1 ; RUN: opt -S -passes=loop-vectorize -force-vector-width=4 %s | FileCheck %s
3 ; CHECK-LABEL: @test_fshl
4 ; CHECK-LABEL: vector.body:
5 ; CHECK-NEXT: [[IDX:%.+]] = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
6 ; CHECK-NEXT: [[IDX0:%.+]] = add i32 %index, 0
7 ; CHECK-NEXT: [[GEP:%.+]] = getelementptr inbounds i16, ptr %src, i32 [[IDX0]]
8 ; CHECK-NEXT: [[GEP0:%.+]] = getelementptr inbounds i16, ptr [[GEP]], i32 0
9 ; CHECK-NEXT: [[WIDE_LOAD:%.+]] = load <4 x i16>, ptr [[GEP0]], align 2
10 ; CHECK-NEXT: [[FSHL:%.+]] = call <4 x i16> @llvm.fshl.v4i16(<4 x i16> [[WIDE_LOAD]], <4 x i16> [[WIDE_LOAD]], <4 x i16> splat (i16 15))
11 ; CHECK-NEXT: [[GEP0:%.+]] = getelementptr inbounds i16, ptr %dst, i32 [[IDX0]]
12 ; CHECK-NEXT: [[GEP1:%.+]] = getelementptr inbounds i16, ptr [[GEP0]], i32 0
13 ; CHECK-NEXT: store <4 x i16> [[FSHL]], ptr [[GEP1]], align 2
14 ; CHECK-NEXT: [[IDX_NEXT:%.+]] = add nuw i32 [[IDX]], 4
15 ; CHECK-NEXT: [[EC:%.+]] = icmp eq i32 [[IDX_NEXT]], %n.vec
16 ; CHECK-NEXT: br i1 [[EC]], label %middle.block, label %vector.body
18 define void @test_fshl(i32 %width, ptr %dst, ptr %src) {
20 br label %for.body9.us.us
22 for.cond6.for.cond.cleanup8_crit_edge.us.us: ; preds = %for.body9.us.us
25 for.body9.us.us: ; preds = %for.body9.us.us, %entry
26 %iv = phi i32 [ 0, %entry ], [ %iv.next, %for.body9.us.us ]
27 %gep = getelementptr inbounds i16, ptr %src, i32 %iv
28 %l = load i16, ptr %gep
29 %conv4.i.us.us = tail call i16 @llvm.fshl.i16(i16 %l, i16 %l, i16 15)
30 %dst.gep = getelementptr inbounds i16, ptr %dst, i32 %iv
31 store i16 %conv4.i.us.us, ptr %dst.gep
32 %iv.next = add nuw i32 %iv, 1
33 %exitcond50 = icmp eq i32 %iv.next, %width
34 br i1 %exitcond50, label %for.cond6.for.cond.cleanup8_crit_edge.us.us, label %for.body9.us.us
37 declare i16 @llvm.fshl.i16(i16, i16, i16)