| blob | df11a44626e3819632770fc469132bab31c58395 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X86
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=X64
5 ;; Use cttz to test if we properly prove never-zero. There is a very
6 ;; simple transform from cttz -> cttz_zero_undef if its operand is
7 ;; known never zero.
8 declare i32 @llvm.cttz.i32(i32, i1)
9 declare i32 @llvm.uadd.sat.i32(i32, i32)
10 declare i32 @llvm.umax.i32(i32, i32)
11 declare i32 @llvm.umin.i32(i32, i32)
12 declare i32 @llvm.smin.i32(i32, i32)
13 declare <4 x i32> @llvm.smin.v4i32(<4 x i32>, <4 x i32>)
14 declare i32 @llvm.smax.i32(i32, i32)
15 declare <4 x i32> @llvm.smax.v4i32(<4 x i32>, <4 x i32>)
16 declare i32 @llvm.bswap.i32(i32)
17 declare i32 @llvm.bitreverse.i32(i32)
18 declare i32 @llvm.ctpop.i32(i32)
19 declare <4 x i32> @llvm.ctpop.v4i32(<4 x i32>)
20 declare i32 @llvm.abs.i32(i32, i1)
21 declare i32 @llvm.fshl.i32(i32, i32, i32)
22 declare i32 @llvm.fshr.i32(i32, i32, i32)
24 define i32 @or_known_nonzero(i32 %x) {
25 ; X86-LABEL: or_known_nonzero:
26 ; X86: # %bb.0:
27 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
28 ; X86-NEXT: orl $1, %eax
29 ; X86-NEXT: rep bsfl %eax, %eax
30 ; X86-NEXT: retl
31 ;
32 ; X64-LABEL: or_known_nonzero:
33 ; X64: # %bb.0:
34 ; X64-NEXT: orl $1, %edi
35 ; X64-NEXT: rep bsfl %edi, %eax
36 ; X64-NEXT: retq
37 %z = or i32 %x, 1
38 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
39 ret i32 %r
40 }
42 define i32 @or_maybe_zero(i32 %x, i32 %y) {
43 ; X86-LABEL: or_maybe_zero:
44 ; X86: # %bb.0:
45 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
46 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
47 ; X86-NEXT: je .LBB1_1
48 ; X86-NEXT: # %bb.2: # %cond.false
49 ; X86-NEXT: rep bsfl %eax, %eax
50 ; X86-NEXT: retl
51 ; X86-NEXT: .LBB1_1:
52 ; X86-NEXT: movl $32, %eax
53 ; X86-NEXT: retl
54 ;
55 ; X64-LABEL: or_maybe_zero:
56 ; X64: # %bb.0:
57 ; X64-NEXT: orl %esi, %edi
58 ; X64-NEXT: je .LBB1_1
59 ; X64-NEXT: # %bb.2: # %cond.false
60 ; X64-NEXT: rep bsfl %edi, %eax
61 ; X64-NEXT: retq
62 ; X64-NEXT: .LBB1_1:
63 ; X64-NEXT: movl $32, %eax
64 ; X64-NEXT: retq
65 %z = or i32 %x, %y
66 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
67 ret i32 %r
68 }
70 define i32 @select_known_nonzero(i1 %c, i32 %x) {
71 ; X86-LABEL: select_known_nonzero:
72 ; X86: # %bb.0:
73 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
74 ; X86-NEXT: orl $1, %eax
75 ; X86-NEXT: testb $1, {{[0-9]+}}(%esp)
76 ; X86-NEXT: movl $122, %ecx
77 ; X86-NEXT: cmovnel %eax, %ecx
78 ; X86-NEXT: rep bsfl %ecx, %eax
79 ; X86-NEXT: retl
80 ;
81 ; X64-LABEL: select_known_nonzero:
82 ; X64: # %bb.0:
83 ; X64-NEXT: orl $1, %esi
84 ; X64-NEXT: testb $1, %dil
85 ; X64-NEXT: movl $122, %eax
86 ; X64-NEXT: cmovnel %esi, %eax
87 ; X64-NEXT: rep bsfl %eax, %eax
88 ; X64-NEXT: retq
89 %y = or i32 %x, 1
90 %z = select i1 %c, i32 %y, i32 122
91 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
92 ret i32 %r
93 }
95 define i32 @select_maybe_zero(i1 %c, i32 %x) {
96 ; X86-LABEL: select_maybe_zero:
97 ; X86: # %bb.0:
98 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
99 ; X86-NEXT: orl $1, %ecx
100 ; X86-NEXT: xorl %eax, %eax
101 ; X86-NEXT: testb $1, {{[0-9]+}}(%esp)
102 ; X86-NEXT: cmovnel %ecx, %eax
103 ; X86-NEXT: testl %eax, %eax
104 ; X86-NEXT: je .LBB3_1
105 ; X86-NEXT: # %bb.2: # %cond.false
106 ; X86-NEXT: rep bsfl %eax, %eax
107 ; X86-NEXT: retl
108 ; X86-NEXT: .LBB3_1:
109 ; X86-NEXT: movl $32, %eax
110 ; X86-NEXT: retl
111 ;
112 ; X64-LABEL: select_maybe_zero:
113 ; X64: # %bb.0:
114 ; X64-NEXT: orl $1, %esi
115 ; X64-NEXT: xorl %eax, %eax
116 ; X64-NEXT: testb $1, %dil
117 ; X64-NEXT: cmovnel %esi, %eax
118 ; X64-NEXT: testl %eax, %eax
119 ; X64-NEXT: je .LBB3_1
120 ; X64-NEXT: # %bb.2: # %cond.false
121 ; X64-NEXT: rep bsfl %eax, %eax
122 ; X64-NEXT: retq
123 ; X64-NEXT: .LBB3_1:
124 ; X64-NEXT: movl $32, %eax
125 ; X64-NEXT: retq
126 %y = or i32 %x, 1
127 %z = select i1 %c, i32 %y, i32 0
128 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
129 ret i32 %r
130 }
132 define i32 @shl_known_nonzero_1s_bit_set(i32 %x) {
133 ; X86-LABEL: shl_known_nonzero_1s_bit_set:
134 ; X86: # %bb.0:
135 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
136 ; X86-NEXT: movl $123, %eax
137 ; X86-NEXT: shll %cl, %eax
138 ; X86-NEXT: rep bsfl %eax, %eax
139 ; X86-NEXT: retl
140 ;
141 ; X64-LABEL: shl_known_nonzero_1s_bit_set:
142 ; X64: # %bb.0:
143 ; X64-NEXT: movl %edi, %ecx
144 ; X64-NEXT: movl $123, %eax
145 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
146 ; X64-NEXT: shll %cl, %eax
147 ; X64-NEXT: rep bsfl %eax, %eax
148 ; X64-NEXT: retq
149 %z = shl i32 123, %x
150 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
151 ret i32 %r
152 }
154 define i32 @shl_known_nonzero_nsw(i32 %x, i32 %yy) {
155 ; X86-LABEL: shl_known_nonzero_nsw:
156 ; X86: # %bb.0:
157 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
158 ; X86-NEXT: movl $256, %eax # imm = 0x100
159 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
160 ; X86-NEXT: shll %cl, %eax
161 ; X86-NEXT: rep bsfl %eax, %eax
162 ; X86-NEXT: retl
163 ;
164 ; X64-LABEL: shl_known_nonzero_nsw:
165 ; X64: # %bb.0:
166 ; X64-NEXT: movl %edi, %ecx
167 ; X64-NEXT: orl $256, %esi # imm = 0x100
168 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
169 ; X64-NEXT: shll %cl, %esi
170 ; X64-NEXT: rep bsfl %esi, %eax
171 ; X64-NEXT: retq
172 %y = or i32 %yy, 256
173 %z = shl nsw i32 %y, %x
174 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
175 ret i32 %r
176 }
178 define i32 @shl_known_nonzero_nuw(i32 %x, i32 %yy) {
179 ; X86-LABEL: shl_known_nonzero_nuw:
180 ; X86: # %bb.0:
181 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
182 ; X86-NEXT: movl $256, %eax # imm = 0x100
183 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
184 ; X86-NEXT: shll %cl, %eax
185 ; X86-NEXT: rep bsfl %eax, %eax
186 ; X86-NEXT: retl
187 ;
188 ; X64-LABEL: shl_known_nonzero_nuw:
189 ; X64: # %bb.0:
190 ; X64-NEXT: movl %edi, %ecx
191 ; X64-NEXT: orl $256, %esi # imm = 0x100
192 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
193 ; X64-NEXT: shll %cl, %esi
194 ; X64-NEXT: rep bsfl %esi, %eax
195 ; X64-NEXT: retq
196 %y = or i32 %yy, 256
197 %z = shl nuw i32 %y, %x
198 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
199 ret i32 %r
200 }
202 define i32 @shl_maybe_zero(i32 %x, i32 %y) {
203 ; X86-LABEL: shl_maybe_zero:
204 ; X86: # %bb.0:
205 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
206 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
207 ; X86-NEXT: shll %cl, %eax
208 ; X86-NEXT: testl %eax, %eax
