"1"} |
| 317 | !9 = !{!"findMinAndMax", !"16", !10, !"fMMInit", !"fMMCombiner", !"fMMOutConverter"} |
| 318 | !10 = !{!"fMMAccumulator", !"9"} |
| 319 | !11 = !{!"fz", !"4", !12, !"fzInit", !"fzCombine", null, !"fzFound"} |
| 320 | !12 = !{!"fzAccum", !"9"} |
| 321 | !13 = !{!"fz2", !"8", !14, !"fz2Init", !"fz2Combine", null, !"fz2Found"} |
| 322 | !14 = !{!"fz2Accum", !"25"} |
| 323 | !15 = !{!"histogram", !"1024", !16, null, !"hsgCombine"} |
| 324 | !16 = !{!"hsgAccum", !"1"} |
| 325 | !17 = !{!"mode", !"1024", !16, null, !"hsgCombine", !"modeOutConvert"} |
| 326 | !18 = !{!19, !19, i64 0} |
| 327 | !19 = !{!"int", !20, i64 0} |
| 328 | !20 = !{!"omnipotent char", !21, i64 0} |
| 329 | !21 = !{!"Simple C/C++ TBAA"} |
| 330 | !22 = !{!23, !23, i64 0} |
| 331 | !23 = !{!"float", !20, i64 0} |
| 332 | !24 = !{i64 0, i64 4, !22, i64 4, i64 4, !18, i64 8, i64 4, !22, i64 12, i64 4, !18} |
| 333 | !25 = !{!20, !20, i64 0} |