Lines Matching refs:VF
6 ; VF is the same as the epilogue vectorization VF. To make these tests target independent we'll use…
8 …ion -force-vector-width=2 -epilogue-vectorization-force-VF=2 -S | FileCheck %s --check-prefix VF-T…
14 ; VF-TWO-CHECK-LABEL: @f1(
15 ; VF-TWO-CHECK-NEXT: entry:
16 ; VF-TWO-CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
17 ; VF-TWO-CHECK-NEXT: br i1 [[CMP1]], label [[ITER_CHECK:%.*]], label [[FOR_END:%.*]]
18 ; VF-TWO-CHECK: iter.check:
19 ; VF-TWO-CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
20 ; VF-TWO-CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 2
21 ; VF-TWO-CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECT…
22 ; VF-TWO-CHECK: vector.main.loop.iter.check:
23 ; VF-TWO-CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 2
24 ; VF-TWO-CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:…
25 ; VF-TWO-CHECK: vector.ph:
26 ; VF-TWO-CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 2
27 ; VF-TWO-CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
28 ; VF-TWO-CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
29 ; VF-TWO-CHECK: vector.body:
30 ; VF-TWO-CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTO…
31 ; VF-TWO-CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
32 ; VF-TWO-CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
33 ; VF-TWO-CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
34 ; VF-TWO-CHECK-NEXT: [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
35 ; VF-TWO-CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP3]], align 4
36 ; VF-TWO-CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[B:%.*]], i64 [[TMP0]]
37 ; VF-TWO-CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0
38 ; VF-TWO-CHECK-NEXT: [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <2 x i32>*
39 ; VF-TWO-CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP6]], align 4
40 ; VF-TWO-CHECK-NEXT: [[TMP7:%.*]] = add nsw <2 x i32> [[WIDE_LOAD]], [[WIDE_LOAD2]]
41 ; VF-TWO-CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
42 ; VF-TWO-CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
43 ; VF-TWO-CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.lo…
44 ; VF-TWO-CHECK: middle.block:
45 ; VF-TWO-CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[TMP7]], i32 1
46 ; VF-TWO-CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
47 ; VF-TWO-CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CH…
48 ; VF-TWO-CHECK: vec.epilog.iter.check:
49 ; VF-TWO-CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
50 ; VF-TWO-CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 2
51 ; VF-TWO-CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[V…
52 ; VF-TWO-CHECK: vec.epilog.ph:
53 ; VF-TWO-CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHEC…
54 ; VF-TWO-CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 2
55 ; VF-TWO-CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF3]]
56 ; VF-TWO-CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
57 ; VF-TWO-CHECK: vec.epilog.vector.body:
58 ; VF-TWO-CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]]…
59 ; VF-TWO-CHECK-NEXT: [[TMP10:%.*]] = add i64 [[OFFSET_IDX]], 0
60 ; VF-TWO-CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP10]]
61 ; VF-TWO-CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP11]], i32 0
62 ; VF-TWO-CHECK-NEXT: [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <2 x i32>*
63 ; VF-TWO-CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, <2 x i32>* [[TMP13]], align 4
64 ; VF-TWO-CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[TMP10]]
65 ; VF-TWO-CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, i32* [[TMP14]], i32 0
66 ; VF-TWO-CHECK-NEXT: [[TMP16:%.*]] = bitcast i32* [[TMP15]] to <2 x i32>*
67 ; VF-TWO-CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x i32>, <2 x i32>* [[TMP16]], align 4
68 ; VF-TWO-CHECK-NEXT: [[TMP17:%.*]] = add nsw <2 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD8]]
69 ; VF-TWO-CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[OFFSET_IDX]], 2
70 ; VF-TWO-CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC4]]
71 ; VF-TWO-CHECK-NEXT: br i1 [[TMP18]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_…
72 ; VF-TWO-CHECK: vec.epilog.middle.block:
73 ; VF-TWO-CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP17]], i32 1
74 ; VF-TWO-CHECK-NEXT: [[CMP_N5:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC4]]
75 ; VF-TWO-CHECK-NEXT: br i1 [[CMP_N5]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]
76 ; VF-TWO-CHECK: vec.epilog.scalar.ph:
77 ; VF-TWO-CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]…
78 ; VF-TWO-CHECK-NEXT: br label [[FOR_BODY:%.*]]
79 ; VF-TWO-CHECK: for.body:
80 ; VF-TWO-CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] …
81 ; VF-TWO-CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[INDVARS_IV…
82 ; VF-TWO-CHECK-NEXT: [[TMP20:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
83 ; VF-TWO-CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[INDVARS_I…
84 ; VF-TWO-CHECK-NEXT: [[TMP21:%.*]] = load i32, i32* [[ARRAYIDX2]], align 4
85 ; VF-TWO-CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP20]], [[TMP21]]
86 ; VF-TWO-CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
87 ; VF-TWO-CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
88 ; VF-TWO-CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.l…
89 ; VF-TWO-CHECK: for.end.loopexit:
90 ; VF-TWO-CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_BODY]] ], [ [[TMP9]], [[MIDDLE…
91 ; VF-TWO-CHECK-NEXT: br label [[FOR_END]]
92 ; VF-TWO-CHECK: for.end:
93 ; VF-TWO-CHECK-NEXT: [[RES_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_…
94 ; VF-TWO-CHECK-NEXT: ret i32 [[RES_0_LCSSA]]