1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+sse,sse2                  < %s | FileCheck %s --check-prefixes=CHECK,X86,V0123,X86-V0123,SSE2,X86-SSE2,BMI1,X86-BMI1,V0,X86-V0
3; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+sse,sse2,+bmi             < %s | FileCheck %s --check-prefixes=CHECK,X86,V0123,X86-V0123,SSE2,X86-SSE2,BMI1,X86-BMI1,V1,X86-V1
4; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+sse,sse2,+bmi,+bmi2       < %s | FileCheck %s --check-prefixes=CHECK,X86,V0123,X86-V0123,SSE2,X86-SSE2,BMI2,X86-BMI2,V2,X86-V2
5; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+sse,sse2,+bmi,+bmi2,+avx2 < %s | FileCheck %s --check-prefixes=CHECK,X86,V0123,X86-V0123,BMI2,X86-BMI2,AVX2,X86-AVX2
6; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,sse2                  < %s | FileCheck %s --check-prefixes=CHECK,X64,V0123,X64-V0123,SSE2,X64-SSE2,BMI1,X64-BMI1,V0,X64-V0
7; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,sse2,+bmi             < %s | FileCheck %s --check-prefixes=CHECK,X64,V0123,X64-V0123,SSE2,X64-SSE2,BMI1,X64-BMI1,V1,X64-V1
8; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,sse2,+bmi,+bmi2       < %s | FileCheck %s --check-prefixes=CHECK,X64,V0123,X64-V0123,SSE2,X64-SSE2,BMI2,X64-BMI2,V2,X64-V2
9; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+sse,sse2,+bmi,+bmi2,+avx2 < %s | FileCheck %s --check-prefixes=CHECK,X64,V0123,X64-V0123,BMI2,X64-BMI2,AVX2,X64-AVX2
10
11; We are looking for the following pattern here:
12;   (X & (C << Y)) ==/!= 0
13; It may be optimal to hoist the constant:
14;   ((X l>> Y) & C) ==/!= 0
15
16;------------------------------------------------------------------------------;
17; A few scalar test
18;------------------------------------------------------------------------------;
19
20; i8 scalar
21
22define i1 @scalar_i8_signbit_eq(i8 %x, i8 %y) nounwind {
23; X86-LABEL: scalar_i8_signbit_eq:
24; X86:       # %bb.0:
25; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
26; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
27; X86-NEXT:    shrb %cl, %al
28; X86-NEXT:    testb $-128, %al
29; X86-NEXT:    sete %al
30; X86-NEXT:    retl
31;
32; X64-LABEL: scalar_i8_signbit_eq:
33; X64:       # %bb.0:
34; X64-NEXT:    movl %esi, %ecx
35; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
36; X64-NEXT:    shrb %cl, %dil
37; X64-NEXT:    testb $-128, %dil
38; X64-NEXT:    sete %al
39; X64-NEXT:    retq
40  %t0 = shl i8 128, %y
41  %t1 = and i8 %t0, %x
42  %res = icmp eq i8 %t1, 0
43  ret i1 %res
44}
45
46define i1 @scalar_i8_lowestbit_eq(i8 %x, i8 %y) nounwind {
47; X86-LABEL: scalar_i8_lowestbit_eq:
48; X86:       # %bb.0:
49; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
50; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
51; X86-NEXT:    btl %eax, %ecx
52; X86-NEXT:    setae %al
53; X86-NEXT:    retl
54;
55; X64-LABEL: scalar_i8_lowestbit_eq:
56; X64:       # %bb.0:
57; X64-NEXT:    btl %esi, %edi
58; X64-NEXT:    setae %al
59; X64-NEXT:    retq
60  %t0 = shl i8 1, %y
61  %t1 = and i8 %t0, %x
62  %res = icmp eq i8 %t1, 0
63  ret i1 %res
64}
65
66define i1 @scalar_i8_bitsinmiddle_eq(i8 %x, i8 %y) nounwind {
67; X86-LABEL: scalar_i8_bitsinmiddle_eq:
68; X86:       # %bb.0:
69; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
70; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
71; X86-NEXT:    shrb %cl, %al
72; X86-NEXT:    testb $24, %al
73; X86-NEXT:    sete %al
74; X86-NEXT:    retl
75;
76; X64-LABEL: scalar_i8_bitsinmiddle_eq:
77; X64:       # %bb.0:
78; X64-NEXT:    movl %esi, %ecx
79; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
80; X64-NEXT:    shrb %cl, %dil
81; X64-NEXT:    testb $24, %dil
82; X64-NEXT:    sete %al
83; X64-NEXT:    retq
84  %t0 = shl i8 24, %y
85  %t1 = and i8 %t0, %x
86  %res = icmp eq i8 %t1, 0
87  ret i1 %res
88}
89
90; i16 scalar
91
92define i1 @scalar_i16_signbit_eq(i16 %x, i16 %y) nounwind {
93; X86-BMI1-LABEL: scalar_i16_signbit_eq:
94; X86-BMI1:       # %bb.0:
95; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
96; X86-BMI1-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
97; X86-BMI1-NEXT:    shrl %cl, %eax
98; X86-BMI1-NEXT:    testl $32768, %eax # imm = 0x8000
99; X86-BMI1-NEXT:    sete %al
100; X86-BMI1-NEXT:    retl
101;
102; X86-BMI2-LABEL: scalar_i16_signbit_eq:
