1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X86
3; RUN: llc < %s -mtriple=x86_64-linux -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X64-LIN
4; RUN: llc < %s -mtriple=x86_64-win32 -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,X64-WIN
5
6; Though it is undefined, we want xor undef,undef to produce zero.
7define <4 x i32> @test1() nounwind {
8; CHECK-LABEL: test1:
9; CHECK:       # %bb.0:
10; CHECK-NEXT:    xorps %xmm0, %xmm0
11; CHECK-NEXT:    ret{{[l|q]}}
12  %tmp = xor <4 x i32> undef, undef
13  ret <4 x i32> %tmp
14}
15
16; Though it is undefined, we want xor undef,undef to produce zero.
17define i32 @test2() nounwind{
18; CHECK-LABEL: test2:
19; CHECK:       # %bb.0:
20; CHECK-NEXT:    xorl %eax, %eax
21; CHECK-NEXT:    ret{{[l|q]}}
22  %tmp = xor i32 undef, undef
23  ret i32 %tmp
24}
25
26define i32 @test3(i32 %a, i32 %b) nounwind  {
27; X86-LABEL: test3:
28; X86:       # %bb.0: # %entry
29; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
30; X86-NEXT:    notl %eax
31; X86-NEXT:    andl {{[0-9]+}}(%esp), %eax
32; X86-NEXT:    shrl %eax
33; X86-NEXT:    retl
34;
35; X64-LIN-LABEL: test3:
36; X64-LIN:       # %bb.0: # %entry
37; X64-LIN-NEXT:    movl %esi, %eax
38; X64-LIN-NEXT:    notl %eax
39; X64-LIN-NEXT:    andl %edi, %eax
40; X64-LIN-NEXT:    shrl %eax
41; X64-LIN-NEXT:    retq
42;
43; X64-WIN-LABEL: test3:
44; X64-WIN:       # %bb.0: # %entry
45; X64-WIN-NEXT:    movl %edx, %eax
46; X64-WIN-NEXT:    notl %eax
47; X64-WIN-NEXT:    andl %ecx, %eax
48; X64-WIN-NEXT:    shrl %eax
49; X64-WIN-NEXT:    retq
50entry:
51  %tmp1not = xor i32 %b, -2
52  %tmp3 = and i32 %tmp1not, %a
53  %tmp4 = lshr i32 %tmp3, 1
54  ret i32 %tmp4
55}
56
57define i32 @test4(i32 %a, i32 %b) nounwind  {
58; X86-LABEL: test4:
59; X86:       # %bb.0: # %entry
60; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
61; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
62; X86-NEXT:    .p2align 4, 0x90
63; X86-NEXT:  .LBB3_1: # %bb
64; X86-NEXT:    # =>This Inner Loop Header: Depth=1
65; X86-NEXT:    xorl %ecx, %eax
66; X86-NEXT:    movl %eax, %edx
67; X86-NEXT:    notl %edx
68; X86-NEXT:    andl %ecx, %edx
69; X86-NEXT:    addl %edx, %edx
70; X86-NEXT:    movl %edx, %ecx
71; X86-NEXT:    jne .LBB3_1
72; X86-NEXT:  # %bb.2: # %bb12
73; X86-NEXT:    retl
74;
75; X64-LIN-LABEL: test4:
76; X64-LIN:       # %bb.0: # %entry
77; X64-LIN-NEXT:    movl %edi, %eax
78; X64-LIN-NEXT:    .p2align 4, 0x90
79; X64-LIN-NEXT:  .LBB3_1: # %bb
80; X64-LIN-NEXT:    # =>This Inner Loop Header: Depth=1
81; X64-LIN-NEXT:    xorl %esi, %eax
82; X64-LIN-NEXT:    movl %eax, %ecx
83; X64-LIN-NEXT:    notl %ecx
84; X64-LIN-NEXT:    andl %esi, %ecx
85; X64-LIN-NEXT:    addl %ecx, %ecx
86; X64-LIN-NEXT:    movl %ecx, %esi
