xref: /llvm-project/llvm/test/CodeGen/X86/umin.ll (revision 7b3bbd83c0c24087072ec5b22a76799ab31f87d5) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s --check-prefixes=X64,SSE
3; RUN: llc < %s -mtriple=x86_64-linux -mattr=avx | FileCheck %s --check-prefixes=X64,AVX,AVX1
4; RUN: llc < %s -mtriple=x86_64-linux -mattr=avx2 | FileCheck %s --check-prefixes=X64,AVX,AVX2
5; RUN: llc < %s -mtriple=i686 -mattr=cmov | FileCheck %s --check-prefix=X86
6
7declare i8 @llvm.umin.i8(i8, i8)
8declare i16 @llvm.umin.i16(i16, i16)
9declare i24 @llvm.umin.i24(i24, i24)
10declare i32 @llvm.umin.i32(i32, i32)
11declare i64 @llvm.umin.i64(i64, i64)
12declare i128 @llvm.umin.i128(i128, i128)
13
14declare <1 x i32> @llvm.umin.v1i32(<1 x i32>, <1 x i32>)
15declare <2 x i32> @llvm.umin.v2i32(<2 x i32>, <2 x i32>)
16declare <3 x i32> @llvm.umin.v3i32(<3 x i32>, <3 x i32>)
17declare <4 x i32> @llvm.umin.v4i32(<4 x i32>, <4 x i32>)
18declare <8 x i32> @llvm.umin.v8i32(<8 x i32>, <8 x i32>)
19
20declare <8 x i16> @llvm.umin.v8i16(<8 x i16>, <8 x i16>)
21declare <16 x i8> @llvm.umin.v16i8(<16 x i8>, <16 x i8>)
22
23define i8 @test_i8(i8 %a, i8 %b) nounwind {
24; X64-LABEL: test_i8:
25; X64:       # %bb.0:
26; X64-NEXT:    movl %esi, %eax
27; X64-NEXT:    cmpb %al, %dil
28; X64-NEXT:    cmovbl %edi, %eax
29; X64-NEXT:    # kill: def $al killed $al killed $eax
30; X64-NEXT:    retq
31;
32; X86-LABEL: test_i8:
33; X86:       # %bb.0:
34; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
35; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
36; X86-NEXT:    cmpb %al, %cl
37; X86-NEXT:    cmovbl %ecx, %eax
38; X86-NEXT:    # kill: def $al killed $al killed $eax
39; X86-NEXT:    retl
40  %r = call i8 @llvm.umin.i8(i8 %a, i8 %b)
41  ret i8 %r
42}
43
44define i16 @test_i16(i16 %a, i16 %b) nounwind {
45; X64-LABEL: test_i16:
46; X64:       # %bb.0:
47; X64-NEXT:    movl %esi, %eax
48; X64-NEXT:    cmpw %ax, %di
49; X64-NEXT:    cmovbl %edi, %eax
50; X64-NEXT:    # kill: def $ax killed $ax killed $eax
51; X64-NEXT:    retq
52;
53; X86-LABEL: test_i16:
54; X86:       # %bb.0:
55; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
56; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
57; X86-NEXT:    cmpw %ax, %cx
58; X86-NEXT:    cmovbl %ecx, %eax
59; X86-NEXT:    # kill: def $ax killed $ax killed $eax
60; X86-NEXT:    retl
61  %r = call i16 @llvm.umin.i16(i16 %a, i16 %b)
62  ret i16 %r
63}
64
65define i24 @test_i24(i24 %a, i24 %b) nounwind {
66; X64-LABEL: test_i24:
67; X64:       # %bb.0:
68; X64-NEXT:    movl %edi, %eax
69; X64-NEXT:    andl $16777215, %esi # imm = 0xFFFFFF
70; X64-NEXT:    andl $16777215, %eax # imm = 0xFFFFFF
71; X64-NEXT:    cmpl %esi, %eax
72; X64-NEXT:    cmovael %esi, %eax
73; X64-NEXT:    retq
74;
75; X86-LABEL: test_i24:
76; X86:       # %bb.0:
77; X86-NEXT:    movl $16777215, %eax # imm = 0xFFFFFF
78; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
79; X86-NEXT:    andl %eax, %ecx
80; X86-NEXT:    andl {{[0-9]+}}(%esp), %eax
81; X86-NEXT:    cmpl %ecx, %eax
82; X86-NEXT:    cmovael %ecx, %eax
83; X86-NEXT:    retl
84  %r = call i24 @llvm.umin.i24(i24 %a, i24 %b)
85  ret i24 %r
86}
87
88define i32 @test_i32(i32 %a, i32 %b) nounwind {
89; X64-LABEL: test_i32:
90; X64:       # %bb.0:
91; X64-NEXT:    movl %esi, %eax
92; X64-NEXT:    cmpl %esi, %edi
93; X64-NEXT:    cmovbl %edi, %eax
94; X64-NEXT:    retq
95;
96; X86-LABEL: test_i32:
97; X86:       # %bb.0:
98; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
99; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
100; X86-NEXT:    cmpl %eax, %ecx
101; X86-NEXT:    cmovbl %ecx, %eax
102; X86-NEXT:    retl
103  %r = call i32 @llvm.umin.i32(i32 %a, i32 %b)
104  ret i32 %r
105}
106
107define i64 @test_i64(i64 %a, i64 %b) nounwind {
108; X64-LABEL: test_i64:
109; X64:       # %bb.0:
110; X64-NEXT:    movq %rsi, %rax
111; X64-NEXT:    cmpq %rsi, %rdi
112; X64-NEXT:    cmovbq %rdi, %rax
113; X64-NEXT:    retq
114;
115; X86-LABEL: test_i64:
116; X86:       # %bb.0:
117; X86-NEXT:    pushl %edi
118; X86-NEXT:    pushl %esi
119; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
120; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
121; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
122; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
123; X86-NEXT:    cmpl %eax, %ecx
124; X86-NEXT:    movl %esi, %edi
125; X86-NEXT:    sbbl %edx, %edi
126; X86-NEXT:    cmovbl %ecx, %eax
127; X86-NEXT:    cmovbl %esi, %edx
128; X86-NEXT:    popl %esi
129; X86-NEXT:    popl %edi
130; X86-NEXT:    retl
131  %r = call i64 @llvm.umin.i64(i64 %a, i64 %b)
132  ret i64 %r
133}
134
135define i128 @test_i128(i128 %a, i128 %b) nounwind {
136; X64-LABEL: test_i128:
137; X64:       # %bb.0:
138; X64-NEXT:    movq %rdx, %rax
139; X64-NEXT:    cmpq %rdx, %rdi
140; X64-NEXT:    movq %rsi, %rdx
141; X64-NEXT:    sbbq %rcx, %rdx
142; X64-NEXT:    cmovbq %rdi, %rax
143; X64-NEXT:    cmovbq %rsi, %rcx
144; X64-NEXT:    movq %rcx, %rdx
145; X64-NEXT:    retq
146;
147; X86-LABEL: test_i128:
148; X86:       # %bb.0:
149; X86-NEXT:    pushl %ebp
150; X86-NEXT:    pushl %ebx
151; X86-NEXT:    pushl %edi
152; X86-NEXT:    pushl %esi
153; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
154; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
155; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
156; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
157; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
158; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
159; X86-NEXT:    cmpl %edx, %ebx
160; X86-NEXT:    sbbl %esi, %ebp
161; X86-NEXT:    movl %eax, %ebp
162; X86-NEXT:    sbbl %ecx, %ebp
163; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
164; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
165; X86-NEXT:    movl %edi, %eax
166; X86-NEXT:    sbbl %ebp, %eax
167; X86-NEXT:    cmovbl %ebx, %edx
168; X86-NEXT:    cmovbl {{[0-9]+}}(%esp), %esi
169; X86-NEXT:    cmovbl {{[0-9]+}}(%esp), %ecx
170; X86-NEXT:    cmovbl %edi, %ebp
171; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
172; X86-NEXT:    movl %ebp, 12(%eax)
173; X86-NEXT:    movl %ecx, 8(%eax)
174; X86-NEXT:    movl %esi, 4(%eax)
175; X86-NEXT:    movl %edx, (%eax)
176; X86-NEXT:    popl %esi
177; X86-NEXT:    popl %edi
178; X86-NEXT:    popl %ebx
179; X86-NEXT:    popl %ebp
180; X86-NEXT:    retl $4
181  %r = call i128 @llvm.umin.i128(i128 %a, i128 %b)
182  ret i128 %r
183}
184
185define <1 x i32> @test_v1i32(<1 x i32> %a, <1 x i32> %b) nounwind {
186; X64-LABEL: test_v1i32:
187; X64:       # %bb.0:
188; X64-NEXT:    movl %esi, %eax
189; X64-NEXT:    cmpl %esi, %edi
190; X64-NEXT:    cmovbl %edi, %eax
191; X64-NEXT:    retq
192;
193; X86-LABEL: test_v1i32:
194; X86:       # %bb.0:
195; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
196; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
197; X86-NEXT:    cmpl %eax, %ecx
198; X86-NEXT:    cmovbl %ecx, %eax
199; X86-NEXT:    retl
200  %r = call <1 x i32> @llvm.umin.v1i32(<1 x i32> %a, <1 x i32> %b)
