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, if none of the bits that are left after the final
6; shift are modified by the mask, we can omit the mask.
7
8; There are many variants to this pattern:
9;   d)  (x & ((-1 << maskNbits) >> maskNbits)) << shiftNbits
10; simplify to:
11;   x << shiftNbits
12; iff (shiftNbits-maskNbits) s>= 0 (i.e. shiftNbits u>= maskNbits)
13
14; Simple tests. We don't care about extra uses.
15
16declare void @use32(i32)
17
18define i32 @t0_basic(i32 %x, i32 %nbits) {
19; CHECK-LABEL: @t0_basic(
20; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
21; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
22; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
23; CHECK-NEXT:    call void @use32(i32 [[T0]])
24; CHECK-NEXT:    call void @use32(i32 [[T1]])
25; CHECK-NEXT:    call void @use32(i32 [[T2]])
26; CHECK-NEXT:    [[T4:%.*]] = shl i32 [[X]], [[NBITS]]
27; CHECK-NEXT:    ret i32 [[T4]]
28;
29  %t0 = shl i32 -1, %nbits
30  %t1 = lshr i32 %t0, %nbits
31  %t2 = and i32 %t1, %x
32  call void @use32(i32 %t0)
33  call void @use32(i32 %t1)
34  call void @use32(i32 %t2)
35  %t4 = shl i32 %t2, %nbits
36  ret i32 %t4
37}
38
39define i32 @t1_bigger_shift(i32 %x, i32 %nbits) {
40; CHECK-LABEL: @t1_bigger_shift(
41; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
42; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
43; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
44; CHECK-NEXT:    [[T3:%.*]] = add i32 [[NBITS]], 1
45; CHECK-NEXT:    call void @use32(i32 [[T0]])
46; CHECK-NEXT:    call void @use32(i32 [[T1]])
47; CHECK-NEXT:    call void @use32(i32 [[T2]])
48; CHECK-NEXT:    call void @use32(i32 [[T3]])
49; CHECK-NEXT:    [[T4:%.*]] = shl i32 [[X]], [[T3]]
50; CHECK-NEXT:    ret i32 [[T4]]
51;
52  %t0 = shl i32 -1, %nbits
53  %t1 = lshr i32 %t0, %nbits
54  %t2 = and i32 %t1, %x
55  %t3 = add i32 %nbits, 1
56  call void @use32(i32 %t0)
57  call void @use32(i32 %t1)
58  call void @use32(i32 %t2)
59  call void @use32(i32 %t3)
60  %t4 = shl i32 %t2, %t3
61  ret i32 %t4
62}
63
64; Vectors
65
66declare void @use3xi32(<3 x i32>)
67
68define <3 x i32> @t2_vec_splat(<3 x i32> %x, <3 x i32> %nbits) {
69; CHECK-LABEL: @t2_vec_splat(
70; CHECK-NEXT:    [[T0:%.*]] = shl <3 x i32> <i32 -1, i32 -1, i32 -1>, [[NBITS:%.*]]
71; CHECK-NEXT:    [[T1:%.*]] = lshr <3 x i32> [[T0]], [[NBITS]]
72; CHECK-NEXT:    [[T2:%.*]] = and <3 x i32> [[T1]], [[X:%.*]]
73; CHECK-NEXT:    [[T3:%.*]] = add <3 x i32> [[NBITS]], <i32 1, i32 1, i32 1>
74; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T0]])
75; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T1]])
76; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T2]])
77; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T3]])
78; CHECK-NEXT:    [[T4:%.*]] = shl <3 x i32> [[X]], [[T3]]
79; CHECK-NEXT:    ret <3 x i32> [[T4]]
80;
81  %t0 = shl <3 x i32> <i32 -1, i32 -1, i32 -1>, %nbits
82  %t1 = lshr <3 x i32> %t0, %nbits
83  %t2 = and <3 x i32> %t1, %x
84  %t3 = add <3 x i32> %nbits, <i32 1, i32 1, i32 1>
85  call void @use3xi32(<3 x i32> %t0)
86  call void @use3xi32(<3 x i32> %t1)
87  call void @use3xi32(<3 x i32> %t2)
88  call void @use3xi32(<3 x i32> %t3)
89  %t4 = shl <3 x i32> %t2, %t3
90  ret <3 x i32> %t4
91}
92
93define <3 x i32> @t3_vec_nonsplat(<3 x i32> %x, <3 x i32> %nbits) {
94; CHECK-LABEL: @t3_vec_nonsplat(
95; CHECK-NEXT:    [[T0:%.*]] = shl <3 x i32> <i32 -1, i32 -1, i32 -1>, [[NBITS:%.*]]
96; CHECK-NEXT:    [[T1:%.*]] = lshr <3 x i32> [[T0]], [[NBITS]]
97; CHECK-NEXT:    [[T2:%.*]] = and <3 x i32> [[T1]], [[X:%.*]]
98; CHECK-NEXT:    [[T3:%.*]] = add <3 x i32> [[NBITS]], <i32 1, i32 0, i32 2>
99; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T0]])
100; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T1]])
101; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T2]])
102; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T3]])
103; CHECK-NEXT:    [[T4:%.*]] = shl <3 x i32> [[X]], [[T3]]
104; CHECK-NEXT:    ret <3 x i32> [[T4]]
105;
106  %t0 = shl <3 x i32> <i32 -1, i32 -1, i32 -1>, %nbits
107  %t1 = lshr <3 x i32> %t0, %nbits
108  %t2 = and <3 x i32> %t1, %x
109  %t3 = add <3 x i32> %nbits, <i32 1, i32 0, i32 2>
110  call void @use3xi32(<3 x i32> %t0)
111  call void @use3xi32(<3 x i32> %t1)
112  call void @use3xi32(<3 x i32> %t2)
113  call void @use3xi32(<3 x i32> %t3)
114  %t4 = shl <3 x i32> %t2, %t3
115  ret <3 x i32> %t4
116}
117
118define <3 x i32> @t4_vec_undef(<3 x i32> %x, <3 x i32> %nbits) {
119; CHECK-LABEL: @t4_vec_undef(
120; CHECK-NEXT:    [[T0:%.*]] = shl <3 x i32> <i32 -1, i32 undef, i32 -1>, [[NBITS:%.*]]
121; CHECK-NEXT:    [[T1:%.*]] = lshr <3 x i32> [[T0]], [[NBITS]]
122; CHECK-NEXT:    [[T2:%.*]] = and <3 x i32> [[T1]], [[X:%.*]]
123; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T0]])
124; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T1]])
125; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[T2]])
126; CHECK-NEXT:    call void @use3xi32(<3 x i32> [[NBITS]])
127; CHECK-NEXT:    [[T4:%.*]] = shl <3 x i32> [[X]], [[NBITS]]
128; CHECK-NEXT:    ret <3 x i32> [[T4]]
129;
130  %t0 = shl <3 x i32> <i32 -1, i32 undef, i32 -1>, %nbits
131  %t1 = lshr <3 x i32> %t0, %nbits
132  %t2 = and <3 x i32> %t1, %x
133  %t3 = add <3 x i32> %nbits, <i32 0, i32 undef, i32 0>
134  call void @use3xi32(<3 x i32> %t0)
135  call void @use3xi32(<3 x i32> %t1)
136  call void @use3xi32(<3 x i32> %t2)
137  call void @use3xi32(<3 x i32> %t3)
138  %t4 = shl <3 x i32> %t2, %t3
139  ret <3 x i32> %t4
140}
141
142; Commutativity
143
144declare i32 @gen32()
145
146define i32 @t5_commutativity0(i32 %nbits) {
147; CHECK-LABEL: @t5_commutativity0(
148; CHECK-NEXT:    [[X:%.*]] = call i32 @gen32()
149; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
150; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
151; CHECK-NEXT:    [[T2:%.*]] = and i32 [[X]], [[T1]]
152; CHECK-NEXT:    call void @use32(i32 [[T0]])
153; CHECK-NEXT:    call void @use32(i32 [[T1]])
154; CHECK-NEXT:    call void @use32(i32 [[T2]])
155; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[X]], [[NBITS]]
156; CHECK-NEXT:    ret i32 [[T3]]
157;
158  %x = call i32 @gen32()
159  %t0 = shl i32 -1, %nbits
160  %t1 = lshr i32 %t0, %nbits
161  %t2 = and i32 %x, %t1 ; swapped order
162  call void @use32(i32 %t0)
163  call void @use32(i32 %t1)
164  call void @use32(i32 %t2)
165  %t3 = shl i32 %t2, %nbits
166  ret i32 %t3
167}
168
169define i32 @t6_commutativity1(i32 %nbits0, i32 %nbits1) {
170; CHECK-LABEL: @t6_commutativity1(
171; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS0:%.*]]
172; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS0]]
173; CHECK-NEXT:    [[T2:%.*]] = shl i32 -1, [[NBITS1:%.*]]
174; CHECK-NEXT:    [[T3:%.*]] = lshr i32 [[T0]], [[NBITS1]]
175; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T3]], [[T1]]
176; CHECK-NEXT:    call void @use32(i32 [[T0]])
177; CHECK-NEXT:    call void @use32(i32 [[T1]])
178; CHECK-NEXT:    call void @use32(i32 [[T2]])
179; CHECK-NEXT:    call void @use32(i32 [[T3]])
180; CHECK-NEXT:    call void @use32(i32 [[T4]])
181; CHECK-NEXT:    [[T5:%.*]] = shl i32 [[T3]], [[NBITS0]]
182; CHECK-NEXT:    ret i32 [[T5]]
183;
184  %t0 = shl i32 -1, %nbits0
185  %t1 = lshr i32 %t0, %nbits0
186  %t2 = shl i32 -1, %nbits1
187  %t3 = lshr i32 %t0, %nbits1
188  %t4 = and i32 %t3, %t1 ; both hands of 'and' could be mask..
