xref: /llvm-project/llvm/test/Transforms/LoopVectorize/X86/intrinsiccost.ll (revision 7f3428d3ed71d87a2088b77b6cab9f3d86544234)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -mcpu=bdver1 -passes=loop-vectorize,instcombine,simplifycfg < %s -S -o - | FileCheck %s --check-prefix=CHECK
3; RUN: opt -mcpu=bdver1 -passes=loop-vectorize -debug-only=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s --check-prefix=CHECK-COST
4; REQUIRES: asserts
5
6target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
7target triple = "x86_64-unknown-linux-gnu"
8
9; CHECK-COST-LABEL: uaddsat
10
11define void @uaddsat(ptr nocapture readonly %pSrc, i16 signext %offset, ptr nocapture noalias %pDst, i32 %blockSize) #0 {
12; CHECK-LABEL: @uaddsat(
13; CHECK-NEXT:  entry:
14; CHECK-NEXT:    [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
15; CHECK-NEXT:    br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[ITER_CHECK:%.*]]
16; CHECK:       iter.check:
17; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[BLOCKSIZE]] to i64
18; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[BLOCKSIZE]], 8
19; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
20; CHECK:       vector.main.loop.iter.check:
21; CHECK-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[BLOCKSIZE]], 64
22; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
23; CHECK:       vector.ph:
24; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP0]], 4294967232
25; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[OFFSET:%.*]], i64 0
26; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
27; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
28; CHECK:       vector.body:
29; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
30; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
31; CHECK-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRC:%.*]], i64 [[OFFSET_IDX]]
32; CHECK-NEXT:    [[OFFSET_IDX2:%.*]] = shl i64 [[INDEX]], 1
33; CHECK-NEXT:    [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i64 [[OFFSET_IDX2]]
34; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 32
35; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 64
36; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 96
37; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <16 x i16>, ptr [[NEXT_GEP]], align 2
38; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <16 x i16>, ptr [[TMP1]], align 2
39; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <16 x i16>, ptr [[TMP2]], align 2
40; CHECK-NEXT:    [[WIDE_LOAD6:%.*]] = load <16 x i16>, ptr [[TMP3]], align 2
41; CHECK-NEXT:    [[TMP4:%.*]] = call <16 x i16> @llvm.uadd.sat.v16i16(<16 x i16> [[WIDE_LOAD]], <16 x i16> [[BROADCAST_SPLAT]])
42; CHECK-NEXT:    [[TMP5:%.*]] = call <16 x i16> @llvm.uadd.sat.v16i16(<16 x i16> [[WIDE_LOAD4]], <16 x i16> [[BROADCAST_SPLAT]])
43; CHECK-NEXT:    [[TMP6:%.*]] = call <16 x i16> @llvm.uadd.sat.v16i16(<16 x i16> [[WIDE_LOAD5]], <16 x i16> [[BROADCAST_SPLAT]])
44; CHECK-NEXT:    [[TMP7:%.*]] = call <16 x i16> @llvm.uadd.sat.v16i16(<16 x i16> [[WIDE_LOAD6]], <16 x i16> [[BROADCAST_SPLAT]])
45; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 32
46; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 64
47; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 96
48; CHECK-NEXT:    store <16 x i16> [[TMP4]], ptr [[NEXT_GEP3]], align 2
49; CHECK-NEXT:    store <16 x i16> [[TMP5]], ptr [[TMP8]], align 2
50; CHECK-NEXT:    store <16 x i16> [[TMP6]], ptr [[TMP9]], align 2
51; CHECK-NEXT:    store <16 x i16> [[TMP7]], ptr [[TMP10]], align 2
52; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
53; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
54; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
55; CHECK:       middle.block:
56; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP0]]
57; CHECK-NEXT:    br i1 [[CMP_N]], label [[WHILE_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
58; CHECK:       vec.epilog.iter.check:
59; CHECK-NEXT:    [[DOTCAST1:%.*]] = trunc nuw i64 [[N_VEC]] to i32
60; CHECK-NEXT:    [[IND_END10:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST1]]
61; CHECK-NEXT:    [[TMP12:%.*]] = shl nuw nsw i64 [[N_VEC]], 1
62; CHECK-NEXT:    [[IND_END12:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[TMP12]]
63; CHECK-NEXT:    [[TMP13:%.*]] = shl nuw nsw i64 [[N_VEC]], 1
64; CHECK-NEXT:    [[IND_END15:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[TMP13]]
65; CHECK-NEXT:    [[N_VEC_REMAINING:%.*]] = and i64 [[TMP0]], 56
66; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
67; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
68; CHECK:       vec.epilog.ph:
69; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
70; CHECK-NEXT:    [[N_VEC8:%.*]] = and i64 [[TMP0]], 4294967288
71; CHECK-NEXT:    [[DOTCAST:%.*]] = trunc nuw i64 [[N_VEC8]] to i32
72; CHECK-NEXT:    [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST]]
73; CHECK-NEXT:    [[TMP14:%.*]] = shl nuw nsw i64 [[N_VEC8]], 1
74; CHECK-NEXT:    [[IND_END11:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[TMP14]]
75; CHECK-NEXT:    [[TMP15:%.*]] = shl nuw nsw i64 [[N_VEC8]], 1
76; CHECK-NEXT:    [[IND_END14:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[TMP15]]
77; CHECK-NEXT:    [[BROADCAST_SPLATINSERT23:%.*]] = insertelement <8 x i16> poison, i16 [[OFFSET]], i64 0
78; CHECK-NEXT:    [[BROADCAST_SPLAT24:%.*]] = shufflevector <8 x i16> [[BROADCAST_SPLATINSERT23]], <8 x i16> poison, <8 x i32> zeroinitializer
79; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
80; CHECK:       vec.epilog.vector.body:
81; CHECK-NEXT:    [[INDEX17:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT25:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
82; CHECK-NEXT:    [[OFFSET_IDX18:%.*]] = shl i64 [[INDEX17]], 1
83; CHECK-NEXT:    [[NEXT_GEP19:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[OFFSET_IDX18]]
84; CHECK-NEXT:    [[OFFSET_IDX20:%.*]] = shl i64 [[INDEX17]], 1
85; CHECK-NEXT:    [[NEXT_GEP21:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[OFFSET_IDX20]]
86; CHECK-NEXT:    [[WIDE_LOAD22:%.*]] = load <8 x i16>, ptr [[NEXT_GEP19]], align 2
87; CHECK-NEXT:    [[TMP16:%.*]] = call <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16> [[WIDE_LOAD22]], <8 x i16> [[BROADCAST_SPLAT24]])
88; CHECK-NEXT:    store <8 x i16> [[TMP16]], ptr [[NEXT_GEP21]], align 2
89; CHECK-NEXT:    [[INDEX_NEXT25]] = add nuw i64 [[INDEX17]], 8
90; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT25]], [[N_VEC8]]
91; CHECK-NEXT:    br i1 [[TMP17]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
92; CHECK:       vec.epilog.middle.block:
93; CHECK-NEXT:    [[CMP_N26:%.*]] = icmp eq i64 [[N_VEC8]], [[TMP0]]
94; CHECK-NEXT:    br i1 [[CMP_N26]], label [[WHILE_END]], label [[VEC_EPILOG_SCALAR_PH]]
95; CHECK:       vec.epilog.scalar.ph:
96; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[BLOCKSIZE]], [[ITER_CHECK]] ], [ [[IND_END10]], [[VEC_EPILOG_ITER_CHECK]] ]
97; CHECK-NEXT:    [[BC_RESUME_VAL13:%.*]] = phi ptr [ [[IND_END11]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[PSRC]], [[ITER_CHECK]] ], [ [[IND_END12]], [[VEC_EPILOG_ITER_CHECK]] ]
98; CHECK-NEXT:    [[BC_RESUME_VAL16:%.*]] = phi ptr [ [[IND_END14]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[PDST]], [[ITER_CHECK]] ], [ [[IND_END15]], [[VEC_EPILOG_ITER_CHECK]] ]
