xref: /llvm-project/llvm/test/CodeGen/AArch64/complex-deinterleaving-mixed-cases.ll (revision 9ef15f4109890b2ad1e7ac714e4398d44b2087ff) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s --mattr=+complxnum,+neon,+fullfp16 -o - | FileCheck %s
3
4target triple = "aarch64"
5
6; Expected to transform
7define <4 x float> @mul_mul(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
8; CHECK-LABEL: mul_mul:
9; CHECK:       // %bb.0: // %entry
10; CHECK-NEXT:    movi v4.2d, #0000000000000000
11; CHECK-NEXT:    movi v3.2d, #0000000000000000
12; CHECK-NEXT:    fcmla v4.4s, v1.4s, v0.4s, #0
13; CHECK-NEXT:    fcmla v4.4s, v1.4s, v0.4s, #90
14; CHECK-NEXT:    fcmla v3.4s, v2.4s, v4.4s, #0
15; CHECK-NEXT:    fcmla v3.4s, v2.4s, v4.4s, #90
16; CHECK-NEXT:    mov v0.16b, v3.16b
17; CHECK-NEXT:    ret
18entry:
19  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
20  %strided.vec151 = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
21  %strided.vec153 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
22  %strided.vec154 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
23  %0 = fmul fast <2 x float> %strided.vec154, %strided.vec151
24  %1 = fmul fast <2 x float> %strided.vec153, %strided.vec
25  %2 = fmul fast <2 x float> %strided.vec154, %strided.vec
26  %3 = fmul fast <2 x float> %strided.vec153, %strided.vec151
27  %4 = fadd fast <2 x float> %3, %2
28  %5 = fsub fast <2 x float> %1, %0
29  %strided.vec156 = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
30  %strided.vec157 = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
31  %6 = fmul fast <2 x float> %4, %strided.vec156
32  %7 = fmul fast <2 x float> %5, %strided.vec157
33  %8 = fadd fast <2 x float> %6, %7
34  %9 = fmul fast <2 x float> %strided.vec156, %5
35  %10 = fmul fast <2 x float> %4, %strided.vec157
36  %11 = fsub fast <2 x float> %9, %10
37  %interleaved.vec = shufflevector <2 x float> %11, <2 x float> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
38  ret <4 x float> %interleaved.vec
39}
40
41; Expected to not transform
42define <4 x float> @add_mul(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
43; CHECK-LABEL: add_mul:
44; CHECK:       // %bb.0: // %entry
45; CHECK-NEXT:    fsub v0.4s, v1.4s, v0.4s
46; CHECK-NEXT:    fsub v1.4s, v1.4s, v2.4s
47; CHECK-NEXT:    ext v3.16b, v2.16b, v2.16b, #8
48; CHECK-NEXT:    ext v4.16b, v0.16b, v0.16b, #8
49; CHECK-NEXT:    ext v5.16b, v1.16b, v1.16b, #8
50; CHECK-NEXT:    zip2 v0.2s, v0.2s, v4.2s
51; CHECK-NEXT:    zip2 v4.2s, v2.2s, v3.2s
52; CHECK-NEXT:    zip1 v1.2s, v1.2s, v5.2s
53; CHECK-NEXT:    zip1 v2.2s, v2.2s, v3.2s
54; CHECK-NEXT:    fmul v5.2s, v4.2s, v0.2s
55; CHECK-NEXT:    fmul v3.2s, v1.2s, v4.2s
56; CHECK-NEXT:    fneg v4.2s, v5.2s
57; CHECK-NEXT:    fmla v3.2s, v0.2s, v2.2s
58; CHECK-NEXT:    fmla v4.2s, v1.2s, v2.2s
59; CHECK-NEXT:    zip1 v0.4s, v4.4s, v3.4s
60; CHECK-NEXT:    ret
61entry:
62  %0 = fsub fast <4 x float> %b, %c
63  %1 = shufflevector <4 x float> %0, <4 x float> poison, <2 x i32> <i32 0, i32 2>
64  %strided.vec58 = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
65  %strided.vec59 = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
66  %2 = fmul fast <2 x float> %1, %strided.vec59
67  %3 = fsub fast <4 x float> %b, %a
68  %4 = shufflevector <4 x float> %3, <4 x float> poison, <2 x i32> <i32 1, i32 3>
69  %5 = fmul fast <2 x float> %strided.vec58, %4
70  %6 = fadd fast <2 x float> %5, %2
