xref: /llvm-project/llvm/test/Analysis/CostModel/ARM/control-flow.ll (revision 68c50b111d74afb9489cf97770fa917d0a1c7f77) 
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 -cost-kind=code-size  -mtriple=thumbv8m.base   | FileCheck %s --check-prefix=CHECK-T1-SIZE
3; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size  -mtriple=thumbv8m.main   | FileCheck %s --check-prefix=CHECK-V8M-SIZE
4; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size  -mtriple=thumbv8.1m.main -mattr=+mve  | FileCheck %s --check-prefix=CHECK-MVE-SIZE
5; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size  -mtriple=armv8a -mattr=+neon  | FileCheck %s --check-prefix=CHECK-NEON-SIZE
6; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency    -mtriple=thumbv8m.base   | FileCheck %s --check-prefix=CHECK-T1-LATENCY
7; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency    -mtriple=thumbv8m.main   | FileCheck %s --check-prefix=CHECK-V8M-LATENCY
8; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency    -mtriple=thumbv8.1m.main | FileCheck %s --check-prefix=CHECK-V8_1M-LATENCY
9; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=throughput -mtriple=thumbv8m.base   | FileCheck %s --check-prefix=CHECK-T1-THROUGHPUT
10; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=throughput -mtriple=thumbv8m.main   | FileCheck %s --check-prefix=CHECK-V8M-THROUGHPUT
11; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=throughput -mtriple=thumbv8.1m.main | FileCheck %s --check-prefix=CHECK-V8_1M-THROUGHPUT
12; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=throughput -mtriple=thumbv8.1m.main -mattr=+mve | FileCheck %s --check-prefix=CHECK-MVE-THROUGHPUT
13; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=throughput -mtriple=armv8a -mattr=+neon | FileCheck %s --check-prefix=CHECK-NEON-THROUGHPUT
14
15target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
16
17define i32 @simple_loop_cost(i32 %N) {
18; CHECK-T1-SIZE-LABEL: 'simple_loop_cost'
19; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
20; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
21; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
22; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
23; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
24; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
25; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
26; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
27; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
28;
29; CHECK-V8M-SIZE-LABEL: 'simple_loop_cost'
30; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
31; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
32; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
33; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
34; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
35; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
36; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
37; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
38; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
39;
40; CHECK-MVE-SIZE-LABEL: 'simple_loop_cost'
41; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
42; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
43; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
44; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
45; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
46; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
47; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
48; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
49; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
50;
51; CHECK-NEON-SIZE-LABEL: 'simple_loop_cost'
52; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
53; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
54; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
55; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
56; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
57; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
58; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
59; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
60; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
61;
62; CHECK-T1-LATENCY-LABEL: 'simple_loop_cost'
63; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
64; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
65; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
66; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
67; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
68; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
69; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
70; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
71; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
72;
73; CHECK-V8M-LATENCY-LABEL: 'simple_loop_cost'
74; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
75; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
76; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
77; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
78; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
79; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
80; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
81; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
82; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
83;
84; CHECK-V8_1M-LATENCY-LABEL: 'simple_loop_cost'
85; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
86; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
87; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
88; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
89; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
90; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
91; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
92; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
93; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
94;
95; CHECK-T1-THROUGHPUT-LABEL: 'simple_loop_cost'
96; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
97; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
98; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
99; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
100; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
101; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
102; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
103; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
104; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
105;
106; CHECK-V8M-THROUGHPUT-LABEL: 'simple_loop_cost'
107; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
108; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
109; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
110; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
111; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
112; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
113; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
114; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
115; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
116;
117; CHECK-V8_1M-THROUGHPUT-LABEL: 'simple_loop_cost'
118; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
119; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
120; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
121; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
122; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
123; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
124; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
125; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
126; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
127;
128; CHECK-MVE-THROUGHPUT-LABEL: 'simple_loop_cost'
129; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
130; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
131; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
132; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
133; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
134; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
135; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
136; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
137; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
138;
139; CHECK-NEON-THROUGHPUT-LABEL: 'simple_loop_cost'
140; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
141; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
142; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
143; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
144; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
145; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
146; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
147; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
148; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
149;
150entry:
151  %zero = icmp eq i32 %N, 0
152  br i1 %zero, label %exit, label %preheader
153
154preheader:
155  br label %loop
156
157loop:
158  %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
159  %iv.next = add nuw i32 %iv, 1
160  %cmp = icmp ne i32 %iv.next, %N
161  br i1 %cmp, label %loop, label %exit
162
163exit:
164  %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
165  ret i32 %res
166}
167
168define i32 @simple_mul_loop(ptr %A, ptr %B, i32 %N) {
169; CHECK-T1-SIZE-LABEL: 'simple_mul_loop'
170; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
171; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
172; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
173; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
174; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
175; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
176; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
177; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
178; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
179; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
180; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
181; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
182; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
183; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
184;
185; CHECK-V8M-SIZE-LABEL: 'simple_mul_loop'
186; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
187; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
188; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
189; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
190; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
191; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
192; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
193; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
194; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
195; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
196; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
197; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
198; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
199; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
200;
201; CHECK-MVE-SIZE-LABEL: 'simple_mul_loop'
202; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
203; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
204; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
205; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
206; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
207; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
208; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
209; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
210; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
211; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
212; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
213; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
214; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
215; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
216;
217; CHECK-NEON-SIZE-LABEL: 'simple_mul_loop'
218; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
219; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
220; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
221; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
222; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
223; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
224; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
225; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
226; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
227; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
228; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
229; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
230; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
231; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
232;
233; CHECK-T1-LATENCY-LABEL: 'simple_mul_loop'
234; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
235; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
236; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
237; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
238; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
239; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i32, ptr %addr.a, align 4
240; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
241; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
242; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
243; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
244; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
245; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
246; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
247; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
248;
249; CHECK-V8M-LATENCY-LABEL: 'simple_mul_loop'
250; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
251; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
252; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
253; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
254; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
255; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i32, ptr %addr.a, align 4
256; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
257; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
258; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
259; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
260; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
261; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
262; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
263; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
264;
265; CHECK-V8_1M-LATENCY-LABEL: 'simple_mul_loop'
266; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
267; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
268; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
269; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
270; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
271; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i32, ptr %addr.a, align 4
272; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
273; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
274; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
275; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
276; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
277; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
278; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
279; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
280;
281; CHECK-T1-THROUGHPUT-LABEL: 'simple_mul_loop'
282; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
283; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
284; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
285; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
286; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
287; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
288; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
289; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
290; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
291; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
292; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
293; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
294; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
295; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
296;
297; CHECK-V8M-THROUGHPUT-LABEL: 'simple_mul_loop'
298; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
299; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
300; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
301; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
302; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
303; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
304; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
305; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
306; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
307; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
308; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
309; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
310; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
311; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
312;
313; CHECK-V8_1M-THROUGHPUT-LABEL: 'simple_mul_loop'
314; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
315; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
316; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
317; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
318; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
319; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
320; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
321; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
322; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
323; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
324; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
325; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
326; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
327; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
328;
329; CHECK-MVE-THROUGHPUT-LABEL: 'simple_mul_loop'
330; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
331; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
332; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
333; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
334; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
335; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
336; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
337; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
338; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
339; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
340; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
341; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
342; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
343; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
344;
345; CHECK-NEON-THROUGHPUT-LABEL: 'simple_mul_loop'
346; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
347; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
348; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
349; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
350; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = getelementptr i32, ptr %A, i32 %iv
351; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i32, ptr %addr.a, align 4
352; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %load, %load
353; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = getelementptr i32, ptr %B, i32 %iv
354; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
355; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
356; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
357; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
358; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
359; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
360;
361entry:
362  %zero = icmp eq i32 %N, 0
363  br i1 %zero, label %exit, label %preheader
364
365preheader:
366  br label %loop
367
368loop:
369  %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
370  %addr.a = getelementptr i32, ptr %A, i32 %iv
371  %load = load i32, ptr %addr.a
372  %mul = mul i32 %load, %load
373  %addr.b = getelementptr i32, ptr %B, i32 %iv
374  store i32 %mul, ptr %addr.b
375  %iv.next = add nuw i32 %iv, 1
376  %cmp = icmp ne i32 %iv.next, %N
377  br i1 %cmp, label %loop, label %exit
378
379exit:
380  %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
381  ret i32 %res
382}
383
384define i32 @simple_mul_ext_lsr_loop(ptr %A, ptr %B, i32 %N) {
385; CHECK-T1-SIZE-LABEL: 'simple_mul_ext_lsr_loop'
386; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
387; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
388; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
389; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
390; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
391; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
392; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
393; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
394; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
395; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
396; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
397; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
398; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
399; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
400; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
401; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
402; CHECK-T1-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
403;
404; CHECK-V8M-SIZE-LABEL: 'simple_mul_ext_lsr_loop'
405; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
406; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
407; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
408; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
409; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
410; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
411; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
412; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
413; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
414; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
415; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
416; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
417; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
418; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
419; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
420; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
421; CHECK-V8M-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
422;
423; CHECK-MVE-SIZE-LABEL: 'simple_mul_ext_lsr_loop'
424; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
425; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
426; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
427; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
428; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
429; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
430; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
431; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
432; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
433; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
434; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
435; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
436; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
437; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
438; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
439; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
440; CHECK-MVE-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
441;
442; CHECK-NEON-SIZE-LABEL: 'simple_mul_ext_lsr_loop'
443; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
444; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
445; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
446; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
447; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
448; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
449; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
450; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
451; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
452; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
453; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
454; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
455; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
456; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
457; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
458; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
459; CHECK-NEON-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
460;
461; CHECK-T1-LATENCY-LABEL: 'simple_mul_ext_lsr_loop'
462; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
463; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
464; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
465; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
466; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
467; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
468; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i16, ptr %addr.a, align 2
469; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
470; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
471; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
472; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
473; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
474; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
475; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
476; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
477; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
478; CHECK-T1-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
479;
480; CHECK-V8M-LATENCY-LABEL: 'simple_mul_ext_lsr_loop'
481; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
482; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
483; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
484; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
485; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
486; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
487; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i16, ptr %addr.a, align 2
488; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
489; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
490; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
491; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
492; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
493; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
494; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
495; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
496; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
497; CHECK-V8M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
498;
499; CHECK-V8_1M-LATENCY-LABEL: 'simple_mul_ext_lsr_loop'
500; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
501; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
502; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
503; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
504; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
505; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
506; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %load = load i16, ptr %addr.a, align 2
507; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
508; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
509; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
510; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
511; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
512; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
513; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
514; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
515; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
516; CHECK-V8_1M-LATENCY-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
517;
518; CHECK-T1-THROUGHPUT-LABEL: 'simple_mul_ext_lsr_loop'
519; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
520; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
521; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
522; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
523; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
524; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
525; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
526; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
527; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
528; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
529; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
530; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
531; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
532; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
533; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
534; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
535; CHECK-T1-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
536;
537; CHECK-V8M-THROUGHPUT-LABEL: 'simple_mul_ext_lsr_loop'
538; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
539; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
540; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
541; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
542; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
543; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
544; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
545; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
546; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
547; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
548; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
549; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
550; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
551; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
552; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
553; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
554; CHECK-V8M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
555;
556; CHECK-V8_1M-THROUGHPUT-LABEL: 'simple_mul_ext_lsr_loop'
557; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
558; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %zero, label %exit, label %preheader
559; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br label %loop
560; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
561; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
562; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
563; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
564; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
565; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
566; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
567; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
568; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
569; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
570; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
571; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: br i1 %cmp, label %loop, label %exit
572; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
573; CHECK-V8_1M-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %res
574;
575; CHECK-MVE-THROUGHPUT-LABEL: 'simple_mul_ext_lsr_loop'
576; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
577; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
578; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
579; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
580; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
581; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
582; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
583; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
584; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
585; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
586; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
587; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
588; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
589; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
590; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
591; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
592; CHECK-MVE-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
593;
594; CHECK-NEON-THROUGHPUT-LABEL: 'simple_mul_ext_lsr_loop'
595; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %zero = icmp eq i32 %N, 0
596; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %zero, label %exit, label %preheader
597; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br label %loop
598; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
599; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
600; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
601; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load i16, ptr %addr.a, align 2
602; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %sext = sext i16 %load to i32
603; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %mul = mul i32 %sext, 7
604; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: store i32 %mul, ptr %addr.b, align 4
605; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %iv.next = add nuw i32 %iv, 1
606; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
607; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
608; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cmp = icmp ne i32 %iv.next, %N
609; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: br i1 %cmp, label %loop, label %exit
610; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
611; CHECK-NEON-THROUGHPUT-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %res
612;
613entry:
614  %zero = icmp eq i32 %N, 0
615  br i1 %zero, label %exit, label %preheader
616
617preheader:
618  br label %loop
619
620loop:
621  %iv = phi i32 [ 0, %preheader ], [ %iv.next, %loop ]
622  %addr.a = phi ptr [ %A, %preheader ], [ %addr.a, %loop ]
623  %addr.b = phi ptr [ %B, %preheader ], [ %addr.b, %loop ]
624  %load = load i16, ptr %addr.a
625  %sext = sext i16 %load to i32
626  %mul = mul i32 %sext, 7
627  store i32 %mul, ptr %addr.b
628  %iv.next = add nuw i32 %iv, 1
629  %addr.a.next = getelementptr i16, ptr %addr.a, i32 1
630  %addr.b.next = getelementptr i32, ptr %addr.b, i32 1
631  %cmp = icmp ne i32 %iv.next, %N
632  br i1 %cmp, label %loop, label %exit
633
634exit:
635  %res = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
636  ret i32 %res
637}
638