53 lines · plain
1; RUN: opt -hexagon-autohvx -passes=loop-vectorize -S < %s | FileCheck %s2; Check that the loop has been interleaved.3; CHECK: store <64 x i32> %interleaved.vec4 5target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048"6target triple = "hexagon"7 8define void @f0(ptr noalias nocapture %a0, ptr noalias nocapture readonly %a1, i32 %a2) #0 {9b0:10 %v0 = icmp eq i32 %a2, 011 br i1 %v0, label %b3, label %b112 13b1: ; preds = %b014 br label %b415 16b2: ; preds = %b417 br label %b318 19b3: ; preds = %b2, %b020 ret void21 22b4: ; preds = %b4, %b123 %v1 = phi i32 [ %v13, %b4 ], [ 0, %b1 ]24 %v2 = getelementptr inbounds i32, ptr %a1, i32 %v125 %v3 = load i32, ptr %v2, align 4, !tbaa !126 %v4 = getelementptr inbounds i32, ptr %a0, i32 %v127 %v5 = load i32, ptr %v4, align 4, !tbaa !128 %v6 = add nsw i32 %v5, %v329 store i32 %v6, ptr %v4, align 4, !tbaa !130 %v7 = or disjoint i32 %v1, 131 %v8 = getelementptr inbounds i32, ptr %a1, i32 %v732 %v9 = load i32, ptr %v8, align 4, !tbaa !133 %v10 = getelementptr inbounds i32, ptr %a0, i32 %v734 %v11 = load i32, ptr %v10, align 4, !tbaa !135 %v12 = add nsw i32 %v11, %v936 store i32 %v12, ptr %v10, align 4, !tbaa !137 %v13 = add nuw nsw i32 %v1, 238 %v14 = icmp eq i32 %v13, %a239 br i1 %v14, label %b2, label %b4, !llvm.loop !540}41 42attributes #0 = { norecurse nounwind "target-cpu"="hexagonv60" "target-features"="+hvx-length128b,+hvxv60" }43 44!llvm.module.flags = !{!0}45 46!0 = !{i32 1, !"wchar_size", i32 4}47!1 = !{!2, !2, i64 0}48!2 = !{!"int", !3, i64 0}49!3 = !{!"omnipotent char", !4, i64 0}50!4 = !{!"Simple C/C++ TBAA"}51!5 = distinct !{!5, !6}52!6 = !{!"llvm.loop.unroll.disable"}53