87; X64-LIN-NEXT:    jne .LBB3_1
88; X64-LIN-NEXT:  # %bb.2: # %bb12
89; X64-LIN-NEXT:    retq
90;
91; X64-WIN-LABEL: test4:
92; X64-WIN:       # %bb.0: # %entry
93; X64-WIN-NEXT:    movl %ecx, %eax
94; X64-WIN-NEXT:    .p2align 4, 0x90
95; X64-WIN-NEXT:  .LBB3_1: # %bb
96; X64-WIN-NEXT:    # =>This Inner Loop Header: Depth=1
97; X64-WIN-NEXT:    xorl %edx, %eax
98; X64-WIN-NEXT:    movl %eax, %ecx
99; X64-WIN-NEXT:    notl %ecx
100; X64-WIN-NEXT:    andl %edx, %ecx
101; X64-WIN-NEXT:    addl %ecx, %ecx
102; X64-WIN-NEXT:    movl %ecx, %edx
103; X64-WIN-NEXT:    jne .LBB3_1
104; X64-WIN-NEXT:  # %bb.2: # %bb12
105; X64-WIN-NEXT:    retq
106entry:
107  br label %bb
108bb:
109  %b_addr.0 = phi i32 [ %b, %entry ], [ %tmp8, %bb ]
110  %a_addr.0 = phi i32 [ %a, %entry ], [ %tmp3, %bb ]
111  %tmp3 = xor i32 %a_addr.0, %b_addr.0
112  %tmp4not = xor i32 %tmp3, 2147483647
113  %tmp6 = and i32 %tmp4not, %b_addr.0
114  %tmp8 = shl i32 %tmp6, 1
115  %tmp10 = icmp eq i32 %tmp8, 0
116  br i1 %tmp10, label %bb12, label %bb
117bb12:
118  ret i32 %tmp3
119}
120
121define i16 @test5(i16 %a, i16 %b) nounwind  {
122; X86-LABEL: test5:
123; X86:       # %bb.0: # %entry
124; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
125; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
126; X86-NEXT:    .p2align 4, 0x90
127; X86-NEXT:  .LBB4_1: # %bb
128; X86-NEXT:    # =>This Inner Loop Header: Depth=1
129; X86-NEXT:    xorl %ecx, %eax
130; X86-NEXT:    movl %eax, %edx
131; X86-NEXT:    notl %edx
132; X86-NEXT:    andl %ecx, %edx
133; X86-NEXT:    addl %edx, %edx
134; X86-NEXT:    testw %dx, %dx
135; X86-NEXT:    movl %edx, %ecx
136; X86-NEXT:    jne .LBB4_1
137; X86-NEXT:  # %bb.2: # %bb12
138; X86-NEXT:    # kill: def $ax killed $ax killed $eax
139; X86-NEXT:    retl
140;
141; X64-LIN-LABEL: test5:
142; X64-LIN:       # %bb.0: # %entry
143; X64-LIN-NEXT:    movl %edi, %eax
144; X64-LIN-NEXT:    .p2align 4, 0x90
145; X64-LIN-NEXT:  .LBB4_1: # %bb
146; X64-LIN-NEXT:    # =>This Inner Loop Header: Depth=1
147; X64-LIN-NEXT:    xorl %esi, %eax
148; X64-LIN-NEXT:    movl %eax, %ecx
149; X64-LIN-NEXT:    notl %ecx
150; X64-LIN-NEXT:    andl %esi, %ecx
151; X64-LIN-NEXT:    addl %ecx, %ecx
152; X64-LIN-NEXT:    testw %cx, %cx
153; X64-LIN-NEXT:    movl %ecx, %esi
154; X64-LIN-NEXT:    jne .LBB4_1
155; X64-LIN-NEXT:  # %bb.2: # %bb12
156; X64-LIN-NEXT:    # kill: def $ax killed $ax killed $eax
157; X64-LIN-NEXT:    retq
158;
159; X64-WIN-LABEL: test5:
160; X64-WIN:       # %bb.0: # %entry
161; X64-WIN-NEXT:    # kill: def $dx killed $dx def $edx
162; X64-WIN-NEXT:    movl %ecx, %eax
163; X64-WIN-NEXT:    .p2align 4, 0x90
164; X64-WIN-NEXT:  .LBB4_1: # %bb
165; X64-WIN-NEXT:    # =>This Inner Loop Header: Depth=1
166; X64-WIN-NEXT:    xorl %edx, %eax
