| blob | b2de7545ff5f51f97366afc8657d337774d71926 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i386-pc-win32 -mcpu=nehalem | FileCheck -check-prefix=WIN32 %s
3 ; RUN: llc < %s -mtriple=x86_64-win32 -mcpu=nehalem | FileCheck -check-prefix=WIN64 %s
4 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=nehalem | FileCheck -check-prefix=NOT_WIN %s
6 declare <16 x float> @func_float16_ptr(<16 x float>, ptr)
7 declare <16 x float> @func_float16(<16 x float>, <16 x float>)
9 ;test calling conventions - input parameters
10 define <16 x float> @testf16_inp(<16 x float> %a, <16 x float> %b) nounwind {
11 ; WIN32-LABEL: testf16_inp:
12 ; WIN32: # %bb.0:
13 ; WIN32-NEXT: pushl %ebp
14 ; WIN32-NEXT: movl %esp, %ebp
15 ; WIN32-NEXT: andl $-16, %esp
16 ; WIN32-NEXT: subl $80, %esp
17 ; WIN32-NEXT: movups 72(%ebp), %xmm4
18 ; WIN32-NEXT: movups 8(%ebp), %xmm3
19 ; WIN32-NEXT: addps %xmm4, %xmm3
20 ; WIN32-NEXT: movups 56(%ebp), %xmm4
21 ; WIN32-NEXT: movups 40(%ebp), %xmm5
22 ; WIN32-NEXT: movups 24(%ebp), %xmm6
23 ; WIN32-NEXT: movl %esp, %eax
24 ; WIN32-NEXT: addps %xmm6, %xmm0
25 ; WIN32-NEXT: addps %xmm5, %xmm1
26 ; WIN32-NEXT: addps %xmm4, %xmm2
27 ; WIN32-NEXT: pushl %eax
28 ; WIN32-NEXT: calll _func_float16_ptr
29 ; WIN32-NEXT: addl $4, %esp
30 ; WIN32-NEXT: addps (%esp), %xmm0
31 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm1
32 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm2
33 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm3
34 ; WIN32-NEXT: movl %ebp, %esp
35 ; WIN32-NEXT: popl %ebp
36 ; WIN32-NEXT: retl
37 ;
38 ; WIN64-LABEL: testf16_inp:
39 ; WIN64: # %bb.0:
40 ; WIN64-NEXT: subq $104, %rsp
41 ; WIN64-NEXT: movaps (%r9), %xmm3
42 ; WIN64-NEXT: movaps (%r8), %xmm2
43 ; WIN64-NEXT: movaps (%rdx), %xmm1
44 ; WIN64-NEXT: movaps (%rcx), %xmm0
45 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rax
46 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rcx
47 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rdx
48 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %r8
49 ; WIN64-NEXT: addps (%r8), %xmm0
50 ; WIN64-NEXT: addps (%rdx), %xmm1
51 ; WIN64-NEXT: addps (%rcx), %xmm2
52 ; WIN64-NEXT: addps (%rax), %xmm3
53 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rcx
54 ; WIN64-NEXT: callq func_float16_ptr
55 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm0
56 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm1
57 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm2
58 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm3
59 ; WIN64-NEXT: addq $104, %rsp
60 ; WIN64-NEXT: retq
61 ;
62 ; NOT_WIN-LABEL: testf16_inp:
63 ; NOT_WIN: ## %bb.0:
64 ; NOT_WIN-NEXT: subq $72, %rsp
65 ; NOT_WIN-NEXT: addps %xmm4, %xmm0
66 ; NOT_WIN-NEXT: addps %xmm5, %xmm1
67 ; NOT_WIN-NEXT: addps %xmm6, %xmm2
68 ; NOT_WIN-NEXT: addps %xmm7, %xmm3
69 ; NOT_WIN-NEXT: movq %rsp, %rdi
70 ; NOT_WIN-NEXT: callq _func_float16_ptr
71 ; NOT_WIN-NEXT: addps (%rsp), %xmm0
72 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm1
73 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm2
74 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm3
75 ; NOT_WIN-NEXT: addq $72, %rsp
76 ; NOT_WIN-NEXT: retq
77 %y = alloca <16 x float>, align 16
78 %x = fadd <16 x float> %a, %b
79 %1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, ptr %y)
80 %2 = load <16 x float>, ptr %y, align 16
81 %3 = fadd <16 x float> %2, %1
82 ret <16 x float> %3
83 }
85 ; test calling conventions - preserved registers
87 ; preserves xmm6-xmm15 on windows, xmm8-xmm15 on other plateforms.
