1 ; RUN: llc < %s -march=x86-64 -mattr=+bmi,+bmi2 | FileCheck %s
3 declare i8 @llvm.cttz.i8(i8, i1) nounwind readnone
4 declare i16 @llvm.cttz.i16(i16, i1) nounwind readnone
5 declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
6 declare i64 @llvm.cttz.i64(i64, i1) nounwind readnone
8 define i8 @t1(i8 %x) nounwind {
9 %tmp = tail call i8 @llvm.cttz.i8( i8 %x, i1 false )
15 define i16 @t2(i16 %x) nounwind {
16 %tmp = tail call i16 @llvm.cttz.i16( i16 %x, i1 false )
22 define i32 @t3(i32 %x) nounwind {
23 %tmp = tail call i32 @llvm.cttz.i32( i32 %x, i1 false )
29 define i32 @tzcnt32_load(i32* %x) nounwind {
30 %x1 = load i32, i32* %x
31 %tmp = tail call i32 @llvm.cttz.i32(i32 %x1, i1 false )
33 ; CHECK-LABEL: tzcnt32_load:
34 ; CHECK: tzcntl ({{.*}})
37 define i64 @t4(i64 %x) nounwind {
38 %tmp = tail call i64 @llvm.cttz.i64( i64 %x, i1 false )
44 define i8 @t5(i8 %x) nounwind {
45 %tmp = tail call i8 @llvm.cttz.i8( i8 %x, i1 true )
51 define i16 @t6(i16 %x) nounwind {
52 %tmp = tail call i16 @llvm.cttz.i16( i16 %x, i1 true )
58 define i32 @t7(i32 %x) nounwind {
59 %tmp = tail call i32 @llvm.cttz.i32( i32 %x, i1 true )
65 define i64 @t8(i64 %x) nounwind {
66 %tmp = tail call i64 @llvm.cttz.i64( i64 %x, i1 true )
72 define i32 @andn32(i32 %x, i32 %y) nounwind readnone {
73 %tmp1 = xor i32 %x, -1
74 %tmp2 = and i32 %y, %tmp1
76 ; CHECK-LABEL: andn32:
80 define i32 @andn32_load(i32 %x, i32* %y) nounwind readnone {
81 %y1 = load i32, i32* %y
82 %tmp1 = xor i32 %x, -1
83 %tmp2 = and i32 %y1, %tmp1
85 ; CHECK-LABEL: andn32_load:
86 ; CHECK: andnl ({{.*}})
89 define i64 @andn64(i64 %x, i64 %y) nounwind readnone {
90 %tmp1 = xor i64 %x, -1
91 %tmp2 = and i64 %tmp1, %y
93 ; CHECK-LABEL: andn64:
97 define i32 @bextr32(i32 %x, i32 %y) nounwind readnone {
98 %tmp = tail call i32 @llvm.x86.bmi.bextr.32(i32 %x, i32 %y)
100 ; CHECK-LABEL: bextr32:
104 define i32 @bextr32_load(i32* %x, i32 %y) nounwind readnone {
105 %x1 = load i32, i32* %x
106 %tmp = tail call i32 @llvm.x86.bmi.bextr.32(i32 %x1, i32 %y)
108 ; CHECK-LABEL: bextr32_load:
109 ; CHECK: bextrl {{.*}}, ({{.*}}), {{.*}}
112 declare i32 @llvm.x86.bmi.bextr.32(i32, i32) nounwind readnone
114 define i32 @bextr32b(i32 %x) nounwind uwtable readnone ssp {
116 %2 = and i32 %1, 4095
118 ; CHECK-LABEL: bextr32b:
122 define i32 @bextr32b_load(i32* %x) nounwind uwtable readnone ssp {
123 %1 = load i32, i32* %x
125 %3 = and i32 %2, 4095
127 ; CHECK-LABEL: bextr32b_load:
128 ; CHECK: bextrl {{.*}}, ({{.*}}), {{.*}}
131 define i64 @bextr64(i64 %x, i64 %y) nounwind readnone {
132 %tmp = tail call i64 @llvm.x86.bmi.bextr.64(i64 %x, i64 %y)