71  %7 = fmul fast <2 x float> %strided.vec58, %1
72  %8 = fmul fast <2 x float> %strided.vec59, %4
73  %9 = fsub fast <2 x float> %7, %8
74  %interleaved.vec = shufflevector <2 x float> %9, <2 x float> %6, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
75  ret <4 x float> %interleaved.vec
76}
77
78; Expected to not transform
79define <4 x float> @mul_mul270_mul(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
80; CHECK-LABEL: mul_mul270_mul:
81; CHECK:       // %bb.0: // %entry
82; CHECK-NEXT:    ext v3.16b, v2.16b, v2.16b, #8
83; CHECK-NEXT:    ext v4.16b, v1.16b, v1.16b, #8
84; CHECK-NEXT:    zip1 v5.2s, v2.2s, v3.2s
85; CHECK-NEXT:    zip1 v6.2s, v1.2s, v4.2s
86; CHECK-NEXT:    zip2 v2.2s, v2.2s, v3.2s
87; CHECK-NEXT:    zip2 v1.2s, v1.2s, v4.2s
88; CHECK-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
89; CHECK-NEXT:    fmul v7.2s, v6.2s, v5.2s
90; CHECK-NEXT:    fneg v4.2s, v7.2s
91; CHECK-NEXT:    zip2 v7.2s, v0.2s, v3.2s
92; CHECK-NEXT:    zip1 v0.2s, v0.2s, v3.2s
93; CHECK-NEXT:    fmla v4.2s, v2.2s, v1.2s
94; CHECK-NEXT:    fmul v1.2s, v1.2s, v5.2s
95; CHECK-NEXT:    fmul v3.2s, v4.2s, v7.2s
96; CHECK-NEXT:    fmla v1.2s, v2.2s, v6.2s
97; CHECK-NEXT:    fmul v2.2s, v4.2s, v0.2s
98; CHECK-NEXT:    fneg v3.2s, v3.2s
99; CHECK-NEXT:    fmla v2.2s, v7.2s, v1.2s
100; CHECK-NEXT:    fmla v3.2s, v0.2s, v1.2s
101; CHECK-NEXT:    zip1 v0.4s, v3.4s, v2.4s
102; CHECK-NEXT:    ret
103entry:
104  %strided.vec = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
105  %strided.vec81 = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
106  %strided.vec83 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
107  %strided.vec84 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
108  %0 = fmul fast <2 x float> %strided.vec84, %strided.vec
109  %1 = fmul fast <2 x float> %strided.vec83, %strided.vec81
110  %2 = fadd fast <2 x float> %1, %0
111  %strided.vec86 = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
112  %strided.vec87 = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
113  %3 = fmul fast <2 x float> %2, %strided.vec87
114  %4 = fmul fast <2 x float> %strided.vec84, %strided.vec81
115  %5 = fmul fast <2 x float> %strided.vec83, %strided.vec
116  %6 = fsub fast <2 x float> %4, %5
117  %7 = fmul fast <2 x float> %6, %strided.vec86
118  %8 = fadd fast <2 x float> %3, %7
119  %9 = fmul fast <2 x float> %2, %strided.vec86
120  %10 = fmul fast <2 x float> %6, %strided.vec87
121  %11 = fsub fast <2 x float> %9, %10
122  %interleaved.vec = shufflevector <2 x float> %11, <2 x float> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
123  ret <4 x float> %interleaved.vec
124}
125
126; (a * b) * a
127; Expected to transform
128define <4 x float> @mul_triangle(<4 x float> %a, <4 x float> %b) {
129; CHECK-LABEL: mul_triangle:
130; CHECK:       // %bb.0: // %entry
131; CHECK-NEXT:    movi v3.2d, #0000000000000000
132; CHECK-NEXT:    movi v2.2d, #0000000000000000
133; CHECK-NEXT:    fcmla v3.4s, v0.4s, v1.4s, #0
134; CHECK-NEXT:    fcmla v3.4s, v0.4s, v1.4s, #90
135; CHECK-NEXT:    fcmla v2.4s, v3.4s, v0.4s, #0
136; CHECK-NEXT:    fcmla v2.4s, v3.4s, v0.4s, #90
137; CHECK-NEXT:    mov v0.16b, v2.16b
138; CHECK-NEXT:    ret
139entry:
140  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
141  %strided.vec35 = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
142  %strided.vec37 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
