1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=slp-vectorizer -mtriple=aarch64-unknown-unknown -mcpu=cortex-a53 -S | FileCheck %s 3 4@b = common global i32 0, align 4 5@a = common global ptr null, align 8 6 7define void @foo(ptr %d, ptr %e) { 8; CHECK-LABEL: @foo( 9; CHECK-NEXT: entry: 10; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [4 x double], ptr [[D:%.*]], i64 2 11; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds [4 x double], ptr [[D]], i64 0, i64 3 12; CHECK-NEXT: [[TMP0:%.*]] = load double, ptr [[ARRAYIDX3]], align 8 13; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds [4 x double], ptr [[E:%.*]], i64 3 14; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds [4 x double], ptr [[D]], i64 1, i64 3 15; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr [[ARRAYIDX15]], align 8 16; CHECK-NEXT: [[ARRAYIDX17:%.*]] = getelementptr inbounds [4 x double], ptr [[D]], i64 3, i64 2 17; CHECK-NEXT: [[TMP2:%.*]] = load double, ptr [[ARRAYIDX17]], align 8 18; CHECK-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds [4 x double], ptr [[E]], i64 2, i64 3 19; CHECK-NEXT: [[TMP3:%.*]] = load double, ptr [[ARRAYIDX19]], align 8 20; CHECK-NEXT: [[TMP4:%.*]] = tail call double @llvm.fmuladd.f64(double [[TMP2]], double [[TMP3]], double [[TMP1]]) 21; CHECK-NEXT: [[ARRAYIDX21:%.*]] = getelementptr inbounds [4 x double], ptr [[D]], i64 3, i64 3 22; CHECK-NEXT: [[TMP5:%.*]] = load double, ptr [[ARRAYIDX21]], align 8 23; CHECK-NEXT: [[ARRAYIDX23:%.*]] = getelementptr inbounds [4 x double], ptr [[E]], i64 3, i64 3 24; CHECK-NEXT: [[TMP6:%.*]] = load double, ptr [[ARRAYIDX23]], align 8 25; CHECK-NEXT: [[TMP7:%.*]] = tail call double @llvm.fmuladd.f64(double [[TMP5]], double [[TMP6]], double [[TMP4]]) 26; CHECK-NEXT: [[TMP8:%.*]] = load ptr, ptr @a, align 8 27; CHECK-NEXT: [[TMP9:%.*]] = load <2 x double>, ptr [[ARRAYIDX]], align 8 28; CHECK-NEXT: [[TMP10:%.*]] = load <2 x double>, ptr [[ARRAYIDX4]], align 8 29; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x double> poison, double [[TMP0]], i32 0 30; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x double> [[TMP11]], <2 x double> poison, <2 x i32> zeroinitializer 31; CHECK-NEXT: [[TMP13:%.*]] = call <2 x double> @llvm.fmuladd.v2f64(<2 x double> [[TMP12]], <2 x double> [[TMP10]], <2 x double> [[TMP9]]) 32; CHECK-NEXT: store <2 x double> [[TMP13]], ptr [[TMP8]], align 8 33; CHECK-NEXT: [[F_SROA_539_0__SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[TMP8]], i64 40 34; CHECK-NEXT: store double [[TMP7]], ptr [[F_SROA_539_0__SROA_IDX]], align 8 35; CHECK-NEXT: store i32 6, ptr @b, align 4 36; CHECK-NEXT: ret void 37; 38entry: 39 %arrayidx = getelementptr inbounds [4 x double], ptr %d, i64 2 40 %0 = load double, ptr %arrayidx, align 8 41 %arrayidx3 = getelementptr inbounds [4 x double], ptr %d, i64 0, i64 3 42 %1 = load double, ptr %arrayidx3, align 8 43 %arrayidx4 = getelementptr inbounds [4 x double], ptr %e, i64 3 44 %2 = load double, ptr %arrayidx4, align 8 45 %3 = tail call double @llvm.fmuladd.f64(double %1, double %2, double %0) 46 %arrayidx8 = getelementptr inbounds [4 x double], ptr %d, i64 2, i64 1 47 %4 = load double, ptr %arrayidx8, align 8 48 %arrayidx12 = getelementptr inbounds [4 x double], ptr %e, i64 3, i64 1 49 %5 = load double, ptr %arrayidx12, align 8 50 %6 = tail call double @llvm.fmuladd.f64(double %1, double %5, double %4) 51 %arrayidx15 = getelementptr inbounds [4 x double], ptr %d, i64 1, i64 3 52 %7 = load double, ptr %arrayidx15, align 8 53 %arrayidx17 = getelementptr inbounds [4 x double], ptr %d, i64 3, i64 2 54 %8 = load double, ptr %arrayidx17, align 8 55 %arrayidx19 = getelementptr inbounds [4 x double], ptr %e, i64 2, i64 3 56 %9 = load double, ptr %arrayidx19, align 8 57 %10 = tail call double @llvm.fmuladd.f64(double %8, double %9, double %7) 58 %arrayidx21 = getelementptr inbounds [4 x double], ptr %d, i64 3, i64 3 59 %11 = load double, ptr %arrayidx21, align 8 60 %arrayidx23 = getelementptr inbounds [4 x double], ptr %e, i64 3, i64 3 61 %12 = load double, ptr %arrayidx23, align 8 62 %13 = tail call double @llvm.fmuladd.f64(double %11, double %12, double %10) 63 %14 = load ptr, ptr @a, align 8 64 store double %3, ptr %14, align 8 65 %f.sroa.4.0..sroa_idx = getelementptr inbounds i8, ptr %14, i64 8 66 store double %6, ptr %f.sroa.4.0..sroa_idx, align 8 67 %f.sroa.539.0..sroa_idx = getelementptr inbounds i8, ptr %14, i64 40 68 store double %13, ptr %f.sroa.539.0..sroa_idx, align 8 69 store i32 6, ptr @b, align 4 70 ret void 71} 72 73declare double @llvm.fmuladd.f64(double, double, double) 74