1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -indvars -mtriple=thumbv8m.base -S %s -o - | FileCheck %s --check-prefix=CHECK-T1
3; RUN: opt -indvars -mtriple=thumbv8m.main -S %s -o - | FileCheck %s --check-prefix=CHECK-T2
4
5define dso_local arm_aapcscc void @arm_conv_fast_q15(i16* %pSrcA, i32 %srcALen, i16* %pSrcB, i32 %srcBLen, i16* %pDst, i16** %store.px, i16** %store.py, i32* %store.res) local_unnamed_addr {
6; CHECK-T1-LABEL: @arm_conv_fast_q15(
7; CHECK-T1-NEXT:  entry:
8; CHECK-T1-NEXT:    [[CMP:%.*]] = icmp ult i32 [[SRCALEN:%.*]], [[SRCBLEN:%.*]]
9; CHECK-T1-NEXT:    [[SRCALEN_SRCBLEN:%.*]] = select i1 [[CMP]], i32 [[SRCALEN]], i32 [[SRCBLEN]]
10; CHECK-T1-NEXT:    [[PSRCB_PSRCA:%.*]] = select i1 [[CMP]], i16* [[PSRCB:%.*]], i16* [[PSRCA:%.*]]
11; CHECK-T1-NEXT:    [[PSRCA_PSRCB:%.*]] = select i1 [[CMP]], i16* [[PSRCA]], i16* [[PSRCB]]
12; CHECK-T1-NEXT:    [[SUB:%.*]] = add i32 [[SRCALEN_SRCBLEN]], -1
13; CHECK-T1-NEXT:    [[CMP41080:%.*]] = icmp eq i32 [[SUB]], 0
14; CHECK-T1-NEXT:    br i1 [[CMP41080]], label [[WHILE_END13:%.*]], label [[WHILE_COND5_PREHEADER_PREHEADER:%.*]]
15; CHECK-T1:       while.cond5.preheader.preheader:
16; CHECK-T1-NEXT:    br label [[WHILE_COND5_PREHEADER:%.*]]
17; CHECK-T1:       while.cond5.preheader:
18; CHECK-T1-NEXT:    [[COUNT_01084:%.*]] = phi i32 [ [[INC:%.*]], [[WHILE_END:%.*]] ], [ 1, [[WHILE_COND5_PREHEADER_PREHEADER]] ]
19; CHECK-T1-NEXT:    [[BLOCKSIZE1_01083:%.*]] = phi i32 [ [[DEC12:%.*]], [[WHILE_END]] ], [ [[SUB]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
20; CHECK-T1-NEXT:    [[PY_01082:%.*]] = phi i16* [ [[ADD_PTR:%.*]], [[WHILE_END]] ], [ [[PSRCA_PSRCB]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
21; CHECK-T1-NEXT:    [[POUT_01081:%.*]] = phi i16* [ [[INCDEC_PTR11:%.*]], [[WHILE_END]] ], [ [[PDST:%.*]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
22; CHECK-T1-NEXT:    br label [[WHILE_BODY7:%.*]]
23; CHECK-T1:       while.body7:
24; CHECK-T1-NEXT:    [[K_01078:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY7]] ], [ [[COUNT_01084]], [[WHILE_COND5_PREHEADER]] ]
25; CHECK-T1-NEXT:    [[SUM_01077:%.*]] = phi i32 [ [[ADD6_I:%.*]], [[WHILE_BODY7]] ], [ 0, [[WHILE_COND5_PREHEADER]] ]
26; CHECK-T1-NEXT:    [[PY_11076:%.*]] = phi i16* [ [[INCDEC_PTR8:%.*]], [[WHILE_BODY7]] ], [ [[PY_01082]], [[WHILE_COND5_PREHEADER]] ]
27; CHECK-T1-NEXT:    [[PX_11075:%.*]] = phi i16* [ [[INCDEC_PTR:%.*]], [[WHILE_BODY7]] ], [ [[PSRCB_PSRCA]], [[WHILE_COND5_PREHEADER]] ]
28; CHECK-T1-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds i16, i16* [[PX_11075]], i32 1
29; CHECK-T1-NEXT:    [[TMP0:%.*]] = load i16, i16* [[PX_11075]], align 2
30; CHECK-T1-NEXT:    [[CONV:%.*]] = sext i16 [[TMP0]] to i32
31; CHECK-T1-NEXT:    [[INCDEC_PTR8]] = getelementptr inbounds i16, i16* [[PY_11076]], i32 -1
32; CHECK-T1-NEXT:    [[TMP1:%.*]] = load i16, i16* [[PY_11076]], align 2
33; CHECK-T1-NEXT:    [[CONV9:%.*]] = sext i16 [[TMP1]] to i32
34; CHECK-T1-NEXT:    [[MUL_I:%.*]] = mul nsw i32 [[CONV9]], [[CONV]]
35; CHECK-T1-NEXT:    [[SHR3_I:%.*]] = ashr i32 [[CONV]], 16
36; CHECK-T1-NEXT:    [[SHR4_I:%.*]] = ashr i32 [[CONV9]], 16
37; CHECK-T1-NEXT:    [[MUL5_I:%.*]] = mul nsw i32 [[SHR4_I]], [[SHR3_I]]
38; CHECK-T1-NEXT:    [[ADD_I:%.*]] = add i32 [[MUL_I]], [[SUM_01077]]
39; CHECK-T1-NEXT:    [[ADD6_I]] = add i32 [[ADD_I]], [[MUL5_I]]
40; CHECK-T1-NEXT:    [[DEC]] = add nsw i32 [[K_01078]], -1
41; CHECK-T1-NEXT:    [[CMP6:%.*]] = icmp eq i32 [[DEC]], 0
42; CHECK-T1-NEXT:    br i1 [[CMP6]], label [[WHILE_END]], label [[WHILE_BODY7]]
43; CHECK-T1:       while.end:
44; CHECK-T1-NEXT:    [[ADD6_I_LCSSA:%.*]] = phi i32 [ [[ADD6_I]], [[WHILE_BODY7]] ]
45; CHECK-T1-NEXT:    [[TMP2:%.*]] = lshr i32 [[ADD6_I_LCSSA]], 15
46; CHECK-T1-NEXT:    [[CONV10:%.*]] = trunc i32 [[TMP2]] to i16
47; CHECK-T1-NEXT:    [[INCDEC_PTR11]] = getelementptr inbounds i16, i16* [[POUT_01081]], i32 1
48; CHECK-T1-NEXT:    store i16 [[CONV10]], i16* [[POUT_01081]], align 2
49; CHECK-T1-NEXT:    [[ADD_PTR]] = getelementptr inbounds i16, i16* [[PSRCA_PSRCB]], i32 [[COUNT_01084]]
50; CHECK-T1-NEXT:    [[INC]] = add nuw nsw i32 [[COUNT_01084]], 1
51; CHECK-T1-NEXT:    [[DEC12]] = add i32 [[BLOCKSIZE1_01083]], -1
52; CHECK-T1-NEXT:    [[CMP3:%.*]] = icmp ult i32 [[COUNT_01084]], 3
53; CHECK-T1-NEXT:    [[CMP4:%.*]] = icmp ne i32 [[DEC12]], 0
54; CHECK-T1-NEXT:    [[TMP3:%.*]] = and i1 [[CMP4]], [[CMP3]]
55; CHECK-T1-NEXT:    br i1 [[TMP3]], label [[WHILE_COND5_PREHEADER]], label [[WHILE_END13_LOOPEXIT:%.*]]
56; CHECK-T1:       while.end13.loopexit:
57; CHECK-T1-NEXT:    [[INCDEC_PTR11_LCSSA:%.*]] = phi i16* [ [[INCDEC_PTR11]], [[WHILE_END]] ]
58; CHECK-T1-NEXT:    [[ADD_PTR_LCSSA:%.*]] = phi i16* [ [[ADD_PTR]], [[WHILE_END]] ]
