1; RUN: opt -slp-vectorizer < %s -S | FileCheck %s
2
3; Verify that the SLP vectorizer is able to figure out that commutativity
4; offers the possibility to splat/broadcast %c and thus make it profitable
5; to vectorize this case
6
7
8; ModuleID = 'bugpoint-reduced-simplified.bc'
9target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
10target triple = "x86_64-apple-macosx10.11.0"
11
12@cle = external unnamed_addr global [32 x i8], align 16
13@cle32 = external unnamed_addr global [32 x i32], align 16
14
15
16; Check that we correctly detect a splat/broadcast by leveraging the
17; commutativity property of `xor`.
18
19; CHECK-LABEL:  @splat
20; CHECK:  store <16 x i8>
21define void @splat(i8 %a, i8 %b, i8 %c) {
22  %1 = xor i8 %c, %a
23  store i8 %1, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 0), align 16
24  %2 = xor i8 %a, %c
25  store i8 %2, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 1)
26  %3 = xor i8 %a, %c
27  store i8 %3, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 2)
28  %4 = xor i8 %a, %c
29  store i8 %4, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 3)
30  %5 = xor i8 %c, %a
31  store i8 %5, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 4)
32  %6 = xor i8 %c, %b
33  store i8 %6, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 5)
34  %7 = xor i8 %c, %a
35  store i8 %7, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 6)
36  %8 = xor i8 %c, %b
37  store i8 %8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 7)
38  %9 = xor i8 %a, %c
39  store i8 %9, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 8)
40  %10 = xor i8 %a, %c
41  store i8 %10, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 9)
42  %11 = xor i8 %a, %c
43  store i8 %11, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 10)
44  %12 = xor i8 %a, %c
45  store i8 %12, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 11)
46  %13 = xor i8 %a, %c
47  store i8 %13, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 12)
48  %14 = xor i8 %a, %c
49  store i8 %14, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 13)
50  %15 = xor i8 %a, %c
51  store i8 %15, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 14)
52  %16 = xor i8 %a, %c
53  store i8 %16, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 15)
54  ret void
55}
56
57
58
59; Check that we correctly detect that we can have the same opcode on one side by
60; leveraging the commutativity property of `xor`.
61
62; CHECK-LABEL:  @same_opcode_on_one_side
63; CHECK:  store <4 x i32>
64define void @same_opcode_on_one_side(i32 %a, i32 %b, i32 %c) {
65  %add1 = add i32 %c, %a
66  %add2 = add i32 %c, %a
67  %add3 = add i32 %a, %c
68  %add4 = add i32 %c, %a
69  %1 = xor i32 %add1, %a
70  store i32 %1, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 0), align 16
71  %2 = xor i32 %b, %add2
72  store i32 %2, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 1)
73  %3 = xor i32 %c, %add3
74  store i32 %3, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 2)
75  %4 = xor i32 %a, %add4
76  store i32 %4, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 3)
77  ret void
78}
79