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
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 -mattr=+v -verify-machineinstrs \
; RUN: -riscv-enable-vl-optimizer=false | FileCheck %s -check-prefixes=CHECK,NOVLOPT
; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 -mattr=+v -verify-machineinstrs \
; RUN: -riscv-enable-vl-optimizer=false | FileCheck %s -check-prefixes=CHECK,NOVLOPT
; RUN: sed 's/iXLen/i32/g' %s | llc -mtriple=riscv32 -mattr=+v -riscv-enable-vl-optimizer \
; RUN: -verify-machineinstrs | FileCheck %s -check-prefixes=CHECK,VLOPT
; RUN: sed 's/iXLen/i64/g' %s | llc -mtriple=riscv64 -mattr=+v -riscv-enable-vl-optimizer \
; RUN: -verify-machineinstrs | FileCheck %s -check-prefixes=CHECK,VLOPT
declare <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>, iXLen)
define <vscale x 4 x i32> @different_imm_vl_with_ta(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; NOVLOPT-LABEL: different_imm_vl_with_ta:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetivli zero, 5, e32, m2, ta, ma
; NOVLOPT-NEXT: vadd.vv v8, v10, v12
; NOVLOPT-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; NOVLOPT-NEXT: vadd.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: different_imm_vl_with_ta:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; VLOPT-NEXT: vadd.vv v8, v10, v12
; VLOPT-NEXT: vadd.vv v8, v8, v10
; VLOPT-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 5)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen 4)
ret <vscale x 4 x i32> %w
}
define <vscale x 4 x i32> @vlmax_and_imm_vl_with_ta(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; NOVLOPT-LABEL: vlmax_and_imm_vl_with_ta:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetvli a0, zero, e32, m2, ta, ma
; NOVLOPT-NEXT: vadd.vv v8, v10, v12
; NOVLOPT-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; NOVLOPT-NEXT: vadd.vv v8, v8, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: vlmax_and_imm_vl_with_ta:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; VLOPT-NEXT: vadd.vv v8, v10, v12
; VLOPT-NEXT: vadd.vv v8, v8, v10
; VLOPT-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen -1)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen 4)
ret <vscale x 4 x i32> %w
}
; Not beneficial to propagate VL since VL is larger in the use side.
define <vscale x 4 x i32> @different_imm_vl_with_ta_larger_vl(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_imm_vl_with_ta_larger_vl:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v10, v12
; CHECK-NEXT: vsetivli zero, 5, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v8, v10
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 4)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen 5)
ret <vscale x 4 x i32> %w
}
define <vscale x 4 x i32> @different_imm_reg_vl_with_ta(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_imm_reg_vl_with_ta:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 4, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v10, v12
; CHECK-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v8, v10
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 4)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen %vl1)
ret <vscale x 4 x i32> %w
}
; Not beneficial to propagate VL since VL is already one.
define <vscale x 4 x i32> @different_imm_vl_with_ta_1(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_imm_vl_with_ta_1:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 1, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v10, v12
; CHECK-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v8, v10
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 1)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen %vl1)
ret <vscale x 4 x i32> %w
}
; Propgate %vl2 to last instruction since it is may smaller than %vl1,
; it's still safe even %vl2 is larger than %vl1, becuase rest of the vector are
; undefined value.
define <vscale x 4 x i32> @different_vl_with_ta(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_vl_with_ta:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli zero, a0, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v10, v8, v10
; CHECK-NEXT: vsetvli zero, a1, e32, m2, ta, ma
; CHECK-NEXT: vadd.vv v8, v10, v8
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> poison, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a,iXLen %vl2)
ret <vscale x 4 x i32> %w
}
; We can propagate VL to a tail-undisturbed policy, provided none of its users
; are passthrus (i.e. read past VL).
define <vscale x 4 x i32> @different_vl_with_tu(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_vl_with_tu:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli zero, a0, e32, m2, tu, ma
; CHECK-NEXT: vmv2r.v v14, v10
; CHECK-NEXT: vadd.vv v14, v10, v12
; CHECK-NEXT: vsetvli zero, a1, e32, m2, tu, ma
; CHECK-NEXT: vadd.vv v8, v14, v10
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen %vl2)
ret <vscale x 4 x i32> %w
}
; We can propagate VL to a tail-undisturbed policy, provided none of its users
; are passthrus (i.e. read past VL).
