1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s
4 define <8 x i32> @cmp00(<8 x float> %a, <8 x float> %b) nounwind {
7 ; CHECK-NEXT: vcmpltps %ymm1, %ymm0, %ymm0
9 %bincmp = fcmp olt <8 x float> %a, %b
10 %s = sext <8 x i1> %bincmp to <8 x i32>
14 define <4 x i64> @cmp01(<4 x double> %a, <4 x double> %b) nounwind {
17 ; CHECK-NEXT: vcmpltpd %ymm1, %ymm0, %ymm0
19 %bincmp = fcmp olt <4 x double> %a, %b
20 %s = sext <4 x i1> %bincmp to <4 x i64>
24 declare void @scale() nounwind
26 define void @render() nounwind {
27 ; CHECK-LABEL: render:
28 ; CHECK: # %bb.0: # %entry
29 ; CHECK-NEXT: pushq %rbp
30 ; CHECK-NEXT: pushq %rbx
31 ; CHECK-NEXT: pushq %rax
32 ; CHECK-NEXT: xorl %eax, %eax
33 ; CHECK-NEXT: testb %al, %al
34 ; CHECK-NEXT: jne .LBB2_6
35 ; CHECK-NEXT: # %bb.1: # %for.cond5.preheader
36 ; CHECK-NEXT: xorl %ebx, %ebx
37 ; CHECK-NEXT: movb $1, %bpl
38 ; CHECK-NEXT: .p2align 4, 0x90
39 ; CHECK-NEXT: .LBB2_2: # %for.cond5
40 ; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
41 ; CHECK-NEXT: testb %bl, %bl
42 ; CHECK-NEXT: jne .LBB2_2
43 ; CHECK-NEXT: # %bb.3: # %for.cond5
44 ; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
45 ; CHECK-NEXT: testb %bpl, %bpl
46 ; CHECK-NEXT: jne .LBB2_2
47 ; CHECK-NEXT: # %bb.4: # %for.body33
48 ; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
49 ; CHECK-NEXT: vucomisd {{\.LCPI.*}}, %xmm0
50 ; CHECK-NEXT: jne .LBB2_5
51 ; CHECK-NEXT: jnp .LBB2_2
52 ; CHECK-NEXT: .LBB2_5: # %if.then
53 ; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
54 ; CHECK-NEXT: callq scale
55 ; CHECK-NEXT: jmp .LBB2_2
56 ; CHECK-NEXT: .LBB2_6: # %for.end52
57 ; CHECK-NEXT: addq $8, %rsp
58 ; CHECK-NEXT: popq %rbx
59 ; CHECK-NEXT: popq %rbp
62 br i1 undef, label %for.cond5, label %for.end52
65 %or.cond = and i1 undef, false
66 br i1 %or.cond, label %for.body33, label %for.cond5
69 br i1 false, label %for.body33, label %for.cond5
72 %tobool = fcmp une double undef, 0.000000e+00
73 br i1 %tobool, label %if.then, label %for.cond30
83 define <8 x i32> @int256_cmp(<8 x i32> %i, <8 x i32> %j) nounwind {
84 ; CHECK-LABEL: int256_cmp:
86 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2
87 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3
88 ; CHECK-NEXT: vpcmpgtd %xmm2, %xmm3, %xmm2
89 ; CHECK-NEXT: vpcmpgtd %xmm0, %xmm1, %xmm0
90 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
92 %bincmp = icmp slt <8 x i32> %i, %j
93 %x = sext <8 x i1> %bincmp to <8 x i32>
97 define <4 x i64> @v4i64_cmp(<4 x i64> %i, <4 x i64> %j) nounwind {
98 ; CHECK-LABEL: v4i64_cmp:
100 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2
101 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3
102 ; CHECK-NEXT: vpcmpgtq %xmm2, %xmm3, %xmm2
103 ; CHECK-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm0
104 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
106 %bincmp = icmp slt <4 x i64> %i, %j
107 %x = sext <4 x i1> %bincmp to <4 x i64>
111 define <16 x i16> @v16i16_cmp(<16 x i16> %i, <16 x i16> %j) nounwind {
112 ; CHECK-LABEL: v16i16_cmp:
114 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2
115 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3
116 ; CHECK-NEXT: vpcmpgtw %xmm2, %xmm3, %xmm2
117 ; CHECK-NEXT: vpcmpgtw %xmm0, %xmm1, %xmm0