201  ret <1 x i32> %r
202}
203
204define <2 x i32> @test_v2i32(<2 x i32> %a, <2 x i32> %b) nounwind {
205; SSE-LABEL: test_v2i32:
206; SSE:       # %bb.0:
207; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
208; SSE-NEXT:    movdqa %xmm0, %xmm3
209; SSE-NEXT:    pxor %xmm2, %xmm3
210; SSE-NEXT:    pxor %xmm1, %xmm2
211; SSE-NEXT:    pcmpgtd %xmm3, %xmm2
212; SSE-NEXT:    pand %xmm2, %xmm0
213; SSE-NEXT:    pandn %xmm1, %xmm2
214; SSE-NEXT:    por %xmm2, %xmm0
215; SSE-NEXT:    retq
216;
217; AVX-LABEL: test_v2i32:
218; AVX:       # %bb.0:
219; AVX-NEXT:    vpminud %xmm1, %xmm0, %xmm0
220; AVX-NEXT:    retq
221;
222; X86-LABEL: test_v2i32:
223; X86:       # %bb.0:
224; X86-NEXT:    pushl %esi
225; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
226; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
227; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
228; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
229; X86-NEXT:    cmpl %eax, %esi
230; X86-NEXT:    cmovbl %esi, %eax
231; X86-NEXT:    cmpl %edx, %ecx
232; X86-NEXT:    cmovbl %ecx, %edx
233; X86-NEXT:    popl %esi
234; X86-NEXT:    retl
235  %r = call <2 x i32> @llvm.umin.v2i32(<2 x i32> %a, <2 x i32> %b)
236  ret <2 x i32> %r
237}
238
239define <3 x i32> @test_v3i32(<3 x i32> %a, <3 x i32> %b) nounwind {
240; SSE-LABEL: test_v3i32:
241; SSE:       # %bb.0:
242; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
243; SSE-NEXT:    movdqa %xmm0, %xmm3
244; SSE-NEXT:    pxor %xmm2, %xmm3
245; SSE-NEXT:    pxor %xmm1, %xmm2
246; SSE-NEXT:    pcmpgtd %xmm3, %xmm2
247; SSE-NEXT:    pand %xmm2, %xmm0
248; SSE-NEXT:    pandn %xmm1, %xmm2
249; SSE-NEXT:    por %xmm2, %xmm0
250; SSE-NEXT:    retq
251;
252; AVX-LABEL: test_v3i32:
253; AVX:       # %bb.0:
254; AVX-NEXT:    vpminud %xmm1, %xmm0, %xmm0
255; AVX-NEXT:    retq
256;
257; X86-LABEL: test_v3i32:
258; X86:       # %bb.0:
259; X86-NEXT:    pushl %ebx
260; X86-NEXT:    pushl %edi
261; X86-NEXT:    pushl %esi
262; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
263; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
264; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
265; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
266; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
267; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
268; X86-NEXT:    cmpl %eax, %ebx
269; X86-NEXT:    cmovbl %ebx, %eax
270; X86-NEXT:    cmpl %edx, %edi
271; X86-NEXT:    cmovbl %edi, %edx
272; X86-NEXT:    cmpl %ecx, %esi
273; X86-NEXT:    cmovbl %esi, %ecx
274; X86-NEXT:    popl %esi
275; X86-NEXT:    popl %edi
276; X86-NEXT:    popl %ebx
277; X86-NEXT:    retl
278  %r = call <3 x i32> @llvm.umin.v3i32(<3 x i32> %a, <3 x i32> %b)
279  ret <3 x i32> %r
280}
281
282define <4 x i32> @test_v4i32(<4 x i32> %a, <4 x i32> %b) nounwind {
283; SSE-LABEL: test_v4i32:
284; SSE:       # %bb.0:
285; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [2147483648,2147483648,2147483648,2147483648]
286; SSE-NEXT:    movdqa %xmm0, %xmm3
287; SSE-NEXT:    pxor %xmm2, %xmm3
288; SSE-NEXT:    pxor %xmm1, %xmm2
289; SSE-NEXT:    pcmpgtd %xmm3, %xmm2
290; SSE-NEXT:    pand %xmm2, %xmm0
291; SSE-NEXT:    pandn %xmm1, %xmm2
292; SSE-NEXT:    por %xmm2, %xmm0
293; SSE-NEXT:    retq
294;
295; AVX-LABEL: test_v4i32:
296; AVX:       # %bb.0:
297; AVX-NEXT:    vpminud %xmm1, %xmm0, %xmm0
298; AVX-NEXT:    retq
299;
300; X86-LABEL: test_v4i32:
301; X86:       # %bb.0:
302; X86-NEXT:    pushl %edi
303; X86-NEXT:    pushl %esi
304; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
305; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
306; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
307; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
308; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
309; X86-NEXT:    cmpl %edi, %eax
310; X86-NEXT:    cmovbl %eax, %edi
311; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
312; X86-NEXT:    cmpl %esi, %eax
313; X86-NEXT:    cmovbl %eax, %esi
314; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
315; X86-NEXT:    cmpl %edx, %eax
316; X86-NEXT:    cmovbl %eax, %edx
317; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
318; X86-NEXT:    cmpl %ecx, %eax
319; X86-NEXT:    cmovbl %eax, %ecx
320; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
321; X86-NEXT:    movl %ecx, 12(%eax)
322; X86-NEXT:    movl %edx, 8(%eax)
323; X86-NEXT:    movl %esi, 4(%eax)
324; X86-NEXT:    movl %edi, (%eax)
325; X86-NEXT:    popl %esi
326; X86-NEXT:    popl %edi
327; X86-NEXT:    retl $4
328  %r = call <4 x i32> @llvm.umin.v4i32(<4 x i32> %a, <4 x i32> %b)
329  ret <4 x i32> %r
330}
331
332define <8 x i32> @test_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind {
333; SSE-LABEL: test_v8i32:
334; SSE:       # %bb.0:
335; SSE-NEXT:    movdqa {{.*#+}} xmm4 = [2147483648,2147483648,2147483648,2147483648]
336; SSE-NEXT:    movdqa %xmm0, %xmm5
337; SSE-NEXT:    pxor %xmm4, %xmm5
338; SSE-NEXT:    movdqa %xmm2, %xmm6
339; SSE-NEXT:    pxor %xmm4, %xmm6
340; SSE-NEXT:    pcmpgtd %xmm5, %xmm6
341; SSE-NEXT:    pand %xmm6, %xmm0
342; SSE-NEXT:    pandn %xmm2, %xmm6
343; SSE-NEXT:    por %xmm6, %xmm0
344; SSE-NEXT:    movdqa %xmm1, %xmm2
345; SSE-NEXT:    pxor %xmm4, %xmm2
346; SSE-NEXT:    pxor %xmm3, %xmm4
347; SSE-NEXT:    pcmpgtd %xmm2, %xmm4
348; SSE-NEXT:    pand %xmm4, %xmm1
349; SSE-NEXT:    pandn %xmm3, %xmm4
350; SSE-NEXT:    por %xmm4, %xmm1
351; SSE-NEXT:    retq
352;
353; AVX1-LABEL: test_v8i32:
354; AVX1:       # %bb.0:
355; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
356; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm3
357; AVX1-NEXT:    vpminud %xmm2, %xmm3, %xmm2
358; AVX1-NEXT:    vpminud %xmm1, %xmm0, %xmm0
359; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
360; AVX1-NEXT:    retq
361;
362; AVX2-LABEL: test_v8i32:
363; AVX2:       # %bb.0:
364; AVX2-NEXT:    vpminud %ymm1, %ymm0, %ymm0
365; AVX2-NEXT:    retq
366;
367; X86-LABEL: test_v8i32:
368; X86:       # %bb.0:
369; X86-NEXT:    pushl %ebp
370; X86-NEXT:    pushl %ebx
371; X86-NEXT:    pushl %edi
372; X86-NEXT:    pushl %esi
373; X86-NEXT:    subl $8, %esp
374; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
375; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
376; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
377; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
378; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
379; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
380; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
381; X86-NEXT:    cmpl %ebp, %eax
382; X86-NEXT:    cmovbl %eax, %ebp
383; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
384; X86-NEXT:    cmpl %ebx, %eax
385; X86-NEXT:    cmovbl %eax, %ebx
386; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
387; X86-NEXT:    cmpl %edi, %eax
388; X86-NEXT:    cmovbl %eax, %edi
389; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
390; X86-NEXT:    cmpl %esi, %eax
391; X86-NEXT:    cmovbl %eax, %esi
392; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