59; CHECK-T1-NEXT:    [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], [[WHILE_END]] ]
60; CHECK-T1-NEXT:    [[DEC12_LCSSA:%.*]] = phi i32 [ [[DEC12]], [[WHILE_END]] ]
61; CHECK-T1-NEXT:    br label [[WHILE_END13]]
62; CHECK-T1:       while.end13:
63; CHECK-T1-NEXT:    [[POUT_0_LCSSA:%.*]] = phi i16* [ [[PDST]], [[ENTRY:%.*]] ], [ [[INCDEC_PTR11_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
64; CHECK-T1-NEXT:    [[PY_0_LCSSA:%.*]] = phi i16* [ [[PSRCA_PSRCB]], [[ENTRY]] ], [ [[ADD_PTR_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
65; CHECK-T1-NEXT:    [[BLOCKSIZE1_0_LCSSA:%.*]] = phi i32 [ [[SUB]], [[ENTRY]] ], [ [[DEC12_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
66; CHECK-T1-NEXT:    [[COUNT_0_LCSSA:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[INC_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
67; CHECK-T1-NEXT:    [[CMP161068:%.*]] = icmp eq i32 [[BLOCKSIZE1_0_LCSSA]], 0
68; CHECK-T1-NEXT:    br i1 [[CMP161068]], label [[EXIT:%.*]], label [[WHILE_BODY18_PREHEADER:%.*]]
69; CHECK-T1:       while.body18.preheader:
70; CHECK-T1-NEXT:    [[ADD_PTR14:%.*]] = getelementptr inbounds i16, i16* [[PY_0_LCSSA]], i32 -1
71; CHECK-T1-NEXT:    br label [[WHILE_BODY18:%.*]]
72; CHECK-T1:       while.body18:
73; CHECK-T1-NEXT:    [[COUNT_11072:%.*]] = phi i32 [ [[INC49:%.*]], [[WHILE_END43:%.*]] ], [ [[COUNT_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
74; CHECK-T1-NEXT:    [[BLOCKSIZE1_11071:%.*]] = phi i32 [ [[DEC50:%.*]], [[WHILE_END43]] ], [ [[BLOCKSIZE1_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
75; CHECK-T1-NEXT:    [[PY_21070:%.*]] = phi i16* [ [[ADD_PTR48:%.*]], [[WHILE_END43]] ], [ [[ADD_PTR14]], [[WHILE_BODY18_PREHEADER]] ]
76; CHECK-T1-NEXT:    [[POUT_11069:%.*]] = phi i16* [ [[INCDEC_PTR46:%.*]], [[WHILE_END43]] ], [ [[POUT_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
77; CHECK-T1-NEXT:    [[SHR19:%.*]] = lshr i32 [[COUNT_11072]], 2
78; CHECK-T1-NEXT:    [[CMP211054:%.*]] = icmp eq i32 [[SHR19]], 0
79; CHECK-T1-NEXT:    br i1 [[CMP211054]], label [[WHILE_END31:%.*]], label [[WHILE_BODY23_PREHEADER:%.*]]
80; CHECK-T1:       while.body23.preheader:
81; CHECK-T1-NEXT:    br label [[WHILE_BODY23:%.*]]
82; CHECK-T1:       while.body23:
83; CHECK-T1-NEXT:    [[K_11058:%.*]] = phi i32 [ [[DEC30:%.*]], [[WHILE_BODY23]] ], [ [[SHR19]], [[WHILE_BODY23_PREHEADER]] ]
84; CHECK-T1-NEXT:    [[SUM_11057:%.*]] = phi i32 [ [[ADD6_I878:%.*]], [[WHILE_BODY23]] ], [ 0, [[WHILE_BODY23_PREHEADER]] ]
85; CHECK-T1-NEXT:    [[PY_31056:%.*]] = phi i16* [ [[ADD_PTR_I884:%.*]], [[WHILE_BODY23]] ], [ [[PY_21070]], [[WHILE_BODY23_PREHEADER]] ]
86; CHECK-T1-NEXT:    [[PX_31055:%.*]] = phi i16* [ [[ADD_PTR_I890:%.*]], [[WHILE_BODY23]] ], [ [[PSRCB_PSRCA]], [[WHILE_BODY23_PREHEADER]] ]
87; CHECK-T1-NEXT:    [[ARRAYIDX_I907:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 1
88; CHECK-T1-NEXT:    [[TMP4:%.*]] = load i16, i16* [[ARRAYIDX_I907]], align 2
89; CHECK-T1-NEXT:    [[TMP5:%.*]] = load i16, i16* [[PX_31055]], align 2
90; CHECK-T1-NEXT:    [[ADD_PTR_I912:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 2
91; CHECK-T1-NEXT:    [[ARRAYIDX_I901:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 1
92; CHECK-T1-NEXT:    [[TMP6:%.*]] = load i16, i16* [[ARRAYIDX_I901]], align 2
93; CHECK-T1-NEXT:    [[TMP7:%.*]] = load i16, i16* [[PY_31056]], align 2
94; CHECK-T1-NEXT:    [[ADD_PTR_I906:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -2
95; CHECK-T1-NEXT:    [[SHR_I892:%.*]] = sext i16 [[TMP5]] to i32
96; CHECK-T1-NEXT:    [[SHR1_I893:%.*]] = sext i16 [[TMP6]] to i32
97; CHECK-T1-NEXT:    [[MUL_I894:%.*]] = mul nsw i32 [[SHR1_I893]], [[SHR_I892]]
98; CHECK-T1-NEXT:    [[SHR2_I895:%.*]] = sext i16 [[TMP4]] to i32
99; CHECK-T1-NEXT:    [[SHR4_I897:%.*]] = sext i16 [[TMP7]] to i32
100; CHECK-T1-NEXT:    [[MUL5_I898:%.*]] = mul nsw i32 [[SHR4_I897]], [[SHR2_I895]]
101; CHECK-T1-NEXT:    [[ADD_I899:%.*]] = add i32 [[MUL_I894]], [[SUM_11057]]
102; CHECK-T1-NEXT:    [[ADD6_I900:%.*]] = add i32 [[ADD_I899]], [[MUL5_I898]]
103; CHECK-T1-NEXT:    [[ARRAYIDX_I885:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 3
104; CHECK-T1-NEXT:    [[TMP8:%.*]] = load i16, i16* [[ARRAYIDX_I885]], align 2
105; CHECK-T1-NEXT:    [[TMP9:%.*]] = load i16, i16* [[ADD_PTR_I912]], align 2
106; CHECK-T1-NEXT:    [[ADD_PTR_I890]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 4
107; CHECK-T1-NEXT:    [[ARRAYIDX_I879:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -1
108; CHECK-T1-NEXT:    [[TMP10:%.*]] = load i16, i16* [[ARRAYIDX_I879]], align 2
109; CHECK-T1-NEXT:    [[TMP11:%.*]] = load i16, i16* [[ADD_PTR_I906]], align 2
