Mercurial > hg > CbC > CbC_llvm

comparison test/CodeGen/X86/sse1.ll @ 120:1172e4bd9c6f

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
update 4.0.0
author mir3636
date Fri, 25 Nov 2016 19:14:25 +0900
parents 60c9769439b8
children 803732b1fca8
comparison
equal deleted inserted replaced
101:34baf5011add 120:1172e4bd9c6f
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
1 ; Tests for SSE1 and below, without SSE2+. 2 ; Tests for SSE1 and below, without SSE2+.
2 ; RUN: llc < %s -mtriple=i386-unknown-unknown -march=x86 -mcpu=pentium3 -O3 | FileCheck %s 3 ; RUN: llc < %s -mtriple=i386-unknown-unknown -mcpu=pentium3 -O3 | FileCheck %s --check-prefix=X32
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -march=x86-64 -mattr=-sse2,+sse -O3 | FileCheck %s 4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-sse2,+sse -O3 | FileCheck %s --check-prefix=X64
4 5
5 ; PR7993 6 ; PR7993
6 ;define <4 x i32> @test3(<4 x i16> %a) nounwind { 7 ;define <4 x i32> @test3(<4 x i16> %a) nounwind {
7 ; %c = sext <4 x i16> %a to <4 x i32> ; <<4 x i32>> [#uses=1] 8 ; %c = sext <4 x i16> %a to <4 x i32> ; <<4 x i32>> [#uses=1]
8 ; ret <4 x i32> %c 9 ; ret <4 x i32> %c
10 11
11 ; This should not emit shuffles to populate the top 2 elements of the 4-element 12 ; This should not emit shuffles to populate the top 2 elements of the 4-element
12 ; vector that this ends up returning. 13 ; vector that this ends up returning.
13 ; rdar://8368414 14 ; rdar://8368414
14 define <2 x float> @test4(<2 x float> %A, <2 x float> %B) nounwind { 15 define <2 x float> @test4(<2 x float> %A, <2 x float> %B) nounwind {
15 ; CHECK-LABEL: test4: 16 ; X32-LABEL: test4:
16 ; CHECK: # BB#0: # %entry 17 ; X32: # BB#0: # %entry
17 ; CHECK-NEXT: movaps %xmm0, %xmm2 18 ; X32-NEXT: movaps %xmm0, %xmm2
18 ; CHECK-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1,2,3] 19 ; X32-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1,2,3]
19 ; CHECK-NEXT: addss %xmm1, %xmm0 20 ; X32-NEXT: addss %xmm1, %xmm0
20 ; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,2,3] 21 ; X32-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
21 ; CHECK-NEXT: subss %xmm1, %xmm2 22 ; X32-NEXT: subss %xmm1, %xmm2
22 ; CHECK-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1] 23 ; X32-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
23 ; CHECK-NEXT: ret 24 ; X32-NEXT: retl
25 ;
26 ; X64-LABEL: test4:
27 ; X64: # BB#0: # %entry
28 ; X64-NEXT: movaps %xmm0, %xmm2
29 ; X64-NEXT: shufps {{.*#+}} xmm2 = xmm2[1,1,2,3]
30 ; X64-NEXT: addss %xmm1, %xmm0
31 ; X64-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,2,3]
32 ; X64-NEXT: subss %xmm1, %xmm2
33 ; X64-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
34 ; X64-NEXT: retq
24 entry: 35 entry:
25 %tmp7 = extractelement <2 x float> %A, i32 0 36 %tmp7 = extractelement <2 x float> %A, i32 0
26 %tmp5 = extractelement <2 x float> %A, i32 1 37 %tmp5 = extractelement <2 x float> %A, i32 1
27 %tmp3 = extractelement <2 x float> %B, i32 0 38 %tmp3 = extractelement <2 x float> %B, i32 0
28 %tmp1 = extractelement <2 x float> %B, i32 1 39 %tmp1 = extractelement <2 x float> %B, i32 1
38 ; is not. We used to ping pong between splitting the vselect for the v4i 49 ; is not. We used to ping pong between splitting the vselect for the v4i
39 ; condition operand and widening the resulting vselect for the v4f32 result. 50 ; condition operand and widening the resulting vselect for the v4f32 result.