209 ; X86-NEXT: je .LBB7_1
210 ; X86-NEXT: # %bb.2: # %cond.false
211 ; X86-NEXT: rep bsfl %eax, %eax
212 ; X86-NEXT: retl
213 ; X86-NEXT: .LBB7_1:
214 ; X86-NEXT: movl $32, %eax
215 ; X86-NEXT: retl
216 ;
217 ; X64-LABEL: shl_maybe_zero:
218 ; X64: # %bb.0:
219 ; X64-NEXT: movl %edi, %ecx
220 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
221 ; X64-NEXT: shll %cl, %esi
222 ; X64-NEXT: testl %esi, %esi
223 ; X64-NEXT: je .LBB7_1
224 ; X64-NEXT: # %bb.2: # %cond.false
225 ; X64-NEXT: rep bsfl %esi, %eax
226 ; X64-NEXT: retq
227 ; X64-NEXT: .LBB7_1:
228 ; X64-NEXT: movl $32, %eax
229 ; X64-NEXT: retq
230 %z = shl nuw nsw i32 %y, %x
231 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
232 ret i32 %r
233 }
235 define i32 @uaddsat_known_nonzero(i32 %x) {
236 ; X86-LABEL: uaddsat_known_nonzero:
237 ; X86: # %bb.0:
238 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
239 ; X86-NEXT: incl %eax
240 ; X86-NEXT: movl $-1, %ecx
241 ; X86-NEXT: cmovnel %eax, %ecx
242 ; X86-NEXT: rep bsfl %ecx, %eax
243 ; X86-NEXT: retl
244 ;
245 ; X64-LABEL: uaddsat_known_nonzero:
246 ; X64: # %bb.0:
247 ; X64-NEXT: incl %edi
248 ; X64-NEXT: movl $-1, %eax
249 ; X64-NEXT: cmovnel %edi, %eax
250 ; X64-NEXT: rep bsfl %eax, %eax
251 ; X64-NEXT: retq
252 %z = call i32 @llvm.uadd.sat.i32(i32 %x, i32 1)
253 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
254 ret i32 %r
255 }
257 define i32 @uaddsat_maybe_zero(i32 %x, i32 %y) {
258 ; X86-LABEL: uaddsat_maybe_zero:
259 ; X86: # %bb.0:
260 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
261 ; X86-NEXT: addl {{[0-9]+}}(%esp), %ecx
262 ; X86-NEXT: movl $-1, %eax
263 ; X86-NEXT: cmovael %ecx, %eax
264 ; X86-NEXT: testl %eax, %eax
265 ; X86-NEXT: je .LBB9_1
266 ; X86-NEXT: # %bb.2: # %cond.false
267 ; X86-NEXT: rep bsfl %eax, %eax
268 ; X86-NEXT: retl
269 ; X86-NEXT: .LBB9_1:
270 ; X86-NEXT: movl $32, %eax
271 ; X86-NEXT: retl
272 ;
273 ; X64-LABEL: uaddsat_maybe_zero:
274 ; X64: # %bb.0:
275 ; X64-NEXT: addl %esi, %edi
276 ; X64-NEXT: movl $-1, %eax
277 ; X64-NEXT: cmovael %edi, %eax
278 ; X64-NEXT: testl %eax, %eax
279 ; X64-NEXT: je .LBB9_1
280 ; X64-NEXT: # %bb.2: # %cond.false
281 ; X64-NEXT: rep bsfl %eax, %eax
282 ; X64-NEXT: retq
283 ; X64-NEXT: .LBB9_1:
284 ; X64-NEXT: movl $32, %eax
285 ; X64-NEXT: retq
286 %z = call i32 @llvm.uadd.sat.i32(i32 %x, i32 %y)
287 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
288 ret i32 %r
289 }
291 define i32 @umax_known_nonzero(i32 %x, i32 %y) {
292 ; X86-LABEL: umax_known_nonzero:
293 ; X86: # %bb.0:
294 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
295 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
296 ; X86-NEXT: movl $4, %edx
297 ; X86-NEXT: shll %cl, %edx
298 ; X86-NEXT: cmpl %edx, %eax
299 ; X86-NEXT: cmoval %eax, %edx
300 ; X86-NEXT: rep bsfl %edx, %eax
301 ; X86-NEXT: retl
302 ;
303 ; X64-LABEL: umax_known_nonzero:
304 ; X64: # %bb.0:
305 ; X64-NEXT: movl %esi, %ecx
306 ; X64-NEXT: movl $4, %eax
307 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
308 ; X64-NEXT: shll %cl, %eax
309 ; X64-NEXT: cmpl %eax, %edi
310 ; X64-NEXT: cmoval %edi, %eax
311 ; X64-NEXT: rep bsfl %eax, %eax
312 ; X64-NEXT: retq
313 %yy = shl nuw i32 4, %y
314 %z = call i32 @llvm.umax.i32(i32 %x, i32 %yy)
315 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
316 ret i32 %r
317 }
319 define i32 @umax_maybe_zero(i32 %x, i32 %y) {
320 ; X86-LABEL: umax_maybe_zero:
321 ; X86: # %bb.0:
322 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
323 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
324 ; X86-NEXT: cmpl %eax, %ecx
325 ; X86-NEXT: cmoval %ecx, %eax
326 ; X86-NEXT: testl %eax, %eax
327 ; X86-NEXT: je .LBB11_1
328 ; X86-NEXT: # %bb.2: # %cond.false
329 ; X86-NEXT: rep bsfl %eax, %eax
330 ; X86-NEXT: retl
331 ; X86-NEXT: .LBB11_1:
332 ; X86-NEXT: movl $32, %eax
333 ; X86-NEXT: retl
334 ;
335 ; X64-LABEL: umax_maybe_zero:
336 ; X64: # %bb.0:
337 ; X64-NEXT: cmpl %esi, %edi
338 ; X64-NEXT: cmoval %edi, %esi
339 ; X64-NEXT: testl %esi, %esi
340 ; X64-NEXT: je .LBB11_1
341 ; X64-NEXT: # %bb.2: # %cond.false
342 ; X64-NEXT: rep bsfl %esi, %eax
343 ; X64-NEXT: retq
344 ; X64-NEXT: .LBB11_1:
345 ; X64-NEXT: movl $32, %eax
346 ; X64-NEXT: retq
347 %z = call i32 @llvm.umax.i32(i32 %x, i32 %y)
348 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
349 ret i32 %r
350 }
352 define i32 @umin_known_nonzero(i32 %xx, i32 %yy) {
353 ; X86-LABEL: umin_known_nonzero:
354 ; X86: # %bb.0:
355 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
356 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
357 ; X86-NEXT: movl $4, %edx
358 ; X86-NEXT: shll %cl, %edx
359 ; X86-NEXT: addl $4, %eax
360 ; X86-NEXT: cmpl %eax, %edx
361 ; X86-NEXT: cmovbl %edx, %eax
362 ; X86-NEXT: rep bsfl %eax, %eax
363 ; X86-NEXT: retl
364 ;
365 ; X64-LABEL: umin_known_nonzero:
366 ; X64: # %bb.0:
367 ; X64-NEXT: movl %edi, %ecx
368 ; X64-NEXT: movl $4, %eax
369 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
370 ; X64-NEXT: shll %cl, %eax
371 ; X64-NEXT: addl $4, %esi
372 ; X64-NEXT: cmpl %esi, %eax
373 ; X64-NEXT: cmovbl %eax, %esi
374 ; X64-NEXT: rep bsfl %esi, %eax
375 ; X64-NEXT: retq
376 %x = shl nuw i32 4, %xx
377 %y = add nuw nsw i32 %yy, 4
378 %z = call i32 @llvm.umin.i32(i32 %x, i32 %y)
379 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
380 ret i32 %r
381 }
383 define i32 @umin_maybe_zero(i32 %x, i32 %y) {
384 ; X86-LABEL: umin_maybe_zero:
385 ; X86: # %bb.0:
386 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
387 ; X86-NEXT: cmpl $54, %ecx
388 ; X86-NEXT: movl $54, %eax
389 ; X86-NEXT: cmovbl %ecx, %eax
390 ; X86-NEXT: testl %eax, %eax
391 ; X86-NEXT: je .LBB13_1
392 ; X86-NEXT: # %bb.2: # %cond.false
393 ; X86-NEXT: rep bsfl %eax, %eax
394 ; X86-NEXT: retl
395 ; X86-NEXT: .LBB13_1:
396 ; X86-NEXT: movl $32, %eax
397 ; X86-NEXT: retl
398 ;
399 ; X64-LABEL: umin_maybe_zero:
400 ; X64: # %bb.0:
401 ; X64-NEXT: cmpl $54, %edi
402 ; X64-NEXT: movl $54, %eax
403 ; X64-NEXT: cmovbl %edi, %eax
404 ; X64-NEXT: testl %eax, %eax
405 ; X64-NEXT: je .LBB13_1
406 ; X64-NEXT: # %bb.2: # %cond.false
407 ; X64-NEXT: rep bsfl %eax, %eax
408 ; X64-NEXT: retq
409 ; X64-NEXT: .LBB13_1:
410 ; X64-NEXT: movl $32, %eax
411 ; X64-NEXT: retq
412 %z = call i32 @llvm.umin.i32(i32 %x, i32 54)
413 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
414 ret i32 %r
415 }
417 define i32 @smin_known_nonzero(i32 %xx, i32 %yy) {
418 ; X86-LABEL: smin_known_nonzero:
419 ; X86: # %bb.0:
420 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
421 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