110; CHECK-T1-NEXT:    [[ADD_PTR_I884]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -4
111; CHECK-T1-NEXT:    [[SHR_I870:%.*]] = sext i16 [[TMP9]] to i32
112; CHECK-T1-NEXT:    [[SHR1_I871:%.*]] = sext i16 [[TMP10]] to i32
113; CHECK-T1-NEXT:    [[MUL_I872:%.*]] = mul nsw i32 [[SHR1_I871]], [[SHR_I870]]
114; CHECK-T1-NEXT:    [[SHR2_I873:%.*]] = sext i16 [[TMP8]] to i32
115; CHECK-T1-NEXT:    [[SHR4_I875:%.*]] = sext i16 [[TMP11]] to i32
116; CHECK-T1-NEXT:    [[MUL5_I876:%.*]] = mul nsw i32 [[SHR4_I875]], [[SHR2_I873]]
117; CHECK-T1-NEXT:    [[ADD_I877:%.*]] = add i32 [[ADD6_I900]], [[MUL_I872]]
118; CHECK-T1-NEXT:    [[ADD6_I878]] = add i32 [[ADD_I877]], [[MUL5_I876]]
119; CHECK-T1-NEXT:    [[DEC30]] = add nsw i32 [[K_11058]], -1
120; CHECK-T1-NEXT:    [[CMP21:%.*]] = icmp eq i32 [[DEC30]], 0
121; CHECK-T1-NEXT:    br i1 [[CMP21]], label [[WHILE_END31_LOOPEXIT:%.*]], label [[WHILE_BODY23]]
122; CHECK-T1:       while.end31.loopexit:
123; CHECK-T1-NEXT:    [[ADD_PTR_I890_LCSSA:%.*]] = phi i16* [ [[ADD_PTR_I890]], [[WHILE_BODY23]] ]
124; CHECK-T1-NEXT:    [[ADD_PTR_I884_LCSSA:%.*]] = phi i16* [ [[ADD_PTR_I884]], [[WHILE_BODY23]] ]
125; CHECK-T1-NEXT:    [[ADD6_I878_LCSSA:%.*]] = phi i32 [ [[ADD6_I878]], [[WHILE_BODY23]] ]
126; CHECK-T1-NEXT:    br label [[WHILE_END31]]
127; CHECK-T1:       while.end31:
128; CHECK-T1-NEXT:    [[PX_3_LCSSA:%.*]] = phi i16* [ [[PSRCB_PSRCA]], [[WHILE_BODY18]] ], [ [[ADD_PTR_I890_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
129; CHECK-T1-NEXT:    [[PY_3_LCSSA:%.*]] = phi i16* [ [[PY_21070]], [[WHILE_BODY18]] ], [ [[ADD_PTR_I884_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
130; CHECK-T1-NEXT:    [[SUM_1_LCSSA:%.*]] = phi i32 [ 0, [[WHILE_BODY18]] ], [ [[ADD6_I878_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
131; CHECK-T1-NEXT:    [[REM:%.*]] = and i32 [[COUNT_11072]], 3
132; CHECK-T1-NEXT:    [[CMP341062:%.*]] = icmp eq i32 [[REM]], 0
133; CHECK-T1-NEXT:    br i1 [[CMP341062]], label [[WHILE_END43]], label [[WHILE_BODY36_PREHEADER:%.*]]
134; CHECK-T1:       while.body36.preheader:
135; CHECK-T1-NEXT:    [[ADD_PTR32:%.*]] = getelementptr inbounds i16, i16* [[PY_3_LCSSA]], i32 1
136; CHECK-T1-NEXT:    br label [[WHILE_BODY36:%.*]]
137; CHECK-T1:       while.body36:
138; CHECK-T1-NEXT:    [[K_21066:%.*]] = phi i32 [ [[DEC42:%.*]], [[WHILE_BODY36]] ], [ [[REM]], [[WHILE_BODY36_PREHEADER]] ]
139; CHECK-T1-NEXT:    [[SUM_21065:%.*]] = phi i32 [ [[ADD6_I868:%.*]], [[WHILE_BODY36]] ], [ [[SUM_1_LCSSA]], [[WHILE_BODY36_PREHEADER]] ]
140; CHECK-T1-NEXT:    [[PY_41064:%.*]] = phi i16* [ [[INCDEC_PTR39:%.*]], [[WHILE_BODY36]] ], [ [[ADD_PTR32]], [[WHILE_BODY36_PREHEADER]] ]
141; CHECK-T1-NEXT:    [[PX_41063:%.*]] = phi i16* [ [[INCDEC_PTR37:%.*]], [[WHILE_BODY36]] ], [ [[PX_3_LCSSA]], [[WHILE_BODY36_PREHEADER]] ]
142; CHECK-T1-NEXT:    [[INCDEC_PTR37]] = getelementptr inbounds i16, i16* [[PX_41063]], i32 1
143; CHECK-T1-NEXT:    [[TMP12:%.*]] = load i16, i16* [[PX_41063]], align 2
144; CHECK-T1-NEXT:    [[CONV38:%.*]] = sext i16 [[TMP12]] to i32
145; CHECK-T1-NEXT:    [[INCDEC_PTR39]] = getelementptr inbounds i16, i16* [[PY_41064]], i32 -1
146; CHECK-T1-NEXT:    [[TMP13:%.*]] = load i16, i16* [[PY_41064]], align 2
147; CHECK-T1-NEXT:    [[CONV40:%.*]] = sext i16 [[TMP13]] to i32
148; CHECK-T1-NEXT:    [[MUL_I863:%.*]] = mul nsw i32 [[CONV40]], [[CONV38]]
149; CHECK-T1-NEXT:    [[SHR3_I864:%.*]] = ashr i32 [[CONV38]], 16
150; CHECK-T1-NEXT:    [[SHR4_I865:%.*]] = ashr i32 [[CONV40]], 16
151; CHECK-T1-NEXT:    [[MUL5_I866:%.*]] = mul nsw i32 [[SHR4_I865]], [[SHR3_I864]]
152; CHECK-T1-NEXT:    [[ADD_I867:%.*]] = add i32 [[MUL_I863]], [[SUM_21065]]
153; CHECK-T1-NEXT:    [[ADD6_I868]] = add i32 [[ADD_I867]], [[MUL5_I866]]
154; CHECK-T1-NEXT:    [[DEC42]] = add nsw i32 [[K_21066]], -1
155; CHECK-T1-NEXT:    [[CMP34:%.*]] = icmp eq i32 [[DEC42]], 0
156; CHECK-T1-NEXT:    br i1 [[CMP34]], label [[WHILE_END43_LOOPEXIT:%.*]], label [[WHILE_BODY36]]
157; CHECK-T1:       while.end43.loopexit:
158; CHECK-T1-NEXT:    [[ADD6_I868_LCSSA:%.*]] = phi i32 [ [[ADD6_I868]], [[WHILE_BODY36]] ]
159; CHECK-T1-NEXT:    br label [[WHILE_END43]]
160; CHECK-T1:       while.end43:
161; CHECK-T1-NEXT:    [[SUM_2_LCSSA:%.*]] = phi i32 [ [[SUM_1_LCSSA]], [[WHILE_END31]] ], [ [[ADD6_I868_LCSSA]], [[WHILE_END43_LOOPEXIT]] ]
162; CHECK-T1-NEXT:    [[TMP14:%.*]] = lshr i32 [[SUM_2_LCSSA]], 15
163; CHECK-T1-NEXT:    [[CONV45:%.*]] = trunc i32 [[TMP14]] to i16
164; CHECK-T1-NEXT:    [[INCDEC_PTR46]] = getelementptr inbounds i16, i16* [[POUT_11069]], i32 1
165; CHECK-T1-NEXT:    store i16 [[CONV45]], i16* [[POUT_11069]], align 2
166; CHECK-T1-NEXT:    [[SUB47:%.*]] = add i32 [[COUNT_11072]], -1
