1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-- | FileCheck %s
4 ; This testcase was written to demonstrate an instruction-selection problem,
5 ; however it also happens to expose a limitation in the DAGCombiner's
6 ; expression reassociation which causes it to miss opportunities for
7 ; constant folding due to the intermediate adds having multiple uses.
8 ; The Reassociate pass has similar limitations. If these limitations are
9 ; fixed, the test commands above will need to be updated to expect fewer
12 @g0 = weak dso_local global [1000 x i32] zeroinitializer, align 32 ; <ptr> [#uses=8]
13 @g1 = weak dso_local global [1000 x i32] zeroinitializer, align 32 ; <ptr> [#uses=7]
15 define dso_local void @foo() {
17 ; CHECK: # %bb.0: # %entry
18 ; CHECK-NEXT: movl g0(%rip), %eax
19 ; CHECK-NEXT: movl g1(%rip), %ecx
20 ; CHECK-NEXT: leal (%rax,%rcx), %edx
21 ; CHECK-NEXT: leal 1(%rax,%rcx), %eax
22 ; CHECK-NEXT: movl %eax, g0+4(%rip)
23 ; CHECK-NEXT: movl g1+4(%rip), %eax
24 ; CHECK-NEXT: leal 1(%rax,%rdx), %ecx
25 ; CHECK-NEXT: leal 2(%rax,%rdx), %eax
26 ; CHECK-NEXT: movl %eax, g0+8(%rip)
27 ; CHECK-NEXT: movl g1+8(%rip), %eax
28 ; CHECK-NEXT: leal 1(%rax,%rcx), %edx
29 ; CHECK-NEXT: leal 2(%rax,%rcx), %eax
30 ; CHECK-NEXT: movl %eax, g0+12(%rip)
31 ; CHECK-NEXT: movl g1+12(%rip), %eax
32 ; CHECK-NEXT: leal 1(%rax,%rdx), %ecx
33 ; CHECK-NEXT: leal 2(%rax,%rdx), %eax
34 ; CHECK-NEXT: movl %eax, g0+16(%rip)
35 ; CHECK-NEXT: movl g1+16(%rip), %eax
36 ; CHECK-NEXT: leal 1(%rax,%rcx), %edx
37 ; CHECK-NEXT: leal 2(%rax,%rcx), %eax
38 ; CHECK-NEXT: movl %eax, g0+20(%rip)
39 ; CHECK-NEXT: movl g1+20(%rip), %eax
40 ; CHECK-NEXT: leal 1(%rax,%rdx), %ecx
41 ; CHECK-NEXT: leal 2(%rax,%rdx), %eax
42 ; CHECK-NEXT: movl %eax, g0+24(%rip)
43 ; CHECK-NEXT: movl g1+24(%rip), %eax
44 ; CHECK-NEXT: leal 2(%rax,%rcx), %eax
45 ; CHECK-NEXT: movl %eax, g0+28(%rip)
48 %tmp4 = load i32, ptr @g0 ; <i32> [#uses=1]
49 %tmp8 = load i32, ptr @g1 ; <i32> [#uses=1]
50 %tmp9 = add i32 %tmp4, 1 ; <i32> [#uses=1]
51 %tmp10 = add i32 %tmp9, %tmp8 ; <i32> [#uses=2]
52 store i32 %tmp10, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 1)
53 %tmp8.1 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 1) ; <i32> [#uses=1]
54 %tmp9.1 = add i32 %tmp10, 1 ; <i32> [#uses=1]
55 %tmp10.1 = add i32 %tmp9.1, %tmp8.1 ; <i32> [#uses=2]
56 store i32 %tmp10.1, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 2)
57 %tmp8.2 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 2) ; <i32> [#uses=1]
58 %tmp9.2 = add i32 %tmp10.1, 1 ; <i32> [#uses=1]
59 %tmp10.2 = add i32 %tmp9.2, %tmp8.2 ; <i32> [#uses=2]
60 store i32 %tmp10.2, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 3)
61 %tmp8.3 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 3) ; <i32> [#uses=1]
62 %tmp9.3 = add i32 %tmp10.2, 1 ; <i32> [#uses=1]
63 %tmp10.3 = add i32 %tmp9.3, %tmp8.3 ; <i32> [#uses=2]
64 store i32 %tmp10.3, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 4)
65 %tmp8.4 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 4) ; <i32> [#uses=1]
66 %tmp9.4 = add i32 %tmp10.3, 1 ; <i32> [#uses=1]
67 %tmp10.4 = add i32 %tmp9.4, %tmp8.4 ; <i32> [#uses=2]
68 store i32 %tmp10.4, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 5)
69 %tmp8.5 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 5) ; <i32> [#uses=1]
70 %tmp9.5 = add i32 %tmp10.4, 1 ; <i32> [#uses=1]
71 %tmp10.5 = add i32 %tmp9.5, %tmp8.5 ; <i32> [#uses=2]
72 store i32 %tmp10.5, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 6)
73 %tmp8.6 = load i32, ptr getelementptr ([1000 x i32], ptr @g1, i32 0, i32 6) ; <i32> [#uses=1]
74 %tmp9.6 = add i32 %tmp10.5, 1 ; <i32> [#uses=1]
75 %tmp10.6 = add i32 %tmp9.6, %tmp8.6 ; <i32> [#uses=1]
76 store i32 %tmp10.6, ptr getelementptr ([1000 x i32], ptr @g0, i32 0, i32 7)