167; X64-WIN-NEXT:    movl %eax, %ecx
168; X64-WIN-NEXT:    notl %ecx
169; X64-WIN-NEXT:    andl %edx, %ecx
170; X64-WIN-NEXT:    addl %ecx, %ecx
171; X64-WIN-NEXT:    testw %cx, %cx
172; X64-WIN-NEXT:    movl %ecx, %edx
173; X64-WIN-NEXT:    jne .LBB4_1
174; X64-WIN-NEXT:  # %bb.2: # %bb12
175; X64-WIN-NEXT:    # kill: def $ax killed $ax killed $eax
176; X64-WIN-NEXT:    retq
177entry:
178  br label %bb
179bb:
180  %b_addr.0 = phi i16 [ %b, %entry ], [ %tmp8, %bb ]
181  %a_addr.0 = phi i16 [ %a, %entry ], [ %tmp3, %bb ]
182  %tmp3 = xor i16 %a_addr.0, %b_addr.0
183  %tmp4not = xor i16 %tmp3, 32767
184  %tmp6 = and i16 %tmp4not, %b_addr.0
185  %tmp8 = shl i16 %tmp6, 1
186  %tmp10 = icmp eq i16 %tmp8, 0
187  br i1 %tmp10, label %bb12, label %bb
188bb12:
189  ret i16 %tmp3
190}
191
192define i8 @test6(i8 %a, i8 %b) nounwind  {
193; X86-LABEL: test6:
194; X86:       # %bb.0: # %entry
195; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
196; X86-NEXT:    movb {{[0-9]+}}(%esp), %al
197; X86-NEXT:    .p2align 4, 0x90
198; X86-NEXT:  .LBB5_1: # %bb
199; X86-NEXT:    # =>This Inner Loop Header: Depth=1
200; X86-NEXT:    xorb %cl, %al
201; X86-NEXT:    movl %eax, %edx
202; X86-NEXT:    notb %dl
203; X86-NEXT:    andb %cl, %dl
204; X86-NEXT:    addb %dl, %dl
205; X86-NEXT:    movl %edx, %ecx
206; X86-NEXT:    jne .LBB5_1
207; X86-NEXT:  # %bb.2: # %bb12
208; X86-NEXT:    retl
209;
210; X64-LIN-LABEL: test6:
211; X64-LIN:       # %bb.0: # %entry
212; X64-LIN-NEXT:    movl %edi, %eax
213; X64-LIN-NEXT:    .p2align 4, 0x90
214; X64-LIN-NEXT:  .LBB5_1: # %bb
215; X64-LIN-NEXT:    # =>This Inner Loop Header: Depth=1
216; X64-LIN-NEXT:    xorb %sil, %al
217; X64-LIN-NEXT:    movl %eax, %ecx
218; X64-LIN-NEXT:    notb %cl
219; X64-LIN-NEXT:    andb %sil, %cl
220; X64-LIN-NEXT:    addb %cl, %cl
221; X64-LIN-NEXT:    movl %ecx, %esi
222; X64-LIN-NEXT:    jne .LBB5_1
223; X64-LIN-NEXT:  # %bb.2: # %bb12
224; X64-LIN-NEXT:    # kill: def $al killed $al killed $eax
225; X64-LIN-NEXT:    retq
226;
227; X64-WIN-LABEL: test6:
228; X64-WIN:       # %bb.0: # %entry
229; X64-WIN-NEXT:    movl %ecx, %eax
230; X64-WIN-NEXT:    .p2align 4, 0x90
231; X64-WIN-NEXT:  .LBB5_1: # %bb
232; X64-WIN-NEXT:    # =>This Inner Loop Header: Depth=1
233; X64-WIN-NEXT:    xorb %dl, %al
234; X64-WIN-NEXT:    movl %eax, %ecx
235; X64-WIN-NEXT:    notb %cl
236; X64-WIN-NEXT:    andb %dl, %cl
237; X64-WIN-NEXT:    addb %cl, %cl
238; X64-WIN-NEXT:    movl %ecx, %edx
239; X64-WIN-NEXT:    jne .LBB5_1
240; X64-WIN-NEXT:  # %bb.2: # %bb12
241; X64-WIN-NEXT:    retq
242entry:
243  br label %bb
244bb:
245  %b_addr.0 = phi i8 [ %b, %entry ], [ %tmp8, %bb ]
246  %a_addr.0 = phi i8 [ %a, %entry ], [ %tmp3, %bb ]
247  %tmp3 = xor i8 %a_addr.0, %b_addr.0
