1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
3; RUN: llc -mtriple=thumbv7em -mcpu=cortex-m4 -O3 %s -o - | FileCheck %s --check-prefix=CHECK-LLC
4; RUN: opt -S -mtriple=armv7-a -arm-parallel-dsp -dce %s -o - | FileCheck %s --check-prefix=CHECK-OPT
5
6define dso_local arm_aapcscc void @complex_dot_prod(i16* nocapture readonly %pSrcA, i16* nocapture readonly %pSrcB, i32* nocapture %realResult, i32* nocapture %imagResult) {
7; CHECK-LLC-LABEL: complex_dot_prod:
8; CHECK-LLC:       @ %bb.0: @ %entry
9; CHECK-LLC-NEXT:    push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}
10; CHECK-LLC-NEXT:    ldr r5, [r0]
11; CHECK-LLC-NEXT:    ldr r7, [r1]
12; CHECK-LLC-NEXT:    ldr.w r10, [r0, #4]
13; CHECK-LLC-NEXT:    ldr.w r8, [r0, #8]
14; CHECK-LLC-NEXT:    ldr.w r12, [r0, #12]
15; CHECK-LLC-NEXT:    ldr r4, [r1, #4]
16; CHECK-LLC-NEXT:    ldr.w r9, [r1, #8]
17; CHECK-LLC-NEXT:    ldr.w lr, [r1, #12]
18; CHECK-LLC-NEXT:    movs r0, #0
19; CHECK-LLC-NEXT:    movs r1, #0
20; CHECK-LLC-NEXT:    smlaldx r0, r1, r5, r7
21; CHECK-LLC-NEXT:    smulbb r6, r7, r5
22; CHECK-LLC-NEXT:    smultt r5, r7, r5
23; CHECK-LLC-NEXT:    asr.w r11, r6, #31
24; CHECK-LLC-NEXT:    subs r6, r6, r5
25; CHECK-LLC-NEXT:    sbc.w r5, r11, r5, asr #31
26; CHECK-LLC-NEXT:    smlaldx r0, r1, r10, r4
27; CHECK-LLC-NEXT:    smlalbb r6, r5, r4, r10
28; CHECK-LLC-NEXT:    smultt r4, r4, r10
29; CHECK-LLC-NEXT:    subs r6, r6, r4
30; CHECK-LLC-NEXT:    sbc.w r4, r5, r4, asr #31
31; CHECK-LLC-NEXT:    smlalbb r6, r4, r9, r8
32; CHECK-LLC-NEXT:    smultt r5, r9, r8
33; CHECK-LLC-NEXT:    subs r6, r6, r5
34; CHECK-LLC-NEXT:    sbc.w r4, r4, r5, asr #31
35; CHECK-LLC-NEXT:    smlaldx r0, r1, r8, r9
36; CHECK-LLC-NEXT:    smlalbb r6, r4, lr, r12
37; CHECK-LLC-NEXT:    smultt r7, lr, r12
38; CHECK-LLC-NEXT:    smlaldx r0, r1, r12, lr
39; CHECK-LLC-NEXT:    subs r6, r6, r7
40; CHECK-LLC-NEXT:    sbc.w r7, r4, r7, asr #31
41; CHECK-LLC-NEXT:    lsrs r6, r6, #6
42; CHECK-LLC-NEXT:    lsrs r0, r0, #6
43; CHECK-LLC-NEXT:    orr.w r7, r6, r7, lsl #26
44; CHECK-LLC-NEXT:    orr.w r0, r0, r1, lsl #26
45; CHECK-LLC-NEXT:    str r7, [r2]
46; CHECK-LLC-NEXT:    str r0, [r3]
47; CHECK-LLC-NEXT:    pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}
48; CHECK-LCC: pop.w {r4, r5, r6, r7, r8, r9, r10, pc}
49;
50; CHECK-OPT-LABEL: @complex_dot_prod(
51; CHECK-OPT-NEXT:  entry:
52; CHECK-OPT-NEXT:    [[TMP0:%.*]] = bitcast i16* [[PSRCA:%.*]] to i32*
53; CHECK-OPT-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 2
54; CHECK-OPT-NEXT:    [[TMP2:%.*]] = trunc i32 [[TMP1]] to i16
55; CHECK-OPT-NEXT:    [[TMP3:%.*]] = sext i16 [[TMP2]] to i32
56; CHECK-OPT-NEXT:    [[TMP4:%.*]] = lshr i32 [[TMP1]], 16
