88 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -passes=loop-vectorize -mtriple=powerpc64le-unknown-unknown \3; RUN: -force-target-max-vector-interleave=1 -mcpu=pwr9 < %s | FileCheck %s4define dso_local void @test(ptr %Arr, i32 signext %Len) {5; CHECK-LABEL: @test(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[CMP1:%.*]] = icmp slt i32 0, [[LEN:%.*]]8; CHECK-NEXT: br i1 [[CMP1]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]]9; CHECK: for.body.lr.ph:10; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[LEN]], 411; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]12; CHECK: vector.ph:13; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[LEN]], 414; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[LEN]], [[N_MOD_VF]]15; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]16; CHECK: vector.body:17; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]18; CHECK-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i6419; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[ARR:%.*]], i64 [[TMP1]]20; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP2]], align 421; CHECK-NEXT: [[TMP4:%.*]] = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> [[WIDE_LOAD]])22; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[TMP2]], align 423; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 424; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]25; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]26; CHECK: middle.block:27; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[LEN]], [[N_VEC]]28; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]]29; CHECK: scalar.ph:30; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]31; CHECK-NEXT: br label [[FOR_BODY:%.*]]32; CHECK: for.cond.for.cond.cleanup_crit_edge:33; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]34; CHECK: for.cond.cleanup:35; CHECK-NEXT: br label [[FOR_END:%.*]]36; CHECK: for.body:37; CHECK-NEXT: [[I_02:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_INC:%.*]] ]38; CHECK-NEXT: [[IDXPROM:%.*]] = sext i32 [[I_02]] to i6439; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[IDXPROM]]40; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[ARRAYIDX]], align 441; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.bswap.i32(i32 [[TMP9]])42; CHECK-NEXT: [[IDXPROM1:%.*]] = sext i32 [[I_02]] to i6443; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[ARR]], i64 [[IDXPROM1]]44; CHECK-NEXT: store i32 [[TMP10]], ptr [[ARRAYIDX2]], align 445; CHECK-NEXT: br label [[FOR_INC]]46; CHECK: for.inc:47; CHECK-NEXT: [[INC]] = add nsw i32 [[I_02]], 148; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[INC]], [[LEN]]49; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE]], !llvm.loop [[LOOP3:![0-9]+]]50; CHECK: for.end:51; CHECK-NEXT: ret void52;53entry:54 %cmp1 = icmp slt i32 0, %Len55 br i1 %cmp1, label %for.body.lr.ph, label %for.cond.cleanup56 57for.body.lr.ph: ; preds = %entry58 br label %for.body59 60for.cond.for.cond.cleanup_crit_edge: ; preds = %for.inc61 br label %for.cond.cleanup62 63for.cond.cleanup: ; preds = %for.cond.for.cond.cleanup_crit_edge, %entry64 br label %for.end65 66for.body: ; preds = %for.body.lr.ph, %for.inc67 %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ]68 %idxprom = sext i32 %i.02 to i6469 %arrayidx = getelementptr inbounds i32, ptr %Arr, i64 %idxprom70 %0 = load i32, ptr %arrayidx, align 471 %1 = call i32 @llvm.bswap.i32(i32 %0)72 %idxprom1 = sext i32 %i.02 to i6473 %arrayidx2 = getelementptr inbounds i32, ptr %Arr, i64 %idxprom174 store i32 %1, ptr %arrayidx2, align 475 br label %for.inc76 77for.inc: ; preds = %for.body78 %inc = add nsw i32 %i.02, 179 %cmp = icmp slt i32 %inc, %Len80 br i1 %cmp, label %for.body, label %for.cond.for.cond.cleanup_crit_edge81 82for.end: ; preds = %for.cond.cleanup83 ret void84}85 86; Function Attrs: nounwind readnone speculatable willreturn87declare i32 @llvm.bswap.i32(i32)88