422 ; X86-NEXT: movl $4, %edx
423 ; X86-NEXT: shll %cl, %edx
424 ; X86-NEXT: addl $4, %eax
425 ; X86-NEXT: cmpl %eax, %edx
426 ; X86-NEXT: cmovll %edx, %eax
427 ; X86-NEXT: rep bsfl %eax, %eax
428 ; X86-NEXT: retl
429 ;
430 ; X64-LABEL: smin_known_nonzero:
431 ; X64: # %bb.0:
432 ; X64-NEXT: movl %edi, %ecx
433 ; X64-NEXT: movl $4, %eax
434 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
435 ; X64-NEXT: shll %cl, %eax
436 ; X64-NEXT: addl $4, %esi
437 ; X64-NEXT: cmpl %esi, %eax
438 ; X64-NEXT: cmovll %eax, %esi
439 ; X64-NEXT: rep bsfl %esi, %eax
440 ; X64-NEXT: retq
441 %x = shl nuw i32 4, %xx
442 %y = add nuw nsw i32 %yy, 4
443 %z = call i32 @llvm.smin.i32(i32 %x, i32 %y)
444 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
445 ret i32 %r
446 }
448 define i32 @smin_known_zero(i32 %x, i32 %y) {
449 ; X86-LABEL: smin_known_zero:
450 ; X86: # %bb.0:
451 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
452 ; X86-NEXT: cmpl $-54, %eax
453 ; X86-NEXT: movl $-54, %ecx
454 ; X86-NEXT: cmovll %eax, %ecx
455 ; X86-NEXT: rep bsfl %ecx, %eax
456 ; X86-NEXT: retl
457 ;
458 ; X64-LABEL: smin_known_zero:
459 ; X64: # %bb.0:
460 ; X64-NEXT: cmpl $-54, %edi
461 ; X64-NEXT: movl $-54, %eax
462 ; X64-NEXT: cmovll %edi, %eax
463 ; X64-NEXT: rep bsfl %eax, %eax
464 ; X64-NEXT: retq
465 %z = call i32 @llvm.smin.i32(i32 %x, i32 -54)
466 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
467 ret i32 %r
468 }
470 define <4 x i32> @smin_known_zero_vec(<4 x i32> %x, <4 x i32> %y) {
471 ; X86-LABEL: smin_known_zero_vec:
472 ; X86: # %bb.0:
473 ; X86-NEXT: movdqa {{.*#+}} xmm1 = [4294967242,4294967273,4294967284,4294967295]
474 ; X86-NEXT: movdqa %xmm1, %xmm2
475 ; X86-NEXT: pcmpgtd %xmm0, %xmm2
476 ; X86-NEXT: pand %xmm2, %xmm0
477 ; X86-NEXT: pandn %xmm1, %xmm2
478 ; X86-NEXT: por %xmm2, %xmm0
479 ; X86-NEXT: pcmpeqd %xmm1, %xmm1
480 ; X86-NEXT: paddd %xmm0, %xmm1
481 ; X86-NEXT: pand %xmm1, %xmm0
482 ; X86-NEXT: pxor %xmm1, %xmm1
483 ; X86-NEXT: pcmpeqd %xmm1, %xmm0
484 ; X86-NEXT: psrld $31, %xmm0
485 ; X86-NEXT: retl
486 ;
487 ; X64-LABEL: smin_known_zero_vec:
488 ; X64: # %bb.0:
489 ; X64-NEXT: vpminsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
490 ; X64-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
491 ; X64-NEXT: vpaddd %xmm1, %xmm0, %xmm1
492 ; X64-NEXT: vpand %xmm1, %xmm0, %xmm0
493 ; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1
494 ; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
495 ; X64-NEXT: vpsrld $31, %xmm0, %xmm0
496 ; X64-NEXT: retq
497 %z = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %x, <4 x i32> <i32 -54, i32 -23, i32 -12, i32 -1>)
498 %r = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %z)
499 %3 = icmp eq <4 x i32> %r, <i32 1, i32 1, i32 1, i32 1>
500 %ret = zext <4 x i1> %3 to <4 x i32>
501 ret <4 x i32> %ret
502 }
504 define i32 @smin_maybe_zero(i32 %x, i32 %y) {
505 ; X86-LABEL: smin_maybe_zero:
506 ; X86: # %bb.0:
507 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
508 ; X86-NEXT: cmpl $54, %ecx
509 ; X86-NEXT: movl $54, %eax
510 ; X86-NEXT: cmovll %ecx, %eax
511 ; X86-NEXT: testl %eax, %eax
512 ; X86-NEXT: je .LBB17_1
513 ; X86-NEXT: # %bb.2: # %cond.false
514 ; X86-NEXT: rep bsfl %eax, %eax
515 ; X86-NEXT: retl
516 ; X86-NEXT: .LBB17_1:
517 ; X86-NEXT: movl $32, %eax
518 ; X86-NEXT: retl
519 ;
520 ; X64-LABEL: smin_maybe_zero:
521 ; X64: # %bb.0:
522 ; X64-NEXT: cmpl $54, %edi
523 ; X64-NEXT: movl $54, %eax
524 ; X64-NEXT: cmovll %edi, %eax
525 ; X64-NEXT: testl %eax, %eax
526 ; X64-NEXT: je .LBB17_1
527 ; X64-NEXT: # %bb.2: # %cond.false
528 ; X64-NEXT: rep bsfl %eax, %eax
529 ; X64-NEXT: retq
530 ; X64-NEXT: .LBB17_1:
531 ; X64-NEXT: movl $32, %eax
532 ; X64-NEXT: retq
533 %z = call i32 @llvm.smin.i32(i32 %x, i32 54)
534 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
535 ret i32 %r
536 }
538 define i32 @smax_known_nonzero(i32 %xx, i32 %yy) {
539 ; X86-LABEL: smax_known_nonzero:
540 ; X86: # %bb.0:
541 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
542 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
543 ; X86-NEXT: movl $4, %edx
544 ; X86-NEXT: shll %cl, %edx
545 ; X86-NEXT: addl $4, %eax
546 ; X86-NEXT: cmpl %eax, %edx
547 ; X86-NEXT: cmovgl %edx, %eax
548 ; X86-NEXT: rep bsfl %eax, %eax
549 ; X86-NEXT: retl
550 ;
551 ; X64-LABEL: smax_known_nonzero:
552 ; X64: # %bb.0:
553 ; X64-NEXT: movl %edi, %ecx
554 ; X64-NEXT: movl $4, %eax
555 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
556 ; X64-NEXT: shll %cl, %eax
557 ; X64-NEXT: addl $4, %esi
558 ; X64-NEXT: cmpl %esi, %eax
559 ; X64-NEXT: cmovgl %eax, %esi
560 ; X64-NEXT: rep bsfl %esi, %eax
561 ; X64-NEXT: retq
562 %x = shl nuw i32 4, %xx
563 %y = add nuw nsw i32 %yy, 4
564 %z = call i32 @llvm.smax.i32(i32 %x, i32 %y)
565 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
566 ret i32 %r
567 }
569 define i32 @smax_maybe_zero(i32 %x, i32 %y) {
570 ; X86-LABEL: smax_maybe_zero:
571 ; X86: # %bb.0:
572 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
573 ; X86-NEXT: cmpl $55, %eax
574 ; X86-NEXT: movl $54, %ecx
575 ; X86-NEXT: cmovgel %eax, %ecx
576 ; X86-NEXT: rep bsfl %ecx, %eax
577 ; X86-NEXT: retl
578 ;
579 ; X64-LABEL: smax_maybe_zero:
580 ; X64: # %bb.0:
581 ; X64-NEXT: cmpl $55, %edi
582 ; X64-NEXT: movl $54, %eax
583 ; X64-NEXT: cmovgel %edi, %eax
584 ; X64-NEXT: rep bsfl %eax, %eax
585 ; X64-NEXT: retq
586 %z = call i32 @llvm.smax.i32(i32 %x, i32 54)
587 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
588 ret i32 %r
589 }
591 define <4 x i32> @smax_known_zero_vec(<4 x i32> %x, <4 x i32> %y) {
592 ; X86-LABEL: smax_known_zero_vec:
593 ; X86: # %bb.0:
594 ; X86-NEXT: movdqa {{.*#+}} xmm1 = [54,23,12,1]
595 ; X86-NEXT: movdqa %xmm0, %xmm2
596 ; X86-NEXT: pcmpgtd %xmm1, %xmm2
597 ; X86-NEXT: pand %xmm2, %xmm0
598 ; X86-NEXT: pandn %xmm1, %xmm2
599 ; X86-NEXT: por %xmm2, %xmm0
600 ; X86-NEXT: pcmpeqd %xmm1, %xmm1
601 ; X86-NEXT: paddd %xmm0, %xmm1
602 ; X86-NEXT: pxor {{\.?LCPI[0-9]+_[0-9]+}}, %xmm1
603 ; X86-NEXT: pxor %xmm1, %xmm0
604 ; X86-NEXT: pcmpgtd %xmm1, %xmm0
605 ; X86-NEXT: psrld $31, %xmm0
606 ; X86-NEXT: retl
607 ;
608 ; X64-LABEL: smax_known_zero_vec:
609 ; X64: # %bb.0:
610 ; X64-NEXT: vpmaxsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
611 ; X64-NEXT: vpcmpeqd %xmm1, %xmm1, %xmm1
612 ; X64-NEXT: vpaddd %xmm1, %xmm0, %xmm1
613 ; X64-NEXT: vpxor %xmm1, %xmm0, %xmm0
614 ; X64-NEXT: vpminud %xmm1, %xmm0, %xmm1
615 ; X64-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
616 ; X64-NEXT: vpandn {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
617 ; X64-NEXT: retq