88 define <16 x float> @testf16_regs(<16 x float> %a, <16 x float> %b) nounwind {
89 ; WIN32-LABEL: testf16_regs:
90 ; WIN32: # %bb.0:
91 ; WIN32-NEXT: pushl %ebp
92 ; WIN32-NEXT: movl %esp, %ebp
93 ; WIN32-NEXT: andl $-16, %esp
94 ; WIN32-NEXT: subl $80, %esp
95 ; WIN32-NEXT: movups 72(%ebp), %xmm6
96 ; WIN32-NEXT: movups 8(%ebp), %xmm3
97 ; WIN32-NEXT: movups 56(%ebp), %xmm7
98 ; WIN32-NEXT: movups 40(%ebp), %xmm5
99 ; WIN32-NEXT: movups 24(%ebp), %xmm4
100 ; WIN32-NEXT: movl %esp, %eax
101 ; WIN32-NEXT: addps %xmm4, %xmm0
102 ; WIN32-NEXT: addps %xmm5, %xmm1
103 ; WIN32-NEXT: addps %xmm7, %xmm2
104 ; WIN32-NEXT: addps %xmm6, %xmm3
105 ; WIN32-NEXT: pushl %eax
106 ; WIN32-NEXT: calll _func_float16_ptr
107 ; WIN32-NEXT: addl $4, %esp
108 ; WIN32-NEXT: movups 72(%ebp), %xmm4
109 ; WIN32-NEXT: addps %xmm4, %xmm3
110 ; WIN32-NEXT: movups 56(%ebp), %xmm4
111 ; WIN32-NEXT: addps %xmm4, %xmm2
112 ; WIN32-NEXT: movups 40(%ebp), %xmm4
113 ; WIN32-NEXT: addps %xmm4, %xmm1
114 ; WIN32-NEXT: movups 24(%ebp), %xmm4
115 ; WIN32-NEXT: addps %xmm4, %xmm0
116 ; WIN32-NEXT: addps (%esp), %xmm0
117 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm1
118 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm2
119 ; WIN32-NEXT: addps {{[0-9]+}}(%esp), %xmm3
120 ; WIN32-NEXT: movl %ebp, %esp
121 ; WIN32-NEXT: popl %ebp
122 ; WIN32-NEXT: retl
123 ;
124 ; WIN64-LABEL: testf16_regs:
125 ; WIN64: # %bb.0:
126 ; WIN64-NEXT: subq $168, %rsp
127 ; WIN64-NEXT: movaps %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
128 ; WIN64-NEXT: movaps %xmm8, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
129 ; WIN64-NEXT: movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
130 ; WIN64-NEXT: movaps %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
131 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rax
132 ; WIN64-NEXT: movaps (%rax), %xmm6
133 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rax
134 ; WIN64-NEXT: movaps (%rax), %xmm7
135 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rax
136 ; WIN64-NEXT: movaps (%rax), %xmm8
137 ; WIN64-NEXT: movq {{[0-9]+}}(%rsp), %rax
138 ; WIN64-NEXT: movaps (%rax), %xmm9
139 ; WIN64-NEXT: movaps (%rcx), %xmm0
140 ; WIN64-NEXT: addps %xmm9, %xmm0
141 ; WIN64-NEXT: movaps (%rdx), %xmm1
142 ; WIN64-NEXT: addps %xmm8, %xmm1
143 ; WIN64-NEXT: movaps (%r8), %xmm2
144 ; WIN64-NEXT: addps %xmm7, %xmm2
145 ; WIN64-NEXT: movaps (%r9), %xmm3
146 ; WIN64-NEXT: addps %xmm6, %xmm3
147 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rcx
148 ; WIN64-NEXT: callq func_float16_ptr
149 ; WIN64-NEXT: addps %xmm6, %xmm3
150 ; WIN64-NEXT: addps %xmm7, %xmm2
151 ; WIN64-NEXT: addps %xmm8, %xmm1
152 ; WIN64-NEXT: addps %xmm9, %xmm0
153 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm0
154 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm1
155 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm2