248  %tmp4not = xor i8 %tmp3, 127
249  %tmp6 = and i8 %tmp4not, %b_addr.0
250  %tmp8 = shl i8 %tmp6, 1
251  %tmp10 = icmp eq i8 %tmp8, 0
252  br i1 %tmp10, label %bb12, label %bb
253bb12:
254  ret i8 %tmp3
255}
256
257define i32 @test7(i32 %a, i32 %b) nounwind  {
258; X86-LABEL: test7:
259; X86:       # %bb.0: # %entry
260; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
261; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
262; X86-NEXT:    .p2align 4, 0x90
263; X86-NEXT:  .LBB6_1: # %bb
264; X86-NEXT:    # =>This Inner Loop Header: Depth=1
265; X86-NEXT:    xorl %ecx, %eax
266; X86-NEXT:    movl %eax, %edx
267; X86-NEXT:    xorl $2147483646, %edx # imm = 0x7FFFFFFE
268; X86-NEXT:    andl %ecx, %edx
269; X86-NEXT:    addl %edx, %edx
270; X86-NEXT:    movl %edx, %ecx
271; X86-NEXT:    jne .LBB6_1
272; X86-NEXT:  # %bb.2: # %bb12
273; X86-NEXT:    retl
274;
275; X64-LIN-LABEL: test7:
276; X64-LIN:       # %bb.0: # %entry
277; X64-LIN-NEXT:    movl %edi, %eax
278; X64-LIN-NEXT:    .p2align 4, 0x90
279; X64-LIN-NEXT:  .LBB6_1: # %bb
280; X64-LIN-NEXT:    # =>This Inner Loop Header: Depth=1
281; X64-LIN-NEXT:    xorl %esi, %eax
282; X64-LIN-NEXT:    movl %eax, %ecx
283; X64-LIN-NEXT:    xorl $2147483646, %ecx # imm = 0x7FFFFFFE
284; X64-LIN-NEXT:    andl %esi, %ecx
285; X64-LIN-NEXT:    addl %ecx, %ecx
286; X64-LIN-NEXT:    movl %ecx, %esi
287; X64-LIN-NEXT:    jne .LBB6_1
288; X64-LIN-NEXT:  # %bb.2: # %bb12
289; X64-LIN-NEXT:    retq
290;
291; X64-WIN-LABEL: test7:
292; X64-WIN:       # %bb.0: # %entry
293; X64-WIN-NEXT:    movl %ecx, %eax
294; X64-WIN-NEXT:    .p2align 4, 0x90
295; X64-WIN-NEXT:  .LBB6_1: # %bb
296; X64-WIN-NEXT:    # =>This Inner Loop Header: Depth=1
297; X64-WIN-NEXT:    xorl %edx, %eax
298; X64-WIN-NEXT:    movl %eax, %ecx
299; X64-WIN-NEXT:    xorl $2147483646, %ecx # imm = 0x7FFFFFFE
300; X64-WIN-NEXT:    andl %edx, %ecx
301; X64-WIN-NEXT:    addl %ecx, %ecx
302; X64-WIN-NEXT:    movl %ecx, %edx
303; X64-WIN-NEXT:    jne .LBB6_1
304; X64-WIN-NEXT:  # %bb.2: # %bb12
305; X64-WIN-NEXT:    retq
306entry:
307  br label %bb
308bb:
309  %b_addr.0 = phi i32 [ %b, %entry ], [ %tmp8, %bb ]
310  %a_addr.0 = phi i32 [ %a, %entry ], [ %tmp3, %bb ]
311  %tmp3 = xor i32 %a_addr.0, %b_addr.0
312  %tmp4not = xor i32 %tmp3, 2147483646
313  %tmp6 = and i32 %tmp4not, %b_addr.0
314  %tmp8 = shl i32 %tmp6, 1
315  %tmp10 = icmp eq i32 %tmp8, 0
316  br i1 %tmp10, label %bb12, label %bb
317bb12:
318  ret i32 %tmp3
319}
320
321; rdar://7553032
322define i32 @test8(i32 %a) nounwind {
323; X86-LABEL: test8:
324; X86:       # %bb.0: # %entry
325; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
326; X86-NEXT:    notl %eax
327; X86-NEXT:    retl
328;
329; X64-LIN-LABEL: test8:
330; X64-LIN:       # %bb.0: # %entry
331; X64-LIN-NEXT:    movl %edi, %eax
332; X64-LIN-NEXT:    notl %eax
333; X64-LIN-NEXT:    retq
334;
335; X64-WIN-LABEL: test8:
336; X64-WIN:       # %bb.0: # %entry
337; X64-WIN-NEXT:    movl %ecx, %eax
338; X64-WIN-NEXT:    notl %eax
339; X64-WIN-NEXT:    retq
340entry:
341  %t1 = sub i32 0, %a
342  %t2 = add i32 %t1, -1
343  ret i32 %t2
344}
345
346define i32 @test9(i32 %a) nounwind {
347; X86-LABEL: test9:
348; X86:       # %bb.0:
349; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
350; X86-NEXT:    notl %eax
351; X86-NEXT:    andl $4096, %eax # imm = 0x1000
352; X86-NEXT:    retl
353;
354; X64-LIN-LABEL: test9:
355; X64-LIN:       # %bb.0:
356; X64-LIN-NEXT:    movl %edi, %eax
357; X64-LIN-NEXT:    notl %eax
358; X64-LIN-NEXT:    andl $4096, %eax # imm = 0x1000
359; X64-LIN-NEXT:    retq
360;
361; X64-WIN-LABEL: test9:
362; X64-WIN:       # %bb.0:
363; X64-WIN-NEXT:    movl %ecx, %eax
364; X64-WIN-NEXT:    notl %eax
365; X64-WIN-NEXT:    andl $4096, %eax # imm = 0x1000
366; X64-WIN-NEXT:    retq
367  %1 = and i32 %a, 4096
368  %2 = xor i32 %1, 4096
369  ret i32 %2
370}
371
372; PR15948
373define <4 x i32> @test10(<4 x i32> %a) nounwind {
374; X86-LABEL: test10:
375; X86:       # %bb.0:
376; X86-NEXT:    andnps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0
377; X86-NEXT:    retl
378;
379; X64-LIN-LABEL: test10:
380; X64-LIN:       # %bb.0:
381; X64-LIN-NEXT:    andnps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
382; X64-LIN-NEXT:    retq
383;
384; X64-WIN-LABEL: test10:
385; X64-WIN:       # %bb.0:
386; X64-WIN-NEXT:    movaps (%rcx), %xmm0
387; X64-WIN-NEXT:    andnps __xmm@00001000000010000000100000001000(%rip), %xmm0
388; X64-WIN-NEXT:    retq
389  %1 = and <4 x i32> %a, <i32 4096, i32 4096, i32 4096, i32 4096>
390  %2 = xor <4 x i32> %1, <i32 4096, i32 4096, i32 4096, i32 4096>
391  ret <4 x i32> %2
392}
393
394define i32 @PR17487(i1 %tobool) {
395; X86-LABEL: PR17487:
396; X86:       # %bb.0:
397; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
398; X86-NEXT:    notb %cl
399; X86-NEXT:    xorl %eax, %eax
400; X86-NEXT:    testb $1, %cl
401; X86-NEXT:    sete %al
402; X86-NEXT:    retl
403;
404; X64-LIN-LABEL: PR17487:
405; X64-LIN:       # %bb.0:
406; X64-LIN-NEXT:    movl %edi, %eax
407; X64-LIN-NEXT:    andl $1, %eax
408; X64-LIN-NEXT:    retq
409;
410; X64-WIN-LABEL: PR17487:
411; X64-WIN:       # %bb.0:
412; X64-WIN-NEXT:    movzbl %cl, %eax
413; X64-WIN-NEXT:    andl $1, %eax
414; X64-WIN-NEXT:    retq
415  %tmp = insertelement <2 x i1> undef, i1 %tobool, i32 1
416  %tmp1 = zext <2 x i1> %tmp to <2 x i64>
417  %tmp2 = xor <2 x i64> %tmp1, <i64 1, i64 1>
418  %tmp3 = extractelement <2 x i64> %tmp2, i32 1
419  %add = add nsw i64 0, %tmp3