143  %strided.vec38 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
144  %0 = fmul fast <2 x float> %strided.vec37, %strided.vec
145  %1 = fmul fast <2 x float> %strided.vec38, %strided.vec35
146  %2 = fsub fast <2 x float> %0, %1
147  %3 = fmul fast <2 x float> %2, %strided.vec35
148  %4 = fmul fast <2 x float> %strided.vec38, %strided.vec
149  %5 = fmul fast <2 x float> %strided.vec35, %strided.vec37
150  %6 = fadd fast <2 x float> %4, %5
151  %7 = fmul fast <2 x float> %6, %strided.vec
152  %8 = fadd fast <2 x float> %3, %7
153  %9 = fmul fast <2 x float> %2, %strided.vec
154  %10 = fmul fast <2 x float> %6, %strided.vec35
155  %11 = fsub fast <2 x float> %9, %10
156  %interleaved.vec = shufflevector <2 x float> %11, <2 x float> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
157  ret <4 x float> %interleaved.vec
158}
159
160
161; d * (b * a) * (c * a)
162; Expected to transform
163define <4 x float> @mul_diamond(<4 x float> %a, <4 x float> %b, <4 x float> %c, <4 x float> %d) {
164; CHECK-LABEL: mul_diamond:
165; CHECK:       // %bb.0: // %entry
166; CHECK-NEXT:    movi v4.2d, #0000000000000000
167; CHECK-NEXT:    movi v5.2d, #0000000000000000
168; CHECK-NEXT:    movi v6.2d, #0000000000000000
169; CHECK-NEXT:    fcmla v4.4s, v0.4s, v1.4s, #0
170; CHECK-NEXT:    fcmla v6.4s, v0.4s, v2.4s, #0
171; CHECK-NEXT:    fcmla v4.4s, v0.4s, v1.4s, #90
172; CHECK-NEXT:    movi v1.2d, #0000000000000000
173; CHECK-NEXT:    fcmla v6.4s, v0.4s, v2.4s, #90
174; CHECK-NEXT:    fcmla v5.4s, v3.4s, v4.4s, #0
175; CHECK-NEXT:    fcmla v5.4s, v3.4s, v4.4s, #90
176; CHECK-NEXT:    fcmla v1.4s, v5.4s, v6.4s, #0
177; CHECK-NEXT:    fcmla v1.4s, v5.4s, v6.4s, #90
178; CHECK-NEXT:    mov v0.16b, v1.16b
179; CHECK-NEXT:    ret
180entry:
181  %a.real = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
182  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
183  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
184  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
185  %c.real = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
186  %c.imag = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
187  %d.real = shufflevector <4 x float> %d, <4 x float> poison, <2 x i32> <i32 0, i32 2>
188  %d.imag = shufflevector <4 x float> %d, <4 x float> poison, <2 x i32> <i32 1, i32 3>
189  %0 = fmul fast <2 x float> %a.imag, %b.real
190  %1 = fmul fast <2 x float> %a.real, %b.imag
191  %2 = fadd fast <2 x float> %1, %0
192  %3 = fmul fast <2 x float> %a.real, %b.real
193  %4 = fmul fast <2 x float> %b.imag, %a.imag
194  %5 = fsub fast <2 x float> %3, %4
195  %6 = fmul fast <2 x float> %d.real, %5
196  %7 = fmul fast <2 x float> %2, %d.imag
197  %8 = fmul fast <2 x float> %d.real, %2
198  %9 = fmul fast <2 x float> %5, %d.imag
199  %10 = fsub fast <2 x float> %6, %7
200  %11 = fadd fast <2 x float> %8, %9
201  %12 = fmul fast <2 x float> %c.real, %a.imag
202  %13 = fmul fast <2 x float> %c.imag, %a.real
203  %14 = fadd fast <2 x float> %13, %12
204  %15 = fmul fast <2 x float> %14, %10
205  %16 = fmul fast <2 x float> %c.real, %a.real
206  %17 = fmul fast <2 x float> %c.imag, %a.imag
207  %18 = fsub fast <2 x float> %16, %17
208  %19 = fmul fast <2 x float> %18, %11
209  %20 = fadd fast <2 x float> %15, %19
210  %21 = fmul fast <2 x float> %18, %10
211  %22 = fmul fast <2 x float> %14, %11
212  %23 = fsub fast <2 x float> %21, %22