167; CHECK-T1-NEXT:    [[ADD_PTR48]] = getelementptr inbounds i16, i16* [[PSRCA_PSRCB]], i32 [[SUB47]]
168; CHECK-T1-NEXT:    [[INC49]] = add i32 [[COUNT_11072]], 1
169; CHECK-T1-NEXT:    [[DEC50]] = add i32 [[BLOCKSIZE1_11071]], -1
170; CHECK-T1-NEXT:    [[CMP16:%.*]] = icmp eq i32 [[DEC50]], 0
171; CHECK-T1-NEXT:    br i1 [[CMP16]], label [[EXIT_LOOPEXIT:%.*]], label [[WHILE_BODY18]]
172; CHECK-T1:       exit.loopexit:
173; CHECK-T1-NEXT:    br label [[EXIT]]
174; CHECK-T1:       exit:
175; CHECK-T1-NEXT:    ret void
176;
177; CHECK-T2-LABEL: @arm_conv_fast_q15(
178; CHECK-T2-NEXT:  entry:
179; CHECK-T2-NEXT:    [[CMP:%.*]] = icmp ult i32 [[SRCALEN:%.*]], [[SRCBLEN:%.*]]
180; CHECK-T2-NEXT:    [[SRCALEN_SRCBLEN:%.*]] = select i1 [[CMP]], i32 [[SRCALEN]], i32 [[SRCBLEN]]
181; CHECK-T2-NEXT:    [[PSRCB_PSRCA:%.*]] = select i1 [[CMP]], i16* [[PSRCB:%.*]], i16* [[PSRCA:%.*]]
182; CHECK-T2-NEXT:    [[PSRCA_PSRCB:%.*]] = select i1 [[CMP]], i16* [[PSRCA]], i16* [[PSRCB]]
183; CHECK-T2-NEXT:    [[SUB:%.*]] = add i32 [[SRCALEN_SRCBLEN]], -1
184; CHECK-T2-NEXT:    [[CMP41080:%.*]] = icmp eq i32 [[SUB]], 0
185; CHECK-T2-NEXT:    br i1 [[CMP41080]], label [[WHILE_END13:%.*]], label [[WHILE_COND5_PREHEADER_PREHEADER:%.*]]
186; CHECK-T2:       while.cond5.preheader.preheader:
187; CHECK-T2-NEXT:    br label [[WHILE_COND5_PREHEADER:%.*]]
188; CHECK-T2:       while.cond5.preheader:
189; CHECK-T2-NEXT:    [[COUNT_01084:%.*]] = phi i32 [ [[INC:%.*]], [[WHILE_END:%.*]] ], [ 1, [[WHILE_COND5_PREHEADER_PREHEADER]] ]
190; CHECK-T2-NEXT:    [[BLOCKSIZE1_01083:%.*]] = phi i32 [ [[DEC12:%.*]], [[WHILE_END]] ], [ [[SUB]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
191; CHECK-T2-NEXT:    [[PY_01082:%.*]] = phi i16* [ [[ADD_PTR:%.*]], [[WHILE_END]] ], [ [[PSRCA_PSRCB]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
192; CHECK-T2-NEXT:    [[POUT_01081:%.*]] = phi i16* [ [[INCDEC_PTR11:%.*]], [[WHILE_END]] ], [ [[PDST:%.*]], [[WHILE_COND5_PREHEADER_PREHEADER]] ]
193; CHECK-T2-NEXT:    br label [[WHILE_BODY7:%.*]]
194; CHECK-T2:       while.body7:
195; CHECK-T2-NEXT:    [[K_01078:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY7]] ], [ [[COUNT_01084]], [[WHILE_COND5_PREHEADER]] ]
196; CHECK-T2-NEXT:    [[SUM_01077:%.*]] = phi i32 [ [[ADD6_I:%.*]], [[WHILE_BODY7]] ], [ 0, [[WHILE_COND5_PREHEADER]] ]
197; CHECK-T2-NEXT:    [[PY_11076:%.*]] = phi i16* [ [[INCDEC_PTR8:%.*]], [[WHILE_BODY7]] ], [ [[PY_01082]], [[WHILE_COND5_PREHEADER]] ]
198; CHECK-T2-NEXT:    [[PX_11075:%.*]] = phi i16* [ [[INCDEC_PTR:%.*]], [[WHILE_BODY7]] ], [ [[PSRCB_PSRCA]], [[WHILE_COND5_PREHEADER]] ]
199; CHECK-T2-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds i16, i16* [[PX_11075]], i32 1
200; CHECK-T2-NEXT:    [[TMP0:%.*]] = load i16, i16* [[PX_11075]], align 2
201; CHECK-T2-NEXT:    [[CONV:%.*]] = sext i16 [[TMP0]] to i32
202; CHECK-T2-NEXT:    [[INCDEC_PTR8]] = getelementptr inbounds i16, i16* [[PY_11076]], i32 -1
203; CHECK-T2-NEXT:    [[TMP1:%.*]] = load i16, i16* [[PY_11076]], align 2
204; CHECK-T2-NEXT:    [[CONV9:%.*]] = sext i16 [[TMP1]] to i32
205; CHECK-T2-NEXT:    [[MUL_I:%.*]] = mul nsw i32 [[CONV9]], [[CONV]]
206; CHECK-T2-NEXT:    [[SHR3_I:%.*]] = ashr i32 [[CONV]], 16
207; CHECK-T2-NEXT:    [[SHR4_I:%.*]] = ashr i32 [[CONV9]], 16
208; CHECK-T2-NEXT:    [[MUL5_I:%.*]] = mul nsw i32 [[SHR4_I]], [[SHR3_I]]
209; CHECK-T2-NEXT:    [[ADD_I:%.*]] = add i32 [[MUL_I]], [[SUM_01077]]
210; CHECK-T2-NEXT:    [[ADD6_I]] = add i32 [[ADD_I]], [[MUL5_I]]
211; CHECK-T2-NEXT:    [[DEC]] = add nsw i32 [[K_01078]], -1
212; CHECK-T2-NEXT:    [[CMP6:%.*]] = icmp eq i32 [[DEC]], 0
213; CHECK-T2-NEXT:    br i1 [[CMP6]], label [[WHILE_END]], label [[WHILE_BODY7]]
214; CHECK-T2:       while.end:
215; CHECK-T2-NEXT:    [[ADD6_I_LCSSA:%.*]] = phi i32 [ [[ADD6_I]], [[WHILE_BODY7]] ]
216; CHECK-T2-NEXT:    [[TMP2:%.*]] = lshr i32 [[ADD6_I_LCSSA]], 15
217; CHECK-T2-NEXT:    [[CONV10:%.*]] = trunc i32 [[TMP2]] to i16
218; CHECK-T2-NEXT:    [[INCDEC_PTR11]] = getelementptr inbounds i16, i16* [[POUT_01081]], i32 1
219; CHECK-T2-NEXT:    store i16 [[CONV10]], i16* [[POUT_01081]], align 2
220; CHECK-T2-NEXT:    [[ADD_PTR]] = getelementptr inbounds i16, i16* [[PSRCA_PSRCB]], i32 [[COUNT_01084]]
221; CHECK-T2-NEXT:    [[INC]] = add nuw nsw i32 [[COUNT_01084]], 1
222; CHECK-T2-NEXT:    [[DEC12]] = add i32 [[BLOCKSIZE1_01083]], -1
223; CHECK-T2-NEXT:    [[CMP3:%.*]] = icmp ult i32 [[COUNT_01084]], 3
224; CHECK-T2-NEXT:    [[CMP4:%.*]] = icmp ne i32 [[DEC12]], 0
225; CHECK-T2-NEXT:    [[TMP3:%.*]] = and i1 [[CMP4]], [[CMP3]]
226; CHECK-T2-NEXT:    br i1 [[TMP3]], label [[WHILE_COND5_PREHEADER]], label [[WHILE_END13_LOOPEXIT:%.*]]
227; CHECK-T2:       while.end13.loopexit:
228; CHECK-T2-NEXT:    [[INCDEC_PTR11_LCSSA:%.*]] = phi i16* [ [[INCDEC_PTR11]], [[WHILE_END]] ]
229; CHECK-T2-NEXT:    [[ADD_PTR_LCSSA:%.*]] = phi i16* [ [[ADD_PTR]], [[WHILE_END]] ]
230; CHECK-T2-NEXT:    [[INC_LCSSA:%.*]] = phi i32 [ [[INC]], [[WHILE_END]] ]
231; CHECK-T2-NEXT:    [[DEC12_LCSSA:%.*]] = phi i32 [ [[DEC12]], [[WHILE_END]] ]
232; CHECK-T2-NEXT:    br label [[WHILE_END13]]
233; CHECK-T2:       while.end13:
234; CHECK-T2-NEXT:    [[POUT_0_LCSSA:%.*]] = phi i16* [ [[PDST]], [[ENTRY:%.*]] ], [ [[INCDEC_PTR11_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
235; CHECK-T2-NEXT:    [[PY_0_LCSSA:%.*]] = phi i16* [ [[PSRCA_PSRCB]], [[ENTRY]] ], [ [[ADD_PTR_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
236; CHECK-T2-NEXT:    [[BLOCKSIZE1_0_LCSSA:%.*]] = phi i32 [ [[SUB]], [[ENTRY]] ], [ [[DEC12_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
237; CHECK-T2-NEXT:    [[COUNT_0_LCSSA:%.*]] = phi i32 [ 1, [[ENTRY]] ], [ [[INC_LCSSA]], [[WHILE_END13_LOOPEXIT]] ]
238; CHECK-T2-NEXT:    [[CMP161068:%.*]] = icmp eq i32 [[BLOCKSIZE1_0_LCSSA]], 0
239; CHECK-T2-NEXT:    br i1 [[CMP161068]], label [[EXIT:%.*]], label [[WHILE_BODY18_PREHEADER:%.*]]
240; CHECK-T2:       while.body18.preheader:
241; CHECK-T2-NEXT:    [[ADD_PTR14:%.*]] = getelementptr inbounds i16, i16* [[PY_0_LCSSA]], i32 -1
242; CHECK-T2-NEXT:    br label [[WHILE_BODY18:%.*]]
243; CHECK-T2:       while.body18:
244; CHECK-T2-NEXT:    [[COUNT_11072:%.*]] = phi i32 [ [[INC49:%.*]], [[WHILE_END43:%.*]] ], [ [[COUNT_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
245; CHECK-T2-NEXT:    [[BLOCKSIZE1_11071:%.*]] = phi i32 [ [[DEC50:%.*]], [[WHILE_END43]] ], [ [[BLOCKSIZE1_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
246; CHECK-T2-NEXT:    [[PY_21070:%.*]] = phi i16* [ [[ADD_PTR48:%.*]], [[WHILE_END43]] ], [ [[ADD_PTR14]], [[WHILE_BODY18_PREHEADER]] ]
247; CHECK-T2-NEXT:    [[POUT_11069:%.*]] = phi i16* [ [[INCDEC_PTR46:%.*]], [[WHILE_END43]] ], [ [[POUT_0_LCSSA]], [[WHILE_BODY18_PREHEADER]] ]
248; CHECK-T2-NEXT:    [[SHR19:%.*]] = lshr i32 [[COUNT_11072]], 2
249; CHECK-T2-NEXT:    [[CMP211054:%.*]] = icmp eq i32 [[SHR19]], 0
250; CHECK-T2-NEXT:    br i1 [[CMP211054]], label [[WHILE_END31:%.*]], label [[WHILE_BODY23_PREHEADER:%.*]]
251; CHECK-T2:       while.body23.preheader:
252; CHECK-T2-NEXT:    br label [[WHILE_BODY23:%.*]]
253; CHECK-T2:       while.body23:
254; CHECK-T2-NEXT:    [[K_11058:%.*]] = phi i32 [ [[DEC30:%.*]], [[WHILE_BODY23]] ], [ [[SHR19]], [[WHILE_BODY23_PREHEADER]] ]
255; CHECK-T2-NEXT:    [[SUM_11057:%.*]] = phi i32 [ [[ADD6_I878:%.*]], [[WHILE_BODY23]] ], [ 0, [[WHILE_BODY23_PREHEADER]] ]
256; CHECK-T2-NEXT:    [[PY_31056:%.*]] = phi i16* [ [[ADD_PTR_I884:%.*]], [[WHILE_BODY23]] ], [ [[PY_21070]], [[WHILE_BODY23_PREHEADER]] ]
257; CHECK-T2-NEXT:    [[PX_31055:%.*]] = phi i16* [ [[ADD_PTR_I890:%.*]], [[WHILE_BODY23]] ], [ [[PSRCB_PSRCA]], [[WHILE_BODY23_PREHEADER]] ]
258; CHECK-T2-NEXT:    [[ARRAYIDX_I907:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 1
259; CHECK-T2-NEXT:    [[TMP4:%.*]] = load i16, i16* [[ARRAYIDX_I907]], align 2
260; CHECK-T2-NEXT:    [[TMP5:%.*]] = load i16, i16* [[PX_31055]], align 2
261; CHECK-T2-NEXT:    [[ADD_PTR_I912:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 2
262; CHECK-T2-NEXT:    [[ARRAYIDX_I901:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 1
263; CHECK-T2-NEXT:    [[TMP6:%.*]] = load i16, i16* [[ARRAYIDX_I901]], align 2
264; CHECK-T2-NEXT:    [[TMP7:%.*]] = load i16, i16* [[PY_31056]], align 2
265; CHECK-T2-NEXT:    [[ADD_PTR_I906:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -2
266; CHECK-T2-NEXT:    [[SHR_I892:%.*]] = sext i16 [[TMP5]] to i32
267; CHECK-T2-NEXT:    [[SHR1_I893:%.*]] = sext i16 [[TMP6]] to i32
268; CHECK-T2-NEXT:    [[MUL_I894:%.*]] = mul nsw i32 [[SHR1_I893]], [[SHR_I892]]
269; CHECK-T2-NEXT:    [[SHR2_I895:%.*]] = sext i16 [[TMP4]] to i32
270; CHECK-T2-NEXT:    [[SHR4_I897:%.*]] = sext i16 [[TMP7]] to i32
271; CHECK-T2-NEXT:    [[MUL5_I898:%.*]] = mul nsw i32 [[SHR4_I897]], [[SHR2_I895]]
272; CHECK-T2-NEXT:    [[ADD_I899:%.*]] = add i32 [[MUL_I894]], [[SUM_11057]]
273; CHECK-T2-NEXT:    [[ADD6_I900:%.*]] = add i32 [[ADD_I899]], [[MUL5_I898]]
274; CHECK-T2-NEXT:    [[ARRAYIDX_I885:%.*]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 3
275; CHECK-T2-NEXT:    [[TMP8:%.*]] = load i16, i16* [[ARRAYIDX_I885]], align 2
276; CHECK-T2-NEXT:    [[TMP9:%.*]] = load i16, i16* [[ADD_PTR_I912]], align 2
277; CHECK-T2-NEXT:    [[ADD_PTR_I890]] = getelementptr inbounds i16, i16* [[PX_31055]], i32 4
278; CHECK-T2-NEXT:    [[ARRAYIDX_I879:%.*]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -1