40 ; PR18036 51 ; PR18036
41 52
42 define <4 x float> @vselect(<4 x float>*%p, <4 x i32> %q) { 53 define <4 x float> @vselect(<4 x float>*%p, <4 x i32> %q) {
43 ; CHECK-LABEL: vselect: 54 ; X32-LABEL: vselect:
44 ; CHECK: ret 55 ; X32: # BB#0: # %entry
56 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
57 ; X32-NEXT: xorps %xmm0, %xmm0
58 ; X32-NEXT: je .LBB1_1
59 ; X32-NEXT: # BB#2: # %entry
60 ; X32-NEXT: xorps %xmm1, %xmm1
61 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
62 ; X32-NEXT: jne .LBB1_5
63 ; X32-NEXT: jmp .LBB1_4
64 ; X32-NEXT: .LBB1_1:
65 ; X32-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
66 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
67 ; X32-NEXT: je .LBB1_4
68 ; X32-NEXT: .LBB1_5: # %entry
69 ; X32-NEXT: xorps %xmm2, %xmm2
70 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
71 ; X32-NEXT: jne .LBB1_8
72 ; X32-NEXT: jmp .LBB1_7
73 ; X32-NEXT: .LBB1_4:
74 ; X32-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
75 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
76 ; X32-NEXT: je .LBB1_7
77 ; X32-NEXT: .LBB1_8: # %entry
78 ; X32-NEXT: xorps %xmm3, %xmm3
79 ; X32-NEXT: jmp .LBB1_9
80 ; X32-NEXT: .LBB1_7:
81 ; X32-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
82 ; X32-NEXT: .LBB1_9: # %entry
83 ; X32-NEXT: cmpl $0, {{[0-9]+}}(%esp)
84 ; X32-NEXT: unpcklps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
85 ; X32-NEXT: jne .LBB1_11
86 ; X32-NEXT: # BB#10:
87 ; X32-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
88 ; X32-NEXT: .LBB1_11: # %entry
89 ; X32-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
90 ; X32-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
91 ; X32-NEXT: retl
92 ;
93 ; X64-LABEL: vselect:
94 ; X64: # BB#0: # %entry
95 ; X64-NEXT: testl %ecx, %ecx
96 ; X64-NEXT: xorps %xmm0, %xmm0
97 ; X64-NEXT: je .LBB1_1
98 ; X64-NEXT: # BB#2: # %entry
99 ; X64-NEXT: xorps %xmm1, %xmm1
100 ; X64-NEXT: testl %edx, %edx
101 ; X64-NEXT: jne .LBB1_5
102 ; X64-NEXT: jmp .LBB1_4
103 ; X64-NEXT: .LBB1_1:
104 ; X64-NEXT: movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
105 ; X64-NEXT: testl %edx, %edx
106 ; X64-NEXT: je .LBB1_4
107 ; X64-NEXT: .LBB1_5: # %entry
108 ; X64-NEXT: xorps %xmm2, %xmm2
109 ; X64-NEXT: testl %r8d, %r8d
110 ; X64-NEXT: jne .LBB1_8
111 ; X64-NEXT: jmp .LBB1_7
112 ; X64-NEXT: .LBB1_4:
113 ; X64-NEXT: movss {{.*#+}} xmm2 = mem[0],zero,zero,zero
114 ; X64-NEXT: testl %r8d, %r8d
115 ; X64-NEXT: je .LBB1_7
116 ; X64-NEXT: .LBB1_8: # %entry
117 ; X64-NEXT: xorps %xmm3, %xmm3
118 ; X64-NEXT: jmp .LBB1_9
119 ; X64-NEXT: .LBB1_7:
120 ; X64-NEXT: movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
121 ; X64-NEXT: .LBB1_9: # %entry
122 ; X64-NEXT: testl %esi, %esi
123 ; X64-NEXT: unpcklps {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
124 ; X64-NEXT: jne .LBB1_11
125 ; X64-NEXT: # BB#10:
126 ; X64-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
127 ; X64-NEXT: .LBB1_11: # %entry
128 ; X64-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
129 ; X64-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
130 ; X64-NEXT: retq
45 entry: 131 entry:
46 %a1 = icmp eq <4 x i32> %q, zeroinitializer 132 %a1 = icmp eq <4 x i32> %q, zeroinitializer
47 %a14 = select <4 x i1> %a1, <4 x float> <float 1.000000e+00, float 2.000000e+00, float 3.000000e+00, float 4.000000e+0> , <4 x float> zeroinitializer 133 %a14 = select <4 x i1> %a1, <4 x float> <float 1.000000e+00, float 2.000000e+00, float 3.000000e+00, float 4.000000e+0> , <4 x float> zeroinitializer
48 ret <4 x float> %a14 134 ret <4 x float> %a14
49 } 135 }
136
137 ; v4i32 isn't legal for SSE1, but this should be cmpps.
