xref: /llvm-project/llvm/test/Analysis/CostModel/X86/reduction.ll (revision 89ca3e72ca03efbbfb5ae9b1c71d81f2d1753521) 
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
2; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
3; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSSE3
4; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
5; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
6; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
7
8; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -costmodel-reduxcost=true -mtriple=x86_64-apple-darwin -mcpu=slm | FileCheck %s --check-prefixes=SLM
9
10; These are old tests for matching reduction costs from extract elements - something that has now been removed.
11
12define 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 poison, i32 poison>
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 poison, i32 poison, i32 poison>
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 0 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 poison, i32 poison>
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 poison, i32 poison, i32 poison>
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 0 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 poison, i32 poison>
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 poison, i32 poison, i32 poison>
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 0 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 poison, i32 poison>
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 poison, i32 poison, i32 poison>
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 0 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; SLM-LABEL: 'reduction_cost_float'
46; SLM-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 poison, i32 poison>
47; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
48; SLM-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 poison, i32 poison, i32 poison>
49; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
50; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
51; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
52;
53  %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
54  %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
55  %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
56  %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
57
58  %r = extractelement <4 x float> %bin.rdx8, i32 0
59  ret float %r
60}
61
62define fastcc i32 @reduction_cost_int(<8 x i32> %rdx) {
63; SSE-LABEL: 'reduction_cost_int'
64; SSE-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
65; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
66; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
67; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
68; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
69; 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
70; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
71; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
72;
73; AVX1-LABEL: 'reduction_cost_int'
74; 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 poison, i32 poison, i32 poison, i32 poison>
75; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
76; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
77; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
78; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
79; 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
80; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
81; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
82;
83; AVX2-LABEL: 'reduction_cost_int'
84; 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 poison, i32 poison, i32 poison, i32 poison>
85; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
86; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
87; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
88; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
89; 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
90; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
91; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
92;
93; SLM-LABEL: 'reduction_cost_int'
94; SLM-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
95; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
96; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
97; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
98; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
99; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
100; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx.3, i32 0
101; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
102;
103  %rdx.shuf = shufflevector <8 x i32> %rdx, <8 x i32> undef,
104  <8 x i32> <i32 4    , i32     5, i32     6, i32     7,
105  i32 undef, i32 undef, i32 undef, i32 undef>
106  %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
107  %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef,
108  <8 x i32> <i32 2    , i32 3,     i32 undef, i32 undef,
109  i32 undef, i32 undef, i32 undef, i32 undef>
110  %bin.rdx.2 = add <8 x i32> %bin.rdx, %rdx.shuf.2
111  %rdx.shuf.3 = shufflevector <8 x i32> %bin.rdx.2, <8 x i32> undef,
112  <8 x i32> <i32 1    , i32 undef, i32 undef, i32 undef,
113  i32 undef, i32 undef, i32 undef, i32 undef>
114  %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
115
116  %r = extractelement <8 x i32> %bin.rdx.3, i32 0
117  ret i32 %r
118}
119
120define fastcc float @pairwise_hadd(<4 x float> %rdx, float %f1) {
121; SSE2-LABEL: 'pairwise_hadd'
122; 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 poison, i32 poison>
123; 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 poison, i32 poison>
124; 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
125; 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 poison, i32 poison, i32 poison>
126; 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 poison, i32 poison, i32 poison>
127; 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
128; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
129; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
130; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
131;
132; SSSE3-LABEL: 'pairwise_hadd'
133; 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 poison, i32 poison>
134; 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 poison, i32 poison>
135; 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
136; 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 poison, i32 poison, i32 poison>
137; 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 poison, i32 poison, i32 poison>
138; 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
139; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
140; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
141; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
142;
143; SSE42-LABEL: 'pairwise_hadd'
144; 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 poison, i32 poison>
145; 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 poison, i32 poison>
146; 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
147; 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 poison, i32 poison, i32 poison>
148; 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 poison, i32 poison, i32 poison>
149; 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
150; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
151; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
152; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
153;
154; AVX-LABEL: 'pairwise_hadd'
155; 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 poison, i32 poison>
156; 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 poison, i32 poison>
157; 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
158; 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 poison, i32 poison, i32 poison>
159; 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 poison, i32 poison, i32 poison>
160; 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
161; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
162; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
163; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
164;
165; SLM-LABEL: 'pairwise_hadd'
166; SLM-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 poison, i32 poison>
167; SLM-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 poison, i32 poison>
168; SLM-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
169; SLM-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 poison, i32 poison, i32 poison>
170; SLM-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 poison, i32 poison, i32 poison>
171; SLM-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
172; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
173; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
174; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
175;
176  %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
177  <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
