3 # (C) Copyright IBM Corporation 2005
6 # Permission is hereby granted, free of charge, to any person obtaining a
7 # copy of this software and associated documentation files (the "Software"),
8 # to deal in the Software without restriction, including without limitation
9 # on the rights to use, copy, modify, merge, publish, distribute, sub
10 # license, and/or sell copies of the Software, and to permit persons to whom
11 # the Software is furnished to do so, subject to the following conditions:
13 # The above copyright notice and this permission notice (including the next
14 # paragraph) shall be included in all copies or substantial portions of the
17 # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 # FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
20 # IBM AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
22 # FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
26 # Ian Romanick <idr@us.ibm.com>
29 import gl_XML
, glX_XML
30 import sys
, getopt
, copy
32 def should_use_push(registers
):
33 for [reg
, offset
] in registers
:
41 def local_size(registers
):
42 # The x86-64 ABI says "the value (%rsp - 8) is always a multiple of
43 # 16 when control is transfered to the function entry point." This
44 # means that the local stack usage must be (16*N)+8 for some value
45 # of N. (16*N)+8 = (8*(2N))+8 = 8*(2N+1). As long as N is odd, we
46 # meet this requirement.
48 N
= (len(registers
) |
1)
52 def save_all_regs(registers
):
54 if not should_use_push(registers
):
55 adjust_stack
= local_size(registers
)
56 print '\tsubq\t$%u, %%rsp' % (adjust_stack
)
58 for [reg
, stack_offset
] in registers
:
59 save_reg( reg
, stack_offset
, adjust_stack
)
63 def restore_all_regs(registers
):
65 if not should_use_push(registers
):
66 adjust_stack
= local_size(registers
)
68 temp
= copy
.deepcopy(registers
)
70 [reg
, stack_offset
] = temp
.pop()
71 restore_reg(reg
, stack_offset
, adjust_stack
)
74 print '\taddq\t$%u, %%rsp' % (adjust_stack
)
78 def save_reg(reg
, offset
, use_move
):
81 print '\tmovq\t%s, (%%rsp)' % (reg
)
83 print '\tmovq\t%s, %u(%%rsp)' % (reg
, offset
)
85 print '\tpushq\t%s' % (reg
)
90 def restore_reg(reg
, offset
, use_move
):
93 print '\tmovq\t(%%rsp), %s' % (reg
)
95 print '\tmovq\t%u(%%rsp), %s' % (offset
, reg
)
97 print '\tpopq\t%s' % (reg
)
102 class PrintGenericStubs(gl_XML
.gl_print_base
):
105 gl_XML
.gl_print_base
.__init
__(self
)
107 self
.name
= "gl_x86-64_asm.py (from Mesa)"
108 self
.license
= license
.bsd_license_template
% ("(C) Copyright IBM Corporation 2005", "IBM")
112 def get_stack_size(self
, f
):
114 for p
in f
.parameterIterator():
115 size
+= p
.get_stack_size()
120 def printRealHeader(self
):
121 print "/* If we build with gcc's -fvisibility=hidden flag, we'll need to change"
122 print " * the symbol visibility mode to 'default'."
125 print '#include "../x86/assyntax.h"'
127 print '#if defined(__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__) >= 303'
128 print '# pragma GCC visibility push(default)'
129 print '# define HIDDEN(x) .hidden x'
131 print '# define HIDDEN(x)'
134 print '# if defined(USE_MGL_NAMESPACE)'
135 print '# define GL_PREFIX(n) GLNAME(CONCAT(mgl,n))'
136 print '# define _glapi_Dispatch _mglapi_Dispatch'
138 print '# define GL_PREFIX(n) GLNAME(CONCAT(gl,n))'
141 print '#if defined(PTHREADS) || defined(USE_XTHREADS) || defined(