1; RUN: opt -instcombine -S < %s | FileCheck %s
2
3%overflow.result = type {i8, i1}
4%ov.result.32 = type { i32, i1 }
5
6
7declare %overflow.result @llvm.uadd.with.overflow.i8(i8, i8) nounwind readnone
8declare %overflow.result @llvm.umul.with.overflow.i8(i8, i8) nounwind readnone
9declare %ov.result.32 @llvm.sadd.with.overflow.i32(i32, i32) nounwind readnone
10declare %ov.result.32 @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone
11declare %ov.result.32 @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone
12declare %ov.result.32 @llvm.usub.with.overflow.i32(i32, i32) nounwind readnone
13declare %ov.result.32 @llvm.smul.with.overflow.i32(i32, i32) nounwind readnone
14declare %ov.result.32 @llvm.umul.with.overflow.i32(i32, i32) nounwind readnone
15declare double @llvm.powi.f64(double, i32) nounwind readonly
16declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
17declare i32 @llvm.ctlz.i32(i32, i1) nounwind readnone
18declare i32 @llvm.ctpop.i32(i32) nounwind readnone
19declare i8 @llvm.ctlz.i8(i8, i1) nounwind readnone
20
21define i8 @uaddtest1(i8 %A, i8 %B) {
22  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 %A, i8 %B)
23  %y = extractvalue %overflow.result %x, 0
24  ret i8 %y
25; CHECK-LABEL: @uaddtest1(
26; CHECK-NEXT: %y = add i8 %A, %B
27; CHECK-NEXT: ret i8 %y
28}
29
30define i8 @uaddtest2(i8 %A, i8 %B, i1* %overflowPtr) {
31  %and.A = and i8 %A, 127
32  %and.B = and i8 %B, 127
33  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 %and.A, i8 %and.B)
34  %y = extractvalue %overflow.result %x, 0
35  %z = extractvalue %overflow.result %x, 1
36  store i1 %z, i1* %overflowPtr
37  ret i8 %y
38; CHECK-LABEL: @uaddtest2(
39; CHECK-NEXT: %and.A = and i8 %A, 127
40; CHECK-NEXT: %and.B = and i8 %B, 127
41; CHECK-NEXT: %x = add nuw i8 %and.A, %and.B
42; CHECK-NEXT: store i1 false, i1* %overflowPtr
43; CHECK-NEXT: ret i8 %x
44}
45
46define i8 @uaddtest3(i8 %A, i8 %B, i1* %overflowPtr) {
47  %or.A = or i8 %A, -128
48  %or.B = or i8 %B, -128
49  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 %or.A, i8 %or.B)
50  %y = extractvalue %overflow.result %x, 0
51  %z = extractvalue %overflow.result %x, 1
52  store i1 %z, i1* %overflowPtr
53  ret i8 %y
54; CHECK-LABEL: @uaddtest3(
55; CHECK-NEXT: %or.A = or i8 %A, -128
56; CHECK-NEXT: %or.B = or i8 %B, -128
57; CHECK-NEXT: %x = add i8 %or.A, %or.B
58; CHECK-NEXT: store i1 true, i1* %overflowPtr
59; CHECK-NEXT: ret i8 %x
60}
61
62define i8 @uaddtest4(i8 %A, i1* %overflowPtr) {
63  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 undef, i8 %A)
64  %y = extractvalue %overflow.result %x, 0
65  %z = extractvalue %overflow.result %x, 1
66  store i1 %z, i1* %overflowPtr
67  ret i8 %y
68; CHECK-LABEL: @uaddtest4(
69; CHECK-NEXT: ret i8 undef
70}
71
72define i8 @uaddtest5(i8 %A, i1* %overflowPtr) {
73  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 0, i8 %A)