213  %interleaved.vec = shufflevector <2 x float> %23, <2 x float> %20, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
214  ret <4 x float> %interleaved.vec
215}
216
217; Expected to transform
218define <4 x float> @mul_add90_mul(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
219; CHECK-LABEL: mul_add90_mul:
220; CHECK:       // %bb.0: // %entry
221; CHECK-NEXT:    movi v3.2d, #0000000000000000
222; CHECK-NEXT:    movi v4.2d, #0000000000000000
223; CHECK-NEXT:    fcmla v4.4s, v0.4s, v2.4s, #0
224; CHECK-NEXT:    fcmla v3.4s, v0.4s, v1.4s, #0
225; CHECK-NEXT:    fcmla v4.4s, v0.4s, v2.4s, #90
226; CHECK-NEXT:    fcmla v3.4s, v0.4s, v1.4s, #90
227; CHECK-NEXT:    fcadd v0.4s, v4.4s, v3.4s, #90
228; CHECK-NEXT:    ret
229entry:
230  %ar = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
231  %ai = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
232  %br = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
233  %bi = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
234  %cr = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
235  %ci = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
236
237  %i6 = fmul fast <2 x float> %br, %ar
238  %i7 = fmul fast <2 x float> %bi, %ai
239  %xr = fsub fast <2 x float> %i6, %i7
240  %i9 = fmul fast <2 x float> %bi, %ar
241  %i10 = fmul fast <2 x float> %br, %ai
242  %xi = fadd fast <2 x float> %i9, %i10
243
244  %j6 = fmul fast <2 x float> %cr, %ar
245  %j7 = fmul fast <2 x float> %ci, %ai
246  %yr = fsub fast <2 x float> %j6, %j7
247  %j9 = fmul fast <2 x float> %ci, %ar
248  %j10 = fmul fast <2 x float> %cr, %ai
249  %yi = fadd fast <2 x float> %j9, %j10
250
251  %zr = fsub fast <2 x float> %yr, %xi
252  %zi = fadd fast <2 x float> %yi, %xr
253  %interleaved.vec = shufflevector <2 x float> %zr, <2 x float> %zi, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
254  ret <4 x float> %interleaved.vec
255}
256
257; Expected to not transform
258define <4 x float> @mul_triangle_addmul(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
259; CHECK-LABEL: mul_triangle_addmul:
260; CHECK:       // %bb.0: // %entry
261; CHECK-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
262; CHECK-NEXT:    ext v4.16b, v1.16b, v1.16b, #8
263; CHECK-NEXT:    zip1 v5.2s, v0.2s, v3.2s
264; CHECK-NEXT:    zip1 v6.2s, v1.2s, v4.2s
265; CHECK-NEXT:    zip2 v1.2s, v1.2s, v4.2s
266; CHECK-NEXT:    ext v4.16b, v2.16b, v2.16b, #8
267; CHECK-NEXT:    zip2 v0.2s, v0.2s, v3.2s
268; CHECK-NEXT:    fmul v7.2s, v6.2s, v5.2s
269; CHECK-NEXT:    fmul v5.2s, v1.2s, v5.2s
270; CHECK-NEXT:    zip1 v3.2s, v2.2s, v4.2s
271; CHECK-NEXT:    zip2 v2.2s, v2.2s, v4.2s
272; CHECK-NEXT:    fmov d4, d7
273; CHECK-NEXT:    fmov d16, d5
274; CHECK-NEXT:    fmls v7.2s, v0.2s, v2.2s
275; CHECK-NEXT:    fmla v5.2s, v0.2s, v3.2s
276; CHECK-NEXT:    fmls v4.2s, v0.2s, v1.2s
277; CHECK-NEXT:    fmla v16.2s, v0.2s, v6.2s
278; CHECK-NEXT:    fsub v0.2s, v7.2s, v16.2s
279; CHECK-NEXT:    fadd v1.2s, v5.2s, v4.2s
280; CHECK-NEXT:    zip1 v0.4s, v0.4s, v1.4s
281; CHECK-NEXT:    ret
282entry:
283  %ar = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
284  %ai = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
285  %br = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
286  %bi = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
287  %cr = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