279; CHECK-T2-NEXT:    [[TMP10:%.*]] = load i16, i16* [[ARRAYIDX_I879]], align 2
280; CHECK-T2-NEXT:    [[TMP11:%.*]] = load i16, i16* [[ADD_PTR_I906]], align 2
281; CHECK-T2-NEXT:    [[ADD_PTR_I884]] = getelementptr inbounds i16, i16* [[PY_31056]], i32 -4
282; CHECK-T2-NEXT:    [[SHR_I870:%.*]] = sext i16 [[TMP9]] to i32
283; CHECK-T2-NEXT:    [[SHR1_I871:%.*]] = sext i16 [[TMP10]] to i32
284; CHECK-T2-NEXT:    [[MUL_I872:%.*]] = mul nsw i32 [[SHR1_I871]], [[SHR_I870]]
285; CHECK-T2-NEXT:    [[SHR2_I873:%.*]] = sext i16 [[TMP8]] to i32
286; CHECK-T2-NEXT:    [[SHR4_I875:%.*]] = sext i16 [[TMP11]] to i32
287; CHECK-T2-NEXT:    [[MUL5_I876:%.*]] = mul nsw i32 [[SHR4_I875]], [[SHR2_I873]]
288; CHECK-T2-NEXT:    [[ADD_I877:%.*]] = add i32 [[ADD6_I900]], [[MUL_I872]]
289; CHECK-T2-NEXT:    [[ADD6_I878]] = add i32 [[ADD_I877]], [[MUL5_I876]]
290; CHECK-T2-NEXT:    [[DEC30]] = add nsw i32 [[K_11058]], -1
291; CHECK-T2-NEXT:    [[CMP21:%.*]] = icmp eq i32 [[DEC30]], 0
292; CHECK-T2-NEXT:    br i1 [[CMP21]], label [[WHILE_END31_LOOPEXIT:%.*]], label [[WHILE_BODY23]]
293; CHECK-T2:       while.end31.loopexit:
294; CHECK-T2-NEXT:    [[ADD_PTR_I890_LCSSA:%.*]] = phi i16* [ [[ADD_PTR_I890]], [[WHILE_BODY23]] ]
295; CHECK-T2-NEXT:    [[ADD_PTR_I884_LCSSA:%.*]] = phi i16* [ [[ADD_PTR_I884]], [[WHILE_BODY23]] ]
296; CHECK-T2-NEXT:    [[ADD6_I878_LCSSA:%.*]] = phi i32 [ [[ADD6_I878]], [[WHILE_BODY23]] ]
297; CHECK-T2-NEXT:    br label [[WHILE_END31]]
298; CHECK-T2:       while.end31:
299; CHECK-T2-NEXT:    [[PX_3_LCSSA:%.*]] = phi i16* [ [[PSRCB_PSRCA]], [[WHILE_BODY18]] ], [ [[ADD_PTR_I890_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
300; CHECK-T2-NEXT:    [[PY_3_LCSSA:%.*]] = phi i16* [ [[PY_21070]], [[WHILE_BODY18]] ], [ [[ADD_PTR_I884_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
301; CHECK-T2-NEXT:    [[SUM_1_LCSSA:%.*]] = phi i32 [ 0, [[WHILE_BODY18]] ], [ [[ADD6_I878_LCSSA]], [[WHILE_END31_LOOPEXIT]] ]
302; CHECK-T2-NEXT:    [[REM:%.*]] = and i32 [[COUNT_11072]], 3
303; CHECK-T2-NEXT:    [[CMP341062:%.*]] = icmp eq i32 [[REM]], 0
304; CHECK-T2-NEXT:    br i1 [[CMP341062]], label [[WHILE_END43]], label [[WHILE_BODY36_PREHEADER:%.*]]
305; CHECK-T2:       while.body36.preheader:
306; CHECK-T2-NEXT:    [[ADD_PTR32:%.*]] = getelementptr inbounds i16, i16* [[PY_3_LCSSA]], i32 1
307; CHECK-T2-NEXT:    br label [[WHILE_BODY36:%.*]]
308; CHECK-T2:       while.body36:
309; CHECK-T2-NEXT:    [[K_21066:%.*]] = phi i32 [ [[DEC42:%.*]], [[WHILE_BODY36]] ], [ [[REM]], [[WHILE_BODY36_PREHEADER]] ]
310; CHECK-T2-NEXT:    [[SUM_21065:%.*]] = phi i32 [ [[ADD6_I868:%.*]], [[WHILE_BODY36]] ], [ [[SUM_1_LCSSA]], [[WHILE_BODY36_PREHEADER]] ]
311; CHECK-T2-NEXT:    [[PY_41064:%.*]] = phi i16* [ [[INCDEC_PTR39:%.*]], [[WHILE_BODY36]] ], [ [[ADD_PTR32]], [[WHILE_BODY36_PREHEADER]] ]
312; CHECK-T2-NEXT:    [[PX_41063:%.*]] = phi i16* [ [[INCDEC_PTR37:%.*]], [[WHILE_BODY36]] ], [ [[PX_3_LCSSA]], [[WHILE_BODY36_PREHEADER]] ]
313; CHECK-T2-NEXT:    [[INCDEC_PTR37]] = getelementptr inbounds i16, i16* [[PX_41063]], i32 1
314; CHECK-T2-NEXT:    [[TMP12:%.*]] = load i16, i16* [[PX_41063]], align 2
315; CHECK-T2-NEXT:    [[CONV38:%.*]] = sext i16 [[TMP12]] to i32
316; CHECK-T2-NEXT:    [[INCDEC_PTR39]] = getelementptr inbounds i16, i16* [[PY_41064]], i32 -1
317; CHECK-T2-NEXT:    [[TMP13:%.*]] = load i16, i16* [[PY_41064]], align 2
318; CHECK-T2-NEXT:    [[CONV40:%.*]] = sext i16 [[TMP13]] to i32
319; CHECK-T2-NEXT:    [[MUL_I863:%.*]] = mul nsw i32 [[CONV40]], [[CONV38]]
320; CHECK-T2-NEXT:    [[SHR3_I864:%.*]] = ashr i32 [[CONV38]], 16
321; CHECK-T2-NEXT:    [[SHR4_I865:%.*]] = ashr i32 [[CONV40]], 16
322; CHECK-T2-NEXT:    [[MUL5_I866:%.*]] = mul nsw i32 [[SHR4_I865]], [[SHR3_I864]]
323; CHECK-T2-NEXT:    [[ADD_I867:%.*]] = add i32 [[MUL_I863]], [[SUM_21065]]
324; CHECK-T2-NEXT:    [[ADD6_I868]] = add i32 [[ADD_I867]], [[MUL5_I866]]
325; CHECK-T2-NEXT:    [[DEC42]] = add nsw i32 [[K_21066]], -1
326; CHECK-T2-NEXT:    [[CMP34:%.*]] = icmp eq i32 [[DEC42]], 0
327; CHECK-T2-NEXT:    br i1 [[CMP34]], label [[WHILE_END43_LOOPEXIT:%.*]], label [[WHILE_BODY36]]
328; CHECK-T2:       while.end43.loopexit:
329; CHECK-T2-NEXT:    [[ADD6_I868_LCSSA:%.*]] = phi i32 [ [[ADD6_I868]], [[WHILE_BODY36]] ]
330; CHECK-T2-NEXT:    br label [[WHILE_END43]]
331; CHECK-T2:       while.end43:
332; CHECK-T2-NEXT:    [[SUM_2_LCSSA:%.*]] = phi i32 [ [[SUM_1_LCSSA]], [[WHILE_END31]] ], [ [[ADD6_I868_LCSSA]], [[WHILE_END43_LOOPEXIT]] ]
333; CHECK-T2-NEXT:    [[TMP14:%.*]] = lshr i32 [[SUM_2_LCSSA]], 15
334; CHECK-T2-NEXT:    [[CONV45:%.*]] = trunc i32 [[TMP14]] to i16
335; CHECK-T2-NEXT:    [[INCDEC_PTR46]] = getelementptr inbounds i16, i16* [[POUT_11069]], i32 1
