| blob | 0acb16c0cd9e3065d789c55cb4906cdaf4330082 |
1 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
2 ; RUN: opt < %s -cost-model -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
3 ; RUN: opt < %s -cost-model -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -analyze -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
4 ; RUN: opt < %s -cost-model -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
5 ; RUN: opt < %s -cost-model -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -analyze -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
6 ; RUN: opt < %s -cost-model -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -analyze -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
8 ; Check that we recognize the tree starting at the extractelement as a
9 ; reduction.
10 ; NOTE: We're only really interested in the extractelement cost, which represents the entire reduction.
12 define fastcc float @reduction_cost_float(<4 x float> %rdx) {
13 ; SSE2-LABEL: 'reduction_cost_float'
14 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
15 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
16 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
17 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
18 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
19 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
20 ;
21 ; SSSE3-LABEL: 'reduction_cost_float'
22 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
23 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
24 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
25 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
26 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
27 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
28 ;
29 ; SSE42-LABEL: 'reduction_cost_float'
30 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
31 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
32 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
33 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
34 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
35 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
36 ;
37 ; AVX-LABEL: 'reduction_cost_float'
38 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
39 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
40 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
41 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
42 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
43 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
44 ;
45 %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
46 %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
47 %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
48 %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
50 %r = extractelement <4 x float> %bin.rdx8, i32 0
51 ret float %r
52 }
54 define fastcc i32 @reduction_cost_int(<8 x i32> %rdx) {
55 ; SSE-LABEL: 'reduction_cost_int'
56 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
57 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
58 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
59 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
60 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.3 = shufflevector <8 x i32> %bin.rdx.2, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
61 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
62 ; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
63 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
64 ;
65 ; AVX1-LABEL: 'reduction_cost_int'
66 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
67 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
68 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
69 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
70 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.3 = shufflevector <8 x i32> %bin.rdx.2, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
71 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
72 ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
73 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
74 ;
75 ; AVX2-LABEL: 'reduction_cost_int'
76 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
77 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
78 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
79 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
80 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.3 = shufflevector <8 x i32> %bin.rdx.2, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
81 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
82 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
83 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
84 ;
85 %rdx.shuf = shufflevector <8 x i32> %rdx, <8 x i32> undef,
86 <8 x i32> <i32 4 , i32 5, i32 6, i32 7,
87 i32 undef, i32 undef, i32 undef, i32 undef>
88 %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
89 %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef,
90 <8 x i32> <i32 2 , i32 3, i32 undef, i32 undef,
91 i32 undef, i32 undef, i32 undef, i32 undef>
92 %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
93 %rdx.shuf.3 = shufflevector <8 x i32> %bin.rdx.2, <8 x i32> undef,
94 <8 x i32> <i32 1 , i32 undef, i32 undef, i32 undef,
95 i32 undef, i32 undef, i32 undef, i32 undef>
96 %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
98 %r = extractelement <8 x i32> %bin.rdx.3, i32 0
99 ret i32 %r
100 }
102 define fastcc float @pairwise_hadd(<4 x float> %rdx, float %f1) {
103 ; SSE2-LABEL: 'pairwise_hadd'
104 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
105 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
106 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
107 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
108 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
109 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
110 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
111 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
112 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
113 ;
114 ; SSSE3-LABEL: 'pairwise_hadd'
115 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
116 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
117 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
118 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
119 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
120 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
121 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
122 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
123 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
124 ;
125 ; SSE42-LABEL: 'pairwise_hadd'
126 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
127 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
128 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
129 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
130 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
131 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
132 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
133 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
134 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
135 ;
136 ; AVX-LABEL: 'pairwise_hadd'
137 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
138 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
139 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
140 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
141 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
142 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
143 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
144 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
145 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
146 ;
147 %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
148 <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
149 %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
150 <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
151 %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
152 %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
153 <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
154 %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
155 <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
156 %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
158 %r = extractelement <4 x float> %bin.rdx.1, i32 0
159 %r2 = fadd float %r, %f1
160 ret float %r2
161 }
163 define fastcc float @pairwise_hadd_assoc(<4 x float> %rdx, float %f1) {
164 ; SSE2-LABEL: 'pairwise_hadd_assoc'
165 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
166 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
167 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
168 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
169 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
170 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
171 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
172 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
173 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
174 ;
175 ; SSSE3-LABEL: 'pairwise_hadd_assoc'
176 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
177 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
178 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
179 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
180 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
181 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
182 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
183 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
184 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
185 ;
186 ; SSE42-LABEL: 'pairwise_hadd_assoc'
187 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
188 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
189 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
190 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
191 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
192 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
193 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
