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
98
99
|
; RUN: llc -march=hexagon -hexagon-hvx-widen=32 < %s | FileCheck %s
; v32i8 -> v32i16
; CHECK-LABEL: f0:
; CHECK: r[[R0:[0-9]+]] = #64
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].h = vunpack(v[[V0]].b)
; CHECK: q[[Q0:[0-3]]] = vsetq(r[[R0]])
; CHECK: if (q[[Q0]]) vmem(r1+#0) = v[[V2]]
define void @f0(ptr %a0, ptr %a1) #0 {
%v0 = load <32 x i8>, ptr %a0, align 128
%v1 = sext <32 x i8> %v0 to <32 x i16>
store <32 x i16> %v1, ptr %a1, align 128
ret void
}
; v32i8 -> v32i32
; CHECK-LABEL: f1:
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].h = vunpack(v[[V0]].b)
; CHECK: v[[V3:[0-9]+]]:[[V4:[0-9]+]].w = vunpack(v[[V2]].h)
; CHECK: vmem(r1+#0) = v[[V4]]
define void @f1(ptr %a0, ptr %a1) #0 {
%v0 = load <32 x i8>, ptr %a0, align 128
%v1 = sext <32 x i8> %v0 to <32 x i32>
store <32 x i32> %v1, ptr %a1, align 128
ret void
}
; v64i8 -> v64i16
; CHECK-LABEL: f2:
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].h = vunpack(v[[V0]].b)
; CHECK: vmem(r1+#0) = v[[V2]]
define void @f2(ptr %a0, ptr %a1) #0 {
%v0 = load <64 x i8>, ptr %a0, align 128
%v1 = sext <64 x i8> %v0 to <64 x i16>
store <64 x i16> %v1, ptr %a1, align 128
ret void
}
; v64i8 -> v64i32
; CHECK-LABEL: f3:
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].h = vunpack(v[[V0]].b)
; CHECK: v[[V3:[0-9]+]]:[[V4:[0-9]+]].w = vunpack(v[[V2]].h)
; CHECK-DAG: vmem(r1+#0) = v[[V4]]
; CHECK-DAG: vmem(r1+#1) = v[[V3]]
define void @f3(ptr %a0, ptr %a1) #0 {
%v0 = load <64 x i8>, ptr %a0, align 128
%v1 = sext <64 x i8> %v0 to <64 x i32>
store <64 x i32> %v1, ptr %a1, align 128
ret void
}
; v16i16 -> v16i32
; CHECK-LABEL: f4:
; CHECK: r[[R0:[0-9]+]] = #64
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].w = vunpack(v[[V0]].h)
; CHECK: q[[Q0:[0-3]]] = vsetq(r[[R0]])
; CHECK: if (q[[Q0]]) vmem(r1+#0) = v[[V2]]
define void @f4(ptr %a0, ptr %a1) #0 {
%v0 = load <16 x i16>, ptr %a0, align 128
%v1 = sext <16 x i16> %v0 to <16 x i32>
store <16 x i32> %v1, ptr %a1, align 128
ret void
}
; v32i16 -> v32i32
; CHECK-LABEL: f5:
; CHECK: v[[V0:[0-9]+]] = vmem(r0+#0)
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].w = vunpack(v[[V0]].h)
; CHECK: vmem(r1+#0) = v[[V2]]
define void @f5(ptr %a0, ptr %a1) #0 {
%v0 = load <32 x i16>, ptr %a0, align 128
%v1 = sext <32 x i16> %v0 to <32 x i32>
store <32 x i32> %v1, ptr %a1, align 128
ret void
}
; v8i8 -> v8i32
; CHECK-LABEL: f6:
; CHECK: r[[R0:[0-9]+]]:[[R1:[0-9]+]] = memd(r0+#0)
; CHECK-DAG: v[[V0:[0-9]+]].w = vinsert(r[[R0]])
; CHECK-DAG: v[[V0]].w = vinsert(r[[R1]])
; CHECK-DAG: q[[Q0:[0-3]]] = vsetq
; CHECK: v[[V1:[0-9]+]]:[[V2:[0-9]+]].h = vunpack(v[[V0]].b)
; CHECK: v[[V3:[0-9]+]]:[[V4:[0-9]+]].w = vunpack(v[[V2]].h)
; CHECK: if (q[[Q0]]) vmem(r1+#0) = v[[V4]]
define void @f6(ptr %a0, ptr %a1) #0 {
%v0 = load <8 x i8>, ptr %a0, align 128
%v1 = sext <8 x i8> %v0 to <8 x i32>
store <8 x i32> %v1, ptr %a1, align 128
ret void
}
attributes #0 = { "target-cpu"="hexagonv65" "target-features"="+hvx,+hvx-length128b,-packets" }
|
