76 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -S -mtriple=x86_64-unknown -mattr=+avx -passes=slp-vectorizer | FileCheck %s3 4 5 6define i32 @jumbled-load(ptr noalias nocapture %in, ptr noalias nocapture %inn, ptr noalias nocapture %out) {7; CHECK-LABEL: @jumbled-load(8; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[IN:%.*]], align 49; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr [[INN:%.*]], align 410; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP4]], <4 x i32> poison, <4 x i32> <i32 2, i32 0, i32 3, i32 1>11; CHECK-NEXT: [[TMP5:%.*]] = mul <4 x i32> [[TMP2]], [[SHUFFLE]]12; CHECK-NEXT: [[SHUFFLE1:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>13; CHECK-NEXT: store <4 x i32> [[SHUFFLE1]], ptr [[OUT:%.*]], align 414; CHECK-NEXT: ret i32 undef15;16 %load.1 = load i32, ptr %in, align 417 %gep.1 = getelementptr inbounds i32, ptr %in, i64 318 %load.2 = load i32, ptr %gep.1, align 419 %gep.2 = getelementptr inbounds i32, ptr %in, i64 120 %load.3 = load i32, ptr %gep.2, align 421 %gep.3 = getelementptr inbounds i32, ptr %in, i64 222 %load.4 = load i32, ptr %gep.3, align 423 %load.5 = load i32, ptr %inn, align 424 %gep.4 = getelementptr inbounds i32, ptr %inn, i64 225 %load.6 = load i32, ptr %gep.4, align 426 %gep.5 = getelementptr inbounds i32, ptr %inn, i64 327 %load.7 = load i32, ptr %gep.5, align 428 %gep.6 = getelementptr inbounds i32, ptr %inn, i64 129 %load.8 = load i32, ptr %gep.6, align 430 %mul.1 = mul i32 %load.3, %load.531 %mul.2 = mul i32 %load.2, %load.832 %mul.3 = mul i32 %load.4, %load.733 %mul.4 = mul i32 %load.1, %load.634 store i32 %mul.1, ptr %out, align 435 %gep.8 = getelementptr inbounds i32, ptr %out, i64 136 store i32 %mul.2, ptr %gep.8, align 437 %gep.9 = getelementptr inbounds i32, ptr %out, i64 238 store i32 %mul.3, ptr %gep.9, align 439 %gep.10 = getelementptr inbounds i32, ptr %out, i64 340 store i32 %mul.4, ptr %gep.10, align 441 42 ret i32 undef43}44 45 46define i32 @jumbled-load-multiuses(ptr noalias nocapture %in, ptr noalias nocapture %out) {47; CHECK-LABEL: @jumbled-load-multiuses(48; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[IN:%.*]], align 449; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> <i32 1, i32 2, i32 0, i32 3>50; CHECK-NEXT: [[TMP4:%.*]] = mul <4 x i32> [[TMP2]], [[TMP3]]51; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP4]], <4 x i32> poison, <4 x i32> <i32 1, i32 3, i32 2, i32 0>52; CHECK-NEXT: store <4 x i32> [[SHUFFLE]], ptr [[OUT:%.*]], align 453; CHECK-NEXT: ret i32 undef54;55 %load.1 = load i32, ptr %in, align 456 %gep.1 = getelementptr inbounds i32, ptr %in, i64 357 %load.2 = load i32, ptr %gep.1, align 458 %gep.2 = getelementptr inbounds i32, ptr %in, i64 159 %load.3 = load i32, ptr %gep.2, align 460 %gep.3 = getelementptr inbounds i32, ptr %in, i64 261 %load.4 = load i32, ptr %gep.3, align 462 %mul.1 = mul i32 %load.3, %load.463 %mul.2 = mul i32 %load.2, %load.264 %mul.3 = mul i32 %load.4, %load.165 %mul.4 = mul i32 %load.1, %load.366 store i32 %mul.1, ptr %out, align 467 %gep.8 = getelementptr inbounds i32, ptr %out, i64 168 store i32 %mul.2, ptr %gep.8, align 469 %gep.9 = getelementptr inbounds i32, ptr %out, i64 270 store i32 %mul.3, ptr %gep.9, align 471 %gep.10 = getelementptr inbounds i32, ptr %out, i64 372 store i32 %mul.4, ptr %gep.10, align 473 74 ret i32 undef75}76