shift-amount-reassociation-with-truncation-ashr.ll
6.28 KB
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
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -instcombine -S | FileCheck %s
; Given pattern:
; (trunc (iSrc x a>> Q) to iDst) a>> K
; we should rewrite it as
; (trunc (iSrc x a>> (Q+K)) to iDst)
; iff (Q+K) is bitwidth(iSrc)-1
; THIS FOLD DOES *NOT* REQUIRE ANY 'nuw'/`nsw` FLAGS!
; Basic scalar test
define i16 @t0(i32 %x, i16 %y) {
; CHECK-LABEL: @t0(
; CHECK-NEXT: [[TMP1:%.*]] = ashr i32 [[X:%.*]], 31
; CHECK-NEXT: [[T5:%.*]] = trunc i32 [[TMP1]] to i16
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = ashr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -1
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}
; Basic vector tests
define <2 x i16> @t1_vec_splat(<2 x i32> %x, <2 x i16> %y) {
; CHECK-LABEL: @t1_vec_splat(
; CHECK-NEXT: [[TMP1:%.*]] = ashr <2 x i32> [[X:%.*]], <i32 31, i32 31>
; CHECK-NEXT: [[T5:%.*]] = trunc <2 x i32> [[TMP1]] to <2 x i16>
; CHECK-NEXT: ret <2 x i16> [[T5]]
;
%t0 = sub <2 x i16> <i16 32, i16 32>, %y
%t1 = zext <2 x i16> %t0 to <2 x i32>
%t2 = ashr <2 x i32> %x, %t1
%t3 = trunc <2 x i32> %t2 to <2 x i16>
%t4 = add <2 x i16> %y, <i16 -1, i16 -1>
%t5 = ashr <2 x i16> %t3, %t4
ret <2 x i16> %t5
}
define <3 x i16> @t3_vec_nonsplat_undef0(<3 x i32> %x, <3 x i16> %y) {
; CHECK-LABEL: @t3_vec_nonsplat_undef0(
; CHECK-NEXT: [[TMP1:%.*]] = ashr <3 x i32> [[X:%.*]], <i32 31, i32 0, i32 31>
; CHECK-NEXT: [[T5:%.*]] = trunc <3 x i32> [[TMP1]] to <3 x i16>
; CHECK-NEXT: ret <3 x i16> [[T5]]
;
%t0 = sub <3 x i16> <i16 32, i16 undef, i16 32>, %y
%t1 = zext <3 x i16> %t0 to <3 x i32>
%t2 = ashr <3 x i32> %x, %t1
%t3 = trunc <3 x i32> %t2 to <3 x i16>
%t4 = add <3 x i16> %y, <i16 -1, i16 -1, i16 -1>
%t5 = ashr <3 x i16> %t3, %t4
ret <3 x i16> %t5
}
define <3 x i16> @t4_vec_nonsplat_undef1(<3 x i32> %x, <3 x i16> %y) {
; CHECK-LABEL: @t4_vec_nonsplat_undef1(
; CHECK-NEXT: [[TMP1:%.*]] = ashr <3 x i32> [[X:%.*]], <i32 31, i32 0, i32 31>
; CHECK-NEXT: [[T5:%.*]] = trunc <3 x i32> [[TMP1]] to <3 x i16>
; CHECK-NEXT: ret <3 x i16> [[T5]]
;
%t0 = sub <3 x i16> <i16 32, i16 32, i16 32>, %y
%t1 = zext <3 x i16> %t0 to <3 x i32>
%t2 = ashr <3 x i32> %x, %t1
%t3 = trunc <3 x i32> %t2 to <3 x i16>
%t4 = add <3 x i16> %y, <i16 -1, i16 undef, i16 -1>
%t5 = ashr <3 x i16> %t3, %t4
ret <3 x i16> %t5
}
define <3 x i16> @t5_vec_nonsplat_undef1(<3 x i32> %x, <3 x i16> %y) {
; CHECK-LABEL: @t5_vec_nonsplat_undef1(
; CHECK-NEXT: [[TMP1:%.*]] = ashr <3 x i32> [[X:%.*]], <i32 31, i32 0, i32 31>
; CHECK-NEXT: [[T5:%.*]] = trunc <3 x i32> [[TMP1]] to <3 x i16>
; CHECK-NEXT: ret <3 x i16> [[T5]]
;
%t0 = sub <3 x i16> <i16 32, i16 undef, i16 32>, %y
%t1 = zext <3 x i16> %t0 to <3 x i32>
%t2 = ashr <3 x i32> %x, %t1
%t3 = trunc <3 x i32> %t2 to <3 x i16>
%t4 = add <3 x i16> %y, <i16 -1, i16 undef, i16 -1>
%t5 = ashr <3 x i16> %t3, %t4
ret <3 x i16> %t5
}
; One-use tests
declare void @use16(i16)
declare void @use32(i32)
define i16 @t6_extrause0(i32 %x, i16 %y) {
; CHECK-LABEL: @t6_extrause0(