103; X86-BMI2:       # %bb.0:
104; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
105; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
106; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
107; X86-BMI2-NEXT:    testl $32768, %eax # imm = 0x8000
108; X86-BMI2-NEXT:    sete %al
109; X86-BMI2-NEXT:    retl
110;
111; X64-BMI1-LABEL: scalar_i16_signbit_eq:
112; X64-BMI1:       # %bb.0:
113; X64-BMI1-NEXT:    movl %esi, %ecx
114; X64-BMI1-NEXT:    movzwl %di, %eax
115; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
116; X64-BMI1-NEXT:    shrl %cl, %eax
117; X64-BMI1-NEXT:    testl $32768, %eax # imm = 0x8000
118; X64-BMI1-NEXT:    sete %al
119; X64-BMI1-NEXT:    retq
120;
121; X64-BMI2-LABEL: scalar_i16_signbit_eq:
122; X64-BMI2:       # %bb.0:
123; X64-BMI2-NEXT:    movzwl %di, %eax
124; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
125; X64-BMI2-NEXT:    testl $32768, %eax # imm = 0x8000
126; X64-BMI2-NEXT:    sete %al
127; X64-BMI2-NEXT:    retq
128  %t0 = shl i16 32768, %y
129  %t1 = and i16 %t0, %x
130  %res = icmp eq i16 %t1, 0
131  ret i1 %res
132}
133
134define i1 @scalar_i16_lowestbit_eq(i16 %x, i16 %y) nounwind {
135; X86-LABEL: scalar_i16_lowestbit_eq:
136; X86:       # %bb.0:
137; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
138; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
139; X86-NEXT:    btl %eax, %ecx
140; X86-NEXT:    setae %al
141; X86-NEXT:    retl
142;
143; X64-LABEL: scalar_i16_lowestbit_eq:
144; X64:       # %bb.0:
145; X64-NEXT:    btl %esi, %edi
146; X64-NEXT:    setae %al
147; X64-NEXT:    retq
148  %t0 = shl i16 1, %y
149  %t1 = and i16 %t0, %x
150  %res = icmp eq i16 %t1, 0
151  ret i1 %res
152}
153
154define i1 @scalar_i16_bitsinmiddle_eq(i16 %x, i16 %y) nounwind {
155; X86-BMI1-LABEL: scalar_i16_bitsinmiddle_eq:
156; X86-BMI1:       # %bb.0:
157; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
158; X86-BMI1-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
159; X86-BMI1-NEXT:    shrl %cl, %eax
160; X86-BMI1-NEXT:    testl $4080, %eax # imm = 0xFF0
161; X86-BMI1-NEXT:    sete %al
162; X86-BMI1-NEXT:    retl
163;
164; X86-BMI2-LABEL: scalar_i16_bitsinmiddle_eq:
165; X86-BMI2:       # %bb.0:
166; X86-BMI2-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
167; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
168; X86-BMI2-NEXT:    shrxl %ecx, %eax, %eax
169; X86-BMI2-NEXT:    testl $4080, %eax # imm = 0xFF0
170; X86-BMI2-NEXT:    sete %al
171; X86-BMI2-NEXT:    retl
172;
173; X64-BMI1-LABEL: scalar_i16_bitsinmiddle_eq:
174; X64-BMI1:       # %bb.0:
175; X64-BMI1-NEXT:    movl %esi, %ecx
176; X64-BMI1-NEXT:    movzwl %di, %eax
177; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
178; X64-BMI1-NEXT:    shrl %cl, %eax
179; X64-BMI1-NEXT:    testl $4080, %eax # imm = 0xFF0
180; X64-BMI1-NEXT:    sete %al
181; X64-BMI1-NEXT:    retq
182;
183; X64-BMI2-LABEL: scalar_i16_bitsinmiddle_eq:
184; X64-BMI2:       # %bb.0:
185; X64-BMI2-NEXT:    movzwl %di, %eax
186; X64-BMI2-NEXT:    shrxl %esi, %eax, %eax
187; X64-BMI2-NEXT:    testl $4080, %eax # imm = 0xFF0
188; X64-BMI2-NEXT:    sete %al
189; X64-BMI2-NEXT:    retq
190  %t0 = shl i16 4080, %y
191  %t1 = and i16 %t0, %x
192  %res = icmp eq i16 %t1, 0
193  ret i1 %res
194}
195
196; i32 scalar
197
198define i1 @scalar_i32_signbit_eq(i32 %x, i32 %y) nounwind {
199; X86-BMI1-LABEL: scalar_i32_signbit_eq:
200; X86-BMI1:       # %bb.0:
201; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
202; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
203; X86-BMI1-NEXT:    shrl %cl, %eax
204; X86-BMI1-NEXT:    testl $-2147483648, %eax # imm = 0x80000000
205; X86-BMI1-NEXT:    sete %al
206; X86-BMI1-NEXT:    retl
207;
208; X86-BMI2-LABEL: scalar_i32_signbit_eq:
209; X86-BMI2:       # %bb.0:
210; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
211; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
212; X86-BMI2-NEXT:    testl $-2147483648, %eax # imm = 0x80000000
213; X86-BMI2-NEXT:    sete %al
214; X86-BMI2-NEXT:    retl
215;
216; X64-BMI1-LABEL: scalar_i32_signbit_eq:
217; X64-BMI1:       # %bb.0:
218; X64-BMI1-NEXT:    movl %esi, %ecx
219; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
220; X64-BMI1-NEXT:    shrl %cl, %edi
221; X64-BMI1-NEXT:    testl $-2147483648, %edi # imm = 0x80000000