74  %y = extractvalue %overflow.result %x, 0
75  %z = extractvalue %overflow.result %x, 1
76  store i1 %z, i1* %overflowPtr
77  ret i8 %y
78; CHECK-LABEL: @uaddtest5(
79; CHECK: ret i8 %A
80}
81
82define i1 @uaddtest6(i8 %A, i8 %B) {
83  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 %A, i8 -4)
84  %z = extractvalue %overflow.result %x, 1
85  ret i1 %z
86; CHECK-LABEL: @uaddtest6(
87; CHECK-NEXT: %z = icmp ugt i8 %A, 3
88; CHECK-NEXT: ret i1 %z
89}
90
91define i8 @uaddtest7(i8 %A, i8 %B) {
92  %x = call %overflow.result @llvm.uadd.with.overflow.i8(i8 %A, i8 %B)
93  %z = extractvalue %overflow.result %x, 0
94  ret i8 %z
95; CHECK-LABEL: @uaddtest7(
96; CHECK-NEXT: %z = add i8 %A, %B
97; CHECK-NEXT: ret i8 %z
98}
99
100; PR20194
101define %ov.result.32 @saddtest_nsw(i8 %a, i8 %b) {
102  %A = sext i8 %a to i32
103  %B = sext i8 %b to i32
104  %x = call %ov.result.32 @llvm.sadd.with.overflow.i32(i32 %A, i32 %B)
105  ret %ov.result.32 %x
106; CHECK-LABEL: @saddtest_nsw
107; CHECK: %x = add nsw i32 %A, %B
108; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
109; CHECK-NEXT:  ret %ov.result.32 %1
110}
111
112define %ov.result.32 @uaddtest_nuw(i32 %a, i32 %b) {
113  %A = and i32 %a, 2147483647
114  %B = and i32 %b, 2147483647
115  %x = call %ov.result.32 @llvm.uadd.with.overflow.i32(i32 %A, i32 %B)
116  ret %ov.result.32 %x
117; CHECK-LABEL: @uaddtest_nuw
118; CHECK: %x = add nuw i32 %A, %B
119; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
120; CHECK-NEXT:  ret %ov.result.32 %1
121}
122
123define %ov.result.32 @ssubtest_nsw(i8 %a, i8 %b) {
124  %A = sext i8 %a to i32
125  %B = sext i8 %b to i32
126  %x = call %ov.result.32 @llvm.ssub.with.overflow.i32(i32 %A, i32 %B)
127  ret %ov.result.32 %x
128; CHECK-LABEL: @ssubtest_nsw
129; CHECK: %x = sub nsw i32 %A, %B
130; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
131; CHECK-NEXT:  ret %ov.result.32 %1
132}
133
134define %ov.result.32 @usubtest_nuw(i32 %a, i32 %b) {
135  %A = or i32 %a, 2147483648
136  %B = and i32 %b, 2147483647
137  %x = call %ov.result.32 @llvm.usub.with.overflow.i32(i32 %A, i32 %B)
138  ret %ov.result.32 %x
139; CHECK-LABEL: @usubtest_nuw
140; CHECK: %x = sub nuw i32 %A, %B
141; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
142; CHECK-NEXT:  ret %ov.result.32 %1
143}
144
145define %ov.result.32 @smultest1_nsw(i32 %a, i32 %b) {
146  %A = and i32 %a, 4095 ; 0xfff
147  %B = and i32 %b, 524287; 0x7ffff
148  %x = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %A, i32 %B)
149  ret %ov.result.32 %x
150; CHECK-LABEL: @smultest1_nsw
151; CHECK: %x = mul nuw nsw i32 %A, %B
152; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
153; CHECK-NEXT:  ret %ov.result.32 %1
154}
155
156define %ov.result.32 @smultest2_nsw(i32 %a, i32 %b) {
157  %A = ashr i32 %a, 16
158  %B = ashr i32 %b, 16
159  %x = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %A, i32 %B)
