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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx512fp16 < %s | FileCheck %s
; Verify that we're folding the load into the math instruction.
; This pattern is generated out of the simplest intrinsics usage:
; _mm_add_ss(a, _mm_load_ss(b));
define <8 x half> @addsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: addsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vaddsh (%rdi), %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = fadd half %a, %b
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
define <8 x half> @subsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: subsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vsubsh (%rdi), %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = fsub half %a, %b
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
define <8 x half> @mulsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: mulsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vmulsh (%rdi), %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = fmul half %a, %b
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
define <8 x half> @divsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: divsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vdivsh (%rdi), %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = fdiv half %a, %b
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
define <8 x half> @minsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: minsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vminsh (%rdi), %xmm0, %xmm1
; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = call nnan half @llvm.minnum.f16(half %a, half %b) readnone
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
define <8 x half> @maxsh(<8 x half> %va, half* %pb) {
; CHECK-LABEL: maxsh:
; CHECK: # %bb.0:
; CHECK-NEXT: vminsh (%rdi), %xmm0, %xmm1
; CHECK-NEXT: vmovsh %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%a = extractelement <8 x half> %va, i32 0
%b = load half, half* %pb
%r = call nnan half @llvm.minnum.f16(half %a, half %b) readnone
%vr = insertelement <8 x half> %va, half %r, i32 0
ret <8 x half> %vr
}
declare half @llvm.minnum.f16(half, half)
declare half @llvm.maxnum.f16(half, half)
|