define <vscale x 4 x i32> @different_imm_vl_with_tu(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; NOVLOPT-LABEL: different_imm_vl_with_tu:
; NOVLOPT: # %bb.0:
; NOVLOPT-NEXT: vsetivli zero, 5, e32, m2, tu, ma
; NOVLOPT-NEXT: vmv2r.v v14, v10
; NOVLOPT-NEXT: vadd.vv v14, v10, v12
; NOVLOPT-NEXT: vsetivli zero, 4, e32, m2, tu, ma
; NOVLOPT-NEXT: vadd.vv v8, v14, v10
; NOVLOPT-NEXT: ret
;
; VLOPT-LABEL: different_imm_vl_with_tu:
; VLOPT: # %bb.0:
; VLOPT-NEXT: vsetivli zero, 4, e32, m2, tu, ma
; VLOPT-NEXT: vmv2r.v v14, v10
; VLOPT-NEXT: vadd.vv v14, v10, v12
; VLOPT-NEXT: vadd.vv v8, v14, v10
; VLOPT-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 5)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %passthru, <vscale x 4 x i32> %v, <vscale x 4 x i32> %a, iXLen 4)
ret <vscale x 4 x i32> %w
}
; We can't reduce the VL as %v is used as a passthru, i.e. the elements past VL
; are demanded.
define <vscale x 4 x i32> @different_vl_as_passthru(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_vl_as_passthru:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli zero, a0, e32, m2, tu, ma
; CHECK-NEXT: vmv2r.v v12, v8
; CHECK-NEXT: vadd.vv v12, v8, v10
; CHECK-NEXT: vsetvli zero, a1, e32, m2, tu, ma
; CHECK-NEXT: vadd.vv v12, v8, v10
; CHECK-NEXT: vmv2r.v v8, v12
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %v, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl2)
ret <vscale x 4 x i32> %w
}
; We can't reduce the VL as %v is used as a passthru, i.e. the elements past VL
; are demanded.
define <vscale x 4 x i32> @different_imm_vl_as_passthru(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen %vl1, iXLen %vl2) {
; CHECK-LABEL: different_imm_vl_as_passthru:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetivli zero, 5, e32, m2, tu, ma
; CHECK-NEXT: vmv2r.v v12, v8
; CHECK-NEXT: vadd.vv v12, v8, v10
; CHECK-NEXT: vsetivli zero, 4, e32, m2, tu, ma
; CHECK-NEXT: vadd.vv v12, v8, v10
; CHECK-NEXT: vmv2r.v v8, v12
; CHECK-NEXT: ret
%v = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 5)
%w = call <vscale x 4 x i32> @llvm.riscv.vadd.nxv4i32.nxv4i32(<vscale x 4 x i32> %v, <vscale x 4 x i32> %a, <vscale x 4 x i32> %b, iXLen 4)
ret <vscale x 4 x i32> %w
}
define <vscale x 4 x i32> @dont_optimize_tied_def(<vscale x 4 x i32> %a, <vscale x 4 x i16> %b, <vscale x 4 x i16> %c, iXLen %vl) {
; CHECK-LABEL: dont_optimize_tied_def:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a1, zero, e16, m1, tu, ma
; CHECK-NEXT: vwmacc.vv v8, v10, v11
; CHECK-NEXT: vsetvli zero, a0, e16, m1, tu, ma
; CHECK-NEXT: vwmacc.vv v8, v10, v11
; CHECK-NEXT: ret
%1 = call <vscale x 4 x i32> @llvm.riscv.vwmacc.nxv4i32.nxv4i16(<vscale x 4 x i32> %a, <vscale x 4 x i16> %b, <vscale x 4 x i16> %c, iXLen -1, iXLen 0)
%2 = call <vscale x 4 x i32> @llvm.riscv.vwmacc.nxv4i32.nxv4i16(<vscale x 4 x i32> %1, <vscale x 4 x i16> %b, <vscale x 4 x i16> %c, iXLen %vl, iXLen 0)
ret <vscale x 4 x i32> %2
}
|