178  %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
179  <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
180  %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
181  %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
182  <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
183  %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
184  <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
185  %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
186
187  %r = extractelement <4 x float> %bin.rdx.1, i32 0
188  %r2 = fadd float %r, %f1
189  ret float %r2
190}
191
192define fastcc float @pairwise_hadd_assoc(<4 x float> %rdx, float %f1) {
193; SSE2-LABEL: 'pairwise_hadd_assoc'
194; 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 poison, i32 poison>
195; 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 poison, i32 poison>
196; 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
197; 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 poison, i32 poison, i32 poison>
198; 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 poison, i32 poison, i32 poison>
199; 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
200; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
201; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
202; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
203;
204; SSSE3-LABEL: 'pairwise_hadd_assoc'
205; 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 poison, i32 poison>
206; 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 poison, i32 poison>
207; 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
208; 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 poison, i32 poison, i32 poison>
209; 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 poison, i32 poison, i32 poison>
210; 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
211; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
212; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
213; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
214;
215; SSE42-LABEL: 'pairwise_hadd_assoc'
216; 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 poison, i32 poison>
217; 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 poison, i32 poison>
218; 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
219; 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 poison, i32 poison, i32 poison>
220; 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 poison, i32 poison, i32 poison>
221; 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
222; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
223; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
224; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
225;
226; AVX-LABEL: 'pairwise_hadd_assoc'
227; 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 poison, i32 poison>
228; 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 poison, i32 poison>
229; 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
230; 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 poison, i32 poison, i32 poison>
231; 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 poison, i32 poison, i32 poison>
232; 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
233; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
234; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
235; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
236;
237; SLM-LABEL: 'pairwise_hadd_assoc'
238; SLM-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 poison, i32 poison>
239; SLM-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 poison, i32 poison>
240; SLM-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
241; SLM-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 poison, i32 poison, i32 poison>
242; SLM-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 poison, i32 poison, i32 poison>
243; SLM-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
244; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
245; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
246; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
247;
248  %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
249  <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
250  %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
251  <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
252  %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.1, %rdx.shuf.0.0
253  %rdx.shuf.1.0 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
254  <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>
255  %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
256  <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
257  %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
258
259  %r = extractelement <4 x float> %bin.rdx.1, i32 0
260  %r2 = fadd float %r, %f1
261  ret float %r2
262}
263
264define fastcc float @pairwise_hadd_skip_first(<4 x float> %rdx, float %f1) {
265; SSE2-LABEL: 'pairwise_hadd_skip_first'
266; 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 poison, i32 poison>
267; 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 poison, i32 poison>
268; 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
269; 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 poison, i32 poison, i32 poison>
270; 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
271; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
272; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
273; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
274;
275; SSSE3-LABEL: 'pairwise_hadd_skip_first'
276; 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 poison, i32 poison>
277; 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 poison, i32 poison>
278; 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
279; 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 poison, i32 poison, i32 poison>
280; 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
281; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
282; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r2 = fadd float %r, %f1
283; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
284;
285; SSE42-LABEL: 'pairwise_hadd_skip_first'
286; 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 poison, i32 poison>
287; 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 poison, i32 poison>
288; 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
289; 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 poison, i32 poison, i32 poison>
290; 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
291; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
292; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
293; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
294;
295; AVX-LABEL: 'pairwise_hadd_skip_first'
296; 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 poison, i32 poison>
297; 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 poison, i32 poison>
298; 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
299; 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 poison, i32 poison, i32 poison>
300; 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
301; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
302; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
303; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
304;
305; SLM-LABEL: 'pairwise_hadd_skip_first'
306; SLM-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 poison, i32 poison>
307; SLM-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 poison, i32 poison>
308; SLM-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
309; SLM-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 poison, i32 poison, i32 poison>
310; SLM-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
311; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx.1, i32 0
312; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r2 = fadd float %r, %f1
313; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r2
314;
315  %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef,
316  <4 x i32> <i32 0, i32 2 , i32 undef, i32 undef>
317  %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef,
318  <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
319  %bin.rdx.0 = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
320  %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
321  <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
322  %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
323
324  %r = extractelement <4 x float> %bin.rdx.1, i32 0
325  %r2 = fadd float %r, %f1
326  ret float %r2
327}
328
329define fastcc double @no_pairwise_reduction2double(<2 x double> %rdx, double %f1) {
330; SSE2-LABEL: 'no_pairwise_reduction2double'
331; 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 poison>
332; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
333; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
334; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
335;
336; SSSE3-LABEL: 'no_pairwise_reduction2double'
337; 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 poison>
338; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
339; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
340; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
341;
342; SSE42-LABEL: 'no_pairwise_reduction2double'
343; 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 poison>
344; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
345; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
346; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
347;
348; AVX-LABEL: 'no_pairwise_reduction2double'
349; 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 poison>
350; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
351; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
352; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
353;
354; SLM-LABEL: 'no_pairwise_reduction2double'
355; SLM-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 poison>
356; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
357; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx, i32 0
358; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
359;
360  %rdx.shuf = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
361  %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
362
363  %r = extractelement <2 x double> %bin.rdx, i32 0
364  ret double %r
365}
366
367define fastcc float @no_pairwise_reduction4float(<4 x float> %rdx, float %f1) {
368; SSE2-LABEL: 'no_pairwise_reduction4float'
369; 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 poison, i32 poison>
370; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
371; 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 poison, i32 poison, i32 poison>
372; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
373; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
374; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
375;
376; SSSE3-LABEL: 'no_pairwise_reduction4float'
377; 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 poison, i32 poison>
378; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
379; 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 poison, i32 poison, i32 poison>
380; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
381; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
382; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
383;
384; SSE42-LABEL: 'no_pairwise_reduction4float'
385; 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 poison, i32 poison>
386; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
387; 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 poison, i32 poison, i32 poison>
388; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
389; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
390; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
391;
392; AVX-LABEL: 'no_pairwise_reduction4float'
393; 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 poison, i32 poison>
394; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
395; 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 poison, i32 poison, i32 poison>
396; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
397; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
398; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
399;
400; SLM-LABEL: 'no_pairwise_reduction4float'
401; SLM-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 poison, i32 poison>
402; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
403; SLM-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 poison, i32 poison, i32 poison>
404; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
405; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
406; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
407;
408  %rdx.shuf = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
409  %bin.rdx = fadd <4 x float> %rdx, %rdx.shuf
410  %rdx.shuf7 = shufflevector <4 x float> %bin.rdx, <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
411  %bin.rdx8 = fadd <4 x float> %bin.rdx, %rdx.shuf7
412
413  %r = extractelement <4 x float> %bin.rdx8, i32 0
414  ret float %r
415}
416
417define fastcc double @no_pairwise_reduction4double(<4 x double> %rdx, double %f1) {
418; SSE2-LABEL: 'no_pairwise_reduction4double'
419; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
420; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
421; SSE2-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 poison, i32 poison, i32 poison>
422; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
423; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
424; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
425;
426; SSSE3-LABEL: 'no_pairwise_reduction4double'
427; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
428; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
429; SSSE3-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 poison, i32 poison, i32 poison>
430; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
431; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
432; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
433;
434; SSE42-LABEL: 'no_pairwise_reduction4double'
435; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
436; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
437; SSE42-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 poison, i32 poison, i32 poison>
438; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
439; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
440; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
441;
442; AVX1-LABEL: 'no_pairwise_reduction4double'
443; 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 poison, i32 poison>
444; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
445; AVX1-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 poison, i32 poison, i32 poison>
446; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
447; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
448; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
449;
450; AVX2-LABEL: 'no_pairwise_reduction4double'
451; 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 poison, i32 poison>
452; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
453; 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 poison, i32 poison, i32 poison>
454; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
455; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
456; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
457;
458; SLM-LABEL: 'no_pairwise_reduction4double'
459; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
460; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
461; SLM-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 poison, i32 poison, i32 poison>
462; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
463; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
464; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
465;
466  %rdx.shuf = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
467  %bin.rdx = fadd <4 x double> %rdx, %rdx.shuf
468  %rdx.shuf7 = shufflevector <4 x double> %bin.rdx, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
469  %bin.rdx8 = fadd <4 x double> %bin.rdx, %rdx.shuf7
470
471  %r = extractelement <4 x double> %bin.rdx8, i32 0
472  ret double %r
473}
474
475define fastcc float @no_pairwise_reduction8float(<8 x float> %rdx, float %f1) {
476; SSE2-LABEL: 'no_pairwise_reduction8float'
477; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
478; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
479; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
480; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
481; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
482; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
483; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
484; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
485;
486; SSSE3-LABEL: 'no_pairwise_reduction8float'
487; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
488; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
489; SSSE3-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
490; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
491; SSSE3-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
492; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
493; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
494; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
495;
496; SSE42-LABEL: 'no_pairwise_reduction8float'
497; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
498; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
499; SSE42-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
500; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
501; SSE42-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
502; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
503; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
504; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
505;
506; AVX1-LABEL: 'no_pairwise_reduction8float'
507; 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 poison, i32 poison, i32 poison, i32 poison>
508; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
509; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
510; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
511; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
512; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
513; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
514; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
515;
516; AVX2-LABEL: 'no_pairwise_reduction8float'
517; 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 poison, i32 poison, i32 poison, i32 poison>
518; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
519; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
520; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
521; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
522; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
523; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
524; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
525;
526; SLM-LABEL: 'no_pairwise_reduction8float'
527; SLM-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
528; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
529; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
530; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
531; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
532; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
533; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx8, i32 0
534; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
535;
536  %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>
537  %bin.rdx4 = fadd <8 x float> %rdx, %rdx.shuf3
538  %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>
539  %bin.rdx = fadd <8 x float> %bin.rdx4, %rdx.shuf
540  %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>
541  %bin.rdx8 = fadd <8 x float> %bin.rdx, %rdx.shuf7
542
543  %r = extractelement <8 x float> %bin.rdx8, i32 0
544  ret float %r
545}
546
547define fastcc i64 @no_pairwise_reduction2i64(<2 x i64> %rdx, i64 %f1) {
548; CHECK-LABEL: 'no_pairwise_reduction2i64'
549; CHECK-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 poison>
550; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
551; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
552; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
553;
554; SLM-LABEL: 'no_pairwise_reduction2i64'
555; SLM-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 poison>
556; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
557; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx, i32 0
558; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
559;
560  %rdx.shuf = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
561  %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
562
563  %r = extractelement <2 x i64> %bin.rdx, i32 0
564  ret i64 %r
565}
566
567define fastcc i32 @no_pairwise_reduction4i32(<4 x i32> %rdx, i32 %f1) {
568; CHECK-LABEL: 'no_pairwise_reduction4i32'
569; CHECK-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 poison, i32 poison>
570; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
571; CHECK-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 poison, i32 poison, i32 poison>
572; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
573; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
574; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
575;
576; SLM-LABEL: 'no_pairwise_reduction4i32'
577; SLM-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 poison, i32 poison>
578; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
579; SLM-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 poison, i32 poison, i32 poison>
580; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
581; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
582; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
583;
584  %rdx.shuf = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
585  %bin.rdx = add <4 x i32> %rdx, %rdx.shuf
586  %rdx.shuf7 = shufflevector <4 x i32> %bin.rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
587  %bin.rdx8 = add <4 x i32> %bin.rdx, %rdx.shuf7
588
589  %r = extractelement <4 x i32> %bin.rdx8, i32 0
590  ret i32 %r
591}
592
593define fastcc i64 @no_pairwise_reduction4i64(<4 x i64> %rdx, i64 %f1) {
594; SSE-LABEL: 'no_pairwise_reduction4i64'
595; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
596; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
597; SSE-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 poison, i32 poison, i32 poison>
598; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
599; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
600; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
601;
602; AVX1-LABEL: 'no_pairwise_reduction4i64'
603; 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 poison, i32 poison>
604; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
605; AVX1-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 poison, i32 poison, i32 poison>
606; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
607; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
608; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
609;
610; AVX2-LABEL: 'no_pairwise_reduction4i64'
611; 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 poison, i32 poison>
612; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
613; 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 poison, i32 poison, i32 poison>
614; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
615; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
616; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
617;
618; SLM-LABEL: 'no_pairwise_reduction4i64'
619; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 poison, i32 poison>
620; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
621; SLM-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 poison, i32 poison, i32 poison>
622; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
623; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
624; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
625;
626  %rdx.shuf = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
627  %bin.rdx = add <4 x i64> %rdx, %rdx.shuf
628  %rdx.shuf7 = shufflevector <4 x i64> %bin.rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
629  %bin.rdx8 = add <4 x i64> %bin.rdx, %rdx.shuf7
630
631  %r = extractelement <4 x i64> %bin.rdx8, i32 0
632  ret i64 %r
633}
634
635define fastcc i16 @no_pairwise_reduction8i16(<8 x i16> %rdx, i16 %f1) {
636; SSE2-LABEL: 'no_pairwise_reduction8i16'
637; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 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 poison, i32 poison, i32 poison, i32 poison>
638; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
639; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %rdx.shuf = shufflevector <8 x i16> %bin.rdx4, <8 x i16> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
640; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
641; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
642; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
643; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
644; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
645;
646; SSSE3-LABEL: 'no_pairwise_reduction8i16'
647; 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 poison, i32 poison, i32 poison, i32 poison>
648; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
649; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
650; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
651; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
652; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
653; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
654; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
655;
656; SSE42-LABEL: 'no_pairwise_reduction8i16'
657; 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 poison, i32 poison, i32 poison, i32 poison>
658; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
659; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
660; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
661; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
662; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
663; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
664; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
665;
666; AVX-LABEL: 'no_pairwise_reduction8i16'
667; 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 poison, i32 poison, i32 poison, i32 poison>
668; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
669; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
670; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
671; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
672; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
673; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
674; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
675;
676; SLM-LABEL: 'no_pairwise_reduction8i16'
677; SLM-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 poison, i32 poison, i32 poison, i32 poison>
678; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
679; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
680; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
681; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
682; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
683; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx8, i32 0
684; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
685;
686  %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>
687  %bin.rdx4 = add <8 x i16> %rdx, %rdx.shuf3
688  %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>
689  %bin.rdx = add <8 x i16> %bin.rdx4, %rdx.shuf
690  %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>
691  %bin.rdx8 = add <8 x i16> %bin.rdx, %rdx.shuf7
692
693  %r = extractelement <8 x i16> %bin.rdx8, i32 0
694  ret i16 %r
695}
696
697define fastcc i32 @no_pairwise_reduction8i32(<8 x i32> %rdx, i32 %f1) {
698; SSE-LABEL: 'no_pairwise_reduction8i32'
699; SSE-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
700; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
701; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
702; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
703; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
704; SSE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
705; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
706; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
707;
708; AVX1-LABEL: 'no_pairwise_reduction8i32'
709; 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 poison, i32 poison, i32 poison, i32 poison>
710; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
711; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
712; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
713; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
714; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
715; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
716; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
717;
718; AVX2-LABEL: 'no_pairwise_reduction8i32'
719; 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 poison, i32 poison, i32 poison, i32 poison>
720; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
721; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
722; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
723; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
724; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
725; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
726; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
727;
728; SLM-LABEL: 'no_pairwise_reduction8i32'
729; SLM-NEXT:  Cost Model: Found an estimated cost of 0 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 poison, i32 poison, i32 poison, i32 poison>
730; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
731; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
732; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
733; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
734; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
735; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx8, i32 0
736; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
737;
738  %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>
739  %bin.rdx4 = add <8 x i32> %rdx, %rdx.shuf3
740  %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>
741  %bin.rdx = add <8 x i32> %bin.rdx4, %rdx.shuf
742  %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>
743  %bin.rdx8 = add <8 x i32> %bin.rdx, %rdx.shuf7
744
745  %r = extractelement <8 x i32> %bin.rdx8, i32 0
746  ret i32 %r
747}
748
749define fastcc double @pairwise_reduction2double(<2 x double> %rdx, double %f1) {
750; SSE2-LABEL: 'pairwise_reduction2double'
751; 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 poison>
752; 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 poison>
753; 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
754; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
755; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
756;
757; SSSE3-LABEL: 'pairwise_reduction2double'
758; 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 poison>
759; 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 poison>
760; 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
761; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
762; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
763;
764; SSE42-LABEL: 'pairwise_reduction2double'
765; 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 poison>
766; 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 poison>
767; 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
768; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
769; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
770;
771; AVX-LABEL: 'pairwise_reduction2double'
772; 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 poison>
773; 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 poison>
774; 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
775; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
776; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
777;
778; SLM-LABEL: 'pairwise_reduction2double'
779; SLM-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 poison>
780; SLM-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 poison>
781; SLM-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
782; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <2 x double> %bin.rdx8, i32 0
783; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
784;
785  %rdx.shuf.1.0 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
786  %rdx.shuf.1.1 = shufflevector <2 x double> %rdx, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
787  %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
788
789  %r = extractelement <2 x double> %bin.rdx8, i32 0
790  ret double %r
791}
792
793define fastcc float @pairwise_reduction4float(<4 x float> %rdx, float %f1) {
794; SSE2-LABEL: 'pairwise_reduction4float'
795; 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 poison, i32 poison>
796; 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 poison, i32 poison>
797; 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
798; 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 poison, i32 poison, i32 poison>
799; 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 poison, i32 poison, i32 poison>
800; 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
801; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
802; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
803;
804; SSSE3-LABEL: 'pairwise_reduction4float'
805; 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 poison, i32 poison>
806; 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 poison, i32 poison>
807; 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
808; 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 poison, i32 poison, i32 poison>
809; 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 poison, i32 poison, i32 poison>
810; 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
811; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
812; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
813;
814; SSE42-LABEL: 'pairwise_reduction4float'
815; 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 poison, i32 poison>
816; 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 poison, i32 poison>
817; 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
818; 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 poison, i32 poison, i32 poison>
819; 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 poison, i32 poison, i32 poison>
820; 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
821; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
822; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
823;
824; AVX-LABEL: 'pairwise_reduction4float'
825; 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 poison, i32 poison>
826; 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 poison, i32 poison>
827; 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
828; 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 poison, i32 poison, i32 poison>
829; 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 poison, i32 poison, i32 poison>
830; 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
831; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
832; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
833;
834; SLM-LABEL: 'pairwise_reduction4float'
835; SLM-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 poison, i32 poison>
