1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -instcombine -S | FileCheck %s
3
4; If we have some pattern that leaves only some low bits set, and then performs
5; left-shift of those bits, we can combine those two shifts into a shift+mask.
6
7; There are many variants to this pattern:
8;   b)  (trunc ((x & (~(-1 << maskNbits))))) << shiftNbits
9; simplify to:
10;   ((trunc(x)) << shiftNbits) & (~(-1 << (maskNbits+shiftNbits)))
11
12; Simple tests.
13
14declare void @use32(i32)
15declare void @use64(i64)
16
17define i32 @t0_basic(i64 %x, i32 %nbits) {
18; CHECK-LABEL: @t0_basic(
19; CHECK-NEXT:    [[T0:%.*]] = add i32 [[NBITS:%.*]], -1
20; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[T0]] to i64
21; CHECK-NEXT:    [[T2:%.*]] = shl i64 -1, [[T1]]
22; CHECK-NEXT:    [[T3:%.*]] = xor i64 [[T2]], -1
23; CHECK-NEXT:    [[T4:%.*]] = sub i32 32, [[NBITS]]
24; CHECK-NEXT:    call void @use32(i32 [[T0]])
25; CHECK-NEXT:    call void @use64(i64 [[T1]])
26; CHECK-NEXT:    call void @use64(i64 [[T2]])
27; CHECK-NEXT:    call void @use64(i64 [[T3]])
28; CHECK-NEXT:    call void @use32(i32 [[T4]])
29; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[X:%.*]] to i32
30; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[TMP1]], [[T4]]
31; CHECK-NEXT:    [[T7:%.*]] = and i32 [[TMP2]], 2147483647
32; CHECK-NEXT:    ret i32 [[T7]]
33;
34  %t0 = add i32 %nbits, -1
35  %t1 = zext i32 %t0 to i64
36  %t2 = shl i64 -1, %t1 ; shifting by nbits-1
37  %t3 = xor i64 %t2, -1
38  %t4 = sub i32 32, %nbits
39
40  call void @use32(i32 %t0)
41  call void @use64(i64 %t1)
42  call void @use64(i64 %t2)
43  call void @use64(i64 %t3)
44  call void @use32(i32 %t4)
45
46  %t5 = and i64 %t3, %x
47  %t6 = trunc i64 %t5 to i32
48  %t7 = shl i32 %t6, %t4
49  ret i32 %t7
50}
51
52; Vectors
53
54declare void @use8xi32(<8 x i32>)
55declare void @use8xi64(<8 x i64>)
56
57define <8 x i32> @t1_vec_splat(<8 x i64> %x, <8 x i32> %nbits) {
58; CHECK-LABEL: @t1_vec_splat(
59; CHECK-NEXT:    [[T0:%.*]] = add <8 x i32> [[NBITS:%.*]], <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
60; CHECK-NEXT:    [[T1:%.*]] = zext <8 x i32> [[T0]] to <8 x i64>
61; CHECK-NEXT:    [[T2:%.*]] = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>, [[T1]]
62; CHECK-NEXT:    [[T3:%.*]] = xor <8 x i64> [[T2]], <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
63; CHECK-NEXT:    [[T4:%.*]] = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, [[NBITS]]
64; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T0]])
65; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T1]])
66; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T2]])
67; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T3]])
68; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T4]])
69; CHECK-NEXT:    [[TMP1:%.*]] = trunc <8 x i64> [[X:%.*]] to <8 x i32>
70; CHECK-NEXT:    [[TMP2:%.*]] = shl <8 x i32> [[TMP1]], [[T4]]
71; CHECK-NEXT:    [[T7:%.*]] = and <8 x i32> [[TMP2]], <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647>
72; CHECK-NEXT:    ret <8 x i32> [[T7]]
73;
74  %t0 = add <8 x i32> %nbits, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>
75  %t1 = zext <8 x i32> %t0 to <8 x i64>
76  %t2 = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>, %t1 ; shifting by nbits-1
77  %t3 = xor <8 x i64> %t2, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
78  %t4 = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, %nbits
79
80  call void @use8xi32(<8 x i32> %t0)
81  call void @use8xi64(<8 x i64> %t1)
82  call void @use8xi64(<8 x i64> %t2)
83  call void @use8xi64(<8 x i64> %t3)
84  call void @use8xi32(<8 x i32> %t4)
85
86  %t5 = and <8 x i64> %t3, %x
87  %t6 = trunc <8 x i64> %t5 to <8 x i32>
88  %t7 = shl <8 x i32> %t6, %t4
89  ret <8 x i32> %t7
90}
91
92define <8 x i32> @t2_vec_splat_undef(<8 x i64> %x, <8 x i32> %nbits) {
93; CHECK-LABEL: @t2_vec_splat_undef(
