89 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: opt -S -O3 < %s | FileCheck %s3 4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"5target triple = "aarch64"6 7; This function (a 16x reduction of a[i] * b[i]) should be vectorized successfully.8 9define dso_local nofpclass(nan inf) float @vmlaq(ptr noundef %0, ptr noundef %1) {10; CHECK-LABEL: define dso_local nofpclass(nan inf) float @vmlaq11; CHECK-SAME: (ptr noundef readonly captures(none) [[TMP0:%.*]], ptr noundef readonly captures(none) [[TMP1:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {12; CHECK-NEXT: [[TMP3:%.*]] = load <16 x float>, ptr [[TMP0]], align 4, !tbaa [[TBAA4:![0-9]+]]13; CHECK-NEXT: [[TMP4:%.*]] = load <16 x float>, ptr [[TMP1]], align 4, !tbaa [[TBAA4]]14; CHECK-NEXT: [[TMP5:%.*]] = fmul fast <16 x float> [[TMP4]], [[TMP3]]15; CHECK-NEXT: [[TMP6:%.*]] = tail call fast float @llvm.vector.reduce.fadd.v16f32(float 0.000000e+00, <16 x float> [[TMP5]])16; CHECK-NEXT: ret float [[TMP6]]17;18 %3 = alloca ptr, align 819 %4 = alloca ptr, align 820 %5 = alloca float, align 421 %6 = alloca i32, align 422 store ptr %0, ptr %3, align 8, !tbaa !423 store ptr %1, ptr %4, align 8, !tbaa !424 call void @llvm.lifetime.start.p0(ptr %5)25 store float 0.000000e+00, ptr %5, align 4, !tbaa !926 call void @llvm.lifetime.start.p0(ptr %6)27 store i32 0, ptr %6, align 4, !tbaa !1128 br label %729 307: ; preds = %25, %231 %8 = load i32, ptr %6, align 4, !tbaa !1132 %9 = icmp slt i32 %8, 1633 br i1 %9, label %11, label %1034 3510: ; preds = %736 call void @llvm.lifetime.end.p0(ptr %6)37 br label %2838 3911: ; preds = %740 %12 = load ptr, ptr %3, align 8, !tbaa !441 %13 = load i32, ptr %6, align 4, !tbaa !1142 %14 = sext i32 %13 to i6443 %15 = getelementptr inbounds float, ptr %12, i64 %1444 %16 = load float, ptr %15, align 4, !tbaa !945 %17 = load ptr, ptr %4, align 8, !tbaa !446 %18 = load i32, ptr %6, align 4, !tbaa !1147 %19 = sext i32 %18 to i6448 %20 = getelementptr inbounds float, ptr %17, i64 %1949 %21 = load float, ptr %20, align 4, !tbaa !950 %22 = fmul fast float %16, %2151 %23 = load float, ptr %5, align 4, !tbaa !952 %24 = fadd fast float %23, %2253 store float %24, ptr %5, align 4, !tbaa !954 br label %2555 5625: ; preds = %1157 %26 = load i32, ptr %6, align 4, !tbaa !1158 %27 = add nsw i32 %26, 159 store i32 %27, ptr %6, align 4, !tbaa !1160 br label %7, !llvm.loop !1361 6228: ; preds = %1063 %29 = load float, ptr %5, align 4, !tbaa !964 call void @llvm.lifetime.end.p0(ptr %5)65 ret float %2966}67 68declare void @llvm.lifetime.start.p0(ptr captures(none))69declare void @llvm.lifetime.end.p0(ptr captures(none))70 71!llvm.module.flags = !{!0, !1, !2}72!llvm.ident = !{!3}73 74!0 = !{i32 1, !"wchar_size", i32 4}75!1 = !{i32 7, !"uwtable", i32 2}76!2 = !{i32 7, !"frame-pointer", i32 1}77!3 = !{!"clang version 22.0.0git"}78!4 = !{!5, !5, i64 0}79!5 = !{!"p1 float", !6, i64 0}80!6 = !{!"any pointer", !7, i64 0}81!7 = !{!"omnipotent char", !8, i64 0}82!8 = !{!"Simple C/C++ TBAA"}83!9 = !{!10, !10, i64 0}84!10 = !{!"float", !7, i64 0}85!11 = !{!12, !12, i64 0}86!12 = !{!"int", !7, i64 0}87!13 = distinct !{!13, !14}88!14 = !{!"llvm.loop.mustprogress"}89