836; SLM-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 poison, i32 poison>
837; SLM-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
838; SLM-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 poison, i32 poison, i32 poison>
839; SLM-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 poison, i32 poison, i32 poison>
840; SLM-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
841; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x float> %bin.rdx8, i32 0
842; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
843;
844  %rdx.shuf.0.0 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
845  %rdx.shuf.0.1 = shufflevector <4 x float> %rdx, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
846  %bin.rdx = fadd <4 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
847  %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>
848  %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>
849  %bin.rdx8 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
850
851  %r = extractelement <4 x float> %bin.rdx8, i32 0
852  ret float %r
853}
854
855define fastcc double @pairwise_reduction4double(<4 x double> %rdx, double %f1) {
856; SSE2-LABEL: 'pairwise_reduction4double'
857; SSE2-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 poison, i32 poison>
858; SSE2-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 poison, i32 poison>
859; 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
860; 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 poison, i32 poison, i32 poison>
861; SSE2-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 poison, i32 poison, i32 poison>
862; 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
863; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
864; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
865;
866; SSSE3-LABEL: 'pairwise_reduction4double'
867; SSSE3-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 poison, i32 poison>
868; SSSE3-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 poison, i32 poison>
869; 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
870; 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 poison, i32 poison, i32 poison>
871; SSSE3-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 poison, i32 poison, i32 poison>
872; 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
873; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
874; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
875;
876; SSE42-LABEL: 'pairwise_reduction4double'
877; SSE42-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 poison, i32 poison>
878; SSE42-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 poison, i32 poison>
879; 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
880; 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 poison, i32 poison, i32 poison>
881; SSE42-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 poison, i32 poison, i32 poison>
882; 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
883; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
884; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
885;
886; AVX1-LABEL: 'pairwise_reduction4double'
887; 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 poison, i32 poison>
888; 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 poison, i32 poison>
889; 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
890; 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 poison, i32 poison, i32 poison>
891; AVX1-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 poison, i32 poison, i32 poison>
892; 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
893; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
894; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
895;
896; AVX2-LABEL: 'pairwise_reduction4double'
897; 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 poison, i32 poison>
898; 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 poison, i32 poison>
899; 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
900; 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 poison, i32 poison, i32 poison>
901; 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 poison, i32 poison, i32 poison>
902; 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
903; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
904; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
905;
906; SLM-LABEL: 'pairwise_reduction4double'
907; SLM-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 poison, i32 poison>
908; SLM-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 poison, i32 poison>
909; SLM-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
910; SLM-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 poison, i32 poison, i32 poison>
911; SLM-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 poison, i32 poison, i32 poison>
912; SLM-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
913; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <4 x double> %bin.rdx8, i32 0
914; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret double %r
915;
916  %rdx.shuf.0.0 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
917  %rdx.shuf.0.1 = shufflevector <4 x double> %rdx, <4 x double> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
918  %bin.rdx = fadd <4 x double> %rdx.shuf.0.0, %rdx.shuf.0.1
919  %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>
920  %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>
921  %bin.rdx8 = fadd <4 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
922
923  %r = extractelement <4 x double> %bin.rdx8, i32 0
924  ret double %r
925}
926
927define fastcc float @pairwise_reduction8float(<8 x float> %rdx, float %f1) {
928; SSE2-LABEL: 'pairwise_reduction8float'
929; SSE2-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 poison, i32 poison, i32 poison, i32 poison>
930; SSE2-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 poison, i32 poison, i32 poison, i32 poison>
931; 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
932; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
933; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
934; 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
935; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
936; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
937; 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
938; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
939; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
940;
941; SSSE3-LABEL: 'pairwise_reduction8float'
942; SSSE3-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 poison, i32 poison, i32 poison, i32 poison>
943; SSSE3-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 poison, i32 poison, i32 poison, i32 poison>
944; 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
945; SSSE3-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
946; SSSE3-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
947; 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
948; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
949; SSSE3-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
950; 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
951; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
952; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
953;
954; SSE42-LABEL: 'pairwise_reduction8float'
955; SSE42-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 poison, i32 poison, i32 poison, i32 poison>
956; SSE42-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 poison, i32 poison, i32 poison, i32 poison>
957; 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
958; SSE42-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
959; SSE42-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
960; 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
961; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
962; SSE42-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
963; 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
964; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
965; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
966;
967; AVX1-LABEL: 'pairwise_reduction8float'
968; 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 poison, i32 poison, i32 poison, i32 poison>
969; 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 poison, i32 poison, i32 poison, i32 poison>
970; 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
971; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
972; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
973; 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
974; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
975; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
976; 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
977; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
978; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
979;
980; AVX2-LABEL: 'pairwise_reduction8float'