160  ret %ov.result.32 %x
161; CHECK-LABEL: @smultest2_nsw
162; CHECK: %x = mul nsw i32 %A, %B
163; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
164; CHECK-NEXT:  ret %ov.result.32 %1
165}
166
167define %ov.result.32 @smultest3_sw(i32 %a, i32 %b) {
168  %A = ashr i32 %a, 16
169  %B = ashr i32 %b, 15
170  %x = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %A, i32 %B)
171  ret %ov.result.32 %x
172; CHECK-LABEL: @smultest3_sw
173; CHECK: %x = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %A, i32 %B)
174; CHECK-NEXT:  ret %ov.result.32 %x
175}
176
177define %ov.result.32 @umultest_nuw(i32 %a, i32 %b) {
178  %A = and i32 %a, 65535 ; 0xffff
179  %B = and i32 %b, 65535 ; 0xffff
180  %x = call %ov.result.32 @llvm.umul.with.overflow.i32(i32 %A, i32 %B)
181  ret %ov.result.32 %x
182; CHECK-LABEL: @umultest_nuw
183; CHECK: %x = mul nuw i32 %A, %B
184; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
185; CHECK-NEXT:  ret %ov.result.32 %1
186}
187
188define i8 @umultest1(i8 %A, i1* %overflowPtr) {
189  %x = call %overflow.result @llvm.umul.with.overflow.i8(i8 0, i8 %A)
190  %y = extractvalue %overflow.result %x, 0
191  %z = extractvalue %overflow.result %x, 1
192  store i1 %z, i1* %overflowPtr
193  ret i8 %y
194; CHECK-LABEL: @umultest1(
195; CHECK-NEXT: store i1 false, i1* %overflowPtr
196; CHECK-NEXT: ret i8 0
197}
198
199define i8 @umultest2(i8 %A, i1* %overflowPtr) {
200  %x = call %overflow.result @llvm.umul.with.overflow.i8(i8 1, i8 %A)
201  %y = extractvalue %overflow.result %x, 0
202  %z = extractvalue %overflow.result %x, 1
203  store i1 %z, i1* %overflowPtr
204  ret i8 %y
205; CHECK-LABEL: @umultest2(
206; CHECK-NEXT: store i1 false, i1* %overflowPtr
207; CHECK-NEXT: ret i8 %A
208}
209
210define i32 @umultest3(i32 %n) nounwind {
211  %shr = lshr i32 %n, 2
212  %mul = call %ov.result.32 @llvm.umul.with.overflow.i32(i32 %shr, i32 3)
213  %ov = extractvalue %ov.result.32 %mul, 1
214  %res = extractvalue %ov.result.32 %mul, 0
215  %ret = select i1 %ov, i32 -1, i32 %res
216  ret i32 %ret
217; CHECK-LABEL: @umultest3(
218; CHECK-NEXT: shr
219; CHECK-NEXT: mul nuw
220; CHECK-NEXT: ret
221}
222
223define i32 @umultest4(i32 %n) nounwind {
224  %shr = lshr i32 %n, 1
225  %mul = call %ov.result.32 @llvm.umul.with.overflow.i32(i32 %shr, i32 4)
226  %ov = extractvalue %ov.result.32 %mul, 1
227  %res = extractvalue %ov.result.32 %mul, 0
228  %ret = select i1 %ov, i32 -1, i32 %res
229  ret i32 %ret
230; CHECK-LABEL: @umultest4(
231; CHECK: umul.with.overflow
232}
233
234define %ov.result.32 @umultest5(i32 %x, i32 %y) nounwind {
235  %or_x = or i32 %x, 2147483648
236  %or_y = or i32 %y, 2147483648
237  %mul = call %ov.result.32 @llvm.umul.with.overflow.i32(i32 %or_x, i32 %or_y)
238  ret %ov.result.32 %mul
239; CHECK-LABEL: @umultest5(
240; CHECK-NEXT: %[[or_x:.*]] = or i32 %x, -2147483648