194 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
195 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
196 ;
197 ; AVX-LABEL: 'pairwise_hadd_assoc'
198 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
199 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
200 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
201 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
202 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
203 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
204 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
205 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
206 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
207 ;
208 %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
209 <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
210 %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
211 <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
212 %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
213 %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
214 <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
215 %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
216 <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
217 %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
219 %r = extractelement <4 x float> %bin.rdx.1, i32 0
220 %r2 = fadd float %r, %f1
221 ret float %r2
222 }
224 define fastcc float @pairwise_hadd_skip_first(<4 x float> %rdx, float %f1) {
225 ; SSE2-LABEL: 'pairwise_hadd_skip_first'
226 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
227 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
228 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
229 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
230 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
231 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
232 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
233 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
234 ;
235 ; SSSE3-LABEL: 'pairwise_hadd_skip_first'
236 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
237 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
238 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
239 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
240 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
241 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
242 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
243 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
244 ;
245 ; SSE42-LABEL: 'pairwise_hadd_skip_first'
246 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
247 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
248 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
249 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
250 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
251 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
252 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
253 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
254 ;
255 ; AVX-LABEL: 'pairwise_hadd_skip_first'
256 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
257 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
258 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
259 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
260 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
261 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
262 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
263 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
264 ;
265 %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
266 <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
267 %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
268 <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
269 %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
270 %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
271 <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
272 %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
274 %r = extractelement <4 x float> %bin.rdx.1, i32 0
275 %r2 = fadd float %r, %f1
276 ret float %r2
277 }
279 define fastcc double @no_pairwise_reduction2double(<2 x double> %rdx, double %f1) {
280 ; SSE2-LABEL: 'no_pairwise_reduction2double'
281 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
282 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
283 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
284 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
285 ;
286 ; SSSE3-LABEL: 'no_pairwise_reduction2double'
287 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
288 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
289 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
290 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
291 ;
292 ; SSE42-LABEL: 'no_pairwise_reduction2double'
293 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
294 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
295 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
296 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
297 ;
298 ; AVX-LABEL: 'no_pairwise_reduction2double'
299 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
300 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
301 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
302 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
303 ;
304 %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
305 %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
307 %r = extractelement <2 x double> %bin.rdx, i32 0
308 ret double %r
309 }
311 define fastcc float @no_pairwise_reduction4float(<4 x float> %rdx, float %f1) {
312 ; SSE2-LABEL: 'no_pairwise_reduction4float'
313 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
314 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
315 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
316 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
317 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
318 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
319 ;
320 ; SSSE3-LABEL: 'no_pairwise_reduction4float'
321 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
322 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
323 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
324 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
325 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
326 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
327 ;
328 ; SSE42-LABEL: 'no_pairwise_reduction4float'
329 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
330 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
331 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
332 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
333 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
334 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
335 ;
336 ; AVX-LABEL: 'no_pairwise_reduction4float'
337 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
338 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
339 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
340 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
341 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
342 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
343 ;
344 %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
345 %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
346 %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
347 %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
349 %r = extractelement <4 x float> %bin.rdx8, i32 0
350 ret float %r
351 }
353 define fastcc double @no_pairwise_reduction4double(<4 x double> %rdx, double %f1) {
354 ; SSE2-LABEL: 'no_pairwise_reduction4double'
355 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
356 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
357 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
358 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
359 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
360 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
361 ;
362 ; SSSE3-LABEL: 'no_pairwise_reduction4double'
363 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
364 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
365 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
366 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
367 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
368 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
369 ;
370 ; SSE42-LABEL: 'no_pairwise_reduction4double'
371 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
372 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
373 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
374 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
375 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
376 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
377 ;
378 ; AVX1-LABEL: 'no_pairwise_reduction4double'
379 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
380 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
381 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
382 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
383 ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
384 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
385 ;
386 ; AVX2-LABEL: 'no_pairwise_reduction4double'
387 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
388 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
389 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
390 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
391 ; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
392 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
393 ;
394 %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
395 %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
396 %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
397 %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
399 %r = extractelement <4 x double> %bin.rdx8, i32 0
400 ret double %r
401 }
403 define fastcc float @no_pairwise_reduction8float(<8 x float> %rdx, float %f1) {
404 ; SSE2-LABEL: 'no_pairwise_reduction8float'