420  %cmp6 = icmp ne i64 %add, 1
421  %conv7 = zext i1 %cmp6 to i32
422  ret i32 %conv7
423}
424
425define i32 @test11(i32 %b) {
426; X86-LABEL: test11:
427; X86:       # %bb.0:
428; X86-NEXT:    movb {{[0-9]+}}(%esp), %cl
429; X86-NEXT:    movl $-2, %eax
430; X86-NEXT:    roll %cl, %eax
431; X86-NEXT:    retl
432;
433; X64-LIN-LABEL: test11:
434; X64-LIN:       # %bb.0:
435; X64-LIN-NEXT:    movl %edi, %ecx
436; X64-LIN-NEXT:    movl $-2, %eax
437; X64-LIN-NEXT:    # kill: def $cl killed $cl killed $ecx
438; X64-LIN-NEXT:    roll %cl, %eax
439; X64-LIN-NEXT:    retq
440;
441; X64-WIN-LABEL: test11:
442; X64-WIN:       # %bb.0:
443; X64-WIN-NEXT:    movl $-2, %eax
444; X64-WIN-NEXT:    # kill: def $cl killed $cl killed $ecx
445; X64-WIN-NEXT:    roll %cl, %eax
446; X64-WIN-NEXT:    retq
447  %shl = shl i32 1, %b
448  %neg = xor i32 %shl, -1
449  ret i32 %neg
450}
451
452%struct.ref_s = type { %union.v, i16, i16 }
453%union.v = type { i64 }
454
455define ptr @test12(ptr %op, i64 %osbot, i64 %intval) {
456; X86-LABEL: test12:
457; X86:       # %bb.0:
458; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
459; X86-NEXT:    notl %eax
460; X86-NEXT:    leal (%eax,%eax,2), %eax
461; X86-NEXT:    shll $2, %eax
462; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
463; X86-NEXT:    retl
464;
465; X64-LIN-LABEL: test12:
466; X64-LIN:       # %bb.0:
467; X64-LIN-NEXT:    notl %edx
468; X64-LIN-NEXT:    movslq %edx, %rax
469; X64-LIN-NEXT:    shlq $4, %rax
470; X64-LIN-NEXT:    addq %rdi, %rax
471; X64-LIN-NEXT:    retq
472;
473; X64-WIN-LABEL: test12:
474; X64-WIN:       # %bb.0:
475; X64-WIN-NEXT:    notl %r8d
476; X64-WIN-NEXT:    movslq %r8d, %rax
477; X64-WIN-NEXT:    shlq $4, %rax
478; X64-WIN-NEXT:    addq %rcx, %rax
479; X64-WIN-NEXT:    retq
480  %neg = shl i64 %intval, 32
481  %sext = xor i64 %neg, -4294967296
482  %idx.ext = ashr exact i64 %sext, 32
483  %add.ptr = getelementptr inbounds %struct.ref_s, ptr %op, i64 %idx.ext
484  ret ptr %add.ptr
485}
486
487define i32 @PR39657(ptr %p, i64 %x) {
488; X86-LABEL: PR39657:
489; X86:       # %bb.0:
490; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
491; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
492; X86-NEXT:    notl %ecx
493; X86-NEXT:    movl (%eax,%ecx,4), %eax
494; X86-NEXT:    retl
495;
496; X64-LIN-LABEL: PR39657:
497; X64-LIN:       # %bb.0:
498; X64-LIN-NEXT:    notq %rsi
499; X64-LIN-NEXT:    movl (%rdi,%rsi,4), %eax
500; X64-LIN-NEXT:    retq
501;
502; X64-WIN-LABEL: PR39657:
503; X64-WIN:       # %bb.0:
504; X64-WIN-NEXT:    notq %rdx
505; X64-WIN-NEXT:    movl (%rcx,%rdx,4), %eax
506; X64-WIN-NEXT:    retq
507  %sh = shl i64 %x, 2
508  %mul = xor i64 %sh, -4
509  %add.ptr = getelementptr inbounds i8, ptr %p, i64 %mul