189  call void @use32(i32 %t0)
190  call void @use32(i32 %t1)
191  call void @use32(i32 %t2)
192  call void @use32(i32 %t3)
193  call void @use32(i32 %t4)
194  %t5 = shl i32 %t4, %nbits0
195  ret i32 %t5
196}
197define i32 @t7_commutativity2(i32 %nbits0, i32 %nbits1) {
198; CHECK-LABEL: @t7_commutativity2(
199; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS0:%.*]]
200; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS0]]
201; CHECK-NEXT:    [[T2:%.*]] = shl i32 -1, [[NBITS1:%.*]]
202; CHECK-NEXT:    [[T3:%.*]] = lshr i32 [[T0]], [[NBITS1]]
203; CHECK-NEXT:    [[T4:%.*]] = and i32 [[T3]], [[T1]]
204; CHECK-NEXT:    call void @use32(i32 [[T0]])
205; CHECK-NEXT:    call void @use32(i32 [[T1]])
206; CHECK-NEXT:    call void @use32(i32 [[T2]])
207; CHECK-NEXT:    call void @use32(i32 [[T3]])
208; CHECK-NEXT:    call void @use32(i32 [[T4]])
209; CHECK-NEXT:    [[T5:%.*]] = shl i32 [[T4]], [[NBITS1]]
210; CHECK-NEXT:    ret i32 [[T5]]
211;
212  %t0 = shl i32 -1, %nbits0
213  %t1 = lshr i32 %t0, %nbits0
214  %t2 = shl i32 -1, %nbits1
215  %t3 = lshr i32 %t0, %nbits1
216  %t4 = and i32 %t3, %t1 ; both hands of 'and' could be mask..
217  call void @use32(i32 %t0)
218  call void @use32(i32 %t1)
219  call void @use32(i32 %t2)
220  call void @use32(i32 %t3)
221  call void @use32(i32 %t4)
222  %t5 = shl i32 %t4, %nbits1
223  ret i32 %t5
224}
225
226; Fast-math flags. We must not preserve them!
227
228define i32 @t8_nuw(i32 %x, i32 %nbits) {
229; CHECK-LABEL: @t8_nuw(
230; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
231; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
232; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
233; CHECK-NEXT:    call void @use32(i32 [[T0]])
234; CHECK-NEXT:    call void @use32(i32 [[T1]])
235; CHECK-NEXT:    call void @use32(i32 [[T2]])
236; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[X]], [[NBITS]]
237; CHECK-NEXT:    ret i32 [[T3]]
238;
239  %t0 = shl i32 -1, %nbits
240  %t1 = lshr i32 %t0, %nbits
241  %t2 = and i32 %t1, %x
242  call void @use32(i32 %t0)
243  call void @use32(i32 %t1)
244  call void @use32(i32 %t2)
245  %t3 = shl nuw i32 %t2, %nbits
246  ret i32 %t3
247}
248
249define i32 @t9_nsw(i32 %x, i32 %nbits) {
250; CHECK-LABEL: @t9_nsw(
251; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
252; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
253; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
254; CHECK-NEXT:    call void @use32(i32 [[T0]])
255; CHECK-NEXT:    call void @use32(i32 [[T1]])
256; CHECK-NEXT:    call void @use32(i32 [[T2]])
257; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[X]], [[NBITS]]
258; CHECK-NEXT:    ret i32 [[T3]]
259;
260  %t0 = shl i32 -1, %nbits
261  %t1 = lshr i32 %t0, %nbits
262  %t2 = and i32 %t1, %x
263  call void @use32(i32 %t0)
264  call void @use32(i32 %t1)
265  call void @use32(i32 %t2)
266  %t3 = shl nsw i32 %t2, %nbits
267  ret i32 %t3
268}
269
270define i32 @t10_nuw_nsw(i32 %x, i32 %nbits) {
271; CHECK-LABEL: @t10_nuw_nsw(
272; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[NBITS:%.*]]
273; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS]]
274; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
275; CHECK-NEXT:    call void @use32(i32 [[T0]])
276; CHECK-NEXT:    call void @use32(i32 [[T1]])
277; CHECK-NEXT:    call void @use32(i32 [[T2]])
278; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[X]], [[NBITS]]
279; CHECK-NEXT:    ret i32 [[T3]]
280;
281  %t0 = shl i32 -1, %nbits
282  %t1 = lshr i32 %t0, %nbits
283  %t2 = and i32 %t1, %x
284  call void @use32(i32 %t0)
285  call void @use32(i32 %t1)
286  call void @use32(i32 %t2)
287  %t3 = shl nuw nsw i32 %t2, %nbits
288  ret i32 %t3
289}
290
291; Special test
292
293declare void @llvm.assume(i1 %cond)
294
295; We can't simplify (%shiftnbits-%masknbits) but we have an assumption.
