xref: /llvm-project/llvm/test/CodeGen/AArch64/complex-deinterleaving-add-mull-fixed-contract.ll (revision 9ef15f4109890b2ad1e7ac714e4398d44b2087ff) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s --mattr=+complxnum,+neon -o - | FileCheck %s
3
4target triple = "aarch64"
5
6; a * b + c
7define <4 x double> @mull_add(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
8; CHECK-LABEL: mull_add:
9; CHECK:       // %bb.0: // %entry
10; CHECK-NEXT:    zip2 v4.2d, v2.2d, v3.2d
11; CHECK-NEXT:    zip2 v5.2d, v0.2d, v1.2d
12; CHECK-NEXT:    zip1 v0.2d, v0.2d, v1.2d
13; CHECK-NEXT:    zip1 v2.2d, v2.2d, v3.2d
14; CHECK-NEXT:    fmul v1.2d, v5.2d, v4.2d
15; CHECK-NEXT:    fmul v3.2d, v0.2d, v4.2d
16; CHECK-NEXT:    fneg v1.2d, v1.2d
17; CHECK-NEXT:    fmla v3.2d, v2.2d, v5.2d
18; CHECK-NEXT:    fmla v1.2d, v2.2d, v0.2d
19; CHECK-NEXT:    fadd v1.2d, v2.2d, v1.2d
20; CHECK-NEXT:    fadd v2.2d, v3.2d, v4.2d
21; CHECK-NEXT:    zip1 v0.2d, v1.2d, v2.2d
22; CHECK-NEXT:    zip2 v1.2d, v1.2d, v2.2d
23; CHECK-NEXT:    ret
24entry:
25  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
26  %strided.vec28 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
27  %strided.vec30 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
28  %strided.vec31 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
29  %0 = fmul contract <2 x double> %strided.vec, %strided.vec31
30  %1 = fmul contract <2 x double> %strided.vec28, %strided.vec30
31  %2 = fadd contract <2 x double> %1, %0
32  %3 = fmul contract <2 x double> %strided.vec, %strided.vec30
33  %4 = fmul contract <2 x double> %strided.vec28, %strided.vec31
34  %5 = fsub contract <2 x double> %3, %4
35  %strided.vec33 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
36  %strided.vec34 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
37  %6 = fadd contract <2 x double> %strided.vec33, %5
38  %7 = fadd contract <2 x double> %2, %strided.vec34
39  %interleaved.vec = shufflevector <2 x double> %6, <2 x double> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
40  ret <4 x double> %interleaved.vec
41}
42
43; a * b + c * d
44define <4 x double> @mul_add_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {
45; CHECK-LABEL: mul_add_mull:
46; CHECK:       // %bb.0: // %entry
47; CHECK-NEXT:    movi v16.2d, #0000000000000000
48; CHECK-NEXT:    movi v17.2d, #0000000000000000
49; CHECK-NEXT:    movi v18.2d, #0000000000000000
50; CHECK-NEXT:    movi v19.2d, #0000000000000000
51; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #0
52; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #0
53; CHECK-NEXT:    fcmla v17.2d, v7.2d, v5.2d, #0
54; CHECK-NEXT:    fcmla v19.2d, v6.2d, v4.2d, #0
55; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #90
56; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #90
57; CHECK-NEXT:    fcmla v17.2d, v7.2d, v5.2d, #90
58; CHECK-NEXT:    fcmla v19.2d, v6.2d, v4.2d, #90
59; CHECK-NEXT:    fadd v1.2d, v18.2d, v17.2d
60; CHECK-NEXT:    fadd v0.2d, v16.2d, v19.2d
61; CHECK-NEXT:    ret
62entry:
63  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
64  %strided.vec51 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
65  %strided.vec53 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
66  %strided.vec54 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
67  %0 = fmul contract <2 x double> %strided.vec, %strided.vec54
68  %1 = fmul contract <2 x double> %strided.vec51, %strided.vec53
69  %2 = fadd contract <2 x double> %1, %0
70  %3 = fmul contract <2 x double> %strided.vec, %strided.vec53
71  %4 = fmul contract <2 x double> %strided.vec51, %strided.vec54
72  %5 = fsub contract <2 x double> %3, %4
73  %strided.vec56 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>
74  %strided.vec57 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>
75  %strided.vec59 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>
76  %strided.vec60 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>
77  %6 = fmul contract <2 x double> %strided.vec56, %strided.vec60
78  %7 = fmul contract <2 x double> %strided.vec57, %strided.vec59
79  %8 = fadd contract <2 x double> %7, %6
80  %9 = fmul contract <2 x double> %strided.vec56, %strided.vec59
81  %10 = fmul contract <2 x double> %strided.vec57, %strided.vec60
82  %11 = fsub contract <2 x double> %9, %10
83  %12 = fadd contract <2 x double> %5, %11
84  %13 = fadd contract <2 x double> %2, %8
