Lines Matching full:bin

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
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
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
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
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
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
58   %r = extractelement <4 x float> %bin.rdx8, i32 0
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
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
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
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
106   %bin.rdx = add <8 x i32> %rdx, %rdx.shuf
107   %rdx.shuf.2 = shufflevector <8 x i32> %bin.rdx, <8 x 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,
114   %bin.rdx.3 = add <8 x i32> %bin.rdx.2, %rdx.shuf.3
116   %r = extractelement <8 x i32> %bin.rdx.3, i32 0
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
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
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
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
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
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,
183   %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
185   %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
187   %r = extractelement <4 x float> %bin.rdx.1, i32 0
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
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
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
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
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
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,
255   %rdx.shuf.1.1 = shufflevector <4 x float> %bin.rdx.0, <4 x float> undef,
257   %bin.rdx.1 = fadd <4 x float> %rdx.shuf.1.0, %rdx.shuf.1.1
259   %r = extractelement <4 x float> %bin.rdx.1, i32 0
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
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
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
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
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
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,
322   %bin.rdx.1 = fadd <4 x float> %bin.rdx.0, %rdx.shuf.1.1
324   %r = extractelement <4 x float> %bin.rdx.1, i32 0
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
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
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
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
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
361   %bin.rdx = fadd <2 x double> %rdx, %rdx.shuf
363   %r = extractelement <2 x double> %bin.rdx, i32 0
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
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
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
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
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
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
413   %r = extractelement <4 x float> %bin.rdx8, i32 0
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
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
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
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
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
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
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
471   %r = extractelement <4 x double> %bin.rdx8, i32 0
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
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
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
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
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
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
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
543   %r = extractelement <8 x float> %bin.rdx8, i32 0
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
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
561   %bin.rdx = add <2 x i64> %rdx, %rdx.shuf
563   %r = extractelement <2 x i64> %bin.rdx, i32 0
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
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
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
589   %r = extractelement <4 x i32> %bin.rdx8, i32 0
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
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
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
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
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
631   %r = extractelement <4 x i64> %bin.rdx8, i32 0
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
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
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
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
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
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
693   %r = extractelement <8 x i16> %bin.rdx8, i32 0
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
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
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
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
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
745   %r = extractelement <8 x i32> %bin.rdx8, i32 0
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
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
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
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
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
787   %bin.rdx8 = fadd <2 x double> %rdx.shuf.1.0, %rdx.shuf.1.1
789   %r = extractelement <2 x double> %bin.rdx8, i32 0
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
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
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
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
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
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
851   %r = extractelement <4 x float> %bin.rdx8, i32 0
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
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
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
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
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
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
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
923   %r = extractelement <4 x double> %bin.rdx8, i32 0
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
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
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
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
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
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
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
1016   %r = extractelement <8 x float> %bin.rdx9, i32 0
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
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
1037   %bin.rdx8 = add <2 x i64> %rdx.shuf.1.0, %rdx.shuf.1.1
1039   %r = extractelement <2 x i64> %bin.rdx8, i32 0
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
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
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
1071   %r = extractelement <4 x i32> %bin.rdx8, i32 0
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
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
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
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
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
1123   %r = extractelement <4 x i64> %bin.rdx8, i32 0
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
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
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
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
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
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
1203   %r = extractelement <8 x i16> %bin.rdx9, i32 0
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
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
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
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
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
1270   %r = extractelement <8 x i32> %bin.rdx9, i32 0