99; CHECK-NEXT:    br label [[WHILE_BODY:%.*]]
100; CHECK:       while.body:
101; CHECK-NEXT:    [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]
102; CHECK-NEXT:    [[PSRC_ADDR_08:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL13]], [[VEC_EPILOG_SCALAR_PH]] ]
103; CHECK-NEXT:    [[PDST_ADDR_07:%.*]] = phi ptr [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL16]], [[VEC_EPILOG_SCALAR_PH]] ]
104; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds nuw i8, ptr [[PSRC_ADDR_08]], i64 2
105; CHECK-NEXT:    [[TMP18:%.*]] = load i16, ptr [[PSRC_ADDR_08]], align 2
106; CHECK-NEXT:    [[TMP19:%.*]] = tail call i16 @llvm.uadd.sat.i16(i16 [[TMP18]], i16 [[OFFSET]])
107; CHECK-NEXT:    [[INCDEC_PTR3]] = getelementptr inbounds nuw i8, ptr [[PDST_ADDR_07]], i64 2
108; CHECK-NEXT:    store i16 [[TMP19]], ptr [[PDST_ADDR_07]], align 2
109; CHECK-NEXT:    [[DEC]] = add i32 [[BLKCNT_09]], -1
110; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0
111; CHECK-NEXT:    br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
112; CHECK:       while.end:
113; CHECK-NEXT:    ret void
114;
115entry:
116  %cmp.not6 = icmp eq i32 %blockSize, 0
117  br i1 %cmp.not6, label %while.end, label %while.body
118
119while.body:                                       ; preds = %entry, %while.body
120  %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]
121  %pSrc.addr.08 = phi ptr [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]
122  %pDst.addr.07 = phi ptr [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]
123  %incdec.ptr = getelementptr inbounds i16, ptr %pSrc.addr.08, i32 1
124  %0 = load i16, ptr %pSrc.addr.08, align 2
125  %1 = tail call i16 @llvm.uadd.sat.i16(i16 %0, i16 %offset)
126  %incdec.ptr3 = getelementptr inbounds i16, ptr %pDst.addr.07, i32 1
127  store i16 %1, ptr %pDst.addr.07, align 2
128  %dec = add i32 %blkCnt.09, -1
129  %cmp.not = icmp eq i32 %dec, 0
130  br i1 %cmp.not, label %while.end, label %while.body
131
132while.end:                                        ; preds = %while.body, %entry
133  ret void
134}
135
136; CHECK-COST-LABEL: fshl
137
138define void @fshl(ptr nocapture readonly %pSrc, i8 signext %offset, ptr nocapture noalias %pDst, i32 %blockSize) #0 {
139; CHECK-LABEL: @fshl(
140; CHECK-NEXT:  entry:
141; CHECK-NEXT:    [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
142; CHECK-NEXT:    br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[ITER_CHECK:%.*]]
143; CHECK:       iter.check:
144; CHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[BLOCKSIZE]] to i64
145; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[BLOCKSIZE]], 8
146; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
147; CHECK:       vector.main.loop.iter.check:
148; CHECK-NEXT:    [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[BLOCKSIZE]], 128
149; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
150; CHECK:       vector.ph:
151; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP0]], 4294967168
152; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <32 x i8> poison, i8 [[OFFSET:%.*]], i64 0
153; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <32 x i8> [[BROADCAST_SPLATINSERT]], <32 x i8> poison, <32 x i32> zeroinitializer
154; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
155; CHECK:       vector.body:
156; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
157; CHECK-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PSRC:%.*]], i64 [[INDEX]]
158; CHECK-NEXT:    [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PDST:%.*]], i64 [[INDEX]]
159; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 32
160; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 64
161; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 96
162; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <32 x i8>, ptr [[NEXT_GEP]], align 2
163; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <32 x i8>, ptr [[TMP1]], align 2
164; CHECK-NEXT:    [[WIDE_LOAD4:%.*]] = load <32 x i8>, ptr [[TMP2]], align 2
165; CHECK-NEXT:    [[WIDE_LOAD5:%.*]] = load <32 x i8>, ptr [[TMP3]], align 2
166; CHECK-NEXT:    [[TMP4:%.*]] = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> [[WIDE_LOAD]], <32 x i8> [[WIDE_LOAD]], <32 x i8> [[BROADCAST_SPLAT]])
167; CHECK-NEXT:    [[TMP5:%.*]] = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> [[WIDE_LOAD3]], <32 x i8> [[WIDE_LOAD3]], <32 x i8> [[BROADCAST_SPLAT]])
168; CHECK-NEXT:    [[TMP6:%.*]] = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> [[WIDE_LOAD4]], <32 x i8> [[WIDE_LOAD4]], <32 x i8> [[BROADCAST_SPLAT]])
169; CHECK-NEXT:    [[TMP7:%.*]] = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> [[WIDE_LOAD5]], <32 x i8> [[WIDE_LOAD5]], <32 x i8> [[BROADCAST_SPLAT]])
170; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 32
171; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 64
172; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 96
173; CHECK-NEXT:    store <32 x i8> [[TMP4]], ptr [[NEXT_GEP2]], align 2
174; CHECK-NEXT:    store <32 x i8> [[TMP5]], ptr [[TMP8]], align 2
175; CHECK-NEXT:    store <32 x i8> [[TMP6]], ptr [[TMP9]], align 2
176; CHECK-NEXT:    store <32 x i8> [[TMP7]], ptr [[TMP10]], align 2
177; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 128
178; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
179; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
180; CHECK:       middle.block:
181; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[TMP0]]
182; CHECK-NEXT:    br i1 [[CMP_N]], label [[WHILE_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
183; CHECK:       vec.epilog.iter.check:
184; CHECK-NEXT:    [[DOTCAST8:%.*]] = trunc nuw i64 [[N_VEC]] to i32
185; CHECK-NEXT:    [[IND_END9:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST8]]
186; CHECK-NEXT:    [[IND_END11:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[N_VEC]]
187; CHECK-NEXT:    [[IND_END14:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[N_VEC]]
188; CHECK-NEXT:    [[N_VEC_REMAINING:%.*]] = and i64 [[TMP0]], 120
189; CHECK-NEXT:    [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
190; CHECK-NEXT:    br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
191; CHECK:       vec.epilog.ph:
192; CHECK-NEXT:    [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
193; CHECK-NEXT:    [[N_VEC7:%.*]] = and i64 [[TMP0]], 4294967288
194; CHECK-NEXT:    [[DOTCAST:%.*]] = trunc nuw i64 [[N_VEC7]] to i32
195; CHECK-NEXT:    [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[DOTCAST]]
196; CHECK-NEXT:    [[IND_END10:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[N_VEC7]]
197; CHECK-NEXT:    [[IND_END13:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[N_VEC7]]
198; CHECK-NEXT:    [[BROADCAST_SPLATINSERT20:%.*]] = insertelement <8 x i8> poison, i8 [[OFFSET]], i64 0
199; CHECK-NEXT:    [[BROADCAST_SPLAT21:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT20]], <8 x i8> poison, <8 x i32> zeroinitializer
200; CHECK-NEXT:    br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
201; CHECK:       vec.epilog.vector.body:
202; CHECK-NEXT:    [[INDEX16:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT22:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
203; CHECK-NEXT:    [[NEXT_GEP17:%.*]] = getelementptr i8, ptr [[PSRC]], i64 [[INDEX16]]