57; CHECK-OPT-NEXT:    [[TMP5:%.*]] = trunc i32 [[TMP4]] to i16
58; CHECK-OPT-NEXT:    [[TMP6:%.*]] = sext i16 [[TMP5]] to i32
59; CHECK-OPT-NEXT:    [[INCDEC_PTR1:%.*]] = getelementptr inbounds i16, i16* [[PSRCA]], i32 2
60; CHECK-OPT-NEXT:    [[TMP7:%.*]] = bitcast i16* [[PSRCB:%.*]] to i32*
61; CHECK-OPT-NEXT:    [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 2
62; CHECK-OPT-NEXT:    [[TMP9:%.*]] = trunc i32 [[TMP8]] to i16
63; CHECK-OPT-NEXT:    [[TMP10:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP1]], i32 [[TMP8]], i64 0)
64; CHECK-OPT-NEXT:    [[TMP11:%.*]] = sext i16 [[TMP9]] to i32
65; CHECK-OPT-NEXT:    [[TMP12:%.*]] = lshr i32 [[TMP8]], 16
66; CHECK-OPT-NEXT:    [[TMP13:%.*]] = trunc i32 [[TMP12]] to i16
67; CHECK-OPT-NEXT:    [[TMP14:%.*]] = sext i16 [[TMP13]] to i32
68; CHECK-OPT-NEXT:    [[INCDEC_PTR3:%.*]] = getelementptr inbounds i16, i16* [[PSRCB]], i32 2
69; CHECK-OPT-NEXT:    [[MUL:%.*]] = mul nsw i32 [[TMP11]], [[TMP3]]
70; CHECK-OPT-NEXT:    [[CONV5:%.*]] = sext i32 [[MUL]] to i64
71; CHECK-OPT-NEXT:    [[MUL13:%.*]] = mul nsw i32 [[TMP14]], [[TMP6]]
72; CHECK-OPT-NEXT:    [[CONV14:%.*]] = sext i32 [[MUL13]] to i64
73; CHECK-OPT-NEXT:    [[SUB:%.*]] = sub nsw i64 [[CONV5]], [[CONV14]]
74; CHECK-OPT-NEXT:    [[TMP15:%.*]] = bitcast i16* [[INCDEC_PTR1]] to i32*
75; CHECK-OPT-NEXT:    [[TMP16:%.*]] = load i32, i32* [[TMP15]], align 2
76; CHECK-OPT-NEXT:    [[TMP17:%.*]] = trunc i32 [[TMP16]] to i16
77; CHECK-OPT-NEXT:    [[TMP18:%.*]] = sext i16 [[TMP17]] to i32
78; CHECK-OPT-NEXT:    [[TMP19:%.*]] = lshr i32 [[TMP16]], 16
79; CHECK-OPT-NEXT:    [[TMP20:%.*]] = trunc i32 [[TMP19]] to i16
80; CHECK-OPT-NEXT:    [[TMP21:%.*]] = sext i16 [[TMP20]] to i32
81; CHECK-OPT-NEXT:    [[INCDEC_PTR21:%.*]] = getelementptr inbounds i16, i16* [[PSRCA]], i32 4
82; CHECK-OPT-NEXT:    [[TMP22:%.*]] = bitcast i16* [[INCDEC_PTR3]] to i32*
83; CHECK-OPT-NEXT:    [[TMP23:%.*]] = load i32, i32* [[TMP22]], align 2
84; CHECK-OPT-NEXT:    [[TMP24:%.*]] = trunc i32 [[TMP23]] to i16
85; CHECK-OPT-NEXT:    [[TMP25:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP16]], i32 [[TMP23]], i64 [[TMP10]])
86; CHECK-OPT-NEXT:    [[TMP26:%.*]] = sext i16 [[TMP24]] to i32
87; CHECK-OPT-NEXT:    [[TMP27:%.*]] = lshr i32 [[TMP23]], 16
88; CHECK-OPT-NEXT:    [[TMP28:%.*]] = trunc i32 [[TMP27]] to i16
89; CHECK-OPT-NEXT:    [[TMP29:%.*]] = sext i16 [[TMP28]] to i32
90; CHECK-OPT-NEXT:    [[INCDEC_PTR23:%.*]] = getelementptr inbounds i16, i16* [[PSRCB]], i32 4
91; CHECK-OPT-NEXT:    [[MUL26:%.*]] = mul nsw i32 [[TMP26]], [[TMP18]]
92; CHECK-OPT-NEXT:    [[CONV27:%.*]] = sext i32 [[MUL26]] to i64
93; CHECK-OPT-NEXT:    [[ADD28:%.*]] = add nsw i64 [[SUB]], [[CONV27]]
94; CHECK-OPT-NEXT:    [[MUL36:%.*]] = mul nsw i32 [[TMP29]], [[TMP21]]