118 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
120 %bincmp = icmp slt <16 x i16> %i, %j
121 %x = sext <16 x i1> %bincmp to <16 x i16>
125 define <32 x i8> @v32i8_cmp(<32 x i8> %i, <32 x i8> %j) nounwind {
126 ; CHECK-LABEL: v32i8_cmp:
128 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm2
129 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm3
130 ; CHECK-NEXT: vpcmpgtb %xmm2, %xmm3, %xmm2
131 ; CHECK-NEXT: vpcmpgtb %xmm0, %xmm1, %xmm0
132 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
134 %bincmp = icmp slt <32 x i8> %i, %j
135 %x = sext <32 x i1> %bincmp to <32 x i8>
139 define <8 x i32> @int256_cmpeq(<8 x i32> %i, <8 x i32> %j) nounwind {
140 ; CHECK-LABEL: int256_cmpeq:
142 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
143 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
144 ; CHECK-NEXT: vpcmpeqd %xmm2, %xmm3, %xmm2
145 ; CHECK-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
146 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
148 %bincmp = icmp eq <8 x i32> %i, %j
149 %x = sext <8 x i1> %bincmp to <8 x i32>
153 define <4 x i64> @v4i64_cmpeq(<4 x i64> %i, <4 x i64> %j) nounwind {
154 ; CHECK-LABEL: v4i64_cmpeq:
156 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
157 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
158 ; CHECK-NEXT: vpcmpeqq %xmm2, %xmm3, %xmm2
159 ; CHECK-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
160 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
162 %bincmp = icmp eq <4 x i64> %i, %j
163 %x = sext <4 x i1> %bincmp to <4 x i64>
167 define <16 x i16> @v16i16_cmpeq(<16 x i16> %i, <16 x i16> %j) nounwind {
168 ; CHECK-LABEL: v16i16_cmpeq:
170 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
171 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
172 ; CHECK-NEXT: vpcmpeqw %xmm2, %xmm3, %xmm2
173 ; CHECK-NEXT: vpcmpeqw %xmm1, %xmm0, %xmm0
174 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
176 %bincmp = icmp eq <16 x i16> %i, %j
177 %x = sext <16 x i1> %bincmp to <16 x i16>
181 define <32 x i8> @v32i8_cmpeq(<32 x i8> %i, <32 x i8> %j) nounwind {
182 ; CHECK-LABEL: v32i8_cmpeq:
184 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
185 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
186 ; CHECK-NEXT: vpcmpeqb %xmm2, %xmm3, %xmm2
187 ; CHECK-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
188 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
190 %bincmp = icmp eq <32 x i8> %i, %j
191 %x = sext <32 x i1> %bincmp to <32 x i8>
197 define i32 @scalarcmpA() uwtable ssp {
198 ; CHECK-LABEL: scalarcmpA:
200 ; CHECK-NEXT: vxorpd %xmm0, %xmm0, %xmm0
201 ; CHECK-NEXT: vcmpeqsd %xmm0, %xmm0, %xmm0
202 ; CHECK-NEXT: vmovq %xmm0, %rax
203 ; CHECK-NEXT: andl $1, %eax
204 ; CHECK-NEXT: # kill: def $eax killed $eax killed $rax
206 %cmp29 = fcmp oeq double undef, 0.000000e+00
207 %res = zext i1 %cmp29 to i32
211 define i32 @scalarcmpB() uwtable ssp {
212 ; CHECK-LABEL: scalarcmpB:
214 ; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
215 ; CHECK-NEXT: vcmpeqss %xmm0, %xmm0, %xmm0
216 ; CHECK-NEXT: vmovd %xmm0, %eax
217 ; CHECK-NEXT: andl $1, %eax
219 %cmp29 = fcmp oeq float undef, 0.000000e+00
220 %res = zext i1 %cmp29 to i32