222; X64-BMI1-NEXT:    sete %al
223; X64-BMI1-NEXT:    retq
224;
225; X64-BMI2-LABEL: scalar_i32_signbit_eq:
226; X64-BMI2:       # %bb.0:
227; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
228; X64-BMI2-NEXT:    testl $-2147483648, %eax # imm = 0x80000000
229; X64-BMI2-NEXT:    sete %al
230; X64-BMI2-NEXT:    retq
231  %t0 = shl i32 2147483648, %y
232  %t1 = and i32 %t0, %x
233  %res = icmp eq i32 %t1, 0
234  ret i1 %res
235}
236
237define i1 @scalar_i32_lowestbit_eq(i32 %x, i32 %y) nounwind {
238; X86-LABEL: scalar_i32_lowestbit_eq:
239; X86:       # %bb.0:
240; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
241; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
242; X86-NEXT:    btl %ecx, %eax
243; X86-NEXT:    setae %al
244; X86-NEXT:    retl
245;
246; X64-LABEL: scalar_i32_lowestbit_eq:
247; X64:       # %bb.0:
248; X64-NEXT:    btl %esi, %edi
249; X64-NEXT:    setae %al
250; X64-NEXT:    retq
251  %t0 = shl i32 1, %y
252  %t1 = and i32 %t0, %x
253  %res = icmp eq i32 %t1, 0
254  ret i1 %res
255}
256
257define i1 @scalar_i32_bitsinmiddle_eq(i32 %x, i32 %y) nounwind {
258; X86-BMI1-LABEL: scalar_i32_bitsinmiddle_eq:
259; X86-BMI1:       # %bb.0:
260; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
261; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
262; X86-BMI1-NEXT:    shrl %cl, %eax
263; X86-BMI1-NEXT:    testl $16776960, %eax # imm = 0xFFFF00
264; X86-BMI1-NEXT:    sete %al
265; X86-BMI1-NEXT:    retl
266;
267; X86-BMI2-LABEL: scalar_i32_bitsinmiddle_eq:
268; X86-BMI2:       # %bb.0:
269; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
270; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %eax
271; X86-BMI2-NEXT:    testl $16776960, %eax # imm = 0xFFFF00
272; X86-BMI2-NEXT:    sete %al
273; X86-BMI2-NEXT:    retl
274;
275; X64-BMI1-LABEL: scalar_i32_bitsinmiddle_eq:
276; X64-BMI1:       # %bb.0:
277; X64-BMI1-NEXT:    movl %esi, %ecx
278; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
279; X64-BMI1-NEXT:    shrl %cl, %edi
280; X64-BMI1-NEXT:    testl $16776960, %edi # imm = 0xFFFF00
281; X64-BMI1-NEXT:    sete %al
282; X64-BMI1-NEXT:    retq
283;
284; X64-BMI2-LABEL: scalar_i32_bitsinmiddle_eq:
285; X64-BMI2:       # %bb.0:
286; X64-BMI2-NEXT:    shrxl %esi, %edi, %eax
287; X64-BMI2-NEXT:    testl $16776960, %eax # imm = 0xFFFF00
288; X64-BMI2-NEXT:    sete %al
289; X64-BMI2-NEXT:    retq
290  %t0 = shl i32 16776960, %y
291  %t1 = and i32 %t0, %x
292  %res = icmp eq i32 %t1, 0
293  ret i1 %res
294}
295
296; i64 scalar
297
298define i1 @scalar_i64_signbit_eq(i64 %x, i64 %y) nounwind {
299; X86-BMI1-LABEL: scalar_i64_signbit_eq:
300; X86-BMI1:       # %bb.0:
301; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
302; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
303; X86-BMI1-NEXT:    shrl %cl, %eax
304; X86-BMI1-NEXT:    xorl %edx, %edx
305; X86-BMI1-NEXT:    testb $32, %cl
306; X86-BMI1-NEXT:    cmovel %eax, %edx
307; X86-BMI1-NEXT:    testl $-2147483648, %edx # imm = 0x80000000
308; X86-BMI1-NEXT:    sete %al
309; X86-BMI1-NEXT:    retl
310;
311; X86-BMI2-LABEL: scalar_i64_signbit_eq:
312; X86-BMI2:       # %bb.0:
313; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
314; X86-BMI2-NEXT:    shrxl %eax, {{[0-9]+}}(%esp), %ecx
315; X86-BMI2-NEXT:    xorl %edx, %edx
316; X86-BMI2-NEXT:    testb $32, %al
317; X86-BMI2-NEXT:    cmovel %ecx, %edx
318; X86-BMI2-NEXT:    testl $-2147483648, %edx # imm = 0x80000000
319; X86-BMI2-NEXT:    sete %al
320; X86-BMI2-NEXT:    retl
321;
322; X64-BMI1-LABEL: scalar_i64_signbit_eq:
323; X64-BMI1:       # %bb.0:
324; X64-BMI1-NEXT:    movq %rsi, %rcx
325; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
326; X64-BMI1-NEXT:    shrq %cl, %rdi
327; X64-BMI1-NEXT:    btq $63, %rdi
328; X64-BMI1-NEXT:    setae %al
329; X64-BMI1-NEXT:    retq
330;
331; X64-BMI2-LABEL: scalar_i64_signbit_eq:
332; X64-BMI2:       # %bb.0:
333; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
334; X64-BMI2-NEXT:    btq $63, %rax
335; X64-BMI2-NEXT:    setae %al
336; X64-BMI2-NEXT:    retq
337  %t0 = shl i64 9223372036854775808, %y
338  %t1 = and i64 %t0, %x
339  %res = icmp eq i64 %t1, 0
340  ret i1 %res
341}
342
343define i1 @scalar_i64_lowestbit_eq(i64 %x, i64 %y) nounwind {
344; X86-BMI1-LABEL: scalar_i64_lowestbit_eq:
345; X86-BMI1:       # %bb.0:
346; X86-BMI1-NEXT:    pushl %esi
347; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
348; X86-BMI1-NEXT:    movl $1, %eax
349; X86-BMI1-NEXT:    xorl %edx, %edx
350; X86-BMI1-NEXT:    xorl %esi, %esi
351; X86-BMI1-NEXT:    shldl %cl, %eax, %esi
352; X86-BMI1-NEXT:    shll %cl, %eax
353; X86-BMI1-NEXT:    testb $32, %cl
354; X86-BMI1-NEXT:    cmovnel %eax, %esi
355; X86-BMI1-NEXT:    cmovnel %edx, %eax
356; X86-BMI1-NEXT:    andl {{[0-9]+}}(%esp), %esi
357; X86-BMI1-NEXT:    andl {{[0-9]+}}(%esp), %eax
358; X86-BMI1-NEXT:    orl %esi, %eax
359; X86-BMI1-NEXT:    sete %al
360; X86-BMI1-NEXT:    popl %esi
361; X86-BMI1-NEXT:    retl
362;
363; X86-BMI2-LABEL: scalar_i64_lowestbit_eq:
364; X86-BMI2:       # %bb.0:
365; X86-BMI2-NEXT:    pushl %esi
366; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
367; X86-BMI2-NEXT:    movl $1, %eax
368; X86-BMI2-NEXT:    xorl %edx, %edx
369; X86-BMI2-NEXT:    xorl %esi, %esi
370; X86-BMI2-NEXT:    shldl %cl, %eax, %esi
371; X86-BMI2-NEXT:    shlxl %ecx, %eax, %eax
372; X86-BMI2-NEXT:    testb $32, %cl
373; X86-BMI2-NEXT:    cmovnel %eax, %esi
374; X86-BMI2-NEXT:    cmovnel %edx, %eax
375; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %esi
376; X86-BMI2-NEXT:    andl {{[0-9]+}}(%esp), %eax
377; X86-BMI2-NEXT:    orl %esi, %eax
378; X86-BMI2-NEXT:    sete %al
379; X86-BMI2-NEXT:    popl %esi
380; X86-BMI2-NEXT:    retl
381;
382; X64-LABEL: scalar_i64_lowestbit_eq:
383; X64:       # %bb.0:
384; X64-NEXT:    btq %rsi, %rdi
385; X64-NEXT:    setae %al
386; X64-NEXT:    retq
387  %t0 = shl i64 1, %y
388  %t1 = and i64 %t0, %x
389  %res = icmp eq i64 %t1, 0
390  ret i1 %res
391}
392
393define i1 @scalar_i64_bitsinmiddle_eq(i64 %x, i64 %y) nounwind {
394; X86-BMI1-LABEL: scalar_i64_bitsinmiddle_eq:
395; X86-BMI1:       # %bb.0:
396; X86-BMI1-NEXT:    pushl %esi
397; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
398; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %eax
399; X86-BMI1-NEXT:    movl {{[0-9]+}}(%esp), %edx
400; X86-BMI1-NEXT:    movl %edx, %esi
401; X86-BMI1-NEXT:    shrl %cl, %esi
402; X86-BMI1-NEXT:    shrdl %cl, %edx, %eax
403; X86-BMI1-NEXT:    xorl %edx, %edx
404; X86-BMI1-NEXT:    testb $32, %cl
405; X86-BMI1-NEXT:    cmovnel %esi, %eax
406; X86-BMI1-NEXT:    cmovel %esi, %edx
407; X86-BMI1-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
408; X86-BMI1-NEXT:    movzwl %dx, %ecx
409; X86-BMI1-NEXT:    orl %eax, %ecx
410; X86-BMI1-NEXT:    sete %al
411; X86-BMI1-NEXT:    popl %esi
412; X86-BMI1-NEXT:    retl
413;
414; X86-BMI2-LABEL: scalar_i64_bitsinmiddle_eq:
415; X86-BMI2:       # %bb.0:
416; X86-BMI2-NEXT:    pushl %esi
417; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
418; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
419; X86-BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
420; X86-BMI2-NEXT:    shrdl %cl, %edx, %eax
421; X86-BMI2-NEXT:    shrxl %ecx, %edx, %edx
422; X86-BMI2-NEXT:    xorl %esi, %esi
423; X86-BMI2-NEXT:    testb $32, %cl
424; X86-BMI2-NEXT:    cmovnel %edx, %eax
425; X86-BMI2-NEXT:    cmovel %edx, %esi
426; X86-BMI2-NEXT:    andl $-65536, %eax # imm = 0xFFFF0000
427; X86-BMI2-NEXT:    movzwl %si, %ecx
428; X86-BMI2-NEXT:    orl %eax, %ecx
429; X86-BMI2-NEXT:    sete %al
430; X86-BMI2-NEXT:    popl %esi
431; X86-BMI2-NEXT:    retl
432;
433; X64-BMI1-LABEL: scalar_i64_bitsinmiddle_eq:
434; X64-BMI1:       # %bb.0:
435; X64-BMI1-NEXT:    movq %rsi, %rcx
436; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $rcx
437; X64-BMI1-NEXT:    shrq %cl, %rdi
438; X64-BMI1-NEXT:    movabsq $281474976645120, %rax # imm = 0xFFFFFFFF0000
439; X64-BMI1-NEXT:    testq %rax, %rdi
440; X64-BMI1-NEXT:    sete %al
441; X64-BMI1-NEXT:    retq
442;
443; X64-BMI2-LABEL: scalar_i64_bitsinmiddle_eq:
444; X64-BMI2:       # %bb.0:
445; X64-BMI2-NEXT:    shrxq %rsi, %rdi, %rax
446; X64-BMI2-NEXT:    movabsq $281474976645120, %rcx # imm = 0xFFFFFFFF0000
447; X64-BMI2-NEXT:    testq %rcx, %rax
448; X64-BMI2-NEXT:    sete %al
449; X64-BMI2-NEXT:    retq
450  %t0 = shl i64 281474976645120, %y
451  %t1 = and i64 %t0, %x
452  %res = icmp eq i64 %t1, 0
453  ret i1 %res
454}
455
456;------------------------------------------------------------------------------;
457; A few trivial vector tests
