1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s
\r
3 ; Check that we do not get excessive spilling from splitting of constant live ranges.
\r
5 ; CHECK-LABEL: PR24139:
\r
6 ; CHECK: # 16-byte Spill
\r
7 ; CHECK-NOT: # 16-byte Spill
\r
10 define <2 x double> @PR24139(<2 x double> %arg, <2 x double> %arg1, <2 x double> %arg2) {
\r
11 %tmp = bitcast <2 x double> %arg to <4 x float>
\r
12 %tmp3 = fmul <4 x float> %tmp, <float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000>
\r
13 %tmp4 = bitcast <2 x double> %arg to <4 x i32>
\r
14 %tmp5 = and <4 x i32> %tmp4, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
15 %tmp6 = or <4 x i32> %tmp5, <i32 1056964608, i32 1056964608, i32 1056964608, i32 1056964608>
\r
16 %tmp7 = bitcast <4 x i32> %tmp6 to <4 x float>
\r
17 %tmp8 = fadd <4 x float> %tmp3, %tmp7
\r
18 %tmp9 = tail call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %tmp8) #2
\r
19 %tmp10 = bitcast <4 x i32> %tmp9 to <2 x i64>
\r
20 %tmp11 = tail call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %tmp9) #2
\r
21 %tmp12 = fmul <4 x float> %tmp11, <float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000>
\r
22 %tmp13 = fsub <4 x float> %tmp, %tmp12
\r
23 %tmp14 = fmul <4 x float> %tmp11, <float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000>
\r
24 %tmp15 = fsub <4 x float> %tmp13, %tmp14
\r
25 %tmp16 = fmul <4 x float> %tmp15, %tmp15
\r
26 %tmp17 = fmul <4 x float> %tmp15, %tmp16
\r
27 %tmp18 = fmul <4 x float> %tmp16, <float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000>
\r
28 %tmp19 = fadd <4 x float> %tmp18, <float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000>
\r
29 %tmp20 = fmul <4 x float> %tmp16, <float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000>
\r
30 %tmp21 = fadd <4 x float> %tmp20, <float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000>
\r
31 %tmp22 = fmul <4 x float> %tmp16, %tmp19
\r
32 %tmp23 = fadd <4 x float> %tmp22, <float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000>
\r
33 %tmp24 = fmul <4 x float> %tmp16, %tmp21
\r
34 %tmp25 = fadd <4 x float> %tmp24, <float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000>
\r
35 %tmp26 = fmul <4 x float> %tmp16, %tmp23
\r
36 %tmp27 = fadd <4 x float> %tmp26, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
\r
37 %tmp28 = fmul <4 x float> %tmp17, %tmp25
\r
38 %tmp29 = fadd <4 x float> %tmp15, %tmp28
\r
39 %tmp30 = and <2 x i64> %tmp10, <i64 4294967297, i64 4294967297>
\r
40 %tmp31 = bitcast <2 x i64> %tmp30 to <4 x i32>
\r
41 %tmp32 = icmp eq <4 x i32> %tmp31, zeroinitializer
\r
42 %tmp33 = sext <4 x i1> %tmp32 to <4 x i32>
\r
43 %tmp34 = bitcast <4 x i32> %tmp33 to <4 x float>
\r
44 %tmp35 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp27, <4 x float> %tmp29, <4 x float> %tmp34) #2
\r
45 %tmp36 = and <2 x i64> %tmp10, <i64 8589934594, i64 8589934594>
\r
46 %tmp37 = bitcast <2 x i64> %tmp36 to <4 x i32>
\r
47 %tmp38 = icmp eq <4 x i32> %tmp37, zeroinitializer
\r
48 %tmp39 = sext <4 x i1> %tmp38 to <4 x i32>
\r
49 %tmp40 = bitcast <4 x float> %tmp35 to <4 x i32>
\r
50 %tmp41 = xor <4 x i32> %tmp40, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
51 %tmp42 = bitcast <4 x i32> %tmp41 to <4 x float>
\r
52 %tmp43 = bitcast <4 x i32> %tmp39 to <4 x float>
\r
53 %tmp44 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp42, <4 x float> %tmp35, <4 x float> %tmp43) #2
\r
54 %tmp45 = bitcast <2 x double> %arg1 to <4 x float>
\r