288  %ci = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
289
290  %i6 = fmul fast <2 x float> %br, %ar
291  %i7 = fmul fast <2 x float> %bi, %ai
292  %xr = fsub fast <2 x float> %i6, %i7
293  %i9 = fmul fast <2 x float> %bi, %ar
294  %i10 = fmul fast <2 x float> %br, %ai
295  %xi = fadd fast <2 x float> %i9, %i10
296
297  ;%j6 = fmul fast <2 x float> %cr, %ar
298  %j7 = fmul fast <2 x float> %ci, %ai
299  %yr = fsub fast <2 x float> %i6, %j7
300  ;%j9 = fmul fast <2 x float> %ci, %ar
301  %j10 = fmul fast <2 x float> %cr, %ai
302  %yi = fadd fast <2 x float> %i9, %j10
303
304  %zr = fsub fast <2 x float> %yr, %xi
305  %zi = fadd fast <2 x float> %yi, %xr
306  %interleaved.vec = shufflevector <2 x float> %zr, <2 x float> %zi, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
307  ret <4 x float> %interleaved.vec
308}
309
310; Expected to not transform
311define <4 x float> @mul_triangle_multiuses(<4 x float> %a, <4 x float> %b, ptr %p) {
312; CHECK-LABEL: mul_triangle_multiuses:
313; CHECK:       // %bb.0: // %entry
314; CHECK-NEXT:    ext v2.16b, v0.16b, v0.16b, #8
315; CHECK-NEXT:    ext v3.16b, v1.16b, v1.16b, #8
316; CHECK-NEXT:    zip2 v4.2s, v0.2s, v2.2s
317; CHECK-NEXT:    zip1 v5.2s, v1.2s, v3.2s
318; CHECK-NEXT:    zip1 v0.2s, v0.2s, v2.2s
319; CHECK-NEXT:    zip2 v1.2s, v1.2s, v3.2s
320; CHECK-NEXT:    fmul v2.2s, v4.2s, v5.2s
321; CHECK-NEXT:    fmul v3.2s, v1.2s, v4.2s
322; CHECK-NEXT:    fmla v2.2s, v0.2s, v1.2s
323; CHECK-NEXT:    fneg v1.2s, v3.2s
324; CHECK-NEXT:    fmul v3.2s, v2.2s, v4.2s
325; CHECK-NEXT:    fmla v1.2s, v0.2s, v5.2s
326; CHECK-NEXT:    fmul v5.2s, v2.2s, v0.2s
327; CHECK-NEXT:    fneg v3.2s, v3.2s
328; CHECK-NEXT:    fmla v5.2s, v4.2s, v1.2s
329; CHECK-NEXT:    fmla v3.2s, v0.2s, v1.2s
330; CHECK-NEXT:    mov v1.d[1], v2.d[0]
331; CHECK-NEXT:    zip1 v0.4s, v3.4s, v5.4s
332; CHECK-NEXT:    str q1, [x0]
333; CHECK-NEXT:    ret
334entry:
335  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
336  %strided.vec35 = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
337  %strided.vec37 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
338  %strided.vec38 = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
339  %0 = fmul fast <2 x float> %strided.vec37, %strided.vec
340  %1 = fmul fast <2 x float> %strided.vec38, %strided.vec35
341  %2 = fsub fast <2 x float> %0, %1
342  %3 = fmul fast <2 x float> %2, %strided.vec35
343  %4 = fmul fast <2 x float> %strided.vec38, %strided.vec
344  %5 = fmul fast <2 x float> %strided.vec35, %strided.vec37
345  %6 = fadd fast <2 x float> %4, %5
346  %otheruse = shufflevector <2 x float> %2, <2 x float> %6, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
347  store <4 x float> %otheruse, ptr %p
348  %7 = fmul fast <2 x float> %6, %strided.vec
349  %8 = fadd fast <2 x float> %3, %7
350  %9 = fmul fast <2 x float> %2, %strided.vec
351  %10 = fmul fast <2 x float> %6, %strided.vec35
352  %11 = fsub fast <2 x float> %9, %10
353  %interleaved.vec = shufflevector <2 x float> %11, <2 x float> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
354  ret <4 x float> %interleaved.vec
355}
356
357; Expected to transform
358define <4 x float> @mul_addequal(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
359; CHECK-LABEL: mul_addequal:
360; CHECK:       // %bb.0: // %entry
361; CHECK-NEXT:    fcmla v2.4s, v1.4s, v0.4s, #0
362; CHECK-NEXT:    fcmla v2.4s, v1.4s, v0.4s, #90
363; CHECK-NEXT:    mov v0.16b, v2.16b