458;------------------------------------------------------------------------------;
459
460define <4 x i1> @vec_4xi32_splat_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
461; X86-SSE2-LABEL: vec_4xi32_splat_eq:
462; X86-SSE2:       # %bb.0:
463; X86-SSE2-NEXT:    pslld $23, %xmm1
464; X86-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
465; X86-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
466; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1]
467; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
468; X86-SSE2-NEXT:    pmuludq %xmm2, %xmm3
469; X86-SSE2-NEXT:    pmuludq %xmm1, %xmm2
470; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
471; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
472; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
473; X86-SSE2-NEXT:    pand %xmm1, %xmm0
474; X86-SSE2-NEXT:    pxor %xmm1, %xmm1
475; X86-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
476; X86-SSE2-NEXT:    retl
477;
478; AVX2-LABEL: vec_4xi32_splat_eq:
479; AVX2:       # %bb.0:
480; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1,1,1,1]
481; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
482; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
483; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
484; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm0
485; AVX2-NEXT:    ret{{[l|q]}}
486;
487; X64-SSE2-LABEL: vec_4xi32_splat_eq:
488; X64-SSE2:       # %bb.0:
489; X64-SSE2-NEXT:    pslld $23, %xmm1
490; X64-SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
491; X64-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
492; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1]
493; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
494; X64-SSE2-NEXT:    pmuludq %xmm2, %xmm3
495; X64-SSE2-NEXT:    pmuludq %xmm1, %xmm2
496; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
497; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
498; X64-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
499; X64-SSE2-NEXT:    pand %xmm1, %xmm0
500; X64-SSE2-NEXT:    pxor %xmm1, %xmm1
501; X64-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
502; X64-SSE2-NEXT:    retq
503  %t0 = shl <4 x i32> <i32 1, i32 1, i32 1, i32 1>, %y
504  %t1 = and <4 x i32> %t0, %x
505  %res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
506  ret <4 x i1> %res
507}
508
509define <4 x i1> @vec_4xi32_nonsplat_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
510; X86-SSE2-LABEL: vec_4xi32_nonsplat_eq:
511; X86-SSE2:       # %bb.0:
512; X86-SSE2-NEXT:    pslld $23, %xmm1
513; X86-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
514; X86-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
515; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,16776960,2147483648]
516; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
517; X86-SSE2-NEXT:    pmuludq %xmm1, %xmm2
518; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
519; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
520; X86-SSE2-NEXT:    pmuludq %xmm3, %xmm1
521; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
522; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
523; X86-SSE2-NEXT:    pand %xmm2, %xmm0
524; X86-SSE2-NEXT:    pxor %xmm1, %xmm1
525; X86-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
526; X86-SSE2-NEXT:    retl
527;
528; AVX2-LABEL: vec_4xi32_nonsplat_eq:
529; AVX2:       # %bb.0:
530; AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,16776960,2147483648]
531; AVX2-NEXT:    vpsllvd %xmm1, %xmm2, %xmm1
532; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
533; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
534; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
535; AVX2-NEXT:    ret{{[l|q]}}
536;
537; X64-SSE2-LABEL: vec_4xi32_nonsplat_eq:
538; X64-SSE2:       # %bb.0:
539; X64-SSE2-NEXT:    pslld $23, %xmm1
540; X64-SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
541; X64-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
542; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [0,1,16776960,2147483648]
543; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm2[1,1,3,3]
544; X64-SSE2-NEXT:    pmuludq %xmm1, %xmm2
545; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
546; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
547; X64-SSE2-NEXT:    pmuludq %xmm3, %xmm1
548; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
549; X64-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
550; X64-SSE2-NEXT:    pand %xmm2, %xmm0
551; X64-SSE2-NEXT:    pxor %xmm1, %xmm1
552; X64-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
553; X64-SSE2-NEXT:    retq
554  %t0 = shl <4 x i32> <i32 0, i32 1, i32 16776960, i32 2147483648>, %y
555  %t1 = and <4 x i32> %t0, %x
556  %res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
557  ret <4 x i1> %res
558}
559
560define <4 x i1> @vec_4xi32_nonsplat_undef0_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
561; X86-SSE2-LABEL: vec_4xi32_nonsplat_undef0_eq:
562; X86-SSE2:       # %bb.0:
563; X86-SSE2-NEXT:    movl $1, %eax
564; X86-SSE2-NEXT:    movd %eax, %xmm2
565; X86-SSE2-NEXT:    pslld $23, %xmm1
566; X86-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
567; X86-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
568; X86-SSE2-NEXT:    pmuludq %xmm1, %xmm2
569; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
570; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
571; X86-SSE2-NEXT:    pmuludq {{\.LCPI.*}}, %xmm1
572; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
573; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
574; X86-SSE2-NEXT:    pand %xmm2, %xmm0
575; X86-SSE2-NEXT:    pxor %xmm1, %xmm1
576; X86-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
577; X86-SSE2-NEXT:    retl
578;
579; AVX2-LABEL: vec_4xi32_nonsplat_undef0_eq:
580; AVX2:       # %bb.0:
581; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1,1,1,1]
582; AVX2-NEXT:    vpxor %xmm3, %xmm3, %xmm3
583; AVX2-NEXT:    vpsrlvd %xmm1, %xmm0, %xmm0
584; AVX2-NEXT:    vpand %xmm2, %xmm0, %xmm0
585; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm0, %xmm0
586; AVX2-NEXT:    ret{{[l|q]}}
587;
588; X64-SSE2-LABEL: vec_4xi32_nonsplat_undef0_eq:
589; X64-SSE2:       # %bb.0:
590; X64-SSE2-NEXT:    movl $1, %eax
591; X64-SSE2-NEXT:    movd %eax, %xmm2
592; X64-SSE2-NEXT:    pslld $23, %xmm1
593; X64-SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
594; X64-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
595; X64-SSE2-NEXT:    pmuludq %xmm1, %xmm2
596; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
597; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
598; X64-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
599; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
600; X64-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
601; X64-SSE2-NEXT:    pand %xmm2, %xmm0
602; X64-SSE2-NEXT:    pxor %xmm1, %xmm1
603; X64-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
604; X64-SSE2-NEXT:    retq
605  %t0 = shl <4 x i32> <i32 1, i32 1, i32 undef, i32 1>, %y
606  %t1 = and <4 x i32> %t0, %x
607  %res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 0, i32 0>
608  ret <4 x i1> %res
609}
610define <4 x i1> @vec_4xi32_nonsplat_undef1_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
611; X86-SSE2-LABEL: vec_4xi32_nonsplat_undef1_eq:
612; X86-SSE2:       # %bb.0:
613; X86-SSE2-NEXT:    pslld $23, %xmm1
614; X86-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
615; X86-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
616; X86-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1]
617; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
618; X86-SSE2-NEXT:    pmuludq %xmm2, %xmm3
619; X86-SSE2-NEXT:    pmuludq %xmm1, %xmm2
620; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
621; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
622; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
623; X86-SSE2-NEXT:    pand %xmm1, %xmm0
624; X86-SSE2-NEXT:    pxor %xmm1, %xmm1
625; X86-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
626; X86-SSE2-NEXT:    retl
627;
628; AVX2-LABEL: vec_4xi32_nonsplat_undef1_eq:
629; AVX2:       # %bb.0:
630; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1,1,1,1]
631; AVX2-NEXT:    vpsllvd %xmm1, %xmm2, %xmm1
632; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
633; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
634; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
635; AVX2-NEXT:    ret{{[l|q]}}
636;
637; X64-SSE2-LABEL: vec_4xi32_nonsplat_undef1_eq:
638; X64-SSE2:       # %bb.0:
639; X64-SSE2-NEXT:    pslld $23, %xmm1
