1; RUN: llc < %s -march=x86-64 -mattr=+bmi,+bmi2 | FileCheck %s
2
3declare i8 @llvm.cttz.i8(i8, i1) nounwind readnone
4declare i16 @llvm.cttz.i16(i16, i1) nounwind readnone
5declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
6declare i64 @llvm.cttz.i64(i64, i1) nounwind readnone
7
8define i8 @t1(i8 %x) nounwind  {
9  %tmp = tail call i8 @llvm.cttz.i8( i8 %x, i1 false )
10  ret i8 %tmp
11; CHECK-LABEL: t1:
12; CHECK: tzcntl
13}
14
15define i16 @t2(i16 %x) nounwind  {
16  %tmp = tail call i16 @llvm.cttz.i16( i16 %x, i1 false )
17  ret i16 %tmp
18; CHECK-LABEL: t2:
19; CHECK: tzcntw
20}
21
22define i32 @t3(i32 %x) nounwind  {
23  %tmp = tail call i32 @llvm.cttz.i32( i32 %x, i1 false )
24  ret i32 %tmp
25; CHECK-LABEL: t3:
26; CHECK: tzcntl
27}
28
29define i32 @tzcnt32_load(i32* %x) nounwind  {
30  %x1 = load i32* %x
31  %tmp = tail call i32 @llvm.cttz.i32(i32 %x1, i1 false )
32  ret i32 %tmp
33; CHECK-LABEL: tzcnt32_load:
34; CHECK: tzcntl ({{.*}})
35}
36
37define i64 @t4(i64 %x) nounwind  {
38  %tmp = tail call i64 @llvm.cttz.i64( i64 %x, i1 false )
39  ret i64 %tmp
40; CHECK-LABEL: t4:
41; CHECK: tzcntq
42}
43
44define i8 @t5(i8 %x) nounwind  {
45  %tmp = tail call i8 @llvm.cttz.i8( i8 %x, i1 true )
46  ret i8 %tmp
47; CHECK-LABEL: t5:
48; CHECK: tzcntl
49}
50
51define i16 @t6(i16 %x) nounwind  {
52  %tmp = tail call i16 @llvm.cttz.i16( i16 %x, i1 true )
53  ret i16 %tmp
54; CHECK-LABEL: t6:
55; CHECK: tzcntw
56}
57
58define i32 @t7(i32 %x) nounwind  {
59  %tmp = tail call i32 @llvm.cttz.i32( i32 %x, i1 true )
60  ret i32 %tmp
61; CHECK-LABEL: t7:
62; CHECK: tzcntl
63}
64
65define i64 @t8(i64 %x) nounwind  {
66  %tmp = tail call i64 @llvm.cttz.i64( i64 %x, i1 true )
67  ret i64 %tmp
68; CHECK-LABEL: t8:
69; CHECK: tzcntq
70}
71
72define i32 @andn32(i32 %x, i32 %y) nounwind readnone {
73  %tmp1 = xor i32 %x, -1
74  %tmp2 = and i32 %y, %tmp1
75  ret i32 %tmp2
76; CHECK-LABEL: andn32:
77; CHECK: andnl
78}
79
80define i32 @andn32_load(i32 %x, i32* %y) nounwind readnone {
81  %y1 = load i32* %y
82  %tmp1 = xor i32 %x, -1
83  %tmp2 = and i32 %y1, %tmp1
84  ret i32 %tmp2
85; CHECK-LABEL: andn32_load:
86; CHECK: andnl ({{.*}})
87}
88
89define i64 @andn64(i64 %x, i64 %y) nounwind readnone {
90  %tmp1 = xor i64 %x, -1
91  %tmp2 = and i64 %tmp1, %y
92  ret i64 %tmp2
93; CHECK-LABEL: andn64:
94; CHECK: andnq
95}
96
97define i32 @bextr32(i32 %x, i32 %y) nounwind readnone {
98  %tmp = tail call i32 @llvm.x86.bmi.bextr.32(i32 %x, i32 %y)
99  ret i32 %tmp
100; CHECK-LABEL: bextr32:
101; CHECK: bextrl
102}
103
104define i32 @bextr32_load(i32* %x, i32 %y) nounwind readnone {
105  %x1 = load i32* %x
106  %tmp = tail call i32 @llvm.x86.bmi.bextr.32(i32 %x1, i32 %y)
107  ret i32 %tmp
108; CHECK-LABEL: bextr32_load:
109; CHECK: bextrl {{.*}}, ({{.*}}), {{.*}}
110}
111
112declare i32 @llvm.x86.bmi.bextr.32(i32, i32) nounwind readnone
113
114define i32 @bextr32b(i32 %x) nounwind uwtable readnone ssp {
115  %1 = lshr i32 %x, 4
116  %2 = and i32 %1, 4095
117  ret i32 %2
118; CHECK-LABEL: bextr32b:
119; CHECK: bextrl
120}
121
122define i32 @bextr32b_load(i32* %x) nounwind uwtable readnone ssp {