510  %load = load i32, ptr %add.ptr, align 4
511  ret i32 %load
512}
513
514define i32 @add_of_not(i32 %x, i32 %y) {
515; X86-LABEL: add_of_not:
516; X86:       # %bb.0:
517; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
518; X86-NEXT:    notl %eax
519; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
520; X86-NEXT:    retl
521;
522; X64-LIN-LABEL: add_of_not:
523; X64-LIN:       # %bb.0:
524; X64-LIN-NEXT:    # kill: def $esi killed $esi def $rsi
525; X64-LIN-NEXT:    # kill: def $edi killed $edi def $rdi
526; X64-LIN-NEXT:    notl %esi
527; X64-LIN-NEXT:    leal (%rsi,%rdi), %eax
528; X64-LIN-NEXT:    retq
529;
530; X64-WIN-LABEL: add_of_not:
531; X64-WIN:       # %bb.0:
532; X64-WIN-NEXT:    # kill: def $edx killed $edx def $rdx
533; X64-WIN-NEXT:    # kill: def $ecx killed $ecx def $rcx
534; X64-WIN-NEXT:    notl %edx
535; X64-WIN-NEXT:    leal (%rdx,%rcx), %eax
536; X64-WIN-NEXT:    retq
537  %t0 = sub i32 %x, %y
538  %r = add i32 %t0, -1
539  ret i32 %r
540}
541
542define i32 @add_of_not_decrement(i32 %x, i32 %y) {
543; X86-LABEL: add_of_not_decrement:
544; X86:       # %bb.0:
545; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
546; X86-NEXT:    notl %eax
547; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
548; X86-NEXT:    retl
549;
550; X64-LIN-LABEL: add_of_not_decrement:
551; X64-LIN:       # %bb.0:
552; X64-LIN-NEXT:    # kill: def $esi killed $esi def $rsi
553; X64-LIN-NEXT:    # kill: def $edi killed $edi def $rdi
554; X64-LIN-NEXT:    notl %esi
555; X64-LIN-NEXT:    leal (%rsi,%rdi), %eax
556; X64-LIN-NEXT:    retq
557;
558; X64-WIN-LABEL: add_of_not_decrement:
559; X64-WIN:       # %bb.0:
560; X64-WIN-NEXT:    # kill: def $edx killed $edx def $rdx
561; X64-WIN-NEXT:    # kill: def $ecx killed $ecx def $rcx
562; X64-WIN-NEXT:    notl %edx
563; X64-WIN-NEXT:    leal (%rdx,%rcx), %eax
564; X64-WIN-NEXT:    retq
565  %t0 = sub i32 %x, %y
566  %r = sub i32 %t0, 1
567  ret i32 %r
568}
569
570define <4 x i32> @vec_add_of_not(<4 x i32> %x, <4 x i32> %y) {
571; X86-LABEL: vec_add_of_not:
572; X86:       # %bb.0:
573; X86-NEXT:    pcmpeqd %xmm2, %xmm2
574; X86-NEXT:    pxor %xmm1, %xmm2
575; X86-NEXT:    paddd %xmm2, %xmm0
576; X86-NEXT:    retl
577;
578; X64-LIN-LABEL: vec_add_of_not:
579; X64-LIN:       # %bb.0:
580; X64-LIN-NEXT:    pcmpeqd %xmm2, %xmm2
581; X64-LIN-NEXT:    pxor %xmm1, %xmm2
582; X64-LIN-NEXT:    paddd %xmm2, %xmm0
583; X64-LIN-NEXT:    retq
584;
585; X64-WIN-LABEL: vec_add_of_not:
586; X64-WIN:       # %bb.0:
587; X64-WIN-NEXT:    pcmpeqd %xmm0, %xmm0
588; X64-WIN-NEXT:    pxor (%rdx), %xmm0
589; X64-WIN-NEXT:    paddd (%rcx), %xmm0
590; X64-WIN-NEXT:    retq
591  %t0 = sub <4 x i32> %x, %y
592  %r = add <4 x i32> %t0, <i32 -1, i32 -1, i32 -1, i32 -1>
593  ret <4 x i32> %r
