1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
|
; RUN: llc < %s -mtriple=i386-linux-gnu -mattr=+avx512fp16 -mattr=+avx512vl -o - | FileCheck %s
; This test checks that only a single je gets generated in the final code
; for lowering the CMOV pseudos that get created for this IR.
; CHECK-LABEL: foo1:
; CHECK: je
; CHECK-NOT: je
define <8 x half> @foo1(i32 %v1, <8 x half> %v2, <8 x half> %v3, <8 x half> %v4) nounwind {
entry:
%cmp = icmp eq i32 %v1, 0
%t1 = select i1 %cmp, <8 x half> %v2, <8 x half> %v3
%t2 = select i1 %cmp, <8 x half> %v3, <8 x half> %v4
%sub = fsub <8 x half> %t1, %t2
ret <8 x half> %sub
}
; This test checks that only a single ja gets generated in the final code
; for lowering the CMOV pseudos that get created for this IR. This combines
; all the supported types together into one long string of selects based
; on the same condition.
; CHECK-LABEL: foo2:
; CHECK: ja
; CHECK-NOT: ja
define void @foo2(i32 %v1,
half %v32, half %v33,
<8 x half> %v52, <8 x half> %v53,
<16 x half> %v122, <16 x half> %v123,
<32 x half> %v132, <32 x half> %v133,
i8 * %dst) nounwind {
entry:
%add.ptr31 = getelementptr inbounds i8, i8* %dst, i32 2
%a31 = bitcast i8* %add.ptr31 to half*
%add.ptr51 = getelementptr inbounds i8, i8* %dst, i32 4
%a51 = bitcast i8* %add.ptr51 to <8 x half>*
%add.ptr121 = getelementptr inbounds i8, i8* %dst, i32 20
%a121 = bitcast i8* %add.ptr121 to <16 x half>*
%add.ptr131 = getelementptr inbounds i8, i8* %dst, i32 52
%a131 = bitcast i8* %add.ptr131 to <32 x half>*
; These operations are necessary, because select of two single use loads
; ends up getting optimized into a select of two leas, followed by a
; single load of the selected address.
%t33 = fadd half %v33, %v32
%t53 = fadd <8 x half> %v53, %v52
%t123 = fadd <16 x half> %v123, %v122
%t133 = fadd <32 x half> %v133, %v132
%cmp = icmp ugt i32 %v1, 31
%t31 = select i1 %cmp, half %v32, half %t33
%t51 = select i1 %cmp, <8 x half> %v52, <8 x half> %t53
%t121 = select i1 %cmp, <16 x half> %v122, <16 x half> %t123
%t131 = select i1 %cmp, <32 x half> %v132, <32 x half> %t133
store half %t31, half* %a31, align 2
store <8 x half> %t51, <8 x half>* %a51, align 16
store <16 x half> %t121, <16 x half>* %a121, align 32
store <32 x half> %t131, <32 x half>* %a131, align 64
ret void
}
; This test checks that only a single jne gets generated in the final code
; for lowering the CMOV pseudos that get created for this IR.
define dso_local <32 x half> @foo3(<32 x half> %a, <32 x half> %b, i1 zeroext %sign) local_unnamed_addr #0 {
; CHECK-LABEL: foo3:
; CHECK: jne
; CHECK-NOT: jne
entry:
%spec.select = select i1 %sign, <32 x half> %a, <32 x half> %b
ret <32 x half> %spec.select
}
; This test checks that only a single jne gets generated in the final code
; for lowering the CMOV pseudos that get created for this IR.
define dso_local <16 x half> @foo4(<16 x half> %a, <16 x half> %b, i1 zeroext %sign) local_unnamed_addr #0 {
; CHECK-LABEL: foo4:
; CHECK: jne
; CHECK-NOT: jne
entry:
%spec.select = select i1 %sign, <16 x half> %a, <16 x half> %b
ret <16 x half> %spec.select
}
; This test checks that only a single jne gets generated in the final code
; for lowering the CMOV pseudos that get created for this IR.
define dso_local <8 x half> @foo5(<8 x half> %a, <8 x half> %b, i1 zeroext %sign) local_unnamed_addr #0 {
; CHECK-LABEL: foo5:
; CHECK: jne
; CHECK-NOT: jne
entry:
%spec.select = select i1 %sign, <8 x half> %a, <8 x half> %b
ret <8 x half> %spec.select
}
|