405 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
406 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
407 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
408 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
409 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
410 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
411 ; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
412 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
413 ;
414 ; SSSE3-LABEL: 'no_pairwise_reduction8float'
415 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
416 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
417 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
418 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
419 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
420 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
421 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
422 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
423 ;
424 ; SSE42-LABEL: 'no_pairwise_reduction8float'
425 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
426 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
427 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
428 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
429 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
430 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
431 ; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
432 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
433 ;
434 ; AVX1-LABEL: 'no_pairwise_reduction8float'
435 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
436 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
437 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
438 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
439 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
440 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
441 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
442 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
443 ;
444 ; AVX2-LABEL: 'no_pairwise_reduction8float'
445 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
446 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
447 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
448 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
449 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
450 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
451 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
452 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
453 ;
454 %rdx.shuf3 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
455 %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
456 %rdx.shuf = shufflevector <8 x float> %bin.rdx4, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
457 %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
458 %rdx.shuf7 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
459 %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
461 %r = extractelement <8 x float> %bin.rdx8, i32 0
462 ret float %r
463 }
465 define fastcc i64 @no_pairwise_reduction2i64(<2 x i64> %rdx, i64 %f1) {
466 ; SSE2-LABEL: 'no_pairwise_reduction2i64'
467 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
468 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
469 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
470 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
471 ;
472 ; SSSE3-LABEL: 'no_pairwise_reduction2i64'
473 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
474 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
475 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
476 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
477 ;
478 ; SSE42-LABEL: 'no_pairwise_reduction2i64'
479 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
480 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
481 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
482 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
483 ;
484 ; AVX-LABEL: 'no_pairwise_reduction2i64'
485 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
486 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
487 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
488 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
489 ;
490 %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
491 %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
493 %r = extractelement <2 x i64> %bin.rdx, i32 0
494 ret i64 %r
495 }
497 define fastcc i32 @no_pairwise_reduction4i32(<4 x i32> %rdx, i32 %f1) {
498 ; SSE2-LABEL: 'no_pairwise_reduction4i32'
499 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
500 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
501 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
502 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
503 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
504 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
505 ;
506 ; SSSE3-LABEL: 'no_pairwise_reduction4i32'
507 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
508 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
509 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
510 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
511 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
512 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
513 ;
514 ; SSE42-LABEL: 'no_pairwise_reduction4i32'
515 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
516 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
517 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
518 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
519 ; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
520 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
521 ;
522 ; AVX-LABEL: 'no_pairwise_reduction4i32'
523 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
524 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
525 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
526 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
527 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
528 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
529 ;
530 %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
531 %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
532 %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
533 %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
535 %r = extractelement <4 x i32> %bin.rdx8, i32 0
536 ret i32 %r
537 }
539 define fastcc i64 @no_pairwise_reduction4i64(<4 x i64> %rdx, i64 %f1) {
540 ; SSE-LABEL: 'no_pairwise_reduction4i64'
541 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
542 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
543 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
544 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
545 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
546 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
547 ;
548 ; AVX1-LABEL: 'no_pairwise_reduction4i64'
549 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
550 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
551 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf7 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
552 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
553 ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
554 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
555 ;
556 ; AVX2-LABEL: 'no_pairwise_reduction4i64'
557 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
558 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
559 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
560 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
561 ; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
562 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
563 ;
564 %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
565 %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
566 %rdx.shuf7 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
567 %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
569 %r = extractelement <4 x i64> %bin.rdx8, i32 0
570 ret i64 %r
571 }
573 define fastcc i16 @no_pairwise_reduction8i16(<8 x i16> %rdx, i16 %f1) {
574 ; SSE2-LABEL: 'no_pairwise_reduction8i16'
575 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf3 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
576 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
577 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
578 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
579 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf7 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
580 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
581 ; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
582 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
583 ;
584 ; SSSE3-LABEL: 'no_pairwise_reduction8i16'
585 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf3 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
586 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
587 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
588 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
589 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
590 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
591 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
592 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
593 ;
594 ; SSE42-LABEL: 'no_pairwise_reduction8i16'
595 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf3 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
596 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
597 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
598 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
599 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
600 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
601 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
602 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
603 ;
604 ; AVX-LABEL: 'no_pairwise_reduction8i16'
605 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf3 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