94; CHECK-NEXT:    [[T0:%.*]] = add <8 x i32> [[NBITS:%.*]], <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 undef, i32 -1>
95; CHECK-NEXT:    [[T1:%.*]] = zext <8 x i32> [[T0]] to <8 x i64>
96; CHECK-NEXT:    [[T2:%.*]] = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 undef, i64 -1>, [[T1]]
97; CHECK-NEXT:    [[T3:%.*]] = xor <8 x i64> [[T2]], <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 undef, i64 -1>
98; CHECK-NEXT:    [[T4:%.*]] = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 undef, i32 32>, [[NBITS]]
99; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T0]])
100; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T1]])
101; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T2]])
102; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T3]])
103; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T4]])
104; CHECK-NEXT:    [[TMP1:%.*]] = trunc <8 x i64> [[X:%.*]] to <8 x i32>
105; CHECK-NEXT:    [[TMP2:%.*]] = shl <8 x i32> [[TMP1]], [[T4]]
106; CHECK-NEXT:    [[T7:%.*]] = and <8 x i32> [[TMP2]], <i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 2147483647, i32 poison, i32 2147483647>
107; CHECK-NEXT:    ret <8 x i32> [[T7]]
108;
109  %t0 = add <8 x i32> %nbits, <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 undef, i32 -1>
110  %t1 = zext <8 x i32> %t0 to <8 x i64>
111  %t2 = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 undef, i64 -1>, %t1 ; shifting by nbits-1
112  %t3 = xor <8 x i64> %t2, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 undef, i64 -1>
113  %t4 = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 undef, i32 32>, %nbits
114
115  call void @use8xi32(<8 x i32> %t0)
116  call void @use8xi64(<8 x i64> %t1)
117  call void @use8xi64(<8 x i64> %t2)
118  call void @use8xi64(<8 x i64> %t3)
119  call void @use8xi32(<8 x i32> %t4)
120
121  %t5 = and <8 x i64> %t3, %x
122  %t6 = trunc <8 x i64> %t5 to <8 x i32>
123  %t7 = shl <8 x i32> %t6, %t4
124  ret <8 x i32> %t7
125}
126
127define <8 x i32> @t3_vec_nonsplat(<8 x i64> %x, <8 x i32> %nbits) {
128; CHECK-LABEL: @t3_vec_nonsplat(
129; CHECK-NEXT:    [[T0:%.*]] = add <8 x i32> [[NBITS:%.*]], <i32 -33, i32 -32, i32 -31, i32 -1, i32 0, i32 1, i32 31, i32 32>
130; CHECK-NEXT:    [[T1:%.*]] = zext <8 x i32> [[T0]] to <8 x i64>
131; CHECK-NEXT:    [[T2:%.*]] = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>, [[T1]]
132; CHECK-NEXT:    [[T3:%.*]] = xor <8 x i64> [[T2]], <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
133; CHECK-NEXT:    [[T4:%.*]] = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, [[NBITS]]
134; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T0]])
135; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T1]])
136; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T2]])
137; CHECK-NEXT:    call void @use8xi64(<8 x i64> [[T3]])
138; CHECK-NEXT:    call void @use8xi32(<8 x i32> [[T4]])
139; CHECK-NEXT:    [[TMP1:%.*]] = trunc <8 x i64> [[X:%.*]] to <8 x i32>
140; CHECK-NEXT:    [[TMP2:%.*]] = shl <8 x i32> [[TMP1]], [[T4]]
141; CHECK-NEXT:    [[T7:%.*]] = and <8 x i32> [[TMP2]], <i32 poison, i32 0, i32 1, i32 2147483647, i32 -1, i32 -1, i32 -1, i32 -1>
142; CHECK-NEXT:    ret <8 x i32> [[T7]]
143;
144  %t0 = add <8 x i32> %nbits, <i32 -33, i32 -32, i32 -31, i32 -1, i32 0, i32 1, i32 31, i32 32>
145  %t1 = zext <8 x i32> %t0 to <8 x i64>
146  %t2 = shl <8 x i64> <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>, %t1 ; shifting by nbits-1
147  %t3 = xor <8 x i64> %t2, <i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1, i64 -1>
148  %t4 = sub <8 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, %nbits
149
150  call void @use8xi32(<8 x i32> %t0)
151  call void @use8xi64(<8 x i64> %t1)
152  call void @use8xi64(<8 x i64> %t2)
153  call void @use8xi64(<8 x i64> %t3)
154  call void @use8xi32(<8 x i32> %t4)
155
156  %t5 = and <8 x i64> %t3, %x
157  %t6 = trunc <8 x i64> %t5 to <8 x i32>
158  %t7 = shl <8 x i32> %t6, %t4
159  ret <8 x i32> %t7
160}
161
162; -1 can be truncated.