981; 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 poison, i32 poison, i32 poison, i32 poison>
982; 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 poison, i32 poison, i32 poison, i32 poison>
983; 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
984; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
985; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
986; 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
987; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
988; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
989; 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
990; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
991; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
992;
993; SLM-LABEL: 'pairwise_reduction8float'
994; SLM-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 poison, i32 poison, i32 poison, i32 poison>
995; SLM-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 poison, i32 poison, i32 poison, i32 poison>
996; SLM-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
997; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
998; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
999; SLM-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
1000; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1001; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1002; SLM-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
1003; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r = extractelement <8 x float> %bin.rdx9, i32 0
1004; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret float %r
1005;
1006  %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>
1007  %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>
1008  %bin.rdx = fadd <8 x float> %rdx.shuf.0.0, %rdx.shuf.0.1
1009  %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>
1010  %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>
1011  %bin.rdx8 = fadd <8 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
1012  %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>
1013  %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>
1014  %bin.rdx9 = fadd <8 x float> %rdx.shuf.2.0, %rdx.shuf.2.1
1015
1016  %r = extractelement <8 x float> %bin.rdx9, i32 0
1017  ret float %r
1018}
1019
1020define fastcc i64 @pairwise_reduction2i64(<2 x i64> %rdx, i64 %f1) {
1021; CHECK-LABEL: 'pairwise_reduction2i64'
1022; CHECK-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 poison>
1023; CHECK-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 poison>
1024; CHECK-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
1025; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
1026; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1027;
1028; SLM-LABEL: 'pairwise_reduction2i64'
1029; SLM-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 poison>
1030; SLM-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 poison>
1031; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1032; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <2 x i64> %bin.rdx8, i32 0
1033; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1034;
1035  %rdx.shuf.1.0 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 0, i32 undef>
1036  %rdx.shuf.1.1 = shufflevector <2 x i64> %rdx, <2 x i64> undef, <2 x i32> <i32 1, i32 undef>
1037  %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1038
1039  %r = extractelement <2 x i64> %bin.rdx8, i32 0
1040  ret i64 %r
1041}
1042
1043define fastcc i32 @pairwise_reduction4i32(<4 x i32> %rdx, i32 %f1) {
1044; CHECK-LABEL: 'pairwise_reduction4i32'
1045; CHECK-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 poison, i32 poison>
1046; CHECK-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 poison, i32 poison>
1047; CHECK-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
1048; CHECK-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 poison, i32 poison, i32 poison>
1049; CHECK-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 poison, i32 poison, i32 poison>
1050; CHECK-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
1051; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
1052; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1053;
1054; SLM-LABEL: 'pairwise_reduction4i32'
1055; SLM-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 poison, i32 poison>
1056; SLM-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 poison, i32 poison>
1057; SLM-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
1058; SLM-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 poison, i32 poison, i32 poison>
1059; SLM-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 poison, i32 poison, i32 poison>
1060; SLM-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
1061; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i32> %bin.rdx8, i32 0
1062; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1063;
1064  %rdx.shuf.0.0 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
1065  %rdx.shuf.0.1 = shufflevector <4 x i32> %rdx, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
1066  %bin.rdx = add <4 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1067  %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>
1068  %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>
1069  %bin.rdx8 = add <4 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1070
1071  %r = extractelement <4 x i32> %bin.rdx8, i32 0
1072  ret i32 %r
1073}
1074
1075define fastcc i64 @pairwise_reduction4i64(<4 x i64> %rdx, i64 %f1) {
1076; SSE-LABEL: 'pairwise_reduction4i64'
1077; SSE-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 poison, i32 poison>
1078; SSE-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 poison, i32 poison>
1079; 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
1080; 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 poison, i32 poison, i32 poison>
1081; SSE-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 poison, i32 poison, i32 poison>
1082; 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
1083; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1084; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1085;
1086; AVX1-LABEL: 'pairwise_reduction4i64'
1087; 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 poison, i32 poison>
1088; 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 poison, i32 poison>
1089; 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
1090; 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 poison, i32 poison, i32 poison>
1091; AVX1-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 poison, i32 poison, i32 poison>
1092; 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
1093; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1094; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1095;
1096; AVX2-LABEL: 'pairwise_reduction4i64'
1097; 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 poison, i32 poison>
1098; 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 poison, i32 poison>
1099; 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
1100; 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 poison, i32 poison, i32 poison>
1101; 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 poison, i32 poison, i32 poison>
1102; 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
1103; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1104; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1105;
1106; SLM-LABEL: 'pairwise_reduction4i64'
1107; SLM-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 poison, i32 poison>
1108; SLM-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 poison, i32 poison>
1109; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
1110; SLM-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 poison, i32 poison, i32 poison>
1111; SLM-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 poison, i32 poison, i32 poison>
1112; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1113; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <4 x i64> %bin.rdx8, i32 0
1114; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %r
1115;
1116  %rdx.shuf.0.0 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
1117  %rdx.shuf.0.1 = shufflevector <4 x i64> %rdx, <4 x i64> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
1118  %bin.rdx = add <4 x i64> %rdx.shuf.0.0, %rdx.shuf.0.1
1119  %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>
1120  %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>
1121  %bin.rdx8 = add <4 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1122
1123  %r = extractelement <4 x i64> %bin.rdx8, i32 0
1124  ret i64 %r
1125}
1126
1127define fastcc i16 @pairwise_reduction8i16(<8 x i16> %rdx, i16 %f1) {
1128; SSE2-LABEL: 'pairwise_reduction8i16'
1129; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 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 poison, i32 poison, i32 poison, i32 poison>