606 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
607 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
608 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
609 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
610 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
611 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
612 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
613 ;
614 %rdx.shuf3 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
615 %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
616 %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
617 %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
618 %rdx.shuf7 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
619 %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
621 %r = extractelement <8 x i16> %bin.rdx8, i32 0
622 ret i16 %r
623 }
625 define fastcc i32 @no_pairwise_reduction8i32(<8 x i32> %rdx, i32 %f1) {
626 ; SSE-LABEL: 'no_pairwise_reduction8i32'
627 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
628 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
629 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x i32> %bin.rdx4, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
630 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
631 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
632 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
633 ; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
634 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
635 ;
636 ; AVX1-LABEL: 'no_pairwise_reduction8i32'
637 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf3 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
638 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
639 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf = shufflevector <8 x i32> %bin.rdx4, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
640 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
641 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf7 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
642 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
643 ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
644 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
645 ;
646 ; AVX2-LABEL: 'no_pairwise_reduction8i32'
647 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf3 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
648 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
649 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf = shufflevector <8 x i32> %bin.rdx4, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
650 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
651 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf7 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
652 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
653 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
654 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
655 ;
656 %rdx.shuf3 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
657 %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
658 %rdx.shuf = shufflevector <8 x i32> %bin.rdx4, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
659 %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
660 %rdx.shuf7 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
661 %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
663 %r = extractelement <8 x i32> %bin.rdx8, i32 0
664 ret i32 %r
665 }
667 define fastcc double @pairwise_reduction2double(<2 x double> %rdx, double %f1) {
668 ; SSE2-LABEL: 'pairwise_reduction2double'
669 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
670 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
671 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
672 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
673 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
674 ;
675 ; SSSE3-LABEL: 'pairwise_reduction2double'
676 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
677 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
678 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
679 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
680 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
681 ;
682 ; SSE42-LABEL: 'pairwise_reduction2double'
683 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
684 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
685 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
686 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
687 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
688 ;
689 ; AVX-LABEL: 'pairwise_reduction2double'
690 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
691 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
692 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
693 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
694 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
695 ;
696 %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
697 %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
698 %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
700 %r = extractelement <2 x double> %bin.rdx8, i32 0
701 ret double %r
702 }
704 define fastcc float @pairwise_reduction4float(<4 x float> %rdx, float %f1) {
705 ; SSE2-LABEL: 'pairwise_reduction4float'
706 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
707 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
708 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
709 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
710 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
711 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
712 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
713 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
714 ;
715 ; SSSE3-LABEL: 'pairwise_reduction4float'
716 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
717 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
718 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
719 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
720 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
721 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
722 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
723 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
724 ;
725 ; SSE42-LABEL: 'pairwise_reduction4float'
726 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
727 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
728 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
729 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
730 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
731 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
732 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
733 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
734 ;
735 ; AVX-LABEL: 'pairwise_reduction4float'
736 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
737 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
738 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
739 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
740 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
741 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
742 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
743 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
744 ;
745 %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
746 %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
747 %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
748 %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
749 %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
750 %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
752 %r = extractelement <4 x float> %bin.rdx8, i32 0
753 ret float %r
754 }
756 define fastcc double @pairwise_reduction4double(<4 x double> %rdx, double %f1) {
757 ; SSE2-LABEL: 'pairwise_reduction4double'
758 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
759 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
760 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
761 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
762 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
763 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
764 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
765 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
766 ;
767 ; SSSE3-LABEL: 'pairwise_reduction4double'
768 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
769 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
770 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
771 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
772 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
773 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
774 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
775 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
776 ;
777 ; SSE42-LABEL: 'pairwise_reduction4double'
778 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
779 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
780 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
781 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
782 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
783 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
784 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
785 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
786 ;
787 ; AVX1-LABEL: 'pairwise_reduction4double'
788 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
789 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
790 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
791 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
792 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
793 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
794 ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
795 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
796 ;
797 ; AVX2-LABEL: 'pairwise_reduction4double'
798 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
799 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
800 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