336; CHECK-T2-NEXT:    store i16 [[CONV45]], i16* [[POUT_11069]], align 2
337; CHECK-T2-NEXT:    [[SUB47:%.*]] = add i32 [[COUNT_11072]], -1
338; CHECK-T2-NEXT:    [[ADD_PTR48]] = getelementptr inbounds i16, i16* [[PSRCA_PSRCB]], i32 [[SUB47]]
339; CHECK-T2-NEXT:    [[INC49]] = add i32 [[COUNT_11072]], 1
340; CHECK-T2-NEXT:    [[DEC50]] = add i32 [[BLOCKSIZE1_11071]], -1
341; CHECK-T2-NEXT:    [[CMP16:%.*]] = icmp eq i32 [[DEC50]], 0
342; CHECK-T2-NEXT:    br i1 [[CMP16]], label [[EXIT_LOOPEXIT:%.*]], label [[WHILE_BODY18]]
343; CHECK-T2:       exit.loopexit:
344; CHECK-T2-NEXT:    br label [[EXIT]]
345; CHECK-T2:       exit:
346; CHECK-T2-NEXT:    ret void
347;
348entry:
349  %cmp = icmp ult i32 %srcALen, %srcBLen
350  %srcALen.srcBLen = select i1 %cmp, i32 %srcALen, i32 %srcBLen
351  %pSrcB.pSrcA = select i1 %cmp, i16* %pSrcB, i16* %pSrcA
352  %pSrcA.pSrcB = select i1 %cmp, i16* %pSrcA, i16* %pSrcB
353  %sub = add i32 %srcALen.srcBLen, -1
354  %cmp41080 = icmp eq i32 %sub, 0
355  br i1 %cmp41080, label %while.end13, label %while.cond5.preheader
356
357while.cond5.preheader:                            ; preds = %while.end, %entry
358  %count.01084 = phi i32 [ %inc, %while.end ], [ 1, %entry ]
359  %blockSize1.01083 = phi i32 [ %dec12, %while.end ], [ %sub, %entry ]
360  %py.01082 = phi i16* [ %add.ptr, %while.end ], [ %pSrcA.pSrcB, %entry ]
361  %pOut.01081 = phi i16* [ %incdec.ptr11, %while.end ], [ %pDst, %entry ]
362  br label %while.body7
363
364while.body7:                                      ; preds = %while.body7, %while.cond5.preheader
365  %k.01078 = phi i32 [ %dec, %while.body7 ], [ %count.01084, %while.cond5.preheader ]
366  %sum.01077 = phi i32 [ %add6.i, %while.body7 ], [ 0, %while.cond5.preheader ]
367  %py.11076 = phi i16* [ %incdec.ptr8, %while.body7 ], [ %py.01082, %while.cond5.preheader ]
368  %px.11075 = phi i16* [ %incdec.ptr, %while.body7 ], [ %pSrcB.pSrcA, %while.cond5.preheader ]
369  %incdec.ptr = getelementptr inbounds i16, i16* %px.11075, i32 1
370  %0 = load i16, i16* %px.11075, align 2
371  %conv = sext i16 %0 to i32
372  %incdec.ptr8 = getelementptr inbounds i16, i16* %py.11076, i32 -1
373  %1 = load i16, i16* %py.11076, align 2
374  %conv9 = sext i16 %1 to i32
375  %mul.i = mul nsw i32 %conv9, %conv
376  %shr3.i = ashr i32 %conv, 16
377  %shr4.i = ashr i32 %conv9, 16
378  %mul5.i = mul nsw i32 %shr4.i, %shr3.i
379  %add.i = add i32 %mul.i, %sum.01077
380  %add6.i = add i32 %add.i, %mul5.i
381  %dec = add nsw i32 %k.01078, -1
382  %cmp6 = icmp eq i32 %dec, 0
383  br i1 %cmp6, label %while.end, label %while.body7
384
385while.end:                                        ; preds = %while.body7
386  %2 = lshr i32 %add6.i, 15
387  %conv10 = trunc i32 %2 to i16
388  %incdec.ptr11 = getelementptr inbounds i16, i16* %pOut.01081, i32 1
389  store i16 %conv10, i16* %pOut.01081, align 2
390  %add.ptr = getelementptr inbounds i16, i16* %pSrcA.pSrcB, i32 %count.01084
391  %inc = add nuw nsw i32 %count.01084, 1
392  %dec12 = add i32 %blockSize1.01083, -1
393  %cmp3 = icmp ult i32 %count.01084, 3
394  %cmp4 = icmp ne i32 %dec12, 0
395  %3 = and i1 %cmp4, %cmp3
396  br i1 %3, label %while.cond5.preheader, label %while.end13
397
398while.end13:                                      ; preds = %while.end, %entry
399  %pOut.0.lcssa = phi i16* [ %pDst, %entry ], [ %incdec.ptr11, %while.end ]
400  %py.0.lcssa = phi i16* [ %pSrcA.pSrcB, %entry ], [ %add.ptr, %while.end ]
401  %blockSize1.0.lcssa = phi i32 [ %sub, %entry ], [ %dec12, %while.end ]
402  %count.0.lcssa = phi i32 [ 1, %entry ], [ %inc, %while.end ]
403  %cmp161068 = icmp eq i32 %blockSize1.0.lcssa, 0
404  br i1 %cmp161068, label %exit, label %while.body18.preheader
405
406while.body18.preheader:                           ; preds = %while.end13
407  %add.ptr14 = getelementptr inbounds i16, i16* %py.0.lcssa, i32 -1
408  br label %while.body18
409
410while.body18:                                     ; preds = %while.end43, %while.body18.preheader
411  %count.11072 = phi i32 [ %inc49, %while.end43 ], [ %count.0.lcssa, %while.body18.preheader ]
412  %blockSize1.11071 = phi i32 [ %dec50, %while.end43 ], [ %blockSize1.0.lcssa, %while.body18.preheader ]
413  %py.21070 = phi i16* [ %add.ptr48, %while.end43 ], [ %add.ptr14, %while.body18.preheader ]
414  %pOut.11069 = phi i16* [ %incdec.ptr46, %while.end43 ], [ %pOut.0.lcssa, %while.body18.preheader ]
415  %shr19 = lshr i32 %count.11072, 2
416  %cmp211054 = icmp eq i32 %shr19, 0
417  br i1 %cmp211054, label %while.end31, label %while.body23
418
419while.body23:                                     ; preds = %while.body23, %while.body18
420  %k.11058 = phi i32 [ %dec30, %while.body23 ], [ %shr19, %while.body18 ]