1130; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 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 poison, i32 poison, i32 poison, i32 poison>
1131; 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
1132; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %rdx.shuf.1.0 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1133; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %rdx.shuf.1.1 = shufflevector <8 x i16> %bin.rdx, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1134; 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
1135; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1136; SSE2-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1137; 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
1138; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1139; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1140;
1141; SSSE3-LABEL: 'pairwise_reduction8i16'
1142; 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 poison, i32 poison, i32 poison, i32 poison>
1143; 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 poison, i32 poison, i32 poison, i32 poison>
1144; 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
1145; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1146; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1147; 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
1148; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1149; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1150; 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
1151; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1152; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1153;
1154; SSE42-LABEL: 'pairwise_reduction8i16'
1155; 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 poison, i32 poison, i32 poison, i32 poison>
1156; 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 poison, i32 poison, i32 poison, i32 poison>
1157; 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
1158; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1159; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1160; 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
1161; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1162; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1163; 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
1164; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1165; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1166;
1167; AVX-LABEL: 'pairwise_reduction8i16'
1168; 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 poison, i32 poison, i32 poison, i32 poison>
1169; 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 poison, i32 poison, i32 poison, i32 poison>
1170; 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
1171; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1172; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1173; 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
1174; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1175; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1176; 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
1177; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1178; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1179;
1180; SLM-LABEL: 'pairwise_reduction8i16'
1181; SLM-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 poison, i32 poison, i32 poison, i32 poison>
1182; SLM-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 poison, i32 poison, i32 poison, i32 poison>
1183; SLM-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
1184; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1185; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1186; SLM-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
1187; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1188; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1189; SLM-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
1190; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i16> %bin.rdx9, i32 0
1191; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %r
1192;
1193  %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>
1194  %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>
1195  %bin.rdx = add <8 x i16> %rdx.shuf.0.0, %rdx.shuf.0.1
1196  %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>
1197  %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>
1198  %bin.rdx8 = add <8 x i16> %rdx.shuf.1.0, %rdx.shuf.1.1
1199  %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>
1200  %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>
1201  %bin.rdx9 = add <8 x i16> %rdx.shuf.2.0, %rdx.shuf.2.1
1202
1203  %r = extractelement <8 x i16> %bin.rdx9, i32 0
1204  ret i16 %r
1205}
1206
1207define fastcc i32 @pairwise_reduction8i32(<8 x i32> %rdx, i32 %f1) {
1208; SSE-LABEL: 'pairwise_reduction8i32'
1209; SSE-NEXT:  Cost Model: Found an estimated cost of 2 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 poison, i32 poison, i32 poison, i32 poison>
1210; SSE-NEXT:  Cost Model: Found an estimated cost of 2 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 poison, i32 poison, i32 poison, i32 poison>
1211; SSE-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
1212; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1213; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1214; SSE-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
1215; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1216; SSE-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1217; SSE-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
1218; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1219; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1220;
1221; AVX1-LABEL: 'pairwise_reduction8i32'
1222; 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 poison, i32 poison, i32 poison, i32 poison>
1223; 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 poison, i32 poison, i32 poison, i32 poison>
1224; 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
1225; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1226; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1227; 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
1228; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1229; AVX1-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1230; 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
1231; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1232; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1233;
1234; AVX2-LABEL: 'pairwise_reduction8i32'
1235; 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 poison, i32 poison, i32 poison, i32 poison>
1236; 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 poison, i32 poison, i32 poison, i32 poison>
1237; 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
1238; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1239; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1240; 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
1241; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1242; 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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1243; 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
1244; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1245; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1246;
1247; SLM-LABEL: 'pairwise_reduction8i32'
1248; SLM-NEXT:  Cost Model: Found an estimated cost of 2 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 poison, i32 poison, i32 poison, i32 poison>
1249; SLM-NEXT:  Cost Model: Found an estimated cost of 2 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 poison, i32 poison, i32 poison, i32 poison>
1250; SLM-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
1251; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1252; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1253; SLM-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
1254; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1255; SLM-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 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
1256; SLM-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
1257; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r = extractelement <8 x i32> %bin.rdx9, i32 0
1258; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %r
1259;
1260  %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>
1261  %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>
1262  %bin.rdx = add <8 x i32> %rdx.shuf.0.0, %rdx.shuf.0.1
1263  %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>
1264  %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>
1265  %bin.rdx8 = add <8 x i32> %rdx.shuf.1.0, %rdx.shuf.1.1
1266  %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>
1267  %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>
1268  %bin.rdx9 = add <8 x i32> %rdx.shuf.2.0, %rdx.shuf.2.1
1269
1270  %r = extractelement <8 x i32> %bin.rdx9, i32 0
1271  ret i32 %r
1272}
1273