; CHECK-NEXT: [[T0:%.*]] = sub i16 32, [[Y:%.*]]
; CHECK-NEXT: [[T1:%.*]] = zext i16 [[T0]] to i32
; CHECK-NEXT: [[T2:%.*]] = ashr i32 [[X:%.*]], [[T1]]
; CHECK-NEXT: [[T3:%.*]] = trunc i32 [[T2]] to i16
; CHECK-NEXT: call void @use16(i16 [[T3]])
; CHECK-NEXT: [[TMP1:%.*]] = ashr i32 [[X]], 31
; CHECK-NEXT: [[T5:%.*]] = trunc i32 [[TMP1]] to i16
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = ashr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -1
call void @use16(i16 %t3)
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}
define i16 @t7_extrause1(i32 %x, i16 %y) {
; CHECK-LABEL: @t7_extrause1(
; CHECK-NEXT: [[T4:%.*]] = add i16 [[Y:%.*]], -1
; CHECK-NEXT: call void @use16(i16 [[T4]])
; CHECK-NEXT: [[TMP1:%.*]] = ashr i32 [[X:%.*]], 31
; CHECK-NEXT: [[T5:%.*]] = trunc i32 [[TMP1]] to i16
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = ashr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -1
call void @use16(i16 %t4)
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}
define i16 @t8_extrause2(i32 %x, i16 %y) {
; CHECK-LABEL: @t8_extrause2(
; CHECK-NEXT: [[T0:%.*]] = sub i16 32, [[Y:%.*]]
; CHECK-NEXT: [[T1:%.*]] = zext i16 [[T0]] to i32
; CHECK-NEXT: [[T2:%.*]] = ashr i32 [[X:%.*]], [[T1]]
; CHECK-NEXT: [[T3:%.*]] = trunc i32 [[T2]] to i16
; CHECK-NEXT: [[T4:%.*]] = add i16 [[Y]], -1
; CHECK-NEXT: call void @use16(i16 [[T3]])
; CHECK-NEXT: call void @use16(i16 [[T4]])
; CHECK-NEXT: [[T5:%.*]] = ashr i16 [[T3]], [[T4]]
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = ashr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -1
call void @use16(i16 %t3)
call void @use16(i16 %t4)
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}
; No 'nuw'/'nsw' flags are to be propagated!
; But we can't test that, such IR does not reach that code.
; Negative tests
; Can only fold if we are extracting the sign bit.
define i16 @t9_ashr(i32 %x, i16 %y) {
; CHECK-LABEL: @t9_ashr(
; CHECK-NEXT: [[T0:%.*]] = sub i16 32, [[Y:%.*]]
; CHECK-NEXT: [[T1:%.*]] = zext i16 [[T0]] to i32
; CHECK-NEXT: [[T2:%.*]] = ashr i32 [[X:%.*]], [[T1]]
; CHECK-NEXT: [[T3:%.*]] = trunc i32 [[T2]] to i16
; CHECK-NEXT: [[T4:%.*]] = add i16 [[Y]], -2
; CHECK-NEXT: [[T5:%.*]] = ashr i16 [[T3]], [[T4]]
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = ashr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -2
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}
; If we have different right-shifts, in general, we can't do anything with it.
define i16 @n10_lshr_ashr(i32 %x, i16 %y) {
; CHECK-LABEL: @n10_lshr_ashr(
; CHECK-NEXT: [[T0:%.*]] = sub i16 32, [[Y:%.*]]
; CHECK-NEXT: [[T1:%.*]] = zext i16 [[T0]] to i32
; CHECK-NEXT: [[T2:%.*]] = lshr i32 [[X:%.*]], [[T1]]
; CHECK-NEXT: [[T3:%.*]] = trunc i32 [[T2]] to i16
; CHECK-NEXT: [[T4:%.*]] = add i16 [[Y]], -1
; CHECK-NEXT: [[T5:%.*]] = ashr i16 [[T3]], [[T4]]
; CHECK-NEXT: ret i16 [[T5]]
;
%t0 = sub i16 32, %y
%t1 = zext i16 %t0 to i32
%t2 = lshr i32 %x, %t1
%t3 = trunc i32 %t2 to i16
%t4 = add i16 %y, -1
%t5 = ashr i16 %t3, %t4
ret i16 %t5
}