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
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple riscv64 -mattr=+m,+v < %s \
; RUN: | FileCheck %s -check-prefixes=RV64,RV64-VLENUNK
; RUN: llc -mtriple riscv32 -mattr=+m,+v < %s \
; RUN: | FileCheck %s -check-prefix=RV32
; RUN: llc -mtriple riscv64 -mattr=+m,+v -riscv-v-vector-bits-min=256 < %s \
; RUN: | FileCheck %s -check-prefixes=RV64,RV64-VLEN256MIN
; RUN: llc -mtriple riscv64 -mattr=+m,+v -riscv-v-vector-bits-max=256 < %s \
; RUN: | FileCheck %s -check-prefixes=RV64,RV64-VLEN256MAX
; RUN: llc -mtriple riscv64 -mattr=+m,+v -riscv-v-vector-bits-min=256 -riscv-v-vector-bits-max=256 < %s \
; RUN: | FileCheck %s -check-prefixes=RV64-VLEN256EXACT
define i64 @vscale_zero() nounwind {
; RV64-LABEL: vscale_zero:
; RV64: # %bb.0: # %entry
; RV64-NEXT: li a0, 0
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_zero:
; RV32: # %bb.0: # %entry
; RV32-NEXT: li a0, 0
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_zero:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 0
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = mul i64 %0, 0
ret i64 %1
}
define i64 @vscale_one() nounwind {
; RV64-LABEL: vscale_one:
; RV64: # %bb.0: # %entry
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: srli a0, a0, 3
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_one:
; RV32: # %bb.0: # %entry
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: srli a0, a0, 3
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_one:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 4
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = mul i64 %0, 1
ret i64 %1
}
define i64 @vscale_uimmpow2xlen() nounwind {
; RV64-LABEL: vscale_uimmpow2xlen:
; RV64: # %bb.0: # %entry
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: slli a0, a0, 3
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_uimmpow2xlen:
; RV32: # %bb.0: # %entry
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: slli a0, a0, 3
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_uimmpow2xlen:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 256
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = mul i64 %0, 64
ret i64 %1
}
define i64 @vscale_non_pow2() nounwind {
; RV64-LABEL: vscale_non_pow2:
; RV64: # %bb.0: # %entry
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: slli a1, a0, 1
; RV64-NEXT: add a0, a1, a0
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_non_pow2:
; RV32: # %bb.0: # %entry
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: slli a1, a0, 1
; RV32-NEXT: add a0, a1, a0
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_non_pow2:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 96
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = mul i64 %0, 24
ret i64 %1
}
; vscale will always be a positive number, but we don't know that until after op
; legalization. The and will be considered a NOP and replaced with its input,
; but not until after the select becomes RISCVISD::SELECT_CC. Make sure we
; simplify this and don't leave behind any code for calculating the select
; condition.
define i64 @vscale_select(i32 %x, i32 %y) {
; RV64-LABEL: vscale_select:
; RV64: # %bb.0:
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: srli a0, a0, 3
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_select:
; RV32: # %bb.0:
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: srli a0, a0, 3
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_select:
; RV64-VLEN256EXACT: # %bb.0:
; RV64-VLEN256EXACT-NEXT: li a0, 4
; RV64-VLEN256EXACT-NEXT: ret
%a = call i64 @llvm.vscale.i64()
%b = and i64 %a, 4294967295
%c = icmp eq i32 %x, %y
%d = select i1 %c, i64 %a, i64 %b
ret i64 %d
}
define i64 @vscale_high_bits_zero() nounwind {
; RV64-LABEL: vscale_high_bits_zero:
; RV64: # %bb.0: # %entry
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: srli a0, a0, 3
; RV64-NEXT: ret
;
; RV32-LABEL: vscale_high_bits_zero:
; RV32: # %bb.0: # %entry
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: srli a0, a0, 3
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_high_bits_zero:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 4
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = and i64 %0, 2047
ret i64 %1
}
define i64 @vscale_masked() nounwind {
; RV64-VLENUNK-LABEL: vscale_masked:
; RV64-VLENUNK: # %bb.0: # %entry
; RV64-VLENUNK-NEXT: csrr a0, vlenb
; RV64-VLENUNK-NEXT: srli a0, a0, 3
; RV64-VLENUNK-NEXT: andi a0, a0, 510
; RV64-VLENUNK-NEXT: ret
;
; RV32-LABEL: vscale_masked:
; RV32: # %bb.0: # %entry
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: srli a0, a0, 3
; RV32-NEXT: andi a0, a0, 510
; RV32-NEXT: li a1, 0
; RV32-NEXT: ret
;
; RV64-VLEN256MIN-LABEL: vscale_masked:
; RV64-VLEN256MIN: # %bb.0: # %entry
; RV64-VLEN256MIN-NEXT: csrr a0, vlenb
; RV64-VLEN256MIN-NEXT: srli a0, a0, 3
; RV64-VLEN256MIN-NEXT: andi a0, a0, 508
; RV64-VLEN256MIN-NEXT: ret
;
; RV64-VLEN256MAX-LABEL: vscale_masked:
; RV64-VLEN256MAX: # %bb.0: # %entry
; RV64-VLEN256MAX-NEXT: csrr a0, vlenb
; RV64-VLEN256MAX-NEXT: srli a0, a0, 3
; RV64-VLEN256MAX-NEXT: ret
;
; RV64-VLEN256EXACT-LABEL: vscale_masked:
; RV64-VLEN256EXACT: # %bb.0: # %entry
; RV64-VLEN256EXACT-NEXT: li a0, 4
; RV64-VLEN256EXACT-NEXT: ret
entry:
%0 = call i64 @llvm.vscale.i64()
%1 = and i64 %0, 511
ret i64 %1
}
declare i64 @llvm.vscale.i64()
|