640; X64-SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
641; X64-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
642; X64-SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1,1,1,1]
643; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[1,1,3,3]
644; X64-SSE2-NEXT:    pmuludq %xmm2, %xmm3
645; X64-SSE2-NEXT:    pmuludq %xmm1, %xmm2
646; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
647; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[0,2,2,3]
648; X64-SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
649; X64-SSE2-NEXT:    pand %xmm1, %xmm0
650; X64-SSE2-NEXT:    pxor %xmm1, %xmm1
651; X64-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
652; X64-SSE2-NEXT:    retq
653  %t0 = shl <4 x i32> <i32 1, i32 1, i32 1, i32 1>, %y
654  %t1 = and <4 x i32> %t0, %x
655  %res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 undef, i32 0>
656  ret <4 x i1> %res
657}
658define <4 x i1> @vec_4xi32_nonsplat_undef2_eq(<4 x i32> %x, <4 x i32> %y) nounwind {
659; X86-SSE2-LABEL: vec_4xi32_nonsplat_undef2_eq:
660; X86-SSE2:       # %bb.0:
661; X86-SSE2-NEXT:    movl $1, %eax
662; X86-SSE2-NEXT:    movd %eax, %xmm2
663; X86-SSE2-NEXT:    pslld $23, %xmm1
664; X86-SSE2-NEXT:    paddd {{\.LCPI.*}}, %xmm1
665; X86-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
666; X86-SSE2-NEXT:    pmuludq %xmm1, %xmm2
667; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
668; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
669; X86-SSE2-NEXT:    pmuludq {{\.LCPI.*}}, %xmm1
670; X86-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
671; X86-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
672; X86-SSE2-NEXT:    pand %xmm2, %xmm0
673; X86-SSE2-NEXT:    pxor %xmm1, %xmm1
674; X86-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
675; X86-SSE2-NEXT:    retl
676;
677; AVX2-LABEL: vec_4xi32_nonsplat_undef2_eq:
678; AVX2:       # %bb.0:
679; AVX2-NEXT:    vpbroadcastd {{.*#+}} xmm2 = [1,1,1,1]
680; AVX2-NEXT:    vpsllvd %xmm1, %xmm2, %xmm1
681; AVX2-NEXT:    vpand %xmm0, %xmm1, %xmm0
682; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
683; AVX2-NEXT:    vpcmpeqd %xmm1, %xmm0, %xmm0
684; AVX2-NEXT:    ret{{[l|q]}}
685;
686; X64-SSE2-LABEL: vec_4xi32_nonsplat_undef2_eq:
687; X64-SSE2:       # %bb.0:
688; X64-SSE2-NEXT:    movl $1, %eax
689; X64-SSE2-NEXT:    movd %eax, %xmm2
690; X64-SSE2-NEXT:    pslld $23, %xmm1
691; X64-SSE2-NEXT:    paddd {{.*}}(%rip), %xmm1
692; X64-SSE2-NEXT:    cvttps2dq %xmm1, %xmm1
693; X64-SSE2-NEXT:    pmuludq %xmm1, %xmm2
694; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
695; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
696; X64-SSE2-NEXT:    pmuludq {{.*}}(%rip), %xmm1
697; X64-SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
698; X64-SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
699; X64-SSE2-NEXT:    pand %xmm2, %xmm0
700; X64-SSE2-NEXT:    pxor %xmm1, %xmm1
701; X64-SSE2-NEXT:    pcmpeqd %xmm1, %xmm0
702; X64-SSE2-NEXT:    retq
703  %t0 = shl <4 x i32> <i32 1, i32 1, i32 undef, i32 1>, %y
704  %t1 = and <4 x i32> %t0, %x
705  %res = icmp eq <4 x i32> %t1, <i32 0, i32 0, i32 undef, i32 0>
706  ret <4 x i1> %res
707}
708
709;------------------------------------------------------------------------------;
710; A special tests
711;------------------------------------------------------------------------------;
712
713define i1 @scalar_i8_signbit_ne(i8 %x, i8 %y) nounwind {
714; X86-LABEL: scalar_i8_signbit_ne:
715; X86:       # %bb.0:
716; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
717; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
718; X86-NEXT:    shrb %cl, %al
719; X86-NEXT:    shrb $7, %al
720; X86-NEXT:    retl
721;
722; X64-LABEL: scalar_i8_signbit_ne:
723; X64:       # %bb.0:
724; X64-NEXT:    movl %esi, %ecx
725; X64-NEXT:    movl %edi, %eax
726; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
727; X64-NEXT:    shrb %cl, %al
728; X64-NEXT:    shrb $7, %al
729; X64-NEXT:    # kill: def $al killed $al killed $eax
730; X64-NEXT:    retq
731  %t0 = shl i8 128, %y
732  %t1 = and i8 %t0, %x
733  %res = icmp ne i8 %t1, 0 ;  we are perfectly happy with 'ne' predicate
734  ret i1 %res
735}
736
737;------------------------------------------------------------------------------;
738; What if X is a constant too?