801 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
802 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
803 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
804 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
805 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret double %r
806 ;
807 %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
808 %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
809 %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
810 %rdx.shuf.1.0 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
811 %rdx.shuf.1.1 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
812 %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
814 %r = extractelement <4 x double> %bin.rdx8, i32 0
815 ret double %r
816 }
818 define fastcc float @pairwise_reduction8float(<8 x float> %rdx, float %f1) {
819 ; SSE2-LABEL: 'pairwise_reduction8float'
820 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
821 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
822 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
823 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
824 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
825 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
826 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
827 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
828 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
829 ; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
830 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
831 ;
832 ; SSSE3-LABEL: 'pairwise_reduction8float'
833 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
834 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
835 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
836 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
837 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
838 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
839 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
840 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
841 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
842 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
843 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
844 ;
845 ; SSE42-LABEL: 'pairwise_reduction8float'
846 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
847 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
848 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
849 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
850 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
851 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
852 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
853 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
854 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
855 ; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
856 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
857 ;
858 ; AVX1-LABEL: 'pairwise_reduction8float'
859 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
860 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
861 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
862 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
863 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
864 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
865 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
866 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
867 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
868 ; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
869 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
870 ;
871 ; AVX2-LABEL: 'pairwise_reduction8float'
872 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
873 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
874 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
875 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
876 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
877 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
878 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
879 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
880 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
881 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
882 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret float %r
883 ;
884 %rdx.shuf.0.0 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6,i32 undef, i32 undef, i32 undef, i32 undef>
885 %rdx.shuf.0.1 = shufflevector <8 x float> %rdx, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
886 %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
887 %rdx.shuf.1.0 = shufflevector <8 x float> %bin.rdx, <8 x float> undef,<8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
888 %rdx.shuf.1.1 = shufflevector <8 x float> %bin.rdx, <8 x float> undef,<8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
889 %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
890 %rdx.shuf.2.0 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef,<8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
891 %rdx.shuf.2.1 = shufflevector <8 x float> %bin.rdx8, <8 x float> undef,<8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
892 %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
894 %r = extractelement <8 x float> %bin.rdx9, i32 0
895 ret float %r
896 }
898 define fastcc i64 @pairwise_reduction2i64(<2 x i64> %rdx, i64 %f1) {
899 ; SSE2-LABEL: 'pairwise_reduction2i64'
900 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
901 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
902 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
903 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
904 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
905 ;
906 ; SSSE3-LABEL: 'pairwise_reduction2i64'
907 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
908 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
909 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
910 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
911 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
912 ;
913 ; SSE42-LABEL: 'pairwise_reduction2i64'
914 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
915 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
916 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
917 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
918 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
919 ;
920 ; AVX-LABEL: 'pairwise_reduction2i64'
921 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
922 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
923 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
924 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
925 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
926 ;
927 %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
928 %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
929 %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
931 %r = extractelement <2 x i64> %bin.rdx8, i32 0
932 ret i64 %r
933 }
935 define fastcc i32 @pairwise_reduction4i32(<4 x i32> %rdx, i32 %f1) {
936 ; SSE2-LABEL: 'pairwise_reduction4i32'
937 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
938 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
939 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
940 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
941 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
942 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
943 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
944 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
945 ;
946 ; SSSE3-LABEL: 'pairwise_reduction4i32'
947 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
948 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
949 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
950 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
951 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
952 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
953 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
954 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
955 ;
956 ; SSE42-LABEL: 'pairwise_reduction4i32'
957 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
958 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
959 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
960 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
961 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
962 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
963 ; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
964 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
965 ;
966 ; AVX-LABEL: 'pairwise_reduction4i32'
967 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
968 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
969 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
970 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
971 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
972 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
973 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
974 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
975 ;
976 %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
977 %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
978 %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
979 %rdx.shuf.1.0 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
980 %rdx.shuf.1.1 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
981 %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
983 %r = extractelement <4 x i32> %bin.rdx8, i32 0
984 ret i32 %r
985 }
987 define fastcc i64 @pairwise_reduction4i64(<4 x i64> %rdx, i64 %f1) {
988 ; SSE-LABEL: 'pairwise_reduction4i64'
989 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
990 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
991 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
992 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
993 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
994 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
995 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