95; CHECK-OPT-NEXT:    [[CONV37:%.*]] = sext i32 [[MUL36]] to i64
96; CHECK-OPT-NEXT:    [[SUB38:%.*]] = sub nsw i64 [[ADD28]], [[CONV37]]
97; CHECK-OPT-NEXT:    [[TMP30:%.*]] = bitcast i16* [[INCDEC_PTR21]] to i32*
98; CHECK-OPT-NEXT:    [[TMP31:%.*]] = load i32, i32* [[TMP30]], align 2
99; CHECK-OPT-NEXT:    [[TMP32:%.*]] = trunc i32 [[TMP31]] to i16
100; CHECK-OPT-NEXT:    [[TMP33:%.*]] = sext i16 [[TMP32]] to i32
101; CHECK-OPT-NEXT:    [[TMP34:%.*]] = lshr i32 [[TMP31]], 16
102; CHECK-OPT-NEXT:    [[TMP35:%.*]] = trunc i32 [[TMP34]] to i16
103; CHECK-OPT-NEXT:    [[TMP36:%.*]] = sext i16 [[TMP35]] to i32
104; CHECK-OPT-NEXT:    [[INCDEC_PTR45:%.*]] = getelementptr inbounds i16, i16* [[PSRCA]], i32 6
105; CHECK-OPT-NEXT:    [[TMP37:%.*]] = bitcast i16* [[INCDEC_PTR23]] to i32*
106; CHECK-OPT-NEXT:    [[TMP38:%.*]] = load i32, i32* [[TMP37]], align 2
107; CHECK-OPT-NEXT:    [[TMP39:%.*]] = trunc i32 [[TMP38]] to i16
108; CHECK-OPT-NEXT:    [[TMP40:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP31]], i32 [[TMP38]], i64 [[TMP25]])
109; CHECK-OPT-NEXT:    [[TMP41:%.*]] = sext i16 [[TMP39]] to i32
110; CHECK-OPT-NEXT:    [[TMP42:%.*]] = lshr i32 [[TMP38]], 16
111; CHECK-OPT-NEXT:    [[TMP43:%.*]] = trunc i32 [[TMP42]] to i16
112; CHECK-OPT-NEXT:    [[TMP44:%.*]] = sext i16 [[TMP43]] to i32
113; CHECK-OPT-NEXT:    [[INCDEC_PTR47:%.*]] = getelementptr inbounds i16, i16* [[PSRCB]], i32 6
114; CHECK-OPT-NEXT:    [[MUL50:%.*]] = mul nsw i32 [[TMP41]], [[TMP33]]
115; CHECK-OPT-NEXT:    [[CONV51:%.*]] = sext i32 [[MUL50]] to i64
116; CHECK-OPT-NEXT:    [[ADD52:%.*]] = add nsw i64 [[SUB38]], [[CONV51]]
117; CHECK-OPT-NEXT:    [[MUL60:%.*]] = mul nsw i32 [[TMP44]], [[TMP36]]
118; CHECK-OPT-NEXT:    [[CONV61:%.*]] = sext i32 [[MUL60]] to i64
119; CHECK-OPT-NEXT:    [[SUB62:%.*]] = sub nsw i64 [[ADD52]], [[CONV61]]
120; CHECK-OPT-NEXT:    [[TMP45:%.*]] = bitcast i16* [[INCDEC_PTR45]] to i32*
121; CHECK-OPT-NEXT:    [[TMP46:%.*]] = load i32, i32* [[TMP45]], align 2
122; CHECK-OPT-NEXT:    [[TMP47:%.*]] = trunc i32 [[TMP46]] to i16
123; CHECK-OPT-NEXT:    [[TMP48:%.*]] = sext i16 [[TMP47]] to i32
124; CHECK-OPT-NEXT:    [[TMP49:%.*]] = lshr i32 [[TMP46]], 16
125; CHECK-OPT-NEXT:    [[TMP50:%.*]] = trunc i32 [[TMP49]] to i16
126; CHECK-OPT-NEXT:    [[TMP51:%.*]] = sext i16 [[TMP50]] to i32
127; CHECK-OPT-NEXT:    [[TMP52:%.*]] = bitcast i16* [[INCDEC_PTR47]] to i32*
128; CHECK-OPT-NEXT:    [[TMP53:%.*]] = load i32, i32* [[TMP52]], align 2
129; CHECK-OPT-NEXT:    [[TMP54:%.*]] = trunc i32 [[TMP53]] to i16
130; CHECK-OPT-NEXT:    [[TMP55:%.*]] = call i64 @llvm.arm.smlaldx(i32 [[TMP46]], i32 [[TMP53]], i64 [[TMP40]])
131; CHECK-OPT-NEXT:    [[TMP56:%.*]] = sext i16 [[TMP54]] to i32