85  %interleaved.vec = shufflevector <2 x double> %12, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
86  ret <4 x double> %interleaved.vec
87}
88
89; a * b - c * d
90define <4 x double> @mul_sub_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {
91; CHECK-LABEL: mul_sub_mull:
92; CHECK:       // %bb.0: // %entry
93; CHECK-NEXT:    movi v16.2d, #0000000000000000
94; CHECK-NEXT:    movi v17.2d, #0000000000000000
95; CHECK-NEXT:    movi v18.2d, #0000000000000000
96; CHECK-NEXT:    movi v19.2d, #0000000000000000
97; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #0
98; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #0
99; CHECK-NEXT:    fcmla v17.2d, v7.2d, v5.2d, #0
100; CHECK-NEXT:    fcmla v19.2d, v6.2d, v4.2d, #0
101; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #90
102; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #90
103; CHECK-NEXT:    fcmla v17.2d, v7.2d, v5.2d, #90
104; CHECK-NEXT:    fcmla v19.2d, v6.2d, v4.2d, #90
105; CHECK-NEXT:    fsub v1.2d, v18.2d, v17.2d
106; CHECK-NEXT:    fsub v0.2d, v16.2d, v19.2d
107; CHECK-NEXT:    ret
108entry:
109  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
110  %strided.vec51 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
111  %strided.vec53 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
112  %strided.vec54 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
113  %0 = fmul contract <2 x double> %strided.vec, %strided.vec54
114  %1 = fmul contract <2 x double> %strided.vec51, %strided.vec53
115  %2 = fadd contract <2 x double> %1, %0
116  %3 = fmul contract <2 x double> %strided.vec, %strided.vec53
117  %4 = fmul contract <2 x double> %strided.vec51, %strided.vec54
118  %5 = fsub contract <2 x double> %3, %4
119  %strided.vec56 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>
120  %strided.vec57 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>
121  %strided.vec59 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>
122  %strided.vec60 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>
123  %6 = fmul contract <2 x double> %strided.vec56, %strided.vec60
124  %7 = fmul contract <2 x double> %strided.vec57, %strided.vec59
125  %8 = fadd contract <2 x double> %7, %6
126  %9 = fmul contract <2 x double> %strided.vec56, %strided.vec59
127  %10 = fmul contract <2 x double> %strided.vec57, %strided.vec60
128  %11 = fsub contract <2 x double> %9, %10
129  %12 = fsub contract <2 x double> %5, %11
130  %13 = fsub contract <2 x double> %2, %8
131  %interleaved.vec = shufflevector <2 x double> %12, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
132  ret <4 x double> %interleaved.vec
133}
134
135; a * b + conj(c) * d
136define <4 x double> @mul_conj_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {
137; CHECK-LABEL: mul_conj_mull:
138; CHECK:       // %bb.0: // %entry
139; CHECK-NEXT:    movi v16.2d, #0000000000000000
140; CHECK-NEXT:    movi v17.2d, #0000000000000000
141; CHECK-NEXT:    movi v18.2d, #0000000000000000
142; CHECK-NEXT:    movi v19.2d, #0000000000000000
143; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #0
144; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #0
145; CHECK-NEXT:    fcmla v17.2d, v5.2d, v7.2d, #0
146; CHECK-NEXT:    fcmla v19.2d, v4.2d, v6.2d, #0
147; CHECK-NEXT:    fcmla v16.2d, v2.2d, v0.2d, #90
148; CHECK-NEXT:    fcmla v18.2d, v3.2d, v1.2d, #90
149; CHECK-NEXT:    fcmla v17.2d, v5.2d, v7.2d, #270
150; CHECK-NEXT:    fcmla v19.2d, v4.2d, v6.2d, #270
151; CHECK-NEXT:    fadd v1.2d, v18.2d, v17.2d
152; CHECK-NEXT:    fadd v0.2d, v16.2d, v19.2d
153; CHECK-NEXT:    ret
154entry:
155  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
156  %strided.vec59 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
157  %strided.vec61 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
158  %strided.vec62 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
159  %0 = fmul contract <2 x double> %strided.vec, %strided.vec62
160  %1 = fmul contract <2 x double> %strided.vec59, %strided.vec61
161  %2 = fadd contract <2 x double> %1, %0
162  %3 = fmul contract <2 x double> %strided.vec, %strided.vec61
163  %4 = fmul contract <2 x double> %strided.vec59, %strided.vec62
164  %5 = fsub contract <2 x double> %3, %4
165  %strided.vec64 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>
166  %strided.vec65 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>