393; X86-NEXT:    cmpl %edx, %eax
394; X86-NEXT:    cmovbl %eax, %edx
395; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
396; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
397; X86-NEXT:    cmpl %ecx, %eax
398; X86-NEXT:    cmovbl %eax, %ecx
399; X86-NEXT:    movl %ecx, (%esp) # 4-byte Spill
400; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
401; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
402; X86-NEXT:    cmpl %edx, %eax
403; X86-NEXT:    cmovbl %eax, %edx
404; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
405; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
406; X86-NEXT:    cmpl %eax, %ecx
407; X86-NEXT:    cmovbl %ecx, %eax
408; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
409; X86-NEXT:    movl %eax, 28(%ecx)
410; X86-NEXT:    movl %edx, 24(%ecx)
411; X86-NEXT:    movl (%esp), %eax # 4-byte Reload
412; X86-NEXT:    movl %eax, 20(%ecx)
413; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
414; X86-NEXT:    movl %eax, 16(%ecx)
415; X86-NEXT:    movl %esi, 12(%ecx)
416; X86-NEXT:    movl %edi, 8(%ecx)
417; X86-NEXT:    movl %ebx, 4(%ecx)
418; X86-NEXT:    movl %ebp, (%ecx)
419; X86-NEXT:    movl %ecx, %eax
420; X86-NEXT:    addl $8, %esp
421; X86-NEXT:    popl %esi
422; X86-NEXT:    popl %edi
423; X86-NEXT:    popl %ebx
424; X86-NEXT:    popl %ebp
425; X86-NEXT:    retl $4
426  %r = call <8 x i32> @llvm.umin.v8i32(<8 x i32> %a, <8 x i32> %b)
427  ret <8 x i32> %r
428}
429
430define <8 x i16> @test_v8i16(<8 x i16> %a, <8 x i16> %b) nounwind {
431; SSE-LABEL: test_v8i16:
432; SSE:       # %bb.0:
433; SSE-NEXT:    movdqa %xmm0, %xmm2
434; SSE-NEXT:    psubusw %xmm1, %xmm2
435; SSE-NEXT:    psubw %xmm2, %xmm0
436; SSE-NEXT:    retq
437;
438; AVX-LABEL: test_v8i16:
439; AVX:       # %bb.0:
440; AVX-NEXT:    vpminuw %xmm1, %xmm0, %xmm0
441; AVX-NEXT:    retq
442;
443; X86-LABEL: test_v8i16:
444; X86:       # %bb.0:
445; X86-NEXT:    pushl %ebp
446; X86-NEXT:    pushl %ebx
447; X86-NEXT:    pushl %edi
448; X86-NEXT:    pushl %esi
449; X86-NEXT:    subl $8, %esp
450; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
451; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
452; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
453; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
454; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
455; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
456; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
457; X86-NEXT:    cmpw %bp, %ax
458; X86-NEXT:    cmovbl %eax, %ebp
459; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
460; X86-NEXT:    cmpw %bx, %ax
461; X86-NEXT:    cmovbl %eax, %ebx
462; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
463; X86-NEXT:    cmpw %di, %ax
464; X86-NEXT:    cmovbl %eax, %edi
465; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
466; X86-NEXT:    cmpw %si, %ax
467; X86-NEXT:    cmovbl %eax, %esi
468; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
469; X86-NEXT:    cmpw %dx, %ax
470; X86-NEXT:    cmovbl %eax, %edx
471; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
472; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
473; X86-NEXT:    cmpw %cx, %ax
474; X86-NEXT:    cmovbl %eax, %ecx
475; X86-NEXT:    movl %ecx, (%esp) # 4-byte Spill
476; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
477; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
478; X86-NEXT:    cmpw %dx, %ax
479; X86-NEXT:    cmovbl %eax, %edx
480; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
481; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
482; X86-NEXT:    cmpw %ax, %cx
483; X86-NEXT:    cmovbl %ecx, %eax
484; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
485; X86-NEXT:    movw %ax, 14(%ecx)
486; X86-NEXT:    movw %dx, 12(%ecx)
487; X86-NEXT:    movl (%esp), %eax # 4-byte Reload
488; X86-NEXT:    movw %ax, 10(%ecx)
489; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
490; X86-NEXT:    movw %ax, 8(%ecx)
491; X86-NEXT:    movw %si, 6(%ecx)
492; X86-NEXT:    movw %di, 4(%ecx)
493; X86-NEXT:    movw %bx, 2(%ecx)
494; X86-NEXT:    movw %bp, (%ecx)
495; X86-NEXT:    movl %ecx, %eax
496; X86-NEXT:    addl $8, %esp
497; X86-NEXT:    popl %esi
498; X86-NEXT:    popl %edi
499; X86-NEXT:    popl %ebx
500; X86-NEXT:    popl %ebp
501; X86-NEXT:    retl $4
502  %r = call <8 x i16> @llvm.umin.v8i16(<8 x i16> %a, <8 x i16> %b)
503  ret <8 x i16> %r
504}
505
506define <16 x i8> @test_v16i8(<16 x i8> %a, <16 x i8> %b) nounwind {
507; SSE-LABEL: test_v16i8:
508; SSE:       # %bb.0:
509; SSE-NEXT:    pminub %xmm1, %xmm0
510; SSE-NEXT:    retq
511;
512; AVX-LABEL: test_v16i8:
513; AVX:       # %bb.0:
514; AVX-NEXT:    vpminub %xmm1, %xmm0, %xmm0
515; AVX-NEXT:    retq
516;
517; X86-LABEL: test_v16i8:
518; X86:       # %bb.0:
519; X86-NEXT:    pushl %ebp
520; X86-NEXT:    pushl %ebx
521; X86-NEXT:    pushl %edi
522; X86-NEXT:    pushl %esi
523; X86-NEXT:    subl $40, %esp
524; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
525; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
526; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
527; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
528; X86-NEXT:    cmpb %bl, %al
529; X86-NEXT:    cmovbl %eax, %ebx
530; X86-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
531; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
532; X86-NEXT:    cmpb %dl, %al
533; X86-NEXT:    cmovbl %eax, %edx
534; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
535; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
536; X86-NEXT:    cmpb %cl, %al
537; X86-NEXT:    cmovbl %eax, %ecx
538; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
539; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
540; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
541; X86-NEXT:    cmpb %cl, %al
542; X86-NEXT:    cmovbl %eax, %ecx
543; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
544; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
545; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
546; X86-NEXT:    cmpb %cl, %al
547; X86-NEXT:    cmovbl %eax, %ecx
548; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
549; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
550; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
551; X86-NEXT:    cmpb %cl, %al
552; X86-NEXT:    cmovbl %eax, %ecx
553; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
554; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
555; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
556; X86-NEXT:    cmpb %cl, %al
557; X86-NEXT:    cmovbl %eax, %ecx
558; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
559; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
560; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
561; X86-NEXT:    cmpb %cl, %al
562; X86-NEXT:    cmovbl %eax, %ecx
563; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
564; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
565; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
566; X86-NEXT:    cmpb %cl, %al
