xref: /llvm-project/llvm/test/CodeGen/AArch64/complex-deinterleaving-splat.ll (revision 3d18c8cd265c0c0bf1d85226c4770a2dd0f86e8f) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s --mattr=+complxnum -o - | FileCheck %s
3
4target triple = "aarch64"
5
6
7; a[i] * b[i] * (11.0 + 3.0.i);
8;
9define <4 x double> @complex_mul_const(<4 x double> %a, <4 x double> %b) {
10; CHECK-LABEL: complex_mul_const:
11; CHECK:       // %bb.0: // %entry
12; CHECK-NEXT:    movi v6.2d, #0000000000000000
13; CHECK-NEXT:    movi v5.2d, #0000000000000000
14; CHECK-NEXT:    adrp x8, .LCPI0_0
15; CHECK-NEXT:    movi v4.2d, #0000000000000000
16; CHECK-NEXT:    fcmla v6.2d, v1.2d, v3.2d, #0
17; CHECK-NEXT:    fcmla v5.2d, v0.2d, v2.2d, #0
18; CHECK-NEXT:    fcmla v6.2d, v1.2d, v3.2d, #90
19; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI0_0]
20; CHECK-NEXT:    fcmla v5.2d, v0.2d, v2.2d, #90
21; CHECK-NEXT:    movi v0.2d, #0000000000000000
22; CHECK-NEXT:    fcmla v4.2d, v6.2d, v1.2d, #0
23; CHECK-NEXT:    fcmla v0.2d, v5.2d, v1.2d, #0
24; CHECK-NEXT:    fcmla v4.2d, v6.2d, v1.2d, #90
25; CHECK-NEXT:    fcmla v0.2d, v5.2d, v1.2d, #90
26; CHECK-NEXT:    mov v1.16b, v4.16b
27; CHECK-NEXT:    ret
28entry:
29  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
30  %strided.vec47 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
31  %strided.vec49 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
32  %strided.vec50 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
33  %0 = fmul fast <2 x double> %strided.vec50, %strided.vec
34  %1 = fmul fast <2 x double> %strided.vec49, %strided.vec47
35  %2 = fadd fast <2 x double> %0, %1
36  %3 = fmul fast <2 x double> %strided.vec49, %strided.vec
37  %4 = fmul fast <2 x double> %strided.vec50, %strided.vec47
38  %5 = fsub fast <2 x double> %3, %4
39  %6 = fmul fast <2 x double> %5, <double 3.000000e+00, double 3.000000e+00>
40  %7 = fmul fast <2 x double> %2, <double 1.100000e+01, double 1.100000e+01>
41  %8 = fadd fast <2 x double> %6, %7
42  %9 = fmul fast <2 x double> %5, <double 1.100000e+01, double 1.100000e+01>
43  %10 = fmul fast <2 x double> %2, <double 3.000000e+00, double 3.000000e+00>
44  %11 = fsub fast <2 x double> %9, %10
45  %interleaved.vec = shufflevector <2 x double> %11, <2 x double> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
46  ret <4 x double> %interleaved.vec
47}
48
49
50; a[i] * b[i] * c;
51;
52define <4 x double> @complex_mul_non_const(<4 x double> %a, <4 x double> %b, [2 x double] %c) {
53; CHECK-LABEL: complex_mul_non_const:
54; CHECK:       // %bb.0: // %entry
55; CHECK-NEXT:    movi v7.2d, #0000000000000000
56; CHECK-NEXT:    movi v6.2d, #0000000000000000
57; CHECK-NEXT:    // kill: def $d5 killed $d5 def $q5
58; CHECK-NEXT:    // kill: def $d4 killed $d4 def $q4
59; CHECK-NEXT:    mov v4.d[1], v5.d[0]
60; CHECK-NEXT:    movi v5.2d, #0000000000000000
61; CHECK-NEXT:    fcmla v7.2d, v1.2d, v3.2d, #0
62; CHECK-NEXT:    fcmla v6.2d, v0.2d, v2.2d, #0
63; CHECK-NEXT:    fcmla v7.2d, v1.2d, v3.2d, #90
64; CHECK-NEXT:    fcmla v6.2d, v0.2d, v2.2d, #90
65; CHECK-NEXT:    movi v0.2d, #0000000000000000
66; CHECK-NEXT:    fcmla v5.2d, v7.2d, v4.2d, #0
67; CHECK-NEXT:    fcmla v0.2d, v6.2d, v4.2d, #0
68; CHECK-NEXT:    fcmla v5.2d, v7.2d, v4.2d, #90
69; CHECK-NEXT:    fcmla v0.2d, v6.2d, v4.2d, #90
70; CHECK-NEXT:    mov v1.16b, v5.16b
71; CHECK-NEXT:    ret
72entry:
73  %c.coerce.fca.1.extract = extractvalue [2 x double] %c, 1
74  %c.coerce.fca.0.extract = extractvalue [2 x double] %c, 0
75  %broadcast.splatinsert = insertelement <2 x double> poison, double %c.coerce.fca.1.extract, i64 0
76  %broadcast.splat = shufflevector <2 x double> %broadcast.splatinsert, <2 x double> poison, <2 x i32> zeroinitializer
77  %broadcast.splatinsert51 = insertelement <2 x double> poison, double %c.coerce.fca.0.extract, i64 0
78  %broadcast.splat52 = shufflevector <2 x double> %broadcast.splatinsert51, <2 x double> poison, <2 x i32> zeroinitializer
79  %strided.vec = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 0, i32 2>
80  %strided.vec47 = shufflevector <4 x double> %a, <4 x double> poison, <2 x i32> <i32 1, i32 3>
81  %strided.vec49 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 0, i32 2>
82  %strided.vec50 = shufflevector <4 x double> %b, <4 x double> poison, <2 x i32> <i32 1, i32 3>
83  %0 = fmul fast <2 x double> %strided.vec50, %strided.vec
84  %1 = fmul fast <2 x double> %strided.vec49, %strided.vec47
85  %2 = fadd fast <2 x double> %0, %1
86  %3 = fmul fast <2 x double> %strided.vec49, %strided.vec
87  %4 = fmul fast <2 x double> %strided.vec50, %strided.vec47
88  %5 = fsub fast <2 x double> %3, %4
89  %6 = fmul fast <2 x double> %5, %broadcast.splat
90  %7 = fmul fast <2 x double> %2, %broadcast.splat52
91  %8 = fadd fast <2 x double> %6, %7
92  %9 = fmul fast <2 x double> %5, %broadcast.splat52
93  %10 = fmul fast <2 x double> %2, %broadcast.splat
94  %11 = fsub fast <2 x double> %9, %10
95  %interleaved.vec = shufflevector <2 x double> %11, <2 x double> %8, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
96  ret <4 x double> %interleaved.vec
97}
98