739;------------------------------------------------------------------------------;
740
741define i1 @scalar_i32_x_is_const_eq(i32 %y) nounwind {
742; X86-BMI1-LABEL: scalar_i32_x_is_const_eq:
743; X86-BMI1:       # %bb.0:
744; X86-BMI1-NEXT:    movb {{[0-9]+}}(%esp), %cl
745; X86-BMI1-NEXT:    movl $-1437226411, %eax # imm = 0xAA55AA55
746; X86-BMI1-NEXT:    shll %cl, %eax
747; X86-BMI1-NEXT:    testb $1, %al
748; X86-BMI1-NEXT:    sete %al
749; X86-BMI1-NEXT:    retl
750;
751; X86-BMI2-LABEL: scalar_i32_x_is_const_eq:
752; X86-BMI2:       # %bb.0:
753; X86-BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
754; X86-BMI2-NEXT:    movl $-1437226411, %ecx # imm = 0xAA55AA55
755; X86-BMI2-NEXT:    shlxl %eax, %ecx, %eax
756; X86-BMI2-NEXT:    testb $1, %al
757; X86-BMI2-NEXT:    sete %al
758; X86-BMI2-NEXT:    retl
759;
760; X64-BMI1-LABEL: scalar_i32_x_is_const_eq:
761; X64-BMI1:       # %bb.0:
762; X64-BMI1-NEXT:    movl %edi, %ecx
763; X64-BMI1-NEXT:    movl $-1437226411, %eax # imm = 0xAA55AA55
764; X64-BMI1-NEXT:    # kill: def $cl killed $cl killed $ecx
765; X64-BMI1-NEXT:    shll %cl, %eax
766; X64-BMI1-NEXT:    testb $1, %al
767; X64-BMI1-NEXT:    sete %al
768; X64-BMI1-NEXT:    retq
769;
770; X64-BMI2-LABEL: scalar_i32_x_is_const_eq:
771; X64-BMI2:       # %bb.0:
772; X64-BMI2-NEXT:    movl $-1437226411, %eax # imm = 0xAA55AA55
773; X64-BMI2-NEXT:    shlxl %edi, %eax, %eax
774; X64-BMI2-NEXT:    testb $1, %al
775; X64-BMI2-NEXT:    sete %al
776; X64-BMI2-NEXT:    retq
777  %t0 = shl i32 2857740885, %y
778  %t1 = and i32 %t0, 1
779  %res = icmp eq i32 %t1, 0
780  ret i1 %res
781}
782define i1 @scalar_i32_x_is_const2_eq(i32 %y) nounwind {
783; X86-LABEL: scalar_i32_x_is_const2_eq:
784; X86:       # %bb.0:
785; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
786; X86-NEXT:    movl $-1437226411, %ecx # imm = 0xAA55AA55
787; X86-NEXT:    btl %eax, %ecx
788; X86-NEXT:    setae %al
789; X86-NEXT:    retl
790;
791; X64-LABEL: scalar_i32_x_is_const2_eq:
792; X64:       # %bb.0:
793; X64-NEXT:    movl $-1437226411, %eax # imm = 0xAA55AA55
794; X64-NEXT:    btl %edi, %eax
795; X64-NEXT:    setae %al
796; X64-NEXT:    retq
797  %t0 = shl i32 1, %y
798  %t1 = and i32 %t0, 2857740885
799  %res = icmp eq i32 %t1, 0
800  ret i1 %res
801}
802
803;------------------------------------------------------------------------------;
804; A few negative tests
805;------------------------------------------------------------------------------;
806
807define i1 @negative_scalar_i8_bitsinmiddle_slt(i8 %x, i8 %y) nounwind {
808; X86-LABEL: negative_scalar_i8_bitsinmiddle_slt:
809; X86:       # %bb.0:
810; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
811; X86-NEXT:    movb $24, %al
812; X86-NEXT:    shlb %cl, %al
813; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
814; X86-NEXT:    shrb $7, %al
815; X86-NEXT:    retl
816;
817; X64-LABEL: negative_scalar_i8_bitsinmiddle_slt:
818; X64:       # %bb.0:
819; X64-NEXT:    movl %esi, %ecx
820; X64-NEXT:    movb $24, %al
821; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
822; X64-NEXT:    shlb %cl, %al
823; X64-NEXT:    andb %dil, %al
824; X64-NEXT:    shrb $7, %al
825; X64-NEXT:    retq
826  %t0 = shl i8 24, %y
827  %t1 = and i8 %t0, %x
828  %res = icmp slt i8 %t1, 0
829  ret i1 %res
830}
831
832define i1 @scalar_i8_signbit_eq_with_nonzero(i8 %x, i8 %y) nounwind {
833; X86-LABEL: scalar_i8_signbit_eq_with_nonzero:
834; X86:       # %bb.0:
835; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
836; X86-NEXT:    movb $-128, %al
837; X86-NEXT:    shlb %cl, %al
838; X86-NEXT:    andb {{[0-9]+}}(%esp), %al
839; X86-NEXT:    cmpb $1, %al
840; X86-NEXT:    sete %al
841; X86-NEXT:    retl
842;
843; X64-LABEL: scalar_i8_signbit_eq_with_nonzero:
844; X64:       # %bb.0:
845; X64-NEXT:    movl %esi, %ecx
846; X64-NEXT:    movb $-128, %al
847; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
848; X64-NEXT:    shlb %cl, %al
849; X64-NEXT:    andb %dil, %al
850; X64-NEXT:    cmpb $1, %al
851; X64-NEXT:    sete %al
852; X64-NEXT:    retq
853  %t0 = shl i8 128, %y
854  %t1 = and i8 %t0, %x
855  %res = icmp eq i8 %t1, 1 ; should be comparing with 0
856  ret i1 %res
857}
858