163
164define i32 @t4_allones_trunc(i64 %x, i32 %nbits) {
165; CHECK-LABEL: @t4_allones_trunc(
166; CHECK-NEXT:    [[T0:%.*]] = add i32 [[NBITS:%.*]], -1
167; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[T0]] to i64
168; CHECK-NEXT:    [[T2:%.*]] = shl i64 -1, [[T1]]
169; CHECK-NEXT:    [[T3:%.*]] = xor i64 [[T2]], 4294967295
170; CHECK-NEXT:    [[T4:%.*]] = sub i32 32, [[NBITS]]
171; CHECK-NEXT:    call void @use32(i32 [[T0]])
172; CHECK-NEXT:    call void @use64(i64 [[T1]])
173; CHECK-NEXT:    call void @use64(i64 [[T2]])
174; CHECK-NEXT:    call void @use64(i64 [[T3]])
175; CHECK-NEXT:    call void @use32(i32 [[T4]])
176; CHECK-NEXT:    [[T5:%.*]] = and i64 [[T3]], [[X:%.*]]
177; CHECK-NEXT:    [[T6:%.*]] = trunc i64 [[T5]] to i32
178; CHECK-NEXT:    [[T7:%.*]] = shl i32 [[T6]], [[T4]]
179; CHECK-NEXT:    ret i32 [[T7]]
180;
181  %t0 = add i32 %nbits, -1
182  %t1 = zext i32 %t0 to i64
183  %t2 = shl i64 -1, %t1 ; shifting by nbits-1
184  %t3 = xor i64 %t2, 4294967295 ; we only care about low 32 bits
185  %t4 = sub i32 32, %nbits
186
187  call void @use32(i32 %t0)
188  call void @use64(i64 %t1)
189  call void @use64(i64 %t2)
190  call void @use64(i64 %t3)
191  call void @use32(i32 %t4)
192
193  %t5 = and i64 %t3, %x
194  %t6 = trunc i64 %t5 to i32
195  %t7 = shl i32 %t6, %t4
196  ret i32 %t7
197}
198
199; Extra uses
200
201define i32 @n5_extrause0(i64 %x, i32 %nbits) {
202; CHECK-LABEL: @n5_extrause0(
203; CHECK-NEXT:    [[T0:%.*]] = add i32 [[NBITS:%.*]], -1
204; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[T0]] to i64
205; CHECK-NEXT:    [[T2:%.*]] = shl i64 -1, [[T1]]
206; CHECK-NEXT:    [[T3:%.*]] = xor i64 [[T2]], -1
207; CHECK-NEXT:    [[T4:%.*]] = sub i32 32, [[NBITS]]
208; CHECK-NEXT:    call void @use32(i32 [[T0]])
209; CHECK-NEXT:    call void @use64(i64 [[T1]])
210; CHECK-NEXT:    call void @use64(i64 [[T2]])
211; CHECK-NEXT:    call void @use64(i64 [[T3]])
212; CHECK-NEXT:    call void @use32(i32 [[T4]])
213; CHECK-NEXT:    [[T5:%.*]] = and i64 [[T3]], [[X:%.*]]
214; CHECK-NEXT:    call void @use64(i64 [[T5]])
215; CHECK-NEXT:    [[T6:%.*]] = trunc i64 [[T5]] to i32
216; CHECK-NEXT:    [[T7:%.*]] = shl i32 [[T6]], [[T4]]
217; CHECK-NEXT:    ret i32 [[T7]]
218;
219  %t0 = add i32 %nbits, -1
220  %t1 = zext i32 %t0 to i64
221  %t2 = shl i64 -1, %t1 ; shifting by nbits-1
222  %t3 = xor i64 %t2, -1
223  %t4 = sub i32 32, %nbits
224
225  call void @use32(i32 %t0)
226  call void @use64(i64 %t1)
227  call void @use64(i64 %t2)
228  call void @use64(i64 %t3)
229  call void @use32(i32 %t4)
230
231  %t5 = and i64 %t3, %x
232  call void @use64(i64 %t5)
233  %t6 = trunc i64 %t5 to i32
234  %t7 = shl i32 %t6, %t4
235  ret i32 %t7
236}
237define i32 @n6_extrause1(i64 %x, i32 %nbits) {