156 ; WIN64-NEXT: addps {{[0-9]+}}(%rsp), %xmm3
157 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
158 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
159 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm8 # 16-byte Reload
160 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 # 16-byte Reload
161 ; WIN64-NEXT: addq $168, %rsp
162 ; WIN64-NEXT: retq
163 ;
164 ; NOT_WIN-LABEL: testf16_regs:
165 ; NOT_WIN: ## %bb.0:
166 ; NOT_WIN-NEXT: subq $72, %rsp
167 ; NOT_WIN-NEXT: movaps %xmm7, %xmm9
168 ; NOT_WIN-NEXT: movaps %xmm6, %xmm10
169 ; NOT_WIN-NEXT: movaps %xmm5, %xmm11
170 ; NOT_WIN-NEXT: movaps %xmm4, %xmm8
171 ; NOT_WIN-NEXT: addps %xmm4, %xmm0
172 ; NOT_WIN-NEXT: addps %xmm5, %xmm1
173 ; NOT_WIN-NEXT: addps %xmm6, %xmm2
174 ; NOT_WIN-NEXT: addps %xmm7, %xmm3
175 ; NOT_WIN-NEXT: movq %rsp, %rdi
176 ; NOT_WIN-NEXT: callq _func_float16_ptr
177 ; NOT_WIN-NEXT: addps %xmm9, %xmm3
178 ; NOT_WIN-NEXT: addps %xmm10, %xmm2
179 ; NOT_WIN-NEXT: addps %xmm11, %xmm1
180 ; NOT_WIN-NEXT: addps %xmm8, %xmm0
181 ; NOT_WIN-NEXT: addps (%rsp), %xmm0
182 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm1
183 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm2
184 ; NOT_WIN-NEXT: addps {{[0-9]+}}(%rsp), %xmm3
185 ; NOT_WIN-NEXT: addq $72, %rsp
186 ; NOT_WIN-NEXT: retq
187 %y = alloca <16 x float>, align 16
188 %x = fadd <16 x float> %a, %b
189 %1 = call intel_ocl_bicc <16 x float> @func_float16_ptr(<16 x float> %x, ptr %y)
190 %2 = load <16 x float>, ptr %y, align 16
191 %3 = fadd <16 x float> %1, %b
192 %4 = fadd <16 x float> %2, %3
193 ret <16 x float> %4
194 }
196 ; test calling conventions - prolog and epilog
197 define intel_ocl_bicc <16 x float> @test_prolog_epilog(<16 x float> %a, <16 x float> %b) nounwind {
198 ; WIN32-LABEL: test_prolog_epilog:
199 ; WIN32: # %bb.0:
200 ; WIN32-NEXT: pushl %ebp
201 ; WIN32-NEXT: movl %esp, %ebp
202 ; WIN32-NEXT: andl $-16, %esp
203 ; WIN32-NEXT: subl $96, %esp
204 ; WIN32-NEXT: movups 8(%ebp), %xmm4
205 ; WIN32-NEXT: movups 24(%ebp), %xmm5
206 ; WIN32-NEXT: movups 40(%ebp), %xmm6
207 ; WIN32-NEXT: movups 56(%ebp), %xmm7
208 ; WIN32-NEXT: movups %xmm7, {{[0-9]+}}(%esp)
209 ; WIN32-NEXT: movups %xmm6, {{[0-9]+}}(%esp)
210 ; WIN32-NEXT: movups %xmm5, {{[0-9]+}}(%esp)
211 ; WIN32-NEXT: movups %xmm4, {{[0-9]+}}(%esp)
212 ; WIN32-NEXT: movups %xmm3, (%esp)
213 ; WIN32-NEXT: calll _func_float16
214 ; WIN32-NEXT: movl %ebp, %esp
215 ; WIN32-NEXT: popl %ebp
216 ; WIN32-NEXT: retl
217 ;
218 ; WIN64-LABEL: test_prolog_epilog:
219 ; WIN64: # %bb.0:
220 ; WIN64-NEXT: subq $232, %rsp
221 ; WIN64-NEXT: movaps %xmm7, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
222 ; WIN64-NEXT: movaps %xmm6, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
223 ; WIN64-NEXT: movaps (%r9), %xmm4
224 ; WIN64-NEXT: movaps (%rdx), %xmm5
225 ; WIN64-NEXT: movaps (%r8), %xmm6
226 ; WIN64-NEXT: movaps (%rcx), %xmm7