618 %z = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %x, <4 x i32> <i32 54, i32 23, i32 12, i32 1>)
619 %r = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %z)
620 %3 = icmp eq <4 x i32> %r, <i32 1, i32 1, i32 1, i32 1>
621 %ret = zext <4 x i1> %3 to <4 x i32>
622 ret <4 x i32> %ret
623 }
625 define i32 @smax_known_zero(i32 %x, i32 %y) {
626 ; X86-LABEL: smax_known_zero:
627 ; X86: # %bb.0:
628 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
629 ; X86-NEXT: testl %ecx, %ecx
630 ; X86-NEXT: movl $-1, %eax
631 ; X86-NEXT: cmovnsl %ecx, %eax
632 ; X86-NEXT: testl %eax, %eax
633 ; X86-NEXT: je .LBB21_1
634 ; X86-NEXT: # %bb.2: # %cond.false
635 ; X86-NEXT: rep bsfl %eax, %eax
636 ; X86-NEXT: retl
637 ; X86-NEXT: .LBB21_1:
638 ; X86-NEXT: movl $32, %eax
639 ; X86-NEXT: retl
640 ;
641 ; X64-LABEL: smax_known_zero:
642 ; X64: # %bb.0:
643 ; X64-NEXT: testl %edi, %edi
644 ; X64-NEXT: movl $-1, %eax
645 ; X64-NEXT: cmovnsl %edi, %eax
646 ; X64-NEXT: testl %eax, %eax
647 ; X64-NEXT: je .LBB21_1
648 ; X64-NEXT: # %bb.2: # %cond.false
649 ; X64-NEXT: rep bsfl %eax, %eax
650 ; X64-NEXT: retq
651 ; X64-NEXT: .LBB21_1:
652 ; X64-NEXT: movl $32, %eax
653 ; X64-NEXT: retq
654 %z = call i32 @llvm.smax.i32(i32 %x, i32 -1)
655 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
656 ret i32 %r
657 }
659 define i32 @rotr_known_nonzero(i32 %xx, i32 %y) {
660 ; X86-LABEL: rotr_known_nonzero:
661 ; X86: # %bb.0:
662 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
663 ; X86-NEXT: movl $256, %eax # imm = 0x100
664 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
665 ; X86-NEXT: rorl %cl, %eax
666 ; X86-NEXT: testl %eax, %eax
667 ; X86-NEXT: je .LBB22_1
668 ; X86-NEXT: # %bb.2: # %cond.false
669 ; X86-NEXT: rep bsfl %eax, %eax
670 ; X86-NEXT: retl
671 ; X86-NEXT: .LBB22_1:
672 ; X86-NEXT: movl $32, %eax
673 ; X86-NEXT: retl
674 ;
675 ; X64-LABEL: rotr_known_nonzero:
676 ; X64: # %bb.0:
677 ; X64-NEXT: movl %esi, %ecx
678 ; X64-NEXT: orl $256, %edi # imm = 0x100
679 ; X64-NEXT: movl %edi, %eax
680 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
681 ; X64-NEXT: rorl %cl, %eax
682 ; X64-NEXT: testl %edi, %edi
683 ; X64-NEXT: je .LBB22_1
684 ; X64-NEXT: # %bb.2: # %cond.false
685 ; X64-NEXT: rep bsfl %eax, %eax
686 ; X64-NEXT: retq
687 ; X64-NEXT: .LBB22_1:
688 ; X64-NEXT: movl $32, %eax
689 ; X64-NEXT: retq
690 %x = or i32 %xx, 256
691 %shr = lshr i32 %x, %y
692 %sub = sub i32 32, %y
693 %shl = shl i32 %x, %sub
694 %z = or i32 %shl, %shr
695 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
696 ret i32 %r
697 }
699 define i32 @rotr_maybe_zero(i32 %x, i32 %y) {
700 ; X86-LABEL: rotr_maybe_zero:
701 ; X86: # %bb.0:
702 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
703 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
704 ; X86-NEXT: rorl %cl, %eax
705 ; X86-NEXT: testl %eax, %eax
706 ; X86-NEXT: je .LBB23_1
707 ; X86-NEXT: # %bb.2: # %cond.false
708 ; X86-NEXT: rep bsfl %eax, %eax
709 ; X86-NEXT: retl
710 ; X86-NEXT: .LBB23_1:
711 ; X86-NEXT: movl $32, %eax
712 ; X86-NEXT: retl
713 ;
714 ; X64-LABEL: rotr_maybe_zero:
715 ; X64: # %bb.0:
716 ; X64-NEXT: movl %esi, %ecx
717 ; X64-NEXT: movl %edi, %eax
718 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
719 ; X64-NEXT: rorl %cl, %eax
720 ; X64-NEXT: testl %edi, %edi
721 ; X64-NEXT: je .LBB23_1
722 ; X64-NEXT: # %bb.2: # %cond.false
723 ; X64-NEXT: rep bsfl %eax, %eax
724 ; X64-NEXT: retq
725 ; X64-NEXT: .LBB23_1:
726 ; X64-NEXT: movl $32, %eax
727 ; X64-NEXT: retq
728 %shr = lshr i32 %x, %y
729 %sub = sub i32 32, %y
730 %shl = shl i32 %x, %sub
731 %z = or i32 %shl, %shr
732 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
733 ret i32 %r
734 }
736 define i32 @rotr_with_fshr_known_nonzero(i32 %xx, i32 %y) {
737 ; X86-LABEL: rotr_with_fshr_known_nonzero:
738 ; X86: # %bb.0:
739 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
740 ; X86-NEXT: movl $256, %eax # imm = 0x100
741 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
742 ; X86-NEXT: rorl %cl, %eax
743 ; X86-NEXT: rep bsfl %eax, %eax
744 ; X86-NEXT: retl
745 ;
746 ; X64-LABEL: rotr_with_fshr_known_nonzero:
747 ; X64: # %bb.0:
748 ; X64-NEXT: movl %esi, %ecx
749 ; X64-NEXT: orl $256, %edi # imm = 0x100
750 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
751 ; X64-NEXT: rorl %cl, %edi
752 ; X64-NEXT: rep bsfl %edi, %eax
753 ; X64-NEXT: retq
754 %x = or i32 %xx, 256
755 %z = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %y)
756 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
757 ret i32 %r
758 }
760 define i32 @rotr_with_fshr_maybe_zero(i32 %x, i32 %y) {
761 ; X86-LABEL: rotr_with_fshr_maybe_zero:
762 ; X86: # %bb.0:
763 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
764 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
765 ; X86-NEXT: rorl %cl, %eax
766 ; X86-NEXT: testl %eax, %eax
767 ; X86-NEXT: je .LBB25_1
768 ; X86-NEXT: # %bb.2: # %cond.false
769 ; X86-NEXT: rep bsfl %eax, %eax
770 ; X86-NEXT: retl
771 ; X86-NEXT: .LBB25_1:
772 ; X86-NEXT: movl $32, %eax
773 ; X86-NEXT: retl
774 ;
775 ; X64-LABEL: rotr_with_fshr_maybe_zero:
776 ; X64: # %bb.0:
777 ; X64-NEXT: movl %esi, %ecx
778 ; X64-NEXT: movl %edi, %eax
779 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
780 ; X64-NEXT: rorl %cl, %eax
781 ; X64-NEXT: testl %edi, %edi
782 ; X64-NEXT: je .LBB25_1
783 ; X64-NEXT: # %bb.2: # %cond.false
784 ; X64-NEXT: rep bsfl %eax, %eax
785 ; X64-NEXT: retq
786 ; X64-NEXT: .LBB25_1:
787 ; X64-NEXT: movl $32, %eax
788 ; X64-NEXT: retq
789 %z = call i32 @llvm.fshr.i32(i32 %x, i32 %x, i32 %y)
790 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
791 ret i32 %r
792 }
794 define i32 @rotl_known_nonzero(i32 %xx, i32 %y) {
795 ; X86-LABEL: rotl_known_nonzero:
796 ; X86: # %bb.0:
797 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
798 ; X86-NEXT: movl $256, %eax # imm = 0x100
799 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
800 ; X86-NEXT: roll %cl, %eax
801 ; X86-NEXT: testl %eax, %eax
802 ; X86-NEXT: je .LBB26_1
803 ; X86-NEXT: # %bb.2: # %cond.false
804 ; X86-NEXT: rep bsfl %eax, %eax
805 ; X86-NEXT: retl
806 ; X86-NEXT: .LBB26_1:
807 ; X86-NEXT: movl $32, %eax
808 ; X86-NEXT: retl
809 ;
810 ; X64-LABEL: rotl_known_nonzero:
811 ; X64: # %bb.0:
812 ; X64-NEXT: movl %esi, %ecx
813 ; X64-NEXT: orl $256, %edi # imm = 0x100
814 ; X64-NEXT: movl %edi, %eax
815 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
816 ; X64-NEXT: roll %cl, %eax
817 ; X64-NEXT: testl %edi, %edi
818 ; X64-NEXT: je .LBB26_1
819 ; X64-NEXT: # %bb.2: # %cond.false
820 ; X64-NEXT: rep bsfl %eax, %eax
821 ; X64-NEXT: retq