55 %tmp46 = fmul <4 x float> %tmp45, <float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000>
\r
56 %tmp47 = bitcast <2 x double> %arg1 to <4 x i32>
\r
57 %tmp48 = and <4 x i32> %tmp47, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
58 %tmp49 = or <4 x i32> %tmp48, <i32 1056964608, i32 1056964608, i32 1056964608, i32 1056964608>
\r
59 %tmp50 = bitcast <4 x i32> %tmp49 to <4 x float>
\r
60 %tmp51 = fadd <4 x float> %tmp46, %tmp50
\r
61 %tmp52 = tail call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %tmp51) #2
\r
62 %tmp53 = bitcast <4 x i32> %tmp52 to <2 x i64>
\r
63 %tmp54 = tail call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %tmp52) #2
\r
64 %tmp55 = fmul <4 x float> %tmp54, <float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000>
\r
65 %tmp56 = fsub <4 x float> %tmp45, %tmp55
\r
66 %tmp57 = fmul <4 x float> %tmp54, <float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000>
\r
67 %tmp58 = fsub <4 x float> %tmp56, %tmp57
\r
68 %tmp59 = fmul <4 x float> %tmp58, %tmp58
\r
69 %tmp60 = fmul <4 x float> %tmp58, %tmp59
\r
70 %tmp61 = fmul <4 x float> %tmp59, <float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000>
\r
71 %tmp62 = fadd <4 x float> %tmp61, <float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000>
\r
72 %tmp63 = fmul <4 x float> %tmp59, <float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000>
\r
73 %tmp64 = fadd <4 x float> %tmp63, <float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000>
\r
74 %tmp65 = fmul <4 x float> %tmp59, %tmp62
\r
75 %tmp66 = fadd <4 x float> %tmp65, <float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000>
\r
76 %tmp67 = fmul <4 x float> %tmp59, %tmp64
\r
77 %tmp68 = fadd <4 x float> %tmp67, <float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000>
\r
78 %tmp69 = fmul <4 x float> %tmp59, %tmp66
\r
79 %tmp70 = fadd <4 x float> %tmp69, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
\r
80 %tmp71 = fmul <4 x float> %tmp60, %tmp68
\r
81 %tmp72 = fadd <4 x float> %tmp58, %tmp71
\r
82 %tmp73 = and <2 x i64> %tmp53, <i64 4294967297, i64 4294967297>
\r
83 %tmp74 = bitcast <2 x i64> %tmp73 to <4 x i32>
\r
84 %tmp75 = icmp eq <4 x i32> %tmp74, zeroinitializer
\r
85 %tmp76 = sext <4 x i1> %tmp75 to <4 x i32>
\r
86 %tmp77 = bitcast <4 x i32> %tmp76 to <4 x float>
\r
87 %tmp78 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp70, <4 x float> %tmp72, <4 x float> %tmp77) #2
\r
88 %tmp79 = and <2 x i64> %tmp53, <i64 8589934594, i64 8589934594>
\r
89 %tmp80 = bitcast <2 x i64> %tmp79 to <4 x i32>
\r
90 %tmp81 = icmp eq <4 x i32> %tmp80, zeroinitializer
\r
91 %tmp82 = sext <4 x i1> %tmp81 to <4 x i32>
\r
92 %tmp83 = bitcast <4 x float> %tmp78 to <4 x i32>
\r
93 %tmp84 = xor <4 x i32> %tmp83, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
94 %tmp85 = bitcast <4 x i32> %tmp84 to <4 x float>
\r
95 %tmp86 = bitcast <4 x i32> %tmp82 to <4 x float>
\r
96 %tmp87 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp85, <4 x float> %tmp78, <4 x float> %tmp86) #2
\r
97 %tmp88 = fadd <4 x float> %tmp44, %tmp87
\r
98 %tmp89 = bitcast <2 x double> %arg2 to <4 x float>
\r
99 %tmp90 = fmul <4 x float> %tmp89, <float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000, float 0x3FE45F3060000000>
\r
100 %tmp91 = bitcast <2 x double> %arg2 to <4 x i32>
\r
101 %tmp92 = and <4 x i32> %tmp91, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
102 %tmp93 = or <4 x i32> %tmp92, <i32 1056964608, i32 1056964608, i32 1056964608, i32 1056964608>