227 ; WIN64-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp)
228 ; WIN64-NEXT: movaps %xmm0, {{[0-9]+}}(%rsp)
229 ; WIN64-NEXT: movaps %xmm2, {{[0-9]+}}(%rsp)
230 ; WIN64-NEXT: movaps %xmm3, {{[0-9]+}}(%rsp)
231 ; WIN64-NEXT: movaps %xmm7, {{[0-9]+}}(%rsp)
232 ; WIN64-NEXT: movaps %xmm6, {{[0-9]+}}(%rsp)
233 ; WIN64-NEXT: movaps %xmm5, {{[0-9]+}}(%rsp)
234 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rax
235 ; WIN64-NEXT: movq %rax, {{[0-9]+}}(%rsp)
236 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rax
237 ; WIN64-NEXT: movq %rax, {{[0-9]+}}(%rsp)
238 ; WIN64-NEXT: movaps %xmm4, {{[0-9]+}}(%rsp)
239 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rax
240 ; WIN64-NEXT: movq %rax, {{[0-9]+}}(%rsp)
241 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rax
242 ; WIN64-NEXT: movq %rax, {{[0-9]+}}(%rsp)
243 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rcx
244 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %rdx
245 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %r8
246 ; WIN64-NEXT: leaq {{[0-9]+}}(%rsp), %r9
247 ; WIN64-NEXT: callq func_float16
248 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm6 # 16-byte Reload
249 ; WIN64-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm7 # 16-byte Reload
250 ; WIN64-NEXT: addq $232, %rsp
251 ; WIN64-NEXT: retq
252 ;
253 ; NOT_WIN-LABEL: test_prolog_epilog:
254 ; NOT_WIN: ## %bb.0:
255 ; NOT_WIN-NEXT: subq $136, %rsp
256 ; NOT_WIN-NEXT: movaps %xmm15, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
257 ; NOT_WIN-NEXT: movaps %xmm14, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
258 ; NOT_WIN-NEXT: movaps %xmm13, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
259 ; NOT_WIN-NEXT: movaps %xmm12, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
260 ; NOT_WIN-NEXT: movaps %xmm11, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
261 ; NOT_WIN-NEXT: movaps %xmm10, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
262 ; NOT_WIN-NEXT: movaps %xmm9, {{[-0-9]+}}(%r{{[sb]}}p) ## 16-byte Spill
263 ; NOT_WIN-NEXT: movaps %xmm8, (%rsp) ## 16-byte Spill
264 ; NOT_WIN-NEXT: callq _func_float16
265 ; NOT_WIN-NEXT: movaps (%rsp), %xmm8 ## 16-byte Reload
266 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm9 ## 16-byte Reload
267 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm10 ## 16-byte Reload
268 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm11 ## 16-byte Reload
269 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm12 ## 16-byte Reload
270 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm13 ## 16-byte Reload
271 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm14 ## 16-byte Reload
272 ; NOT_WIN-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm15 ## 16-byte Reload
273 ; NOT_WIN-NEXT: addq $136, %rsp
274 ; NOT_WIN-NEXT: retq
275 %c = call <16 x float> @func_float16(<16 x float> %a, <16 x float> %b)
276 ret <16 x float> %c
277 }