132; CHECK-OPT-NEXT:    [[TMP57:%.*]] = lshr i32 [[TMP53]], 16
133; CHECK-OPT-NEXT:    [[TMP58:%.*]] = trunc i32 [[TMP57]] to i16
134; CHECK-OPT-NEXT:    [[TMP59:%.*]] = sext i16 [[TMP58]] to i32
135; CHECK-OPT-NEXT:    [[MUL74:%.*]] = mul nsw i32 [[TMP56]], [[TMP48]]
136; CHECK-OPT-NEXT:    [[CONV75:%.*]] = sext i32 [[MUL74]] to i64
137; CHECK-OPT-NEXT:    [[ADD76:%.*]] = add nsw i64 [[SUB62]], [[CONV75]]
138; CHECK-OPT-NEXT:    [[MUL84:%.*]] = mul nsw i32 [[TMP59]], [[TMP51]]
139; CHECK-OPT-NEXT:    [[CONV85:%.*]] = sext i32 [[MUL84]] to i64
140; CHECK-OPT-NEXT:    [[SUB86:%.*]] = sub nsw i64 [[ADD76]], [[CONV85]]
141; CHECK-OPT-NEXT:    [[TMP60:%.*]] = lshr i64 [[SUB86]], 6
142; CHECK-OPT-NEXT:    [[CONV92:%.*]] = trunc i64 [[TMP60]] to i32
143; CHECK-OPT-NEXT:    store i32 [[CONV92]], i32* [[REALRESULT:%.*]], align 4
144; CHECK-OPT-NEXT:    [[TMP61:%.*]] = lshr i64 [[TMP55]], 6
145; CHECK-OPT-NEXT:    [[CONV94:%.*]] = trunc i64 [[TMP61]] to i32
146; CHECK-OPT-NEXT:    store i32 [[CONV94]], i32* [[IMAGRESULT:%.*]], align 4
147; CHECK-OPT-NEXT:    ret void
148entry:
149  %incdec.ptr = getelementptr inbounds i16, i16* %pSrcA, i32 1
150  %0 = load i16, i16* %pSrcA, align 2
151  %incdec.ptr1 = getelementptr inbounds i16, i16* %pSrcA, i32 2
152  %1 = load i16, i16* %incdec.ptr, align 2
153  %incdec.ptr2 = getelementptr inbounds i16, i16* %pSrcB, i32 1
154  %2 = load i16, i16* %pSrcB, align 2
155  %incdec.ptr3 = getelementptr inbounds i16, i16* %pSrcB, i32 2
156  %3 = load i16, i16* %incdec.ptr2, align 2
157  %conv = sext i16 %0 to i32
158  %conv4 = sext i16 %2 to i32
159  %mul = mul nsw i32 %conv4, %conv
160  %conv5 = sext i32 %mul to i64
161  %conv7 = sext i16 %3 to i32
162  %mul8 = mul nsw i32 %conv7, %conv
163  %conv9 = sext i32 %mul8 to i64
164  %conv11 = sext i16 %1 to i32
165  %mul13 = mul nsw i32 %conv7, %conv11
166  %conv14 = sext i32 %mul13 to i64
167  %sub = sub nsw i64 %conv5, %conv14
168  %mul17 = mul nsw i32 %conv4, %conv11
169  %conv18 = sext i32 %mul17 to i64
170  %add19 = add nsw i64 %conv9, %conv18
171  %incdec.ptr20 = getelementptr inbounds i16, i16* %pSrcA, i32 3
172  %4 = load i16, i16* %incdec.ptr1, align 2
173  %incdec.ptr21 = getelementptr inbounds i16, i16* %pSrcA, i32 4
174  %5 = load i16, i16* %incdec.ptr20, align 2
175  %incdec.ptr22 = getelementptr inbounds i16, i16* %pSrcB, i32 3
176  %6 = load i16, i16* %incdec.ptr3, align 2
177  %incdec.ptr23 = getelementptr inbounds i16, i16* %pSrcB, i32 4
178  %7 = load i16, i16* %incdec.ptr22, align 2
179  %conv24 = sext i16 %4 to i32
180  %conv25 = sext i16 %6 to i32
181  %mul26 = mul nsw i32 %conv25, %conv24
182  %conv27 = sext i32 %mul26 to i64
183  %add28 = add nsw i64 %sub, %conv27
184  %conv30 = sext i16 %7 to i32