\r
103 %tmp94 = bitcast <4 x i32> %tmp93 to <4 x float>
\r
104 %tmp95 = fadd <4 x float> %tmp90, %tmp94
\r
105 %tmp96 = tail call <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float> %tmp95) #2
\r
106 %tmp97 = bitcast <4 x i32> %tmp96 to <2 x i64>
\r
107 %tmp98 = tail call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %tmp96) #2
\r
108 %tmp99 = fmul <4 x float> %tmp98, <float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000, float 0x3FF921FB40000000>
\r
109 %tmp100 = fsub <4 x float> %tmp89, %tmp99
\r
110 %tmp101 = fmul <4 x float> %tmp98, <float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000, float 0x3E74442D00000000>
\r
111 %tmp102 = fsub <4 x float> %tmp100, %tmp101
\r
112 %tmp103 = fmul <4 x float> %tmp102, %tmp102
\r
113 %tmp104 = fmul <4 x float> %tmp102, %tmp103
\r
114 %tmp105 = fmul <4 x float> %tmp103, <float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000, float 0xBF56493260000000>
\r
115 %tmp106 = fadd <4 x float> %tmp105, <float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000, float 0x3FA55406C0000000>
\r
116 %tmp107 = fmul <4 x float> %tmp103, <float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000, float 0xBF29918DC0000000>
\r
117 %tmp108 = fadd <4 x float> %tmp107, <float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000, float 0x3F81106840000000>
\r
118 %tmp109 = fmul <4 x float> %tmp103, %tmp106
\r
119 %tmp110 = fadd <4 x float> %tmp109, <float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000, float 0xBFDFFFFBE0000000>
\r
120 %tmp111 = fmul <4 x float> %tmp103, %tmp108
\r
121 %tmp112 = fadd <4 x float> %tmp111, <float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000, float 0xBFC5555420000000>
\r
122 %tmp113 = fmul <4 x float> %tmp103, %tmp110
\r
123 %tmp114 = fadd <4 x float> %tmp113, <float 1.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>
\r
124 %tmp115 = fmul <4 x float> %tmp104, %tmp112
\r
125 %tmp116 = fadd <4 x float> %tmp102, %tmp115
\r
126 %tmp117 = and <2 x i64> %tmp97, <i64 4294967297, i64 4294967297>
\r
127 %tmp118 = bitcast <2 x i64> %tmp117 to <4 x i32>
\r
128 %tmp119 = icmp eq <4 x i32> %tmp118, zeroinitializer
\r
129 %tmp120 = sext <4 x i1> %tmp119 to <4 x i32>
\r
130 %tmp121 = bitcast <4 x i32> %tmp120 to <4 x float>
\r
131 %tmp122 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp114, <4 x float> %tmp116, <4 x float> %tmp121) #2
\r
132 %tmp123 = and <2 x i64> %tmp97, <i64 8589934594, i64 8589934594>
\r
133 %tmp124 = bitcast <2 x i64> %tmp123 to <4 x i32>
\r
134 %tmp125 = icmp eq <4 x i32> %tmp124, zeroinitializer
\r
135 %tmp126 = sext <4 x i1> %tmp125 to <4 x i32>
\r
136 %tmp127 = bitcast <4 x float> %tmp122 to <4 x i32>
\r
137 %tmp128 = xor <4 x i32> %tmp127, <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>
\r
138 %tmp129 = bitcast <4 x i32> %tmp128 to <4 x float>
\r
139 %tmp130 = bitcast <4 x i32> %tmp126 to <4 x float>
\r
140 %tmp131 = tail call <4 x float> @llvm.x86.sse41.blendvps(<4 x float> %tmp129, <4 x float> %tmp122, <4 x float> %tmp130) #2
\r
141 %tmp132 = fadd <4 x float> %tmp88, %tmp131
\r
142 %tmp133 = bitcast <4 x float> %tmp132 to <2 x double>
\r
143 ret <2 x double> %tmp133
\r
146 declare <4 x i32> @llvm.x86.sse2.cvttps2dq(<4 x float>)
\r
147 declare <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32>)
\r
148 declare <4 x float> @llvm.x86.sse41.blendvps(<4 x float>, <4 x float>, <4 x float>)
\r