238; CHECK-LABEL: @n6_extrause1(
239; CHECK-NEXT:    [[T0:%.*]] = add i32 [[NBITS:%.*]], -1
240; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[T0]] to i64
241; CHECK-NEXT:    [[T2:%.*]] = shl i64 -1, [[T1]]
242; CHECK-NEXT:    [[T3:%.*]] = xor i64 [[T2]], -1
243; CHECK-NEXT:    [[T4:%.*]] = sub i32 32, [[NBITS]]
244; CHECK-NEXT:    call void @use32(i32 [[T0]])
245; CHECK-NEXT:    call void @use64(i64 [[T1]])
246; CHECK-NEXT:    call void @use64(i64 [[T2]])
247; CHECK-NEXT:    call void @use64(i64 [[T3]])
248; CHECK-NEXT:    call void @use32(i32 [[T4]])
249; CHECK-NEXT:    [[T5:%.*]] = and i64 [[T3]], [[X:%.*]]
250; CHECK-NEXT:    [[T6:%.*]] = trunc i64 [[T5]] to i32
251; CHECK-NEXT:    call void @use32(i32 [[T6]])
252; CHECK-NEXT:    [[T7:%.*]] = shl i32 [[T6]], [[T4]]
253; CHECK-NEXT:    ret i32 [[T7]]
254;
255  %t0 = add i32 %nbits, -1
256  %t1 = zext i32 %t0 to i64
257  %t2 = shl i64 -1, %t1 ; shifting by nbits-1
258  %t3 = xor i64 %t2, -1
259  %t4 = sub i32 32, %nbits
260
261  call void @use32(i32 %t0)
262  call void @use64(i64 %t1)
263  call void @use64(i64 %t2)
264  call void @use64(i64 %t3)
265  call void @use32(i32 %t4)
266
267  %t5 = and i64 %t3, %x
268  %t6 = trunc i64 %t5 to i32
269  call void @use32(i32 %t6)
270  %t7 = shl i32 %t6, %t4
271  ret i32 %t7
272}
273define i32 @n7_extrause2(i64 %x, i32 %nbits) {
274; CHECK-LABEL: @n7_extrause2(
275; CHECK-NEXT:    [[T0:%.*]] = add i32 [[NBITS:%.*]], -1
276; CHECK-NEXT:    [[T1:%.*]] = zext i32 [[T0]] to i64
277; CHECK-NEXT:    [[T2:%.*]] = shl i64 -1, [[T1]]
278; CHECK-NEXT:    [[T3:%.*]] = xor i64 [[T2]], -1
279; CHECK-NEXT:    [[T4:%.*]] = sub i32 32, [[NBITS]]
280; CHECK-NEXT:    call void @use32(i32 [[T0]])
281; CHECK-NEXT:    call void @use64(i64 [[T1]])
282; CHECK-NEXT:    call void @use64(i64 [[T2]])
283; CHECK-NEXT:    call void @use64(i64 [[T3]])
284; CHECK-NEXT:    call void @use32(i32 [[T4]])
285; CHECK-NEXT:    [[T5:%.*]] = and i64 [[T3]], [[X:%.*]]
286; CHECK-NEXT:    call void @use64(i64 [[T5]])
287; CHECK-NEXT:    [[T6:%.*]] = trunc i64 [[T5]] to i32
288; CHECK-NEXT:    call void @use32(i32 [[T6]])
289; CHECK-NEXT:    [[T7:%.*]] = shl i32 [[T6]], [[T4]]
290; CHECK-NEXT:    ret i32 [[T7]]
291;
292  %t0 = add i32 %nbits, -1
293  %t1 = zext i32 %t0 to i64
294  %t2 = shl i64 -1, %t1 ; shifting by nbits-1
295  %t3 = xor i64 %t2, -1
296  %t4 = sub i32 32, %nbits
297
298  call void @use32(i32 %t0)
299  call void @use64(i64 %t1)
300  call void @use64(i64 %t2)
301  call void @use64(i64 %t3)
302  call void @use32(i32 %t4)
303
304  %t5 = and i64 %t3, %x
305  call void @use64(i64 %t5)
306  %t6 = trunc i64 %t5 to i32
307  call void @use32(i32 %t6)
308  %t7 = shl i32 %t6, %t4
309  ret i32 %t7
310}
311