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
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: opt -S -loop-reduce < %s | FileCheck %s --check-prefix=IR
; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s --check-prefix=ASM
; Note: To update this test, please run utils/update_test_checks.py and utils/update_llc_test_checks.py separately on opt/llc run line.
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
target triple = "aarch64-linux-gnu"
; These tests check that the IR coming out of LSR does not cast input/output pointer from ptr to ptr type.
; And scaled-index addressing mode is leveraged in the generated assembly, i.e. ld1h { z1.h }, p0/z, [x0, x8, lsl #1].
define void @ld_st_nxv8i16(ptr %in, ptr %out) {
; IR-LABEL: @ld_st_nxv8i16(
; IR-NEXT: entry:
; IR-NEXT: br label [[LOOP_PH:%.*]]
; IR: loop.ph:
; IR-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
; IR-NEXT: [[SCALED_VF:%.*]] = shl i64 [[VSCALE]], 3
; IR-NEXT: br label [[LOOP:%.*]]
; IR: loop:
; IR-NEXT: [[INDVAR:%.*]] = phi i64 [ 0, [[LOOP_PH]] ], [ [[INDVAR_NEXT:%.*]], [[LOOP]] ]
; IR-NEXT: [[TMP0:%.*]] = shl i64 [[INDVAR]], 1
; IR-NEXT: [[UGLYGEP1:%.*]] = getelementptr i8, ptr [[IN:%.*]], i64 [[TMP0]]
; IR-NEXT: [[TMP1:%.*]] = shl i64 [[INDVAR]], 1
; IR-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, ptr [[OUT:%.*]], i64 [[TMP1]]
; IR-NEXT: [[VAL:%.*]] = load <vscale x 8 x i16>, ptr [[UGLYGEP1]], align 16
; IR-NEXT: [[ADDP_VEC:%.*]] = add <vscale x 8 x i16> [[VAL]], splat (i16 3)
; IR-NEXT: store <vscale x 8 x i16> [[ADDP_VEC]], ptr [[UGLYGEP]], align 16
; IR-NEXT: [[INDVAR_NEXT]] = add nsw i64 [[INDVAR]], [[SCALED_VF]]
; IR-NEXT: [[EXIT_COND:%.*]] = icmp eq i64 [[INDVAR_NEXT]], 1024
; IR-NEXT: br i1 [[EXIT_COND]], label [[LOOP_EXIT:%.*]], label [[LOOP]]
; IR: loop.exit:
; IR-NEXT: br label [[EXIT:%.*]]
; IR: exit:
; IR-NEXT: ret void
;
; ASM-LABEL: ld_st_nxv8i16:
; ASM: // %bb.0: // %entry
; ASM-NEXT: ptrue p0.h
; ASM-NEXT: mov x8, xzr
; ASM-NEXT: cnth x9
; ASM-NEXT: .LBB0_1: // %loop
; ASM-NEXT: // =>This Inner Loop Header: Depth=1
; ASM-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
; ASM-NEXT: add z0.h, z0.h, #3
; ASM-NEXT: st1h { z0.h }, p0, [x1, x8, lsl #1]
; ASM-NEXT: add x8, x8, x9
; ASM-NEXT: cmp x8, #1024
; ASM-NEXT: b.ne .LBB0_1
; ASM-NEXT: // %bb.2: // %exit
; ASM-NEXT: ret
entry:
br label %loop.ph
loop.ph:
%vscale = call i64 @llvm.vscale.i64()
%scaled_vf = shl i64 %vscale, 3
br label %loop
loop: ; preds = %loop, %loop.ph
%indvar = phi i64 [ 0, %loop.ph ], [ %indvar.next, %loop ]
%ptr.in = getelementptr inbounds i16, ptr %in, i64 %indvar
%ptr.out = getelementptr inbounds i16, ptr %out, i64 %indvar
%val = load <vscale x 8 x i16>, ptr %ptr.in, align 16
%addp_vec = add <vscale x 8 x i16> %val, splat (i16 3)
store <vscale x 8 x i16> %addp_vec, ptr %ptr.out, align 16
%indvar.next = add nsw i64 %indvar, %scaled_vf
%exit.cond = icmp eq i64 %indvar.next, 1024
br i1 %exit.cond, label %loop.exit, label %loop