822 ; X64-NEXT: .LBB26_1:
823 ; X64-NEXT: movl $32, %eax
824 ; X64-NEXT: retq
825 %x = or i32 %xx, 256
826 %shl = shl i32 %x, %y
827 %sub = sub i32 32, %y
828 %shr = lshr i32 %x, %sub
829 %z = or i32 %shr, %shl
830 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
831 ret i32 %r
832 }
834 define i32 @rotl_maybe_zero(i32 %x, i32 %y) {
835 ; X86-LABEL: rotl_maybe_zero:
836 ; X86: # %bb.0:
837 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
838 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
839 ; X86-NEXT: roll %cl, %eax
840 ; X86-NEXT: testl %eax, %eax
841 ; X86-NEXT: je .LBB27_1
842 ; X86-NEXT: # %bb.2: # %cond.false
843 ; X86-NEXT: rep bsfl %eax, %eax
844 ; X86-NEXT: retl
845 ; X86-NEXT: .LBB27_1:
846 ; X86-NEXT: movl $32, %eax
847 ; X86-NEXT: retl
848 ;
849 ; X64-LABEL: rotl_maybe_zero:
850 ; X64: # %bb.0:
851 ; X64-NEXT: movl %esi, %ecx
852 ; X64-NEXT: movl %edi, %eax
853 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
854 ; X64-NEXT: roll %cl, %eax
855 ; X64-NEXT: testl %edi, %edi
856 ; X64-NEXT: je .LBB27_1
857 ; X64-NEXT: # %bb.2: # %cond.false
858 ; X64-NEXT: rep bsfl %eax, %eax
859 ; X64-NEXT: retq
860 ; X64-NEXT: .LBB27_1:
861 ; X64-NEXT: movl $32, %eax
862 ; X64-NEXT: retq
863 %shl = shl i32 %x, %y
864 %sub = sub i32 32, %y
865 %shr = lshr i32 %x, %sub
866 %z = or i32 %shr, %shl
867 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
868 ret i32 %r
869 }
871 define i32 @rotl_with_fshl_known_nonzero(i32 %xx, i32 %y) {
872 ; X86-LABEL: rotl_with_fshl_known_nonzero:
873 ; X86: # %bb.0:
874 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
875 ; X86-NEXT: movl $256, %eax # imm = 0x100
876 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
877 ; X86-NEXT: roll %cl, %eax
878 ; X86-NEXT: rep bsfl %eax, %eax
879 ; X86-NEXT: retl
880 ;
881 ; X64-LABEL: rotl_with_fshl_known_nonzero:
882 ; X64: # %bb.0:
883 ; X64-NEXT: movl %esi, %ecx
884 ; X64-NEXT: orl $256, %edi # imm = 0x100
885 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
886 ; X64-NEXT: roll %cl, %edi
887 ; X64-NEXT: rep bsfl %edi, %eax
888 ; X64-NEXT: retq
889 %x = or i32 %xx, 256
890 %z = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %y)
891 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
892 ret i32 %r
893 }
895 define i32 @rotl_with_fshl_maybe_zero(i32 %x, i32 %y) {
896 ; X86-LABEL: rotl_with_fshl_maybe_zero:
897 ; X86: # %bb.0:
898 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
899 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
900 ; X86-NEXT: roll %cl, %eax
901 ; X86-NEXT: testl %eax, %eax
902 ; X86-NEXT: je .LBB29_1
903 ; X86-NEXT: # %bb.2: # %cond.false
904 ; X86-NEXT: rep bsfl %eax, %eax
905 ; X86-NEXT: retl
906 ; X86-NEXT: .LBB29_1:
907 ; X86-NEXT: movl $32, %eax
908 ; X86-NEXT: retl
909 ;
910 ; X64-LABEL: rotl_with_fshl_maybe_zero:
911 ; X64: # %bb.0:
912 ; X64-NEXT: movl %esi, %ecx
913 ; X64-NEXT: movl %edi, %eax
914 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
915 ; X64-NEXT: roll %cl, %eax
916 ; X64-NEXT: testl %edi, %edi
917 ; X64-NEXT: je .LBB29_1
918 ; X64-NEXT: # %bb.2: # %cond.false
919 ; X64-NEXT: rep bsfl %eax, %eax
920 ; X64-NEXT: retq
921 ; X64-NEXT: .LBB29_1:
922 ; X64-NEXT: movl $32, %eax
923 ; X64-NEXT: retq
924 %z = call i32 @llvm.fshl.i32(i32 %x, i32 %x, i32 %y)
925 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
926 ret i32 %r
927 }
929 define i32 @sra_known_nonzero_sign_bit_set(i32 %x) {
930 ; X86-LABEL: sra_known_nonzero_sign_bit_set:
931 ; X86: # %bb.0:
932 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
933 ; X86-NEXT: movl $-2147360405, %eax # imm = 0x8001E16B
934 ; X86-NEXT: sarl %cl, %eax
935 ; X86-NEXT: rep bsfl %eax, %eax
936 ; X86-NEXT: retl
937 ;
938 ; X64-LABEL: sra_known_nonzero_sign_bit_set:
939 ; X64: # %bb.0:
940 ; X64-NEXT: movl %edi, %ecx
941 ; X64-NEXT: movl $-2147360405, %eax # imm = 0x8001E16B
942 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
943 ; X64-NEXT: sarl %cl, %eax
944 ; X64-NEXT: rep bsfl %eax, %eax
945 ; X64-NEXT: retq
946 %z = ashr i32 2147606891, %x
947 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
948 ret i32 %r
949 }
951 define i32 @sra_known_nonzero_exact(i32 %x, i32 %yy) {
952 ; X86-LABEL: sra_known_nonzero_exact:
953 ; X86: # %bb.0:
954 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
955 ; X86-NEXT: movl $256, %eax # imm = 0x100
956 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
957 ; X86-NEXT: sarl %cl, %eax
958 ; X86-NEXT: rep bsfl %eax, %eax
959 ; X86-NEXT: retl
960 ;
961 ; X64-LABEL: sra_known_nonzero_exact:
962 ; X64: # %bb.0:
963 ; X64-NEXT: movl %edi, %ecx
964 ; X64-NEXT: orl $256, %esi # imm = 0x100
965 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
966 ; X64-NEXT: sarl %cl, %esi
967 ; X64-NEXT: rep bsfl %esi, %eax
968 ; X64-NEXT: retq
969 %y = or i32 %yy, 256
970 %z = ashr exact i32 %y, %x
971 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
972 ret i32 %r
973 }
975 define i32 @sra_maybe_zero(i32 %x, i32 %y) {
976 ; X86-LABEL: sra_maybe_zero:
977 ; X86: # %bb.0:
978 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
979 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
980 ; X86-NEXT: sarl %cl, %eax
981 ; X86-NEXT: testl %eax, %eax
982 ; X86-NEXT: je .LBB32_1
983 ; X86-NEXT: # %bb.2: # %cond.false
984 ; X86-NEXT: rep bsfl %eax, %eax
985 ; X86-NEXT: retl
986 ; X86-NEXT: .LBB32_1:
987 ; X86-NEXT: movl $32, %eax
988 ; X86-NEXT: retl
989 ;
990 ; X64-LABEL: sra_maybe_zero:
991 ; X64: # %bb.0:
992 ; X64-NEXT: movl %edi, %ecx
993 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
994 ; X64-NEXT: sarl %cl, %esi
995 ; X64-NEXT: testl %esi, %esi
996 ; X64-NEXT: je .LBB32_1
997 ; X64-NEXT: # %bb.2: # %cond.false
998 ; X64-NEXT: rep bsfl %esi, %eax
999 ; X64-NEXT: retq
1000 ; X64-NEXT: .LBB32_1:
1001 ; X64-NEXT: movl $32, %eax
1002 ; X64-NEXT: retq
1003 %z = ashr exact i32 %y, %x
1004 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1005 ret i32 %r
1006 }
1008 define i32 @srl_known_nonzero_sign_bit_set(i32 %x) {
1009 ; X86-LABEL: srl_known_nonzero_sign_bit_set:
1010 ; X86: # %bb.0:
1011 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1012 ; X86-NEXT: movl $-2147360405, %eax # imm = 0x8001E16B
1013 ; X86-NEXT: shrl %cl, %eax
1014 ; X86-NEXT: rep bsfl %eax, %eax
1015 ; X86-NEXT: retl
1016 ;
1017 ; X64-LABEL: srl_known_nonzero_sign_bit_set:
1018 ; X64: # %bb.0:
1019 ; X64-NEXT: movl %edi, %ecx
1020 ; X64-NEXT: movl $-2147360405, %eax # imm = 0x8001E16B
1021 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1022 ; X64-NEXT: shrl %cl, %eax
1023 ; X64-NEXT: rep bsfl %eax, %eax
1024 ; X64-NEXT: retq