204; CHECK-NEXT:    [[NEXT_GEP18:%.*]] = getelementptr i8, ptr [[PDST]], i64 [[INDEX16]]
205; CHECK-NEXT:    [[WIDE_LOAD19:%.*]] = load <8 x i8>, ptr [[NEXT_GEP17]], align 2
206; CHECK-NEXT:    [[TMP12:%.*]] = call <8 x i8> @llvm.fshl.v8i8(<8 x i8> [[WIDE_LOAD19]], <8 x i8> [[WIDE_LOAD19]], <8 x i8> [[BROADCAST_SPLAT21]])
207; CHECK-NEXT:    store <8 x i8> [[TMP12]], ptr [[NEXT_GEP18]], align 2
208; CHECK-NEXT:    [[INDEX_NEXT22]] = add nuw i64 [[INDEX16]], 8
209; CHECK-NEXT:    [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT22]], [[N_VEC7]]
210; CHECK-NEXT:    br i1 [[TMP13]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
211; CHECK:       vec.epilog.middle.block:
212; CHECK-NEXT:    [[CMP_N23:%.*]] = icmp eq i64 [[N_VEC7]], [[TMP0]]
213; CHECK-NEXT:    br i1 [[CMP_N23]], label [[WHILE_END]], label [[VEC_EPILOG_SCALAR_PH]]
214; CHECK:       vec.epilog.scalar.ph:
215; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[BLOCKSIZE]], [[ITER_CHECK]] ], [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ]
216; CHECK-NEXT:    [[BC_RESUME_VAL12:%.*]] = phi ptr [ [[IND_END10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[PSRC]], [[ITER_CHECK]] ], [ [[IND_END11]], [[VEC_EPILOG_ITER_CHECK]] ]
217; CHECK-NEXT:    [[BC_RESUME_VAL15:%.*]] = phi ptr [ [[IND_END13]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[PDST]], [[ITER_CHECK]] ], [ [[IND_END14]], [[VEC_EPILOG_ITER_CHECK]] ]
218; CHECK-NEXT:    br label [[WHILE_BODY:%.*]]
219; CHECK:       while.body:
220; CHECK-NEXT:    [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]
221; CHECK-NEXT:    [[PSRC_ADDR_08:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL12]], [[VEC_EPILOG_SCALAR_PH]] ]
222; CHECK-NEXT:    [[PDST_ADDR_07:%.*]] = phi ptr [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL15]], [[VEC_EPILOG_SCALAR_PH]] ]
223; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds nuw i8, ptr [[PSRC_ADDR_08]], i64 1
224; CHECK-NEXT:    [[TMP14:%.*]] = load i8, ptr [[PSRC_ADDR_08]], align 2
225; CHECK-NEXT:    [[TMP15:%.*]] = tail call i8 @llvm.fshl.i8(i8 [[TMP14]], i8 [[TMP14]], i8 [[OFFSET]])
226; CHECK-NEXT:    [[INCDEC_PTR3]] = getelementptr inbounds nuw i8, ptr [[PDST_ADDR_07]], i64 1
227; CHECK-NEXT:    store i8 [[TMP15]], ptr [[PDST_ADDR_07]], align 2
228; CHECK-NEXT:    [[DEC]] = add i32 [[BLKCNT_09]], -1
229; CHECK-NEXT:    [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0
230; CHECK-NEXT:    br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
231; CHECK:       while.end:
232; CHECK-NEXT:    ret void
233;
234entry:
235  %cmp.not6 = icmp eq i32 %blockSize, 0
236  br i1 %cmp.not6, label %while.end, label %while.body
237
238while.body:                                       ; preds = %entry, %while.body
239  %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]
240  %pSrc.addr.08 = phi ptr [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]
241  %pDst.addr.07 = phi ptr [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]
242  %incdec.ptr = getelementptr inbounds i8, ptr %pSrc.addr.08, i32 1
243  %0 = load i8, ptr %pSrc.addr.08, align 2
244  %1 = tail call i8 @llvm.fshl.i8(i8 %0, i8 %0, i8 %offset)
245  %incdec.ptr3 = getelementptr inbounds i8, ptr %pDst.addr.07, i32 1
246  store i8 %1, ptr %pDst.addr.07, align 2
247  %dec = add i32 %blkCnt.09, -1
248  %cmp.not = icmp eq i32 %dec, 0
249  br i1 %cmp.not, label %while.end, label %while.body
250
251while.end:                                        ; preds = %while.body, %entry
252  ret void
253}
254
255declare i16 @llvm.uadd.sat.i16(i16, i16)
256declare i8 @llvm.fshl.i8(i8, i8, i8)
257
258;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
259; CHECK-COST: {{.*}}
260