123  %1 = load i32* %x
124  %2 = lshr i32 %1, 4
125  %3 = and i32 %2, 4095
126  ret i32 %3
127; CHECK-LABEL: bextr32b_load:
128; CHECK: bextrl {{.*}}, ({{.*}}), {{.*}}
129}
130
131define i64 @bextr64(i64 %x, i64 %y) nounwind readnone {
132  %tmp = tail call i64 @llvm.x86.bmi.bextr.64(i64 %x, i64 %y)
133  ret i64 %tmp
134; CHECK-LABEL: bextr64:
135; CHECK: bextrq
136}
137
138declare i64 @llvm.x86.bmi.bextr.64(i64, i64) nounwind readnone
139
140define i64 @bextr64b(i64 %x) nounwind uwtable readnone ssp {
141  %1 = lshr i64 %x, 4
142  %2 = and i64 %1, 4095
143  ret i64 %2
144; CHECK-LABEL: bextr64b:
145; CHECK: bextrq
146}
147
148define i32 @bzhi32(i32 %x, i32 %y) nounwind readnone {
149  %tmp = tail call i32 @llvm.x86.bmi.bzhi.32(i32 %x, i32 %y)
150  ret i32 %tmp
151; CHECK-LABEL: bzhi32:
152; CHECK: bzhil
153}
154
155define i32 @bzhi32_load(i32* %x, i32 %y) nounwind readnone {
156  %x1 = load i32* %x
157  %tmp = tail call i32 @llvm.x86.bmi.bzhi.32(i32 %x1, i32 %y)
158  ret i32 %tmp
159; CHECK-LABEL: bzhi32_load:
160; CHECK: bzhil {{.*}}, ({{.*}}), {{.*}}
161}
162
163declare i32 @llvm.x86.bmi.bzhi.32(i32, i32) nounwind readnone
164
165define i64 @bzhi64(i64 %x, i64 %y) nounwind readnone {
166  %tmp = tail call i64 @llvm.x86.bmi.bzhi.64(i64 %x, i64 %y)
167  ret i64 %tmp
168; CHECK-LABEL: bzhi64:
169; CHECK: bzhiq
170}
171
172declare i64 @llvm.x86.bmi.bzhi.64(i64, i64) nounwind readnone
173
174define i32 @bzhi32b(i32 %x, i8 zeroext %index) #0 {
175entry:
176  %conv = zext i8 %index to i32
177  %shl = shl i32 1, %conv
178  %sub = add nsw i32 %shl, -1
179  %and = and i32 %sub, %x
180  ret i32 %and
181; CHECK-LABEL: bzhi32b:
182; CHECK: bzhil
183}
184
185define i32 @bzhi32b_load(i32* %w, i8 zeroext %index) #0 {
186entry:
187  %x = load i32* %w
188  %conv = zext i8 %index to i32
189  %shl = shl i32 1, %conv
190  %sub = add nsw i32 %shl, -1
191  %and = and i32 %sub, %x
192  ret i32 %and
193; CHECK-LABEL: bzhi32b_load:
194; CHECK: bzhil {{.*}}, ({{.*}}), {{.*}}
195}
196
197define i32 @bzhi32c(i32 %x, i8 zeroext %index) #0 {
198entry:
199  %conv = zext i8 %index to i32
200  %shl = shl i32 1, %conv
201  %sub = add nsw i32 %shl, -1
202  %and = and i32 %x, %sub
203  ret i32 %and
204; CHECK-LABEL: bzhi32c:
205; CHECK: bzhil
206}
207
208define i64 @bzhi64b(i64 %x, i8 zeroext %index) #0 {
209entry:
210  %conv = zext i8 %index to i64
211  %shl = shl i64 1, %conv
212  %sub = add nsw i64 %shl, -1
213  %and = and i64 %x, %sub
214  ret i64 %and
215; CHECK-LABEL: bzhi64b:
216; CHECK: bzhiq
217}
218
219define i64 @bzhi64_constant_mask(i64 %x) #0 {
220entry:
221  %and = and i64 %x, 4611686018427387903
222  ret i64 %and
223; CHECK-LABEL: bzhi64_constant_mask:
224; CHECK: movb    $62, %al
225; CHECK: bzhiq   %rax, %r[[ARG1:di|cx]], %rax
226}
227
228define i64 @bzhi64_small_constant_mask(i64 %x) #0 {
229entry:
230  %and = and i64 %x, 2147483647
231  ret i64 %and
232; CHECK-LABEL: bzhi64_small_constant_mask:
233; CHECK: andq  $2147483647, %r[[ARG1]]
234}
235
236define i32 @blsi32(i32 %x) nounwind readnone {
237  %tmp = sub i32 0, %x
238  %tmp2 = and i32 %x, %tmp
239  ret i32 %tmp2
240; CHECK-LABEL: blsi32:
241; CHECK: blsil
242}
243
244define i32 @blsi32_load(i32* %x) nounwind readnone {
245  %x1 = load i32* %x
246  %tmp = sub i32 0, %x1
247  %tmp2 = and i32 %x1, %tmp
248  ret i32 %tmp2
249; CHECK-LABEL: blsi32_load:
250; CHECK: blsil ({{.*}})
251}
252
253define i64 @blsi64(i64 %x) nounwind readnone {
254  %tmp = sub i64 0, %x
255  %tmp2 = and i64 %tmp, %x
256  ret i64 %tmp2
257; CHECK-LABEL: blsi64:
258; CHECK: blsiq
259}
260
261define i32 @blsmsk32(i32 %x) nounwind readnone {
262  %tmp = sub i32 %x, 1
263  %tmp2 = xor i32 %x, %tmp
264  ret i32 %tmp2
265; CHECK-LABEL: blsmsk32:
266; CHECK: blsmskl
267}
268
269define i32 @blsmsk32_load(i32* %x) nounwind readnone {
270  %x1 = load i32* %x
271  %tmp = sub i32 %x1, 1
272  %tmp2 = xor i32 %x1, %tmp
273  ret i32 %tmp2
274; CHECK-LABEL: blsmsk32_load:
275; CHECK: blsmskl ({{.*}})
276}
277
278define i64 @blsmsk64(i64 %x) nounwind readnone {
279  %tmp = sub i64 %x, 1
280  %tmp2 = xor i64 %tmp, %x
281  ret i64 %tmp2
282; CHECK-LABEL: blsmsk64:
283; CHECK: blsmskq
284}
285
286define i32 @blsr32(i32 %x) nounwind readnone {
287  %tmp = sub i32 %x, 1
288  %tmp2 = and i32 %x, %tmp
289  ret i32 %tmp2
290; CHECK-LABEL: blsr32:
291; CHECK: blsrl
292}
293
294define i32 @blsr32_load(i32* %x) nounwind readnone {
295  %x1 = load i32* %x
296  %tmp = sub i32 %x1, 1
297  %tmp2 = and i32 %x1, %tmp
298  ret i32 %tmp2
299; CHECK-LABEL: blsr32_load:
300; CHECK: blsrl ({{.*}})
301}
302
303define i64 @blsr64(i64 %x) nounwind readnone {
304  %tmp = sub i64 %x, 1
305  %tmp2 = and i64 %tmp, %x
306  ret i64 %tmp2
307; CHECK-LABEL: blsr64:
308; CHECK: blsrq
309}
310
311define i32 @pdep32(i32 %x, i32 %y) nounwind readnone {
312  %tmp = tail call i32 @llvm.x86.bmi.pdep.32(i32 %x, i32 %y)
313  ret i32 %tmp
314; CHECK-LABEL: pdep32:
315; CHECK: pdepl
316}
317
318define i32 @pdep32_load(i32 %x, i32* %y) nounwind readnone {
319  %y1 = load i32* %y
320  %tmp = tail call i32 @llvm.x86.bmi.pdep.32(i32 %x, i32 %y1)
321  ret i32 %tmp
322; CHECK-LABEL: pdep32_load:
323; CHECK: pdepl ({{.*}})
324}
325
326declare i32 @llvm.x86.bmi.pdep.32(i32, i32) nounwind readnone
327
328define i64 @pdep64(i64 %x, i64 %y) nounwind readnone {
329  %tmp = tail call i64 @llvm.x86.bmi.pdep.64(i64 %x, i64 %y)
330  ret i64 %tmp
331; CHECK-LABEL: pdep64:
332; CHECK: pdepq
333}
334
335declare i64 @llvm.x86.bmi.pdep.64(i64, i64) nounwind readnone
336
337define i32 @pext32(i32 %x, i32 %y) nounwind readnone {
338  %tmp = tail call i32 @llvm.x86.bmi.pext.32(i32 %x, i32 %y)
339  ret i32 %tmp
340; CHECK-LABEL: pext32:
341; CHECK: pextl
342}
343
344define i32 @pext32_load(i32 %x, i32* %y) nounwind readnone {
345  %y1 = load i32* %y
346  %tmp = tail call i32 @llvm.x86.bmi.pext.32(i32 %x, i32 %y1)
347  ret i32 %tmp
348; CHECK-LABEL: pext32_load:
349; CHECK: pextl ({{.*}})
350}
351
352declare i32 @llvm.x86.bmi.pext.32(i32, i32) nounwind readnone
353
354define i64 @pext64(i64 %x, i64 %y) nounwind readnone {
355  %tmp = tail call i64 @llvm.x86.bmi.pext.64(i64 %x, i64 %y)
356  ret i64 %tmp
357; CHECK-LABEL: pext64:
358; CHECK: pextq
359}
360
361declare i64 @llvm.x86.bmi.pext.64(i64, i64) nounwind readnone
362
363