1025 %z = lshr i32 2147606891, %x
1026 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1027 ret i32 %r
1028 }
1030 define i32 @srl_known_nonzero_exact(i32 %x, i32 %yy) {
1031 ; X86-LABEL: srl_known_nonzero_exact:
1032 ; X86: # %bb.0:
1033 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1034 ; X86-NEXT: movl $256, %eax # imm = 0x100
1035 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
1036 ; X86-NEXT: shrl %cl, %eax
1037 ; X86-NEXT: rep bsfl %eax, %eax
1038 ; X86-NEXT: retl
1039 ;
1040 ; X64-LABEL: srl_known_nonzero_exact:
1041 ; X64: # %bb.0:
1042 ; X64-NEXT: movl %edi, %ecx
1043 ; X64-NEXT: orl $256, %esi # imm = 0x100
1044 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1045 ; X64-NEXT: shrl %cl, %esi
1046 ; X64-NEXT: rep bsfl %esi, %eax
1047 ; X64-NEXT: retq
1048 %y = or i32 %yy, 256
1049 %z = lshr exact i32 %y, %x
1050 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1051 ret i32 %r
1052 }
1054 define i32 @srl_maybe_zero(i32 %x, i32 %y) {
1055 ; X86-LABEL: srl_maybe_zero:
1056 ; X86: # %bb.0:
1057 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1058 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1059 ; X86-NEXT: shrl %cl, %eax
1060 ; X86-NEXT: testl %eax, %eax
1061 ; X86-NEXT: je .LBB35_1
1062 ; X86-NEXT: # %bb.2: # %cond.false
1063 ; X86-NEXT: rep bsfl %eax, %eax
1064 ; X86-NEXT: retl
1065 ; X86-NEXT: .LBB35_1:
1066 ; X86-NEXT: movl $32, %eax
1067 ; X86-NEXT: retl
1068 ;
1069 ; X64-LABEL: srl_maybe_zero:
1070 ; X64: # %bb.0:
1071 ; X64-NEXT: movl %edi, %ecx
1072 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1073 ; X64-NEXT: shrl %cl, %esi
1074 ; X64-NEXT: testl %esi, %esi
1075 ; X64-NEXT: je .LBB35_1
1076 ; X64-NEXT: # %bb.2: # %cond.false
1077 ; X64-NEXT: rep bsfl %esi, %eax
1078 ; X64-NEXT: retq
1079 ; X64-NEXT: .LBB35_1:
1080 ; X64-NEXT: movl $32, %eax
1081 ; X64-NEXT: retq
1082 %z = lshr exact i32 %y, %x
1083 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1084 ret i32 %r
1085 }
1087 define i32 @udiv_known_nonzero(i32 %xx, i32 %y) {
1088 ; X86-LABEL: udiv_known_nonzero:
1089 ; X86: # %bb.0:
1090 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1091 ; X86-NEXT: orl $64, %eax
1092 ; X86-NEXT: xorl %edx, %edx
1093 ; X86-NEXT: divl {{[0-9]+}}(%esp)
1094 ; X86-NEXT: rep bsfl %eax, %eax
1095 ; X86-NEXT: retl
1096 ;
1097 ; X64-LABEL: udiv_known_nonzero:
1098 ; X64: # %bb.0:
1099 ; X64-NEXT: movl %edi, %eax
1100 ; X64-NEXT: orl $64, %eax
1101 ; X64-NEXT: xorl %edx, %edx
1102 ; X64-NEXT: divl %esi
1103 ; X64-NEXT: rep bsfl %eax, %eax
1104 ; X64-NEXT: retq
1105 %x = or i32 %xx, 64
1106 %z = udiv exact i32 %x, %y
1107 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1108 ret i32 %r
1109 }
1111 define i32 @udiv_maybe_zero(i32 %x, i32 %y) {
1112 ; X86-LABEL: udiv_maybe_zero:
1113 ; X86: # %bb.0:
1114 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1115 ; X86-NEXT: xorl %edx, %edx
1116 ; X86-NEXT: divl {{[0-9]+}}(%esp)
1117 ; X86-NEXT: testl %eax, %eax
1118 ; X86-NEXT: je .LBB37_1
1119 ; X86-NEXT: # %bb.2: # %cond.false
1120 ; X86-NEXT: rep bsfl %eax, %eax
1121 ; X86-NEXT: retl
1122 ; X86-NEXT: .LBB37_1:
1123 ; X86-NEXT: movl $32, %eax
1124 ; X86-NEXT: retl
1125 ;
1126 ; X64-LABEL: udiv_maybe_zero:
1127 ; X64: # %bb.0:
1128 ; X64-NEXT: movl %edi, %eax
1129 ; X64-NEXT: xorl %edx, %edx
1130 ; X64-NEXT: divl %esi
1131 ; X64-NEXT: testl %eax, %eax
1132 ; X64-NEXT: je .LBB37_1
1133 ; X64-NEXT: # %bb.2: # %cond.false
1134 ; X64-NEXT: rep bsfl %eax, %eax
1135 ; X64-NEXT: retq
1136 ; X64-NEXT: .LBB37_1:
1137 ; X64-NEXT: movl $32, %eax
1138 ; X64-NEXT: retq
1139 %z = udiv exact i32 %x, %y
1140 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1141 ret i32 %r
1142 }
1144 define i32 @sdiv_known_nonzero(i32 %xx, i32 %y) {
1145 ; X86-LABEL: sdiv_known_nonzero:
1146 ; X86: # %bb.0:
1147 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1148 ; X86-NEXT: orl $64, %eax
1149 ; X86-NEXT: cltd
1150 ; X86-NEXT: idivl {{[0-9]+}}(%esp)
1151 ; X86-NEXT: rep bsfl %eax, %eax
1152 ; X86-NEXT: retl
1153 ;
1154 ; X64-LABEL: sdiv_known_nonzero:
1155 ; X64: # %bb.0:
1156 ; X64-NEXT: movl %edi, %eax
1157 ; X64-NEXT: orl $64, %eax
1158 ; X64-NEXT: cltd
1159 ; X64-NEXT: idivl %esi
1160 ; X64-NEXT: rep bsfl %eax, %eax
1161 ; X64-NEXT: retq
1162 %x = or i32 %xx, 64
1163 %z = sdiv exact i32 %x, %y
1164 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1165 ret i32 %r
1166 }
1168 define i32 @sdiv_maybe_zero(i32 %x, i32 %y) {
1169 ; X86-LABEL: sdiv_maybe_zero:
1170 ; X86: # %bb.0:
1171 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1172 ; X86-NEXT: cltd
1173 ; X86-NEXT: idivl {{[0-9]+}}(%esp)
1174 ; X86-NEXT: testl %eax, %eax
1175 ; X86-NEXT: je .LBB39_1
1176 ; X86-NEXT: # %bb.2: # %cond.false
1177 ; X86-NEXT: rep bsfl %eax, %eax
1178 ; X86-NEXT: retl
1179 ; X86-NEXT: .LBB39_1:
1180 ; X86-NEXT: movl $32, %eax
1181 ; X86-NEXT: retl
1182 ;
1183 ; X64-LABEL: sdiv_maybe_zero:
1184 ; X64: # %bb.0:
1185 ; X64-NEXT: movl %edi, %eax
1186 ; X64-NEXT: cltd
1187 ; X64-NEXT: idivl %esi
1188 ; X64-NEXT: testl %eax, %eax
1189 ; X64-NEXT: je .LBB39_1
1190 ; X64-NEXT: # %bb.2: # %cond.false
1191 ; X64-NEXT: rep bsfl %eax, %eax
1192 ; X64-NEXT: retq
1193 ; X64-NEXT: .LBB39_1:
1194 ; X64-NEXT: movl $32, %eax
1195 ; X64-NEXT: retq
1196 %z = sdiv exact i32 %x, %y
1197 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1198 ret i32 %r
1199 }
1201 define i32 @add_known_nonzero(i32 %xx, i32 %y) {
1202 ; X86-LABEL: add_known_nonzero:
1203 ; X86: # %bb.0:
1204 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1205 ; X86-NEXT: orl $1, %eax
1206 ; X86-NEXT: addl {{[0-9]+}}(%esp), %eax
1207 ; X86-NEXT: rep bsfl %eax, %eax
1208 ; X86-NEXT: retl
1209 ;
1210 ; X64-LABEL: add_known_nonzero:
1211 ; X64: # %bb.0:
1212 ; X64-NEXT: orl $1, %edi
1213 ; X64-NEXT: addl %esi, %edi
1214 ; X64-NEXT: rep bsfl %edi, %eax
1215 ; X64-NEXT: retq
1216 %x = or i32 %xx, 1
1217 %z = add nuw i32 %x, %y
1218 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1219 ret i32 %r
1220 }
1222 define i32 @add_maybe_zero(i32 %xx, i32 %y) {
1223 ; X86-LABEL: add_maybe_zero:
1224 ; X86: # %bb.0:
1225 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1226 ; X86-NEXT: orl $1, %eax
1227 ; X86-NEXT: addl {{[0-9]+}}(%esp), %eax
1228 ; X86-NEXT: je .LBB41_1
1229 ; X86-NEXT: # %bb.2: # %cond.false
1230 ; X86-NEXT: rep bsfl %eax, %eax
1231 ; X86-NEXT: retl
1232 ; X86-NEXT: .LBB41_1:
1233 ; X86-NEXT: movl $32, %eax
1234 ; X86-NEXT: retl
1235 ;
1236 ; X64-LABEL: add_maybe_zero:
1237 ; X64: # %bb.0:
1238 ; X64-NEXT: orl $1, %edi
1239 ; X64-NEXT: addl %esi, %edi
1240 ; X64-NEXT: je .LBB41_1