167  %strided.vec67 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>
168  %strided.vec68 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>
169  %6 = fmul contract <2 x double> %strided.vec64, %strided.vec68
170  %7 = fmul contract <2 x double> %strided.vec65, %strided.vec67
171  %8 = fsub contract <2 x double> %6, %7
172  %9 = fmul contract <2 x double> %strided.vec64, %strided.vec67
173  %10 = fmul contract <2 x double> %strided.vec65, %strided.vec68
174  %11 = fadd contract <2 x double> %9, %10
175  %12 = fadd contract <2 x double> %5, %11
176  %13 = fadd contract <2 x double> %2, %8
177  %interleaved.vec = shufflevector <2 x double> %12, <2 x double> %13, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
178  ret <4 x double> %interleaved.vec
179}
180
181; a + b + 1i * c * d
182define <4 x double> @mul_add_rot_mull(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {
183; CHECK-LABEL: mul_add_rot_mull:
184; CHECK:       // %bb.0: // %entry
185; CHECK-NEXT:    movi v16.2d, #0xffffffffffffffff
186; CHECK-NEXT:    zip2 v17.2d, v4.2d, v5.2d
187; CHECK-NEXT:    movi v18.2d, #0000000000000000
188; CHECK-NEXT:    zip1 v4.2d, v4.2d, v5.2d
189; CHECK-NEXT:    zip2 v19.2d, v0.2d, v1.2d
190; CHECK-NEXT:    zip2 v20.2d, v2.2d, v3.2d
191; CHECK-NEXT:    zip1 v0.2d, v0.2d, v1.2d
192; CHECK-NEXT:    zip1 v2.2d, v2.2d, v3.2d
193; CHECK-NEXT:    fneg v16.2d, v16.2d
194; CHECK-NEXT:    fmul v1.2d, v19.2d, v20.2d
195; CHECK-NEXT:    fmul v3.2d, v0.2d, v20.2d
196; CHECK-NEXT:    mov v5.16b, v16.16b
197; CHECK-NEXT:    bsl v16.16b, v18.16b, v4.16b
198; CHECK-NEXT:    fneg v1.2d, v1.2d
199; CHECK-NEXT:    fmla v3.2d, v2.2d, v19.2d
200; CHECK-NEXT:    bsl v5.16b, v18.16b, v17.16b
201; CHECK-NEXT:    fsub v16.2d, v16.2d, v17.2d
202; CHECK-NEXT:    fmla v1.2d, v2.2d, v0.2d
203; CHECK-NEXT:    fadd v4.2d, v4.2d, v5.2d
204; CHECK-NEXT:    zip2 v5.2d, v6.2d, v7.2d
205; CHECK-NEXT:    zip1 v6.2d, v6.2d, v7.2d
206; CHECK-NEXT:    fmul v17.2d, v4.2d, v5.2d
207; CHECK-NEXT:    fmul v5.2d, v16.2d, v5.2d
208; CHECK-NEXT:    fneg v7.2d, v17.2d
209; CHECK-NEXT:    fmla v5.2d, v4.2d, v6.2d
210; CHECK-NEXT:    fmla v7.2d, v6.2d, v16.2d
211; CHECK-NEXT:    fadd v2.2d, v3.2d, v5.2d
212; CHECK-NEXT:    fadd v1.2d, v1.2d, v7.2d
213; CHECK-NEXT:    zip1 v0.2d, v1.2d, v2.2d
214; CHECK-NEXT:    zip2 v1.2d, v1.2d, v2.2d
215; CHECK-NEXT:    ret
216entry:
217  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
218  %strided.vec77 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
219  %strided.vec79 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
220  %strided.vec80 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
221  %0 = fmul contract <2 x double> %strided.vec, %strided.vec80
222  %1 = fmul contract <2 x double> %strided.vec77, %strided.vec79
223  %2 = fadd contract <2 x double> %1, %0
224  %3 = fmul contract <2 x double> %strided.vec, %strided.vec79
225  %4 = fmul contract <2 x double> %strided.vec77, %strided.vec80
226  %5 = fsub contract <2 x double> %3, %4
227  %strided.vec82 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 0, i32 2>
228  %strided.vec83 = shufflevector <4 x double> %c, <4 x double> poison, <2 x i32> <i32 1, i32 3>
229  %6 = tail call contract <2 x double> @llvm.copysign.v2f64(<2 x double> zeroinitializer, <2 x double> %strided.vec83)
230  %7 = fadd contract <2 x double> %strided.vec82, %6
231  %8 = tail call contract <2 x double> @llvm.copysign.v2f64(<2 x double> zeroinitializer, <2 x double> %strided.vec82)
232  %9 = fsub contract <2 x double> %8, %strided.vec83
233  %strided.vec85 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 0, i32 2>
234  %strided.vec86 = shufflevector <4 x double> %d, <4 x double> poison, <2 x i32> <i32 1, i32 3>
235  %10 = fmul contract <2 x double> %9, %strided.vec86
236  %11 = fmul contract <2 x double> %strided.vec85, %7
237  %12 = fadd contract <2 x double> %11, %10
238  %13 = fmul contract <2 x double> %9, %strided.vec85
239  %14 = fmul contract <2 x double> %7, %strided.vec86
240  %15 = fsub contract <2 x double> %13, %14
241  %16 = fadd contract <2 x double> %5, %15
242  %17 = fadd contract <2 x double> %2, %12
243  %interleaved.vec = shufflevector <2 x double> %16, <2 x double> %17, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
244  ret <4 x double> %interleaved.vec
245}
246
247declare <2 x double> @llvm.copysign.v2f64(<2 x double>, <2 x double>)
248