996 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
997 ;
998 ; AVX1-LABEL: 'pairwise_reduction4i64'
999 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
1000 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
1001 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
1002 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
1003 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
1004 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1005 ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1006 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1007 ;
1008 ; AVX2-LABEL: 'pairwise_reduction4i64'
1009 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
1010 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
1011 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
1012 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.1.0 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
1013 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
1014 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1015 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1016 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1017 ;
1018 %rdx.shuf.0.0 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
1019 %rdx.shuf.0.1 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
1020 %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
1021 %rdx.shuf.1.0 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
1022 %rdx.shuf.1.1 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
1023 %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1025 %r = extractelement <4 x i64> %bin.rdx8, i32 0
1026 ret i64 %r
1027 }
1029 define fastcc i16 @pairwise_reduction8i16(<8 x i16> %rdx, i16 %f1) {
1030 ; SSE2-LABEL: 'pairwise_reduction8i16'
1031 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1032 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1033 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1034 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1035 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1036 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1037 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1038 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1039 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1040 ; SSE2-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1041 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1042 ;
1043 ; SSSE3-LABEL: 'pairwise_reduction8i16'
1044 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1045 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1046 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1047 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1048 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1049 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1050 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1051 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1052 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1053 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1054 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1055 ;
1056 ; SSE42-LABEL: 'pairwise_reduction8i16'
1057 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1058 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1059 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1060 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1061 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1062 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1063 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1064 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1065 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1066 ; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1067 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1068 ;
1069 ; AVX-LABEL: 'pairwise_reduction8i16'
1070 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1071 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1072 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1073 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1074 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1075 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1076 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1077 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1078 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1079 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1080 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1081 ;
1082 %rdx.shuf.0.0 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6,i32 undef, i32 undef, i32 undef, i32 undef>
1083 %rdx.shuf.0.1 = shufflevector <8 x i16> %rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
1084 %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1085 %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef,<8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1086 %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef,<8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1087 %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1088 %rdx.shuf.2.0 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef,<8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1089 %rdx.shuf.2.1 = shufflevector <8 x i16> %bin.rdx8, <8 x i16> undef,<8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1090 %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1092 %r = extractelement <8 x i16> %bin.rdx9, i32 0
1093 ret i16 %r
1094 }
1096 define fastcc i32 @pairwise_reduction8i32(<8 x i32> %rdx, i32 %f1) {
1097 ; SSE2-LABEL: 'pairwise_reduction8i32'
1098 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1099 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1100 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1101 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1102 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1103 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1104 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1105 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1106 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1107 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1108 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1109 ;
1110 ; SSSE3-LABEL: 'pairwise_reduction8i32'
1111 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1112 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1113 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1114 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1115 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1116 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1117 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1118 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1119 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1120 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1121 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1122 ;
1123 ; SSE42-LABEL: 'pairwise_reduction8i32'
1124 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1125 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1126 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1127 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1128 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1129 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1130 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1131 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1132 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1133 ; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1134 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1135 ;
1136 ; AVX1-LABEL: 'pairwise_reduction8i32'
1137 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1138 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1139 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1140 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1141 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1142 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1143 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1144 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1145 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1146 ; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1147 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1148 ;
1149 ; AVX2-LABEL: 'pairwise_reduction8i32'
1150 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
1151 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
1152 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1153 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1154 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1155 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1156 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1157 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef, <8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1158 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1159 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1160 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1161 ;
1162 %rdx.shuf.0.0 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6,i32 undef, i32 undef, i32 undef, i32 undef>
1163 %rdx.shuf.0.1 = shufflevector <8 x i32> %rdx, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7,i32 undef, i32 undef, i32 undef, i32 undef>
1164 %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1165 %rdx.shuf.1.0 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef,<8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1166 %rdx.shuf.1.1 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef,<8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1167 %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1168 %rdx.shuf.2.0 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef,<8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1169 %rdx.shuf.2.1 = shufflevector <8 x i32> %bin.rdx8, <8 x i32> undef,<8 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
1170 %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1172 %r = extractelement <8 x i32> %bin.rdx9, i32 0
1173 ret i32 %r
1174 }