567; X86-NEXT:    cmovbl %eax, %ecx
568; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
569; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
570; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
571; X86-NEXT:    cmpb %cl, %al
572; X86-NEXT:    cmovbl %eax, %ecx
573; X86-NEXT:    movl %ecx, (%esp) # 4-byte Spill
574; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
575; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
576; X86-NEXT:    cmpb %cl, %al
577; X86-NEXT:    cmovbl %eax, %ecx
578; X86-NEXT:    movl %ecx, %ebp
579; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
580; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
581; X86-NEXT:    cmpb %cl, %al
582; X86-NEXT:    cmovbl %eax, %ecx
583; X86-NEXT:    movl %ecx, %edi
584; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
585; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
586; X86-NEXT:    cmpb %cl, %al
587; X86-NEXT:    cmovbl %eax, %ecx
588; X86-NEXT:    movl %ecx, %esi
589; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
590; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
591; X86-NEXT:    cmpb %bl, %al
592; X86-NEXT:    cmovbl %eax, %ebx
593; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
594; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
595; X86-NEXT:    cmpb %dl, %al
596; X86-NEXT:    cmovbl %eax, %edx
597; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
598; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
599; X86-NEXT:    cmpb %cl, %al
600; X86-NEXT:    cmovbl %eax, %ecx
601; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
602; X86-NEXT:    movb %cl, 15(%eax)
603; X86-NEXT:    movb %dl, 14(%eax)
604; X86-NEXT:    movb %bl, 13(%eax)
605; X86-NEXT:    movl %esi, %ecx
606; X86-NEXT:    movb %cl, 12(%eax)
607; X86-NEXT:    movl %edi, %ecx
608; X86-NEXT:    movb %cl, 11(%eax)
609; X86-NEXT:    movl %ebp, %ecx
610; X86-NEXT:    movb %cl, 10(%eax)
611; X86-NEXT:    movl (%esp), %ecx # 4-byte Reload
612; X86-NEXT:    movb %cl, 9(%eax)
613; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
614; X86-NEXT:    movb %cl, 8(%eax)
615; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
616; X86-NEXT:    movb %cl, 7(%eax)
617; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
618; X86-NEXT:    movb %cl, 6(%eax)
619; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
620; X86-NEXT:    movb %cl, 5(%eax)
621; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
622; X86-NEXT:    movb %cl, 4(%eax)
623; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
624; X86-NEXT:    movb %cl, 3(%eax)
625; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
626; X86-NEXT:    movb %cl, 2(%eax)
627; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
628; X86-NEXT:    movb %cl, 1(%eax)
629; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
630; X86-NEXT:    movb %cl, (%eax)
631; X86-NEXT:    addl $40, %esp
632; X86-NEXT:    popl %esi
633; X86-NEXT:    popl %edi
634; X86-NEXT:    popl %ebx
635; X86-NEXT:    popl %ebp
636; X86-NEXT:    retl $4
637  %r = call <16 x i8> @llvm.umin.v16i8(<16 x i8> %a, <16 x i8> %b)
638  ret <16 x i8> %r
639}
640
641define i16 @test_signbits_i16(i16 %a, i16 %b) nounwind {
642; X64-LABEL: test_signbits_i16:
643; X64:       # %bb.0:
644; X64-NEXT:    movswl %si, %eax
645; X64-NEXT:    movswl %di, %ecx
646; X64-NEXT:    shrl $15, %ecx
647; X64-NEXT:    shrl $8, %eax
648; X64-NEXT:    cmpw %ax, %cx
649; X64-NEXT:    cmovbl %ecx, %eax
650; X64-NEXT:    # kill: def $ax killed $ax killed $eax