loop.exit: ; preds = %loop
br label %exit
exit:
ret void
}
define void @masked_ld_st_nxv8i16(ptr %in, ptr %out, i64 %n) {
; IR-LABEL: @masked_ld_st_nxv8i16(
; IR-NEXT: entry:
; IR-NEXT: br label [[LOOP_PH:%.*]]
; IR: loop.ph:
; IR-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
; IR-NEXT: [[SCALED_VF:%.*]] = shl i64 [[VSCALE]], 3
; IR-NEXT: br label [[LOOP:%.*]]
; IR: loop:
; IR-NEXT: [[INDVAR:%.*]] = phi i64 [ 0, [[LOOP_PH]] ], [ [[INDVAR_NEXT:%.*]], [[LOOP]] ]
; IR-NEXT: [[TMP0:%.*]] = shl i64 [[INDVAR]], 1
; IR-NEXT: [[UGLYGEP1:%.*]] = getelementptr i8, ptr [[IN:%.*]], i64 [[TMP0]]
; IR-NEXT: [[TMP1:%.*]] = shl i64 [[INDVAR]], 1
; IR-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, ptr [[OUT:%.*]], i64 [[TMP1]]
; IR-NEXT: [[VAL:%.*]] = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0(ptr [[UGLYGEP1]], i32 4, <vscale x 8 x i1> splat (i1 true), <vscale x 8 x i16> poison)
; IR-NEXT: [[ADDP_VEC:%.*]] = add <vscale x 8 x i16> [[VAL]], splat (i16 3)
; IR-NEXT: call void @llvm.masked.store.nxv8i16.p0(<vscale x 8 x i16> [[ADDP_VEC]], ptr [[UGLYGEP]], i32 4, <vscale x 8 x i1> splat (i1 true))
; IR-NEXT: [[INDVAR_NEXT]] = add nsw i64 [[INDVAR]], [[SCALED_VF]]
; IR-NEXT: [[EXIT_COND:%.*]] = icmp eq i64 [[N:%.*]], [[INDVAR_NEXT]]
; IR-NEXT: br i1 [[EXIT_COND]], label [[LOOP_EXIT:%.*]], label [[LOOP]]
; IR: loop.exit:
; IR-NEXT: br label [[EXIT:%.*]]
; IR: exit:
; IR-NEXT: ret void
;
; ASM-LABEL: masked_ld_st_nxv8i16:
; ASM: // %bb.0: // %entry
; ASM-NEXT: ptrue p0.h
; ASM-NEXT: mov x8, xzr
; ASM-NEXT: cnth x9
; ASM-NEXT: .LBB1_1: // %loop
; ASM-NEXT: // =>This Inner Loop Header: Depth=1
; ASM-NEXT: ld1h { z0.h }, p0/z, [x0, x8, lsl #1]
; ASM-NEXT: add z0.h, z0.h, #3
; ASM-NEXT: st1h { z0.h }, p0, [x1, x8, lsl #1]
; ASM-NEXT: add x8, x8, x9
; ASM-NEXT: cmp x2, x8
; ASM-NEXT: b.ne .LBB1_1
; ASM-NEXT: // %bb.2: // %exit
; ASM-NEXT: ret
entry:
br label %loop.ph
loop.ph:
%vscale = call i64 @llvm.vscale.i64()
%scaled_vf = shl i64 %vscale, 3
br label %loop
loop: ; preds = %loop, %loop.ph
%indvar = phi i64 [ 0, %loop.ph ], [ %indvar.next, %loop ]
%ptr.in = getelementptr inbounds i16, ptr %in, i64 %indvar
%ptr.out = getelementptr inbounds i16, ptr %out, i64 %indvar
%val = call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0(ptr %ptr.in, i32 4, <vscale x 8 x i1> splat (i1 true), <vscale x 8 x i16> poison)
%addp_vec = add <vscale x 8 x i16> %val, splat (i16 3)
call void @llvm.masked.store.nxv8i16.p0(<vscale x 8 x i16> %addp_vec, ptr %ptr.out, i32 4, <vscale x 8 x i1> splat (i1 true))
%indvar.next = add nsw i64 %indvar, %scaled_vf
%exit.cond = icmp eq i64 %indvar.next, %n
br i1 %exit.cond, label %loop.exit, label %loop
loop.exit: ; preds = %loop
br label %exit
exit:
ret void
}
declare i64 @llvm.vscale.i64()
declare <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0(ptr, i32 immarg, <vscale x 8 x i1>, <vscale x 8 x i16>)
declare void @llvm.masked.store.nxv8i16.p0(<vscale x 8 x i16>, ptr, i32 immarg, <vscale x 8 x i1>)
|