594}
595
596define <4 x i32> @vec_add_of_not_decrement(<4 x i32> %x, <4 x i32> %y) {
597; X86-LABEL: vec_add_of_not_decrement:
598; X86:       # %bb.0:
599; X86-NEXT:    pcmpeqd %xmm2, %xmm2
600; X86-NEXT:    pxor %xmm1, %xmm2
601; X86-NEXT:    paddd %xmm2, %xmm0
602; X86-NEXT:    retl
603;
604; X64-LIN-LABEL: vec_add_of_not_decrement:
605; X64-LIN:       # %bb.0:
606; X64-LIN-NEXT:    pcmpeqd %xmm2, %xmm2
607; X64-LIN-NEXT:    pxor %xmm1, %xmm2
608; X64-LIN-NEXT:    paddd %xmm2, %xmm0
609; X64-LIN-NEXT:    retq
610;
611; X64-WIN-LABEL: vec_add_of_not_decrement:
612; X64-WIN:       # %bb.0:
613; X64-WIN-NEXT:    pcmpeqd %xmm0, %xmm0
614; X64-WIN-NEXT:    pxor (%rdx), %xmm0
615; X64-WIN-NEXT:    paddd (%rcx), %xmm0
616; X64-WIN-NEXT:    retq
617  %t0 = sub <4 x i32> %x, %y
618  %r = sub <4 x i32> %t0, <i32 1, i32 1, i32 1, i32 1>
619  ret <4 x i32> %r
620}
621
622define <4 x i32> @vec_add_of_not_with_undef(<4 x i32> %x, <4 x i32> %y) {
623; X86-LABEL: vec_add_of_not_with_undef:
624; X86:       # %bb.0:
625; X86-NEXT:    psubd %xmm1, %xmm0
626; X86-NEXT:    pcmpeqd %xmm1, %xmm1
627; X86-NEXT:    paddd %xmm1, %xmm0
628; X86-NEXT:    retl
629;
630; X64-LIN-LABEL: vec_add_of_not_with_undef:
631; X64-LIN:       # %bb.0:
632; X64-LIN-NEXT:    psubd %xmm1, %xmm0
633; X64-LIN-NEXT:    pcmpeqd %xmm1, %xmm1
634; X64-LIN-NEXT:    paddd %xmm1, %xmm0
635; X64-LIN-NEXT:    retq
636;
637; X64-WIN-LABEL: vec_add_of_not_with_undef:
638; X64-WIN:       # %bb.0:
639; X64-WIN-NEXT:    movdqa (%rcx), %xmm1
640; X64-WIN-NEXT:    psubd (%rdx), %xmm1
641; X64-WIN-NEXT:    pcmpeqd %xmm0, %xmm0
642; X64-WIN-NEXT:    paddd %xmm1, %xmm0
643; X64-WIN-NEXT:    retq
644  %t0 = sub <4 x i32> %x, %y
645  %r = add <4 x i32> %t0, <i32 -1, i32 undef, i32 -1, i32 -1>
646  ret <4 x i32> %r
647}
648
649define <4 x i32> @vec_add_of_not_with_undef_decrement(<4 x i32> %x, <4 x i32> %y) {
650; X86-LABEL: vec_add_of_not_with_undef_decrement:
651; X86:       # %bb.0:
652; X86-NEXT:    psubd %xmm1, %xmm0
653; X86-NEXT:    pcmpeqd %xmm1, %xmm1
654; X86-NEXT:    psubd %xmm1, %xmm0
655; X86-NEXT:    retl
656;
657; X64-LIN-LABEL: vec_add_of_not_with_undef_decrement:
658; X64-LIN:       # %bb.0:
659; X64-LIN-NEXT:    psubd %xmm1, %xmm0
660; X64-LIN-NEXT:    pcmpeqd %xmm1, %xmm1
661; X64-LIN-NEXT:    psubd %xmm1, %xmm0
662; X64-LIN-NEXT:    retq
663;
664; X64-WIN-LABEL: vec_add_of_not_with_undef_decrement:
665; X64-WIN:       # %bb.0:
666; X64-WIN-NEXT:    movdqa (%rcx), %xmm0
667; X64-WIN-NEXT:    psubd (%rdx), %xmm0
668; X64-WIN-NEXT:    pcmpeqd %xmm1, %xmm1
669; X64-WIN-NEXT:    psubd %xmm1, %xmm0
670; X64-WIN-NEXT:    retq
671  %t0 = sub <4 x i32> %x, %y
672  %r = add <4 x i32> %t0, <i32 1, i32 undef, i32 1, i32 1>
673  ret <4 x i32> %r
674}
675