296define i32 @t11_assume_uge(i32 %x, i32 %masknbits, i32 %shiftnbits) {
297; CHECK-LABEL: @t11_assume_uge(
298; CHECK-NEXT:    [[CMP:%.*]] = icmp uge i32 [[SHIFTNBITS:%.*]], [[MASKNBITS:%.*]]
299; CHECK-NEXT:    call void @llvm.assume(i1 [[CMP]])
300; CHECK-NEXT:    [[T0:%.*]] = shl i32 -1, [[MASKNBITS]]
301; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[MASKNBITS]]
302; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X:%.*]]
303; CHECK-NEXT:    call void @use32(i32 [[T0]])
304; CHECK-NEXT:    call void @use32(i32 [[T1]])
305; CHECK-NEXT:    call void @use32(i32 [[T2]])
306; CHECK-NEXT:    [[T4:%.*]] = shl i32 [[T2]], [[SHIFTNBITS]]
307; CHECK-NEXT:    ret i32 [[T4]]
308;
309  %cmp = icmp uge i32 %shiftnbits, %masknbits
310  call void @llvm.assume(i1 %cmp)
311  %t0 = shl i32 -1, %masknbits
312  %t1 = lshr i32 %t0, %masknbits
313  %t2 = and i32 %t1, %x
314  call void @use32(i32 %t0)
315  call void @use32(i32 %t1)
316  call void @use32(i32 %t2)
317  %t4 = shl i32 %t2, %shiftnbits
318  ret i32 %t4
319}
320
321; Negative tests
322
323define i32 @n12_different_shamts0(i32 %x, i32 %nbits0, i32 %nbits1) {
324; CHECK-LABEL: @n12_different_shamts0(
325; CHECK-NEXT:    [[T0:%.*]] = shl i32 [[X:%.*]], [[NBITS0:%.*]]
326; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS1:%.*]]
327; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X]]
328; CHECK-NEXT:    call void @use32(i32 [[T0]])
329; CHECK-NEXT:    call void @use32(i32 [[T1]])
330; CHECK-NEXT:    call void @use32(i32 [[T2]])
331; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[T2]], [[NBITS0]]
332; CHECK-NEXT:    ret i32 [[T3]]
333;
334  %t0 = shl i32 %x, %nbits0 ; different shift amts
335  %t1 = lshr i32 %t0, %nbits1 ; different shift amts
336  %t2 = and i32 %t1, %x
337  call void @use32(i32 %t0)
338  call void @use32(i32 %t1)
339  call void @use32(i32 %t2)
340  %t3 = shl i32 %t2, %nbits0
341  ret i32 %t3
342}
343
344define i32 @n13_different_shamts1(i32 %x, i32 %nbits0, i32 %nbits1) {
345; CHECK-LABEL: @n13_different_shamts1(
346; CHECK-NEXT:    [[T0:%.*]] = shl i32 [[X:%.*]], [[NBITS0:%.*]]
347; CHECK-NEXT:    [[T1:%.*]] = lshr i32 [[T0]], [[NBITS1:%.*]]
348; CHECK-NEXT:    [[T2:%.*]] = and i32 [[T1]], [[X]]
349; CHECK-NEXT:    call void @use32(i32 [[T0]])
350; CHECK-NEXT:    call void @use32(i32 [[T1]])
351; CHECK-NEXT:    call void @use32(i32 [[T2]])
352; CHECK-NEXT:    [[T3:%.*]] = shl i32 [[T2]], [[NBITS1]]
353; CHECK-NEXT:    ret i32 [[T3]]
354;
355  %t0 = shl i32 %x, %nbits0 ; different shift amts
356  %t1 = lshr i32 %t0, %nbits1 ; different shift amts
357  %t2 = and i32 %t1, %x
358  call void @use32(i32 %t0)
359  call void @use32(i32 %t1)
360  call void @use32(i32 %t2)
361  %t3 = shl i32 %t2, %nbits1
362  ret i32 %t3
363}
364