134 ; CHECK-LABEL: bextr64:
138 declare i64 @llvm.x86.bmi.bextr.64(i64, i64) nounwind readnone
140 define i64 @bextr64b(i64 %x) nounwind uwtable readnone ssp {
142 %2 = and i64 %1, 4095
144 ; CHECK-LABEL: bextr64b:
148 define i64 @bextr64b_load(i64* %x) {
149 %1 = load i64, i64* %x, align 8
151 %3 = and i64 %2, 4095
153 ; CHECK-LABEL: bextr64b_load:
154 ; CHECK: bextrq {{.*}}, ({{.*}}), {{.*}}
157 define i32 @non_bextr32(i32 %x) {
159 %shr = lshr i32 %x, 2
160 %and = and i32 %shr, 111
162 ; CHECK-LABEL: non_bextr32:
167 define i64 @non_bextr64(i64 %x) {
169 %shr = lshr i64 %x, 2
170 %and = and i64 %shr, 8589934590
172 ; CHECK-LABEL: non_bextr64:
174 ; CHECK: movabsq $8589934590
178 define i32 @bzhi32(i32 %x, i32 %y) nounwind readnone {
179 %tmp = tail call i32 @llvm.x86.bmi.bzhi.32(i32 %x, i32 %y)
181 ; CHECK-LABEL: bzhi32:
185 define i32 @bzhi32_load(i32* %x, i32 %y) nounwind readnone {
186 %x1 = load i32, i32* %x
187 %tmp = tail call i32 @llvm.x86.bmi.bzhi.32(i32 %x1, i32 %y)
189 ; CHECK-LABEL: bzhi32_load:
190 ; CHECK: bzhil {{.*}}, ({{.*}}), {{.*}}
193 declare i32 @llvm.x86.bmi.bzhi.32(i32, i32) nounwind readnone
195 define i64 @bzhi64(i64 %x, i64 %y) nounwind readnone {
196 %tmp = tail call i64 @llvm.x86.bmi.bzhi.64(i64 %x, i64 %y)
198 ; CHECK-LABEL: bzhi64:
202 declare i64 @llvm.x86.bmi.bzhi.64(i64, i64) nounwind readnone
204 define i32 @bzhi32b(i32 %x, i8 zeroext %index) #0 {
206 %conv = zext i8 %index to i32
207 %shl = shl i32 1, %conv
208 %sub = add nsw i32 %shl, -1
209 %and = and i32 %sub, %x
211 ; CHECK-LABEL: bzhi32b:
215 define i32 @bzhi32b_load(i32* %w, i8 zeroext %index) #0 {
217 %x = load i32, i32* %w
218 %conv = zext i8 %index to i32
219 %shl = shl i32 1, %conv
220 %sub = add nsw i32 %shl, -1
221 %and = and i32 %sub, %x
223 ; CHECK-LABEL: bzhi32b_load:
224 ; CHECK: bzhil {{.*}}, ({{.*}}), {{.*}}
227 define i32 @bzhi32c(i32 %x, i8 zeroext %index) #0 {
229 %conv = zext i8 %index to i32
230 %shl = shl i32 1, %conv
231 %sub = add nsw i32 %shl, -1
232 %and = and i32 %x, %sub
234 ; CHECK-LABEL: bzhi32c:
238 define i64 @bzhi64b(i64 %x, i8 zeroext %index) #0 {
240 %conv = zext i8 %index to i64
241 %shl = shl i64 1, %conv
242 %sub = add nsw i64 %shl, -1
243 %and = and i64 %x, %sub
245 ; CHECK-LABEL: bzhi64b:
249 define i64 @bzhi64_constant_mask(i64 %x) #0 {
251 %and = and i64 %x, 4611686018427387903
253 ; CHECK-LABEL: bzhi64_constant_mask:
254 ; CHECK: movb $62, %al
255 ; CHECK: bzhiq %rax, %r[[ARG1:di|cx]], %rax
258 define i64 @bzhi64_small_constant_mask(i64 %x) #0 {
260 %and = and i64 %x, 2147483647
262 ; CHECK-LABEL: bzhi64_small_constant_mask:
263 ; CHECK: andl $2147483647, %e[[ARG1]]
266 define i32 @blsi32(i32 %x) nounwind readnone {
268 %tmp2 = and i32 %x, %tmp
270 ; CHECK-LABEL: blsi32:
274 define i32 @blsi32_load(i32* %x) nounwind readnone {
275 %x1 = load i32, i32* %x
276 %tmp = sub i32 0, %x1
277 %tmp2 = and i32 %x1, %tmp
279 ; CHECK-LABEL: blsi32_load:
280 ; CHECK: blsil ({{.*}})
283 define i64 @blsi64(i64 %x) nounwind readnone {
285 %tmp2 = and i64 %tmp, %x
287 ; CHECK-LABEL: blsi64:
291 define i32 @blsmsk32(i32 %x) nounwind readnone {
293 %tmp2 = xor i32 %x, %tmp
295 ; CHECK-LABEL: blsmsk32:
299 define i32 @blsmsk32_load(i32* %x) nounwind readnone {
300 %x1 = load i32, i32* %x
301 %tmp = sub i32 %x1, 1
302 %tmp2 = xor i32 %x1, %tmp
304 ; CHECK-LABEL: blsmsk32_load:
305 ; CHECK: blsmskl ({{.*}})
308 define i64 @blsmsk64(i64 %x) nounwind readnone {
310 %tmp2 = xor i64 %tmp, %x
312 ; CHECK-LABEL: blsmsk64:
316 define i32 @blsr32(i32 %x) nounwind readnone {
318 %tmp2 = and i32 %x, %tmp
320 ; CHECK-LABEL: blsr32:
324 define i32 @blsr32_load(i32* %x) nounwind readnone {
325 %x1 = load i32, i32* %x
326 %tmp = sub i32 %x1, 1
327 %tmp2 = and i32 %x1, %tmp
329 ; CHECK-LABEL: blsr32_load:
330 ; CHECK: blsrl ({{.*}})
333 define i64 @blsr64(i64 %x) nounwind readnone {
335 %tmp2 = and i64 %tmp, %x
337 ; CHECK-LABEL: blsr64:
341 define i32 @pdep32(i32 %x, i32 %y) nounwind readnone {
342 %tmp = tail call i32 @llvm.x86.bmi.pdep.32(i32 %x, i32 %y)
344 ; CHECK-LABEL: pdep32:
348 define i32 @pdep32_load(i32 %x, i32* %y) nounwind readnone {
349 %y1 = load i32, i32* %y
350 %tmp = tail call i32 @llvm.x86.bmi.pdep.32(i32 %x, i32 %y1)
352 ; CHECK-LABEL: pdep32_load:
353 ; CHECK: pdepl ({{.*}})
356 declare i32 @llvm.x86.bmi.pdep.32(i32, i32) nounwind readnone
358 define i64 @pdep64(i64 %x, i64 %y) nounwind readnone {
359 %tmp = tail call i64 @llvm.x86.bmi.pdep.64(i64 %x, i64 %y)
361 ; CHECK-LABEL: pdep64:
365 declare i64 @llvm.x86.bmi.pdep.64(i64, i64) nounwind readnone
367 define i32 @pext32(i32 %x, i32 %y) nounwind readnone {
368 %tmp = tail call i32 @llvm.x86.bmi.pext.32(i32 %x, i32 %y)
370 ; CHECK-LABEL: pext32:
374 define i32 @pext32_load(i32 %x, i32* %y) nounwind readnone {
375 %y1 = load i32, i32* %y
376 %tmp = tail call i32 @llvm.x86.bmi.pext.32(i32 %x, i32 %y1)
378 ; CHECK-LABEL: pext32_load:
379 ; CHECK: pextl ({{.*}})
382 declare i32 @llvm.x86.bmi.pext.32(i32, i32) nounwind readnone
384 define i64 @pext64(i64 %x, i64 %y) nounwind readnone {
385 %tmp = tail call i64 @llvm.x86.bmi.pext.64(i64 %x, i64 %y)
387 ; CHECK-LABEL: pext64:
391 declare i64 @llvm.x86.bmi.pext.64(i64, i64) nounwind readnone