1241 ; X64-NEXT: # %bb.2: # %cond.false
1242 ; X64-NEXT: rep bsfl %edi, %eax
1243 ; X64-NEXT: retq
1244 ; X64-NEXT: .LBB41_1:
1245 ; X64-NEXT: movl $32, %eax
1246 ; X64-NEXT: retq
1247 %x = or i32 %xx, 1
1248 %z = add nsw i32 %x, %y
1249 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1250 ret i32 %r
1251 }
1253 define i32 @sub_known_nonzero_neg_case(i32 %xx) {
1254 ; X86-LABEL: sub_known_nonzero_neg_case:
1255 ; X86: # %bb.0:
1256 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1257 ; X86-NEXT: movl $256, %eax # imm = 0x100
1258 ; X86-NEXT: shll %cl, %eax
1259 ; X86-NEXT: negl %eax
1260 ; X86-NEXT: rep bsfl %eax, %eax
1261 ; X86-NEXT: retl
1262 ;
1263 ; X64-LABEL: sub_known_nonzero_neg_case:
1264 ; X64: # %bb.0:
1265 ; X64-NEXT: movl %edi, %ecx
1266 ; X64-NEXT: movl $256, %eax # imm = 0x100
1267 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1268 ; X64-NEXT: shll %cl, %eax
1269 ; X64-NEXT: negl %eax
1270 ; X64-NEXT: rep bsfl %eax, %eax
1271 ; X64-NEXT: retq
1272 %x = shl nuw nsw i32 256, %xx
1273 %z = sub i32 0, %x
1274 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1275 ret i32 %r
1276 }
1278 define i32 @sub_known_nonzero_ne_case(i32 %xx, i32 %yy) {
1279 ; X86-LABEL: sub_known_nonzero_ne_case:
1280 ; X86: # %bb.0:
1281 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1282 ; X86-NEXT: movl %eax, %ecx
1283 ; X86-NEXT: orl $64, %ecx
1284 ; X86-NEXT: andl $-65, %eax
1285 ; X86-NEXT: subl %ecx, %eax
1286 ; X86-NEXT: rep bsfl %eax, %eax
1287 ; X86-NEXT: retl
1288 ;
1289 ; X64-LABEL: sub_known_nonzero_ne_case:
1290 ; X64: # %bb.0:
1291 ; X64-NEXT: movl %edi, %eax
1292 ; X64-NEXT: orl $64, %eax
1293 ; X64-NEXT: andl $-65, %edi
1294 ; X64-NEXT: subl %eax, %edi
1295 ; X64-NEXT: rep bsfl %edi, %eax
1296 ; X64-NEXT: retq
1297 %x = or i32 %xx, 64
1298 %y = and i32 %xx, -65
1299 %z = sub i32 %y, %x
1300 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1301 ret i32 %r
1302 }
1304 define i32 @sub_maybe_zero(i32 %x) {
1305 ; X86-LABEL: sub_maybe_zero:
1306 ; X86: # %bb.0:
1307 ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
1308 ; X86-NEXT: movl %ecx, %eax
1309 ; X86-NEXT: orl $64, %eax
1310 ; X86-NEXT: subl %ecx, %eax
1311 ; X86-NEXT: je .LBB44_1
1312 ; X86-NEXT: # %bb.2: # %cond.false
1313 ; X86-NEXT: rep bsfl %eax, %eax
1314 ; X86-NEXT: retl
1315 ; X86-NEXT: .LBB44_1:
1316 ; X86-NEXT: movl $32, %eax
1317 ; X86-NEXT: retl
1318 ;
1319 ; X64-LABEL: sub_maybe_zero:
1320 ; X64: # %bb.0:
1321 ; X64-NEXT: movl %edi, %eax
1322 ; X64-NEXT: orl $64, %eax
1323 ; X64-NEXT: subl %edi, %eax
1324 ; X64-NEXT: je .LBB44_1
1325 ; X64-NEXT: # %bb.2: # %cond.false
1326 ; X64-NEXT: rep bsfl %eax, %eax
1327 ; X64-NEXT: retq
1328 ; X64-NEXT: .LBB44_1:
1329 ; X64-NEXT: movl $32, %eax
1330 ; X64-NEXT: retq
1331 %y = or i32 %x, 64
1332 %z = sub i32 %y, %x
1333 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1334 ret i32 %r
1335 }
1337 define i32 @sub_maybe_zero2(i32 %x) {
1338 ; X86-LABEL: sub_maybe_zero2:
1339 ; X86: # %bb.0:
1340 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1341 ; X86-NEXT: negl %eax
1342 ; X86-NEXT: je .LBB45_1
1343 ; X86-NEXT: # %bb.2: # %cond.false
1344 ; X86-NEXT: rep bsfl %eax, %eax
1345 ; X86-NEXT: retl
1346 ; X86-NEXT: .LBB45_1:
1347 ; X86-NEXT: movl $32, %eax
1348 ; X86-NEXT: retl
1349 ;
1350 ; X64-LABEL: sub_maybe_zero2:
1351 ; X64: # %bb.0:
1352 ; X64-NEXT: negl %edi
1353 ; X64-NEXT: je .LBB45_1
1354 ; X64-NEXT: # %bb.2: # %cond.false
1355 ; X64-NEXT: rep bsfl %edi, %eax
1356 ; X64-NEXT: retq
1357 ; X64-NEXT: .LBB45_1:
1358 ; X64-NEXT: movl $32, %eax
1359 ; X64-NEXT: retq
1360 %z = sub i32 0, %x
1361 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1362 ret i32 %r
1363 }
1365 define i32 @mul_known_nonzero_nsw(i32 %x, i32 %yy) {
1366 ; X86-LABEL: mul_known_nonzero_nsw:
1367 ; X86: # %bb.0:
1368 ; X86-NEXT: movl $256, %eax # imm = 0x100
1369 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
1370 ; X86-NEXT: imull {{[0-9]+}}(%esp), %eax
1371 ; X86-NEXT: testl %eax, %eax
1372 ; X86-NEXT: je .LBB46_1
1373 ; X86-NEXT: # %bb.2: # %cond.false
1374 ; X86-NEXT: rep bsfl %eax, %eax
1375 ; X86-NEXT: retl
1376 ; X86-NEXT: .LBB46_1:
1377 ; X86-NEXT: movl $32, %eax
1378 ; X86-NEXT: retl
1379 ;
1380 ; X64-LABEL: mul_known_nonzero_nsw:
1381 ; X64: # %bb.0:
1382 ; X64-NEXT: orl $256, %esi # imm = 0x100
1383 ; X64-NEXT: imull %edi, %esi
1384 ; X64-NEXT: testl %esi, %esi
1385 ; X64-NEXT: je .LBB46_1
1386 ; X64-NEXT: # %bb.2: # %cond.false
1387 ; X64-NEXT: rep bsfl %esi, %eax
1388 ; X64-NEXT: retq
1389 ; X64-NEXT: .LBB46_1:
1390 ; X64-NEXT: movl $32, %eax
1391 ; X64-NEXT: retq
1392 %y = or i32 %yy, 256
1393 %z = mul nsw i32 %y, %x
1394 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1395 ret i32 %r
1396 }
1398 define i32 @mul_known_nonzero_nuw(i32 %x, i32 %yy) {
1399 ; X86-LABEL: mul_known_nonzero_nuw:
1400 ; X86: # %bb.0:
1401 ; X86-NEXT: movl $256, %eax # imm = 0x100
1402 ; X86-NEXT: orl {{[0-9]+}}(%esp), %eax
1403 ; X86-NEXT: imull {{[0-9]+}}(%esp), %eax
1404 ; X86-NEXT: testl %eax, %eax
1405 ; X86-NEXT: je .LBB47_1
1406 ; X86-NEXT: # %bb.2: # %cond.false
1407 ; X86-NEXT: rep bsfl %eax, %eax
1408 ; X86-NEXT: retl
1409 ; X86-NEXT: .LBB47_1:
1410 ; X86-NEXT: movl $32, %eax
1411 ; X86-NEXT: retl
1412 ;
1413 ; X64-LABEL: mul_known_nonzero_nuw:
1414 ; X64: # %bb.0:
1415 ; X64-NEXT: orl $256, %esi # imm = 0x100
1416 ; X64-NEXT: imull %edi, %esi
1417 ; X64-NEXT: testl %esi, %esi
1418 ; X64-NEXT: je .LBB47_1
1419 ; X64-NEXT: # %bb.2: # %cond.false
1420 ; X64-NEXT: rep bsfl %esi, %eax
1421 ; X64-NEXT: retq
1422 ; X64-NEXT: .LBB47_1:
1423 ; X64-NEXT: movl $32, %eax
1424 ; X64-NEXT: retq
1425 %y = or i32 %yy, 256
1426 %z = mul nuw i32 %y, %x
1427 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1428 ret i32 %r
1429 }
1431 define i32 @mul_maybe_zero(i32 %x, i32 %y) {
1432 ; X86-LABEL: mul_maybe_zero:
1433 ; X86: # %bb.0:
1434 ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
1435 ; X86-NEXT: imull {{[0-9]+}}(%esp), %eax
1436 ; X86-NEXT: testl %eax, %eax
1437 ; X86-NEXT: je .LBB48_1
1438 ; X86-NEXT: # %bb.2: # %cond.false
1439 ; X86-NEXT: rep bsfl %eax, %eax
1440 ; X86-NEXT: retl
1441 ; X86-NEXT: .LBB48_1:
1442 ; X86-NEXT: movl $32, %eax
1443 ; X86-NEXT: retl
1444 ;
1445 ; X64-LABEL: mul_maybe_zero:
1446 ; X64: # %bb.0:
1447 ; X64-NEXT: imull %esi, %edi
1448 ; X64-NEXT: testl %edi, %edi
1449 ; X64-NEXT: je .LBB48_1
1450 ; X64-NEXT: # %bb.2: # %cond.false
1451 ; X64-NEXT: rep bsfl %edi, %eax
1452 ; X64-NEXT: retq
1453 ; X64-NEXT: .LBB48_1:
1454 ; X64-NEXT: movl $32, %eax
1455 ; X64-NEXT: retq
1456 %z = mul nuw nsw i32 %y, %x