138
139 define <4 x float> @PR28044(<4 x float> %a0, <4 x float> %a1) nounwind {
140 ; X32-LABEL: PR28044:
141 ; X32: # BB#0:
142 ; X32-NEXT: cmpeqps %xmm1, %xmm0
143 ; X32-NEXT: retl
144 ;
145 ; X64-LABEL: PR28044:
146 ; X64: # BB#0:
147 ; X64-NEXT: cmpeqps %xmm1, %xmm0
148 ; X64-NEXT: retq
149 %cmp = fcmp oeq <4 x float> %a0, %a1
150 %sext = sext <4 x i1> %cmp to <4 x i32>
151 %res = bitcast <4 x i32> %sext to <4 x float>
152 ret <4 x float> %res
153 }
154
155 ; Don't crash trying to do the impossible: an integer vector comparison doesn't exist, so we must scalarize.
156 ; https://llvm.org/bugs/show_bug.cgi?id=30512
157
158 define <4 x i32> @PR30512(<4 x i32> %x, <4 x i32> %y) nounwind {
159 ; X32-LABEL: PR30512:
160 ; X32: # BB#0:
161 ; X32-NEXT: pushl %ebp
162 ; X32-NEXT: pushl %ebx
163 ; X32-NEXT: pushl %edi
164 ; X32-NEXT: pushl %esi
165 ; X32-NEXT: movl {{[0-9]+}}(%esp), %ebp
166 ; X32-NEXT: movl {{[0-9]+}}(%esp), %esi
167 ; X32-NEXT: movl {{[0-9]+}}(%esp), %edi
168 ; X32-NEXT: movl {{[0-9]+}}(%esp), %ebx
169 ; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
170 ; X32-NEXT: xorl %ecx, %ecx
171 ; X32-NEXT: cmpl {{[0-9]+}}(%esp), %edx
172 ; X32-NEXT: sete %cl
173 ; X32-NEXT: xorl %edx, %edx
174 ; X32-NEXT: cmpl {{[0-9]+}}(%esp), %ebx
175 ; X32-NEXT: sete %dl
176 ; X32-NEXT: xorl %ebx, %ebx
177 ; X32-NEXT: cmpl {{[0-9]+}}(%esp), %edi
178 ; X32-NEXT: sete %bl
179 ; X32-NEXT: xorl %eax, %eax
180 ; X32-NEXT: cmpl {{[0-9]+}}(%esp), %esi
181 ; X32-NEXT: sete %al
182 ; X32-NEXT: movl %eax, 12(%ebp)
183 ; X32-NEXT: movl %ebx, 8(%ebp)
184 ; X32-NEXT: movl %edx, 4(%ebp)
185 ; X32-NEXT: movl %ecx, (%ebp)
186 ; X32-NEXT: movl %ebp, %eax
187 ; X32-NEXT: popl %esi
188 ; X32-NEXT: popl %edi
189 ; X32-NEXT: popl %ebx
190 ; X32-NEXT: popl %ebp
191 ; X32-NEXT: retl $4
192 ;
193 ; X64-LABEL: PR30512:
194 ; X64: # BB#0:
195 ; X64-NEXT: xorl %eax, %eax
196 ; X64-NEXT: cmpl %r9d, %esi
197 ; X64-NEXT: sete %al
198 ; X64-NEXT: xorl %esi, %esi
199 ; X64-NEXT: cmpl {{[0-9]+}}(%rsp), %edx
200 ; X64-NEXT: sete %sil
201 ; X64-NEXT: xorl %edx, %edx
202 ; X64-NEXT: cmpl {{[0-9]+}}(%rsp), %ecx
203 ; X64-NEXT: sete %dl
204 ; X64-NEXT: xorl %ecx, %ecx
205 ; X64-NEXT: cmpl {{[0-9]+}}(%rsp), %r8d
206 ; X64-NEXT: sete %cl
207 ; X64-NEXT: movl %ecx, 12(%rdi)
208 ; X64-NEXT: movl %edx, 8(%rdi)
209 ; X64-NEXT: movl %esi, 4(%rdi)
210 ; X64-NEXT: movl %eax, (%rdi)
211 ; X64-NEXT: movq %rdi, %rax
212 ; X64-NEXT: retq
213 %cmp = icmp eq <4 x i32> %x, %y
214 %zext = zext <4 x i1> %cmp to <4 x i32>
215 ret <4 x i32> %zext
216 }
217