364; CHECK-NEXT:    ret
365entry:
366  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
367  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
368  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
369  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
370  %0 = fmul fast <2 x float> %b.imag, %strided.vec
371  %1 = fmul fast <2 x float> %b.real, %a.imag
372  %2 = fadd fast <2 x float> %1, %0
373  %3 = fmul fast <2 x float> %b.real, %strided.vec
374  %4 = fmul fast <2 x float> %a.imag, %b.imag
375  %5 = fsub fast <2 x float> %3, %4
376  %c.real = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
377  %c.imag = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
378  %6 = fadd fast <2 x float> %5, %c.real
379  %7 = fadd fast <2 x float> %2, %c.imag
380  %interleaved.vec = shufflevector <2 x float> %6, <2 x float> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
381  ret <4 x float> %interleaved.vec
382}
383
384; Expected to transform
385define <4 x float> @mul_subequal(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
386; CHECK-LABEL: mul_subequal:
387; CHECK:       // %bb.0: // %entry
388; CHECK-NEXT:    movi v3.2d, #0000000000000000
389; CHECK-NEXT:    fcmla v3.4s, v1.4s, v0.4s, #0
390; CHECK-NEXT:    fcmla v3.4s, v1.4s, v0.4s, #90
391; CHECK-NEXT:    fsub v0.4s, v3.4s, v2.4s
392; CHECK-NEXT:    ret
393entry:
394  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
395  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
396  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
397  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
398  %0 = fmul fast <2 x float> %b.imag, %strided.vec
399  %1 = fmul fast <2 x float> %b.real, %a.imag
400  %2 = fadd fast <2 x float> %1, %0
401  %3 = fmul fast <2 x float> %b.real, %strided.vec
402  %4 = fmul fast <2 x float> %a.imag, %b.imag
403  %5 = fsub fast <2 x float> %3, %4
404  %c.real = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
405  %c.imag = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
406  %6 = fsub fast <2 x float> %5, %c.real
407  %7 = fsub fast <2 x float> %2, %c.imag
408  %interleaved.vec = shufflevector <2 x float> %6, <2 x float> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
409  ret <4 x float> %interleaved.vec
410}
411
412
413; Expected to transform
414define <4 x float> @mul_mulequal(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
415; CHECK-LABEL: mul_mulequal:
416; CHECK:       // %bb.0: // %entry
417; CHECK-NEXT:    movi v3.2d, #0000000000000000
418; CHECK-NEXT:    fcmla v3.4s, v1.4s, v0.4s, #0
419; CHECK-NEXT:    fcmla v3.4s, v1.4s, v0.4s, #90
420; CHECK-NEXT:    fmul v0.4s, v3.4s, v2.4s
421; CHECK-NEXT:    ret
422entry:
423  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
424  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
425  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
426  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
427  %0 = fmul fast <2 x float> %b.imag, %strided.vec
428  %1 = fmul fast <2 x float> %b.real, %a.imag
429  %2 = fadd fast <2 x float> %1, %0
430  %3 = fmul fast <2 x float> %b.real, %strided.vec
431  %4 = fmul fast <2 x float> %a.imag, %b.imag
432  %5 = fsub fast <2 x float> %3, %4
433  %c.real = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
434  %c.imag = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
435  %6 = fmul fast <2 x float> %5, %c.real