185  %mul31 = mul nsw i32 %conv30, %conv24
186  %conv32 = sext i32 %mul31 to i64
187  %conv34 = sext i16 %5 to i32
188  %mul36 = mul nsw i32 %conv30, %conv34
189  %conv37 = sext i32 %mul36 to i64
190  %sub38 = sub nsw i64 %add28, %conv37
191  %mul41 = mul nsw i32 %conv25, %conv34
192  %conv42 = sext i32 %mul41 to i64
193  %add33 = add nsw i64 %add19, %conv42
194  %add43 = add nsw i64 %add33, %conv32
195  %incdec.ptr44 = getelementptr inbounds i16, i16* %pSrcA, i32 5
196  %8 = load i16, i16* %incdec.ptr21, align 2
197  %incdec.ptr45 = getelementptr inbounds i16, i16* %pSrcA, i32 6
198  %9 = load i16, i16* %incdec.ptr44, align 2
199  %incdec.ptr46 = getelementptr inbounds i16, i16* %pSrcB, i32 5
200  %10 = load i16, i16* %incdec.ptr23, align 2
201  %incdec.ptr47 = getelementptr inbounds i16, i16* %pSrcB, i32 6
202  %11 = load i16, i16* %incdec.ptr46, align 2
203  %conv48 = sext i16 %8 to i32
204  %conv49 = sext i16 %10 to i32
205  %mul50 = mul nsw i32 %conv49, %conv48
206  %conv51 = sext i32 %mul50 to i64
207  %add52 = add nsw i64 %sub38, %conv51
208  %conv54 = sext i16 %11 to i32
209  %mul55 = mul nsw i32 %conv54, %conv48
210  %conv56 = sext i32 %mul55 to i64
211  %conv58 = sext i16 %9 to i32
212  %mul60 = mul nsw i32 %conv54, %conv58
213  %conv61 = sext i32 %mul60 to i64
214  %sub62 = sub nsw i64 %add52, %conv61
215  %mul65 = mul nsw i32 %conv49, %conv58
216  %conv66 = sext i32 %mul65 to i64
217  %add57 = add nsw i64 %add43, %conv66
218  %add67 = add nsw i64 %add57, %conv56
219  %incdec.ptr68 = getelementptr inbounds i16, i16* %pSrcA, i32 7
220  %12 = load i16, i16* %incdec.ptr45, align 2
221  %13 = load i16, i16* %incdec.ptr68, align 2
222  %incdec.ptr70 = getelementptr inbounds i16, i16* %pSrcB, i32 7
223  %14 = load i16, i16* %incdec.ptr47, align 2
224  %15 = load i16, i16* %incdec.ptr70, align 2
225  %conv72 = sext i16 %12 to i32
226  %conv73 = sext i16 %14 to i32
227  %mul74 = mul nsw i32 %conv73, %conv72
228  %conv75 = sext i32 %mul74 to i64
229  %add76 = add nsw i64 %sub62, %conv75
230  %conv78 = sext i16 %15 to i32
231  %mul79 = mul nsw i32 %conv78, %conv72
232  %conv80 = sext i32 %mul79 to i64
233  %conv82 = sext i16 %13 to i32
234  %mul84 = mul nsw i32 %conv78, %conv82
235  %conv85 = sext i32 %mul84 to i64
236  %sub86 = sub nsw i64 %add76, %conv85
237  %mul89 = mul nsw i32 %conv73, %conv82
238  %conv90 = sext i32 %mul89 to i64
239  %add81 = add nsw i64 %add67, %conv90
240  %add91 = add nsw i64 %add81, %conv80
241  %16 = lshr i64 %sub86, 6
242  %conv92 = trunc i64 %16 to i32
243  store i32 %conv92, i32* %realResult, align 4
244  %17 = lshr i64 %add91, 6
245  %conv94 = trunc i64 %17 to i32
246  store i32 %conv94, i32* %imagResult, align 4
247  ret void
248}
249