241; CHECK-NEXT: %[[or_y:.*]] = or i32 %y, -2147483648
242; CHECK-NEXT: %[[mul:.*]] = mul i32 %[[or_x]], %[[or_y]]
243; CHECK-NEXT: %[[ret:.*]] = insertvalue %ov.result.32 { i32 undef, i1 true }, i32 %[[mul]], 0
244; CHECK-NEXT: ret %ov.result.32 %[[ret]]
245}
246
247define void @powi(double %V, double *%P) {
248entry:
249  %A = tail call double @llvm.powi.f64(double %V, i32 -1) nounwind
250  store volatile double %A, double* %P
251
252  %B = tail call double @llvm.powi.f64(double %V, i32 0) nounwind
253  store volatile double %B, double* %P
254
255  %C = tail call double @llvm.powi.f64(double %V, i32 1) nounwind
256  store volatile double %C, double* %P
257  ret void
258; CHECK-LABEL: @powi(
259; CHECK: %A = fdiv double 1.0{{.*}}, %V
260; CHECK: store volatile double %A,
261; CHECK: store volatile double 1.0
262; CHECK: store volatile double %V
263}
264
265define i32 @cttz(i32 %a) {
266entry:
267  %or = or i32 %a, 8
268  %and = and i32 %or, -8
269  %count = tail call i32 @llvm.cttz.i32(i32 %and, i1 true) nounwind readnone
270  ret i32 %count
271; CHECK-LABEL: @cttz(
272; CHECK-NEXT: entry:
273; CHECK-NEXT: ret i32 3
274}
275
276define i8 @ctlz(i8 %a) {
277entry:
278  %or = or i8 %a, 32
279  %and = and i8 %or, 63
280  %count = tail call i8 @llvm.ctlz.i8(i8 %and, i1 true) nounwind readnone
281  ret i8 %count
282; CHECK-LABEL: @ctlz(
283; CHECK-NEXT: entry:
284; CHECK-NEXT: ret i8 2
285}
286
287define void @cmp.simplify(i32 %a, i32 %b, i1* %c) {
288entry:
289  %lz = tail call i32 @llvm.ctlz.i32(i32 %a, i1 false) nounwind readnone
290  %lz.cmp = icmp eq i32 %lz, 32
291  store volatile i1 %lz.cmp, i1* %c
292  %tz = tail call i32 @llvm.cttz.i32(i32 %a, i1 false) nounwind readnone
293  %tz.cmp = icmp ne i32 %tz, 32
294  store volatile i1 %tz.cmp, i1* %c
295  %pop = tail call i32 @llvm.ctpop.i32(i32 %b) nounwind readnone
296  %pop.cmp = icmp eq i32 %pop, 0
297  store volatile i1 %pop.cmp, i1* %c
298  ret void
299; CHECK: @cmp.simplify
300; CHECK-NEXT: entry:
301; CHECK-NEXT: %lz.cmp = icmp eq i32 %a, 0
302; CHECK-NEXT: store volatile i1 %lz.cmp, i1* %c
303; CHECK-NEXT: %tz.cmp = icmp ne i32 %a, 0
304; CHECK-NEXT: store volatile i1 %tz.cmp, i1* %c
305; CHECK-NEXT: %pop.cmp = icmp eq i32 %b, 0
306; CHECK-NEXT: store volatile i1 %pop.cmp, i1* %c
307}
308
309define i32 @cttz_simplify1a(i32 %x) nounwind readnone ssp {
310  %tmp1 = tail call i32 @llvm.ctlz.i32(i32 %x, i1 false)
311  %shr3 = lshr i32 %tmp1, 5
312  ret i32 %shr3
313
314; CHECK-LABEL: @cttz_simplify1a(
315; CHECK: icmp eq i32 %x, 0
316; CHECK-NEXT: zext i1
317; CHECK-NEXT: ret i32
318}
319
320define i32 @cttz_simplify1b(i32 %x) nounwind readnone ssp {
321  %tmp1 = tail call i32 @llvm.ctlz.i32(i32 %x, i1 true)
322  %shr3 = lshr i32 %tmp1, 5
323  ret i32 %shr3
324
325; CHECK-LABEL: @cttz_simplify1b(
326; CHECK-NEXT: ret i32 0
327}
328
329define i32 @ctlz_undef(i32 %Value) nounwind {