421  %sum.11057 = phi i32 [ %add6.i878, %while.body23 ], [ 0, %while.body18 ]
422  %py.31056 = phi i16* [ %add.ptr.i884, %while.body23 ], [ %py.21070, %while.body18 ]
423  %px.31055 = phi i16* [ %add.ptr.i890, %while.body23 ], [ %pSrcB.pSrcA, %while.body18 ]
424  %arrayidx.i907 = getelementptr inbounds i16, i16* %px.31055, i32 1
425  %4 = load i16, i16* %arrayidx.i907, align 2
426  %5 = load i16, i16* %px.31055, align 2
427  %add.ptr.i912 = getelementptr inbounds i16, i16* %px.31055, i32 2
428  %arrayidx.i901 = getelementptr inbounds i16, i16* %py.31056, i32 1
429  %6 = load i16, i16* %arrayidx.i901, align 2
430  %7 = load i16, i16* %py.31056, align 2
431  %add.ptr.i906 = getelementptr inbounds i16, i16* %py.31056, i32 -2
432  %shr.i892 = sext i16 %5 to i32
433  %shr1.i893 = sext i16 %6 to i32
434  %mul.i894 = mul nsw i32 %shr1.i893, %shr.i892
435  %shr2.i895 = sext i16 %4 to i32
436  %shr4.i897 = sext i16 %7 to i32
437  %mul5.i898 = mul nsw i32 %shr4.i897, %shr2.i895
438  %add.i899 = add i32 %mul.i894, %sum.11057
439  %add6.i900 = add i32 %add.i899, %mul5.i898
440  %arrayidx.i885 = getelementptr inbounds i16, i16* %px.31055, i32 3
441  %8 = load i16, i16* %arrayidx.i885, align 2
442  %9 = load i16, i16* %add.ptr.i912, align 2
443  %add.ptr.i890 = getelementptr inbounds i16, i16* %px.31055, i32 4
444  %arrayidx.i879 = getelementptr inbounds i16, i16* %py.31056, i32 -1
445  %10 = load i16, i16* %arrayidx.i879, align 2
446  %11 = load i16, i16* %add.ptr.i906, align 2
447  %add.ptr.i884 = getelementptr inbounds i16, i16* %py.31056, i32 -4
448  %shr.i870 = sext i16 %9 to i32
449  %shr1.i871 = sext i16 %10 to i32
450  %mul.i872 = mul nsw i32 %shr1.i871, %shr.i870
451  %shr2.i873 = sext i16 %8 to i32
452  %shr4.i875 = sext i16 %11 to i32
453  %mul5.i876 = mul nsw i32 %shr4.i875, %shr2.i873
454  %add.i877 = add i32 %add6.i900, %mul.i872
455  %add6.i878 = add i32 %add.i877, %mul5.i876
456  %dec30 = add nsw i32 %k.11058, -1
457  %cmp21 = icmp eq i32 %dec30, 0
458  br i1 %cmp21, label %while.end31, label %while.body23
459
460while.end31:                                      ; preds = %while.body23, %while.body18
461  %px.3.lcssa = phi i16* [ %pSrcB.pSrcA, %while.body18 ], [ %add.ptr.i890, %while.body23 ]
462  %py.3.lcssa = phi i16* [ %py.21070, %while.body18 ], [ %add.ptr.i884, %while.body23 ]
463  %sum.1.lcssa = phi i32 [ 0, %while.body18 ], [ %add6.i878, %while.body23 ]
464  %rem = and i32 %count.11072, 3
465  %cmp341062 = icmp eq i32 %rem, 0
466  br i1 %cmp341062, label %while.end43, label %while.body36.preheader
467
468while.body36.preheader:                           ; preds = %while.end31
469  %add.ptr32 = getelementptr inbounds i16, i16* %py.3.lcssa, i32 1
470  br label %while.body36
471
472while.body36:                                     ; preds = %while.body36, %while.body36.preheader
473  %k.21066 = phi i32 [ %dec42, %while.body36 ], [ %rem, %while.body36.preheader ]
474  %sum.21065 = phi i32 [ %add6.i868, %while.body36 ], [ %sum.1.lcssa, %while.body36.preheader ]
475  %py.41064 = phi i16* [ %incdec.ptr39, %while.body36 ], [ %add.ptr32, %while.body36.preheader ]
476  %px.41063 = phi i16* [ %incdec.ptr37, %while.body36 ], [ %px.3.lcssa, %while.body36.preheader ]
477  %incdec.ptr37 = getelementptr inbounds i16, i16* %px.41063, i32 1
478  %12 = load i16, i16* %px.41063, align 2
479  %conv38 = sext i16 %12 to i32
480  %incdec.ptr39 = getelementptr inbounds i16, i16* %py.41064, i32 -1
481  %13 = load i16, i16* %py.41064, align 2
482  %conv40 = sext i16 %13 to i32
483  %mul.i863 = mul nsw i32 %conv40, %conv38
484  %shr3.i864 = ashr i32 %conv38, 16
485  %shr4.i865 = ashr i32 %conv40, 16
486  %mul5.i866 = mul nsw i32 %shr4.i865, %shr3.i864
487  %add.i867 = add i32 %mul.i863, %sum.21065
488  %add6.i868 = add i32 %add.i867, %mul5.i866
489  %dec42 = add nsw i32 %k.21066, -1
490  %cmp34 = icmp eq i32 %dec42, 0
491  br i1 %cmp34, label %while.end43, label %while.body36
492
493while.end43:                                      ; preds = %while.body36, %while.end31
494  %sum.2.lcssa = phi i32 [ %sum.1.lcssa, %while.end31 ], [ %add6.i868, %while.body36 ]
495  %14 = lshr i32 %sum.2.lcssa, 15
496  %conv45 = trunc i32 %14 to i16
497  %incdec.ptr46 = getelementptr inbounds i16, i16* %pOut.11069, i32 1
498  store i16 %conv45, i16* %pOut.11069, align 2
499  %sub47 = add i32 %count.11072, -1
500  %add.ptr48 = getelementptr inbounds i16, i16* %pSrcA.pSrcB, i32 %sub47
501  %inc49 = add i32 %count.11072, 1
502  %dec50 = add i32 %blockSize1.11071, -1
503  %cmp16 = icmp eq i32 %dec50, 0
504  br i1 %cmp16, label %exit, label %while.body18
505
506exit:                                             ; preds = %while.end43, %while.end13
507  ret void
508}
509