80 lines · plain
1; RUN: opt -mtriple=hexagon-- -passes='loop(hexagon-vlcr),adce' -S %s | FileCheck %s2 3; CHECK: %.hexagon.vlcr = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B4; ModuleID = 'hexagon_vector_loop_carried_reuse.c'5source_filename = "hexagon_vector_loop_carried_reuse.c"6target 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"7target triple = "hexagon"8 9@W = external local_unnamed_addr global i32, align 410 11; Function Attrs: nounwind12define void @foo(ptr noalias nocapture readonly %src, ptr noalias nocapture %dst, i32 %stride) local_unnamed_addr #0 {13entry:14 %add.ptr = getelementptr inbounds i8, ptr %src, i32 %stride15 %mul = mul nsw i32 %stride, 216 %add.ptr1 = getelementptr inbounds i8, ptr %src, i32 %mul17 %0 = load i32, ptr @W, align 4, !tbaa !118 %cmp55 = icmp sgt i32 %0, 019 br i1 %cmp55, label %for.body.lr.ph, label %for.end20 21for.body.lr.ph: ; preds = %entry22 %1 = load <32 x i32>, ptr %add.ptr1, align 128, !tbaa !523 %incdec.ptr4 = getelementptr inbounds i8, ptr %add.ptr1, i32 12824 %2 = load <32 x i32>, ptr %add.ptr, align 128, !tbaa !525 %incdec.ptr2 = getelementptr inbounds i8, ptr %add.ptr, i32 12826 %3 = load <32 x i32>, ptr %src, align 128, !tbaa !527 %incdec.ptr = getelementptr inbounds i8, ptr %src, i32 12828 br label %for.body29 30for.body: ; preds = %for.body.lr.ph, %for.body31 %out.063 = phi ptr [ %dst, %for.body.lr.ph ], [ %incdec.ptr18, %for.body ]32 %p2.062 = phi ptr [ %incdec.ptr4, %for.body.lr.ph ], [ %incdec.ptr10, %for.body ]33 %p1.061 = phi ptr [ %incdec.ptr2, %for.body.lr.ph ], [ %incdec.ptr8, %for.body ]34 %p0.060 = phi ptr [ %incdec.ptr, %for.body.lr.ph ], [ %incdec.ptr6, %for.body ]35 %i.059 = phi i32 [ 0, %for.body.lr.ph ], [ %add, %for.body ]36 %a.sroa.0.058 = phi <32 x i32> [ %3, %for.body.lr.ph ], [ %4, %for.body ]37 %b.sroa.0.057 = phi <32 x i32> [ %2, %for.body.lr.ph ], [ %5, %for.body ]38 %c.sroa.0.056 = phi <32 x i32> [ %1, %for.body.lr.ph ], [ %6, %for.body ]39 %incdec.ptr6 = getelementptr inbounds <32 x i32>, ptr %p0.060, i32 140 %4 = load <32 x i32>, ptr %p0.060, align 128, !tbaa !541 %incdec.ptr8 = getelementptr inbounds <32 x i32>, ptr %p1.061, i32 142 %5 = load <32 x i32>, ptr %p1.061, align 128, !tbaa !543 %incdec.ptr10 = getelementptr inbounds <32 x i32>, ptr %p2.062, i32 144 %6 = load <32 x i32>, ptr %p2.062, align 128, !tbaa !545 %7 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %a.sroa.0.058, <32 x i32> %b.sroa.0.057)46 %8 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %7, <32 x i32> %c.sroa.0.056)47 %9 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %4, <32 x i32> %5)48 %10 = tail call <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32> %9, <32 x i32> %6)49 %11 = tail call <32 x i32> @llvm.hexagon.V6.valignbi.128B(<32 x i32> %10, <32 x i32> %8, i32 1)50 %incdec.ptr18 = getelementptr inbounds <32 x i32>, ptr %out.063, i32 151 store <32 x i32> %11, ptr %out.063, align 128, !tbaa !552 %add = add nuw nsw i32 %i.059, 12853 %cmp = icmp slt i32 %add, %054 br i1 %cmp, label %for.body, label %for.end.loopexit55 56for.end.loopexit: ; preds = %for.body57 br label %for.end58 59for.end: ; preds = %for.end.loopexit, %entry60 ret void61}62 63; Function Attrs: nounwind readnone64declare <32 x i32> @llvm.hexagon.V6.vmaxub.128B(<32 x i32>, <32 x i32>) #165 66; Function Attrs: nounwind readnone67declare <32 x i32> @llvm.hexagon.V6.valignbi.128B(<32 x i32>, <32 x i32>, i32) #168 69attributes #0 = { nounwind "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="+hvx,+hvx-length128b,-long-calls" "unsafe-fp-math"="false" "use-soft-float"="false" }70attributes #1 = { nounwind readnone }71 72!llvm.ident = !{!0}73 74!0 = !{!"QuIC LLVM Hexagon Clang version hexagon-clang-82-2622 (based on LLVM 5.0.0)"}75!1 = !{!2, !2, i64 0}76!2 = !{!"int", !3, i64 0}77!3 = !{!"omnipotent char", !4, i64 0}78!4 = !{!"Simple C/C++ TBAA"}79!5 = !{!3, !3, i64 0}80