651; X64-NEXT:    retq
652;
653; X86-LABEL: test_signbits_i16:
654; X86:       # %bb.0:
655; X86-NEXT:    movsbl {{[0-9]+}}(%esp), %ecx
656; X86-NEXT:    movswl {{[0-9]+}}(%esp), %eax
657; X86-NEXT:    shrl $15, %eax
658; X86-NEXT:    cmpw %cx, %ax
659; X86-NEXT:    cmovael %ecx, %eax
660; X86-NEXT:    # kill: def $ax killed $ax killed $eax
661; X86-NEXT:    retl
662  %ax = ashr i16 %a, 15
663  %bx = ashr i16 %b, 8
664  %r = call i16 @llvm.umin.i16(i16 %ax, i16 %bx)
665  ret i16 %r
666}
667
668define i32 @test_signbits_i32(i32 %a, i32 %b) nounwind {
669; X64-LABEL: test_signbits_i32:
670; X64:       # %bb.0:
671; X64-NEXT:    movl %esi, %eax
672; X64-NEXT:    sarl $16, %edi
673; X64-NEXT:    sarl $17, %eax
674; X64-NEXT:    cmpl %eax, %edi
675; X64-NEXT:    cmovbl %edi, %eax
676; X64-NEXT:    retq
677;
678; X86-LABEL: test_signbits_i32:
679; X86:       # %bb.0:
680; X86-NEXT:    movswl {{[0-9]+}}(%esp), %ecx
681; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
682; X86-NEXT:    sarl $17, %eax
683; X86-NEXT:    cmpl %eax, %ecx
684; X86-NEXT:    cmovbl %ecx, %eax
685; X86-NEXT:    retl
686  %ax = ashr i32 %a, 16
687  %bx = ashr i32 %b, 17
688  %r = call i32 @llvm.umin.i32(i32 %ax, i32 %bx)
689  ret i32 %r
690}
691
692define i64 @test_signbits_i64(i64 %a, i64 %b) nounwind {
693; X64-LABEL: test_signbits_i64:
694; X64:       # %bb.0:
695; X64-NEXT:    movq %rsi, %rax
696; X64-NEXT:    sarq $32, %rdi
697; X64-NEXT:    sarq $32, %rax
698; X64-NEXT:    cmpq %rax, %rdi
699; X64-NEXT:    cmovbq %rdi, %rax
700; X64-NEXT:    retq
701;
702; X86-LABEL: test_signbits_i64:
703; X86:       # %bb.0:
704; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
705; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
706; X86-NEXT:    cmpl %eax, %ecx
707; X86-NEXT:    cmovbl %ecx, %eax
708; X86-NEXT:    movl %eax, %edx
709; X86-NEXT:    sarl $31, %edx
710; X86-NEXT:    retl
711  %ax = ashr i64 %a, 32
712  %bx = ashr i64 %b, 32
713  %r = call i64 @llvm.umin.i64(i64 %ax, i64 %bx)
714  ret i64 %r
715}
716
717define i128 @test_signbits_i128(i128 %a, i128 %b) nounwind {
718; X64-LABEL: test_signbits_i128:
719; X64:       # %bb.0:
720; X64-NEXT:    movq %rcx, %rax
721; X64-NEXT:    sarq $28, %rax
722; X64-NEXT:    cmpq %rax, %rsi
723; X64-NEXT:    cmovbq %rsi, %rax
724; X64-NEXT:    movq %rax, %rdx
725; X64-NEXT:    sarq $63, %rdx
726; X64-NEXT:    retq
727;
728; X86-LABEL: test_signbits_i128:
729; X86:       # %bb.0:
730; X86-NEXT:    pushl %ebx
731; X86-NEXT:    pushl %edi
732; X86-NEXT:    pushl %esi
733; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
734; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
735; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
736; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
737; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
738; X86-NEXT:    shrdl $28, %edi, %ecx
739; X86-NEXT:    sarl $28, %edi
740; X86-NEXT:    cmpl %ecx, %esi
741; X86-NEXT:    movl %edx, %ebx
742; X86-NEXT:    sbbl %edi, %ebx
743; X86-NEXT:    cmovbl %esi, %ecx
744; X86-NEXT:    cmovbl %edx, %edi
745; X86-NEXT:    movl %edi, 4(%eax)
746; X86-NEXT:    sarl $31, %edi
747; X86-NEXT:    movl %ecx, (%eax)
748; X86-NEXT:    movl %edi, 12(%eax)
749; X86-NEXT:    movl %edi, 8(%eax)
750; X86-NEXT:    popl %esi
751; X86-NEXT:    popl %edi
752; X86-NEXT:    popl %ebx
753; X86-NEXT:    retl $4
754  %ax = ashr i128 %a, 64
755  %bx = ashr i128 %b, 92
756  %r = call i128 @llvm.umin.i128(i128 %ax, i128 %bx)
757  ret i128 %r
758}
759