94 lines · plain
1; Test that MergeConsecutiveStores() does not during DAG combining2; incorrectly drop a chain dependency to a store previously chained to3; one of two combined loads.4;5; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s6 7@A = common global [2048 x float] zeroinitializer, align 48 9; Function Attrs: nounwind10define signext i32 @main(i32 signext %argc, ptr nocapture readnone %argv) #0 {11entry:12 br label %for.body13 14for.body: ; preds = %for.body, %entry15 %indvars.iv24 = phi i64 [ 0, %entry ], [ %indvars.iv.next25, %for.body ]16 %sum.018 = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ]17 %0 = trunc i64 %indvars.iv24 to i3218 %conv = sitofp i32 %0 to float19 %arrayidx = getelementptr inbounds [2048 x float], ptr @A, i64 0, i64 %indvars.iv2420 store float %conv, ptr %arrayidx, align 421 %add = fadd float %sum.018, %conv22 %indvars.iv.next25 = add nuw nsw i64 %indvars.iv24, 123 %exitcond26 = icmp eq i64 %indvars.iv.next25, 204824 br i1 %exitcond26, label %for.end, label %for.body25 26for.end: ; preds = %for.body27 br label %for.body.3.lr.ph.i.preheader28 29for.body.3.lr.ph.i.preheader: ; preds = %complex_transpose.exit, %for.end30 %i.116 = phi i32 [ 0, %for.end ], [ %inc9, %complex_transpose.exit ]31 br label %for.body.3.lr.ph.i32 33for.body.3.lr.ph.i: ; preds = %for.body.3.lr.ph.i.preheader, %for.inc.40.i34 %indvars.iv19 = phi i32 [ 1, %for.body.3.lr.ph.i.preheader ], [ %indvars.iv.next20, %for.inc.40.i ]35 %indvars.iv57.i = phi i64 [ 1, %for.body.3.lr.ph.i.preheader ], [ %indvars.iv.next58.i, %for.inc.40.i ]36 %1 = shl nsw i64 %indvars.iv57.i, 137 %2 = shl nsw i64 %indvars.iv57.i, 638 br label %for.body.3.i39 40for.body.3.i: ; preds = %for.body.3.i, %for.body.3.lr.ph.i41; CHECK-LABEL: .LBB0_5:42; CHECK-NOT: stfh %r{{.*}}, 0(%r{{.*}})43; CHECK: lg %r{{.*}}, 8(%r{{.*}})44; Overlapping load should go before the store45 %indvars.iv.i = phi i64 [ 0, %for.body.3.lr.ph.i ], [ %indvars.iv.next.i, %for.body.3.i ]46 %3 = shl nsw i64 %indvars.iv.i, 647 %4 = add nuw nsw i64 %3, %148 %arrayidx.i = getelementptr inbounds [2048 x float], ptr @A, i64 0, i64 %449 %5 = load i32, ptr %arrayidx.i, align 450 %arrayidx9.i = getelementptr inbounds float, ptr getelementptr inbounds ([2048 x float], ptr @A, i64 0, i64 1), i64 %451 %6 = load i32, ptr %arrayidx9.i, align 452 %7 = shl nsw i64 %indvars.iv.i, 153 %8 = add nuw nsw i64 %7, %254 %arrayidx14.i = getelementptr inbounds [2048 x float], ptr @A, i64 0, i64 %855 %9 = load i32, ptr %arrayidx14.i, align 456 %arrayidx19.i = getelementptr inbounds float, ptr getelementptr inbounds ([2048 x float], ptr @A, i64 0, i64 1), i64 %857 %10 = load i32, ptr %arrayidx19.i, align 458 store i32 %5, ptr %arrayidx14.i, align 459 store i32 %6, ptr %arrayidx19.i, align 460 store i32 %9, ptr %arrayidx.i, align 461 store i32 %10, ptr %arrayidx9.i, align 462 %indvars.iv.next.i = add nuw nsw i64 %indvars.iv.i, 163 %lftr.wideiv = trunc i64 %indvars.iv.next.i to i3264 %exitcond21 = icmp eq i32 %lftr.wideiv, %indvars.iv1965 br i1 %exitcond21, label %for.inc.40.i, label %for.body.3.i66 67for.inc.40.i: ; preds = %for.body.3.i68 %indvars.iv.next58.i = add nuw nsw i64 %indvars.iv57.i, 169 %indvars.iv.next20 = add nuw nsw i32 %indvars.iv19, 170 %exitcond22 = icmp eq i64 %indvars.iv.next58.i, 3271 br i1 %exitcond22, label %complex_transpose.exit, label %for.body.3.lr.ph.i72 73complex_transpose.exit: ; preds = %for.inc.40.i74 %inc9 = add nuw nsw i32 %i.116, 175 %exitcond23 = icmp eq i32 %inc9, 1076 br i1 %exitcond23, label %for.body.14.preheader, label %for.body.3.lr.ph.i.preheader77 78for.body.14.preheader: ; preds = %complex_transpose.exit79 br label %for.body.1480 81for.body.14: ; preds = %for.body.14.preheader, %for.body.1482 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body.14 ], [ 0, %for.body.14.preheader ]83 %sum.115 = phi float [ %add17, %for.body.14 ], [ 0.000000e+00, %for.body.14.preheader ]84 %arrayidx16 = getelementptr inbounds [2048 x float], ptr @A, i64 0, i64 %indvars.iv85 %11 = load float, ptr %arrayidx16, align 486 %add17 = fadd float %sum.115, %1187 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 188 %exitcond = icmp eq i64 %indvars.iv.next, 204889 br i1 %exitcond, label %for.end.20, label %for.body.1490 91for.end.20: ; preds = %for.body.1492 ret i32 093}94