330  %ctlz = call i32 @llvm.ctlz.i32(i32 0, i1 true)
331  ret i32 %ctlz
332
333; CHECK-LABEL: @ctlz_undef(
334; CHECK-NEXT: ret i32 undef
335}
336
337define i32 @cttz_undef(i32 %Value) nounwind {
338  %cttz = call i32 @llvm.cttz.i32(i32 0, i1 true)
339  ret i32 %cttz
340
341; CHECK-LABEL: @cttz_undef(
342; CHECK-NEXT: ret i32 undef
343}
344
345define i32 @ctlz_select(i32 %Value) nounwind {
346  %tobool = icmp ne i32 %Value, 0
347  %ctlz = call i32 @llvm.ctlz.i32(i32 %Value, i1 true)
348  %s = select i1 %tobool, i32 %ctlz, i32 32
349  ret i32 %s
350
351; CHECK-LABEL: @ctlz_select(
352; CHECK-NEXT: call i32 @llvm.ctlz.i32(i32 %Value, i1 false)
353; CHECK-NEXT: ret i32
354}
355
356define i32 @cttz_select(i32 %Value) nounwind {
357  %tobool = icmp ne i32 %Value, 0
358  %cttz = call i32 @llvm.cttz.i32(i32 %Value, i1 true)
359  %s = select i1 %tobool, i32 %cttz, i32 32
360  ret i32 %s
361
362; CHECK-LABEL: @cttz_select(
363; CHECK-NEXT: call i32 @llvm.cttz.i32(i32 %Value, i1 false)
364; CHECK-NEXT: ret i32
365}
366
367; CHECK-LABEL: @overflow_div_add(
368; CHECK: ret i1 false
369define i1 @overflow_div_add(i32 %v1, i32 %v2) nounwind {
370entry:
371  %div = sdiv i32 %v1, 2
372  %t = call %ov.result.32 @llvm.sadd.with.overflow.i32(i32 %div, i32 1)
373  %obit = extractvalue %ov.result.32 %t, 1
374  ret i1 %obit
375}
376
377; CHECK-LABEL: @overflow_div_sub(
378; CHECK: ret i1 false
379define i1 @overflow_div_sub(i32 %v1, i32 %v2) nounwind {
380entry:
381  ; Check cases where the known sign bits are larger than the word size.
382  %a = ashr i32 %v1, 18
383  %div = sdiv i32 %a, 65536
384  %t = call %ov.result.32 @llvm.ssub.with.overflow.i32(i32 %div, i32 1)
385  %obit = extractvalue %ov.result.32 %t, 1
386  ret i1 %obit
387}
388
389; CHECK-LABEL: @overflow_mod_mul(
390; CHECK: ret i1 false
391define i1 @overflow_mod_mul(i32 %v1, i32 %v2) nounwind {
392entry:
393  %rem = srem i32 %v1, 1000
394  %t = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %rem, i32 %rem)
395  %obit = extractvalue %ov.result.32 %t, 1
396  ret i1 %obit
397}
398
399; CHECK-LABEL: @overflow_mod_overflow_mul(
400; CHECK-NOT: ret i1 false
401define i1 @overflow_mod_overflow_mul(i32 %v1, i32 %v2) nounwind {
402entry:
403  %rem = srem i32 %v1, 65537
404  ; This may overflow because the result of the mul operands may be greater than 16bits
405  ; and the result greater than 32.
406  %t = call %ov.result.32 @llvm.smul.with.overflow.i32(i32 %rem, i32 %rem)
407  %obit = extractvalue %ov.result.32 %t, 1
408  ret i1 %obit
409}
410
411define %ov.result.32 @ssubtest_reorder(i8 %a) {
412  %A = sext i8 %a to i32
413  %x = call %ov.result.32 @llvm.ssub.with.overflow.i32(i32 0, i32 %A)
414  ret %ov.result.32 %x
415; CHECK-LABEL: @ssubtest_reorder
416; CHECK: %x = sub nsw i32 0, %A
417; CHECK-NEXT: %1 = insertvalue %ov.result.32 { i32 undef, i1 false }, i32 %x, 0
418; CHECK-NEXT:  ret %ov.result.32 %1
419}
420