436  %7 = fmul fast <2 x float> %2, %c.imag
437  %interleaved.vec = shufflevector <2 x float> %6, <2 x float> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
438  ret <4 x float> %interleaved.vec
439}
440
441; Expected to not transform
442define <4 x float> @mul_divequal(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
443; CHECK-LABEL: mul_divequal:
444; CHECK:       // %bb.0: // %entry
445; CHECK-NEXT:    ext v3.16b, v0.16b, v0.16b, #8
446; CHECK-NEXT:    ext v4.16b, v1.16b, v1.16b, #8
447; CHECK-NEXT:    zip2 v5.2s, v0.2s, v3.2s
448; CHECK-NEXT:    zip2 v6.2s, v1.2s, v4.2s
449; CHECK-NEXT:    zip1 v0.2s, v0.2s, v3.2s
450; CHECK-NEXT:    zip1 v1.2s, v1.2s, v4.2s
451; CHECK-NEXT:    ext v3.16b, v2.16b, v2.16b, #8
452; CHECK-NEXT:    fmul v7.2s, v5.2s, v6.2s
453; CHECK-NEXT:    fneg v4.2s, v7.2s
454; CHECK-NEXT:    zip1 v7.2s, v2.2s, v3.2s
455; CHECK-NEXT:    zip2 v2.2s, v2.2s, v3.2s
456; CHECK-NEXT:    fmla v4.2s, v0.2s, v1.2s
457; CHECK-NEXT:    fmul v0.2s, v6.2s, v0.2s
458; CHECK-NEXT:    fmla v0.2s, v5.2s, v1.2s
459; CHECK-NEXT:    fdiv v4.2s, v4.2s, v7.2s
460; CHECK-NEXT:    fdiv v0.2s, v0.2s, v2.2s
461; CHECK-NEXT:    zip1 v0.4s, v4.4s, v0.4s
462; CHECK-NEXT:    ret
463entry:
464  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
465  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
466  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
467  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
468  %0 = fmul fast <2 x float> %b.imag, %strided.vec
469  %1 = fmul fast <2 x float> %b.real, %a.imag
470  %2 = fadd fast <2 x float> %1, %0
471  %3 = fmul fast <2 x float> %b.real, %strided.vec
472  %4 = fmul fast <2 x float> %a.imag, %b.imag
473  %5 = fsub fast <2 x float> %3, %4
474  %c.real = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 0, i32 2>
475  %c.imag = shufflevector <4 x float> %c, <4 x float> poison, <2 x i32> <i32 1, i32 3>
476  %6 = fdiv fast <2 x float> %5, %c.real
477  %7 = fdiv fast <2 x float> %2, %c.imag
478  %interleaved.vec = shufflevector <2 x float> %6, <2 x float> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
479  ret <4 x float> %interleaved.vec
480}
481
482; Expected to transform
483define <4 x float> @mul_negequal(<4 x float> %a, <4 x float> %b) {
484; CHECK-LABEL: mul_negequal:
485; CHECK:       // %bb.0: // %entry
486; CHECK-NEXT:    movi v2.2d, #0000000000000000
487; CHECK-NEXT:    fcmla v2.4s, v1.4s, v0.4s, #180
488; CHECK-NEXT:    fcmla v2.4s, v1.4s, v0.4s, #270
489; CHECK-NEXT:    mov v0.16b, v2.16b
490; CHECK-NEXT:    ret
491entry:
492  %strided.vec = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 0, i32 2>
493  %a.imag = shufflevector <4 x float> %a, <4 x float> poison, <2 x i32> <i32 1, i32 3>
494  %b.real = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 0, i32 2>
495  %b.imag = shufflevector <4 x float> %b, <4 x float> poison, <2 x i32> <i32 1, i32 3>
496  %0 = fmul fast <2 x float> %b.imag, %strided.vec
497  %1 = fmul fast <2 x float> %b.real, %a.imag
498  %2 = fadd fast <2 x float> %1, %0
499  %3 = fmul fast <2 x float> %b.real, %strided.vec
500  %4 = fmul fast <2 x float> %a.imag, %b.imag
501  %5 = fsub fast <2 x float> %3, %4
502  %6 = fneg fast <2 x float> %5
503  %7 = fneg fast <2 x float> %2
504  %interleaved.vec = shufflevector <2 x float> %6, <2 x float> %7, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
505  ret <4 x float> %interleaved.vec
506}
507