1457 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1458 ret i32 %r
1459 }
1461 define i32 @bitcast_known_nonzero(<2 x i16> %xx) {
1462 ; X86-LABEL: bitcast_known_nonzero:
1463 ; X86: # %bb.0:
1464 ; X86-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
1465 ; X86-NEXT: pslld $23, %xmm0
1466 ; X86-NEXT: paddd {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
1467 ; X86-NEXT: cvttps2dq %xmm0, %xmm0
1468 ; X86-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
1469 ; X86-NEXT: pmullw {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 # [256,256,u,u,u,u,u,u]
1470 ; X86-NEXT: movd %xmm0, %eax
1471 ; X86-NEXT: bsfl %eax, %ecx
1472 ; X86-NEXT: movl $32, %eax
1473 ; X86-NEXT: cmovnel %ecx, %eax
1474 ; X86-NEXT: retl
1475 ;
1476 ; X64-LABEL: bitcast_known_nonzero:
1477 ; X64: # %bb.0:
1478 ; X64-NEXT: vpmovzxwd {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
1479 ; X64-NEXT: vpslld $23, %xmm0, %xmm0
1480 ; X64-NEXT: vpaddd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0
1481 ; X64-NEXT: vcvttps2dq %xmm0, %xmm0
1482 ; X64-NEXT: vpackusdw %xmm0, %xmm0, %xmm0
1483 ; X64-NEXT: vpmullw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 # [256,256,u,u,u,u,u,u]
1484 ; X64-NEXT: vmovd %xmm0, %eax
1485 ; X64-NEXT: bsfl %eax, %ecx
1486 ; X64-NEXT: movl $32, %eax
1487 ; X64-NEXT: cmovnel %ecx, %eax
1488 ; X64-NEXT: retq
1489 %x = shl nuw nsw <2 x i16> <i16 256, i16 256>, %xx
1490 %z = bitcast <2 x i16> %x to i32
1491 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1492 ret i32 %r
1493 }
1495 define i32 @bitcast_maybe_zero(<2 x i16> %x) {
1496 ; X86-LABEL: bitcast_maybe_zero:
1497 ; X86: # %bb.0:
1498 ; X86-NEXT: movd %xmm0, %eax
1499 ; X86-NEXT: testl %eax, %eax
1500 ; X86-NEXT: je .LBB50_1
1501 ; X86-NEXT: # %bb.2: # %cond.false
1502 ; X86-NEXT: rep bsfl %eax, %eax
1503 ; X86-NEXT: retl
1504 ; X86-NEXT: .LBB50_1:
1505 ; X86-NEXT: movl $32, %eax
1506 ; X86-NEXT: retl
1507 ;
1508 ; X64-LABEL: bitcast_maybe_zero:
1509 ; X64: # %bb.0:
1510 ; X64-NEXT: vmovd %xmm0, %eax
1511 ; X64-NEXT: testl %eax, %eax
1512 ; X64-NEXT: je .LBB50_1
1513 ; X64-NEXT: # %bb.2: # %cond.false
1514 ; X64-NEXT: rep bsfl %eax, %eax
1515 ; X64-NEXT: retq
1516 ; X64-NEXT: .LBB50_1:
1517 ; X64-NEXT: movl $32, %eax
1518 ; X64-NEXT: retq
1519 %z = bitcast <2 x i16> %x to i32
1520 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1521 ret i32 %r
1522 }
1524 define i32 @bitcast_from_float(float %x) {
1525 ; X86-LABEL: bitcast_from_float:
1526 ; X86: # %bb.0:
1527 ; X86-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero
1528 ; X86-NEXT: movd %xmm0, %eax
1529 ; X86-NEXT: testl %eax, %eax
1530 ; X86-NEXT: je .LBB51_1
1531 ; X86-NEXT: # %bb.2: # %cond.false
1532 ; X86-NEXT: rep bsfl %eax, %eax
1533 ; X86-NEXT: retl
1534 ; X86-NEXT: .LBB51_1:
1535 ; X86-NEXT: movl $32, %eax
1536 ; X86-NEXT: retl
1537 ;
1538 ; X64-LABEL: bitcast_from_float:
1539 ; X64: # %bb.0:
1540 ; X64-NEXT: vmovd %xmm0, %eax
1541 ; X64-NEXT: testl %eax, %eax
1542 ; X64-NEXT: je .LBB51_1
1543 ; X64-NEXT: # %bb.2: # %cond.false
1544 ; X64-NEXT: rep bsfl %eax, %eax
1545 ; X64-NEXT: retq
1546 ; X64-NEXT: .LBB51_1:
1547 ; X64-NEXT: movl $32, %eax
1548 ; X64-NEXT: retq
1549 %z = bitcast float %x to i32
1550 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1551 ret i32 %r
1552 }
1554 define i32 @zext_known_nonzero(i16 %xx) {
1555 ; X86-LABEL: zext_known_nonzero:
1556 ; X86: # %bb.0:
1557 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1558 ; X86-NEXT: movl $256, %eax # imm = 0x100
1559 ; X86-NEXT: shll %cl, %eax
1560 ; X86-NEXT: movzwl %ax, %eax
1561 ; X86-NEXT: rep bsfl %eax, %eax
1562 ; X86-NEXT: retl
1563 ;
1564 ; X64-LABEL: zext_known_nonzero:
1565 ; X64: # %bb.0:
1566 ; X64-NEXT: movl %edi, %ecx
1567 ; X64-NEXT: movl $256, %eax # imm = 0x100
1568 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1569 ; X64-NEXT: shll %cl, %eax
1570 ; X64-NEXT: movzwl %ax, %eax
1571 ; X64-NEXT: rep bsfl %eax, %eax
1572 ; X64-NEXT: retq
1573 %x = shl nuw nsw i16 256, %xx
1574 %z = zext i16 %x to i32
1575 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1576 ret i32 %r
1577 }
1579 define i32 @zext_maybe_zero(i16 %x) {
1580 ; X86-LABEL: zext_maybe_zero:
1581 ; X86: # %bb.0:
1582 ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
1583 ; X86-NEXT: testw %ax, %ax
1584 ; X86-NEXT: je .LBB53_1
1585 ; X86-NEXT: # %bb.2: # %cond.false
1586 ; X86-NEXT: movzwl %ax, %eax
1587 ; X86-NEXT: rep bsfl %eax, %eax
1588 ; X86-NEXT: retl
1589 ; X86-NEXT: .LBB53_1:
1590 ; X86-NEXT: movl $32, %eax
1591 ; X86-NEXT: retl
1592 ;
1593 ; X64-LABEL: zext_maybe_zero:
1594 ; X64: # %bb.0:
1595 ; X64-NEXT: testw %di, %di
1596 ; X64-NEXT: je .LBB53_1
1597 ; X64-NEXT: # %bb.2: # %cond.false
1598 ; X64-NEXT: movzwl %di, %eax
1599 ; X64-NEXT: rep bsfl %eax, %eax
1600 ; X64-NEXT: retq
1601 ; X64-NEXT: .LBB53_1:
1602 ; X64-NEXT: movl $32, %eax
1603 ; X64-NEXT: retq
1604 %z = zext i16 %x to i32
1605 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1606 ret i32 %r
1607 }
1609 define i32 @sext_known_nonzero(i16 %xx) {
1610 ; X86-LABEL: sext_known_nonzero:
1611 ; X86: # %bb.0:
1612 ; X86-NEXT: movzbl {{[0-9]+}}(%esp), %ecx
1613 ; X86-NEXT: movl $256, %eax # imm = 0x100
1614 ; X86-NEXT: shll %cl, %eax
1615 ; X86-NEXT: movzwl %ax, %eax
1616 ; X86-NEXT: rep bsfl %eax, %eax
1617 ; X86-NEXT: retl
1618 ;
1619 ; X64-LABEL: sext_known_nonzero:
1620 ; X64: # %bb.0:
1621 ; X64-NEXT: movl %edi, %ecx
1622 ; X64-NEXT: movl $256, %eax # imm = 0x100
1623 ; X64-NEXT: # kill: def $cl killed $cl killed $ecx
1624 ; X64-NEXT: shll %cl, %eax
1625 ; X64-NEXT: movzwl %ax, %eax
1626 ; X64-NEXT: rep bsfl %eax, %eax
1627 ; X64-NEXT: retq
1628 %x = shl nuw nsw i16 256, %xx
1629 %z = sext i16 %x to i32
1630 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1631 ret i32 %r
1632 }
1634 define i32 @sext_maybe_zero(i16 %x) {
1635 ; X86-LABEL: sext_maybe_zero:
1636 ; X86: # %bb.0:
1637 ; X86-NEXT: movswl {{[0-9]+}}(%esp), %eax
1638 ; X86-NEXT: testl %eax, %eax
1639 ; X86-NEXT: je .LBB55_1
1640 ; X86-NEXT: # %bb.2: # %cond.false
1641 ; X86-NEXT: rep bsfl %eax, %eax
1642 ; X86-NEXT: retl
1643 ; X86-NEXT: .LBB55_1:
1644 ; X86-NEXT: movl $32, %eax
1645 ; X86-NEXT: retl
1646 ;
1647 ; X64-LABEL: sext_maybe_zero:
1648 ; X64: # %bb.0:
1649 ; X64-NEXT: testw %di, %di
1650 ; X64-NEXT: je .LBB55_1
1651 ; X64-NEXT: # %bb.2: # %cond.false
1652 ; X64-NEXT: movswl %di, %eax
1653 ; X64-NEXT: rep bsfl %eax, %eax
1654 ; X64-NEXT: retq
1655 ; X64-NEXT: .LBB55_1:
1656 ; X64-NEXT: movl $32, %eax
1657 ; X64-NEXT: retq
1658 %z = sext i16 %x to i32
1659 %r = call i32 @llvm.cttz.i32(i32 %z, i1 false)
1660 ret i32 %r
1661 }