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
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
; RUN: opt -passes=slp-vectorizer -mtriple=arm64-apple-macosx -S %s | FileCheck %s
declare float @llvm.maximumnum.f32(float, float)
declare float @llvm.minimumnum.f32(float, float)
; TODO: Need reduction version of maximumnum/minimumnum.
define float @reduction_v4f32_maximumnum(ptr %p) {
; CHECK-LABEL: define float @reduction_v4f32_maximumnum
; CHECK-SAME: (ptr [[P:%.*]]) {
; CHECK-NEXT: [[G1:%.*]] = getelementptr inbounds float, ptr [[P]], i64 1
; CHECK-NEXT: [[G2:%.*]] = getelementptr inbounds float, ptr [[P]], i64 2
; CHECK-NEXT: [[G3:%.*]] = getelementptr inbounds float, ptr [[P]], i64 3
; CHECK-NEXT: [[T0:%.*]] = load float, ptr [[P]], align 4
; CHECK-NEXT: [[T1:%.*]] = load float, ptr [[G1]], align 4
; CHECK-NEXT: [[T2:%.*]] = load float, ptr [[G2]], align 4
; CHECK-NEXT: [[T3:%.*]] = load float, ptr [[G3]], align 4
; CHECK-NEXT: [[M1:%.*]] = tail call float @llvm.maximumnum.f32(float [[T1]], float [[T0]])
; CHECK-NEXT: [[M2:%.*]] = tail call float @llvm.maximumnum.f32(float [[T2]], float [[M1]])
; CHECK-NEXT: [[M3:%.*]] = tail call float @llvm.maximumnum.f32(float [[T3]], float [[M2]])
; CHECK-NEXT: ret float [[M3]]
;
%g1 = getelementptr inbounds float, ptr %p, i64 1
%g2 = getelementptr inbounds float, ptr %p, i64 2
%g3 = getelementptr inbounds float, ptr %p, i64 3
%t0 = load float, ptr %p, align 4
%t1 = load float, ptr %g1, align 4
%t2 = load float, ptr %g2, align 4
%t3 = load float, ptr %g3, align 4
%m1 = tail call float @llvm.maximumnum.f32(float %t1, float %t0)
%m2 = tail call float @llvm.maximumnum.f32(float %t2, float %m1)
%m3 = tail call float @llvm.maximumnum.f32(float %t3, float %m2)
ret float %m3
}
define float @reduction_v4f64_minimumnum(ptr %p) {
; CHECK-LABEL: define float @reduction_v4f64_minimumnum
; CHECK-SAME: (ptr [[P:%.*]]) {
; CHECK-NEXT: [[G1:%.*]] = getelementptr inbounds float, ptr [[P]], i64 1
; CHECK-NEXT: [[G2:%.*]] = getelementptr inbounds float, ptr [[P]], i64 2
; CHECK-NEXT: [[G3:%.*]] = getelementptr inbounds float, ptr [[P]], i64 3
; CHECK-NEXT: [[T0:%.*]] = load float, ptr [[P]], align 4
; CHECK-NEXT: [[T1:%.*]] = load float, ptr [[G1]], align 4
; CHECK-NEXT: [[T2:%.*]] = load float, ptr [[G2]], align 4
; CHECK-NEXT: [[T3:%.*]] = load float, ptr [[G3]], align 4
; CHECK-NEXT: [[M1:%.*]] = tail call float @llvm.minimumnum.f32(float [[T1]], float [[T0]])
; CHECK-NEXT: [[M2:%.*]] = tail call float @llvm.minimumnum.f32(float [[T2]], float [[M1]])
; CHECK-NEXT: [[M3:%.*]] = tail call float @llvm.minimumnum.f32(float [[T3]], float [[M2]])
; CHECK-NEXT: ret float [[M3]]
;
%g1 = getelementptr inbounds float, ptr %p, i64 1
%g2 = getelementptr inbounds float, ptr %p, i64 2
%g3 = getelementptr inbounds float, ptr %p, i64 3
%t0 = load float, ptr %p, align 4
%t1 = load float, ptr %g1, align 4
%t2 = load float, ptr %g2, align 4
%t3 = load float, ptr %g3, align 4
%m1 = tail call float @llvm.minimumnum.f32(float %t1, float %t0)
%m2 = tail call float @llvm.minimumnum.f32(float %t2, float %m1)
%m3 = tail call float @llvm.minimumnum.f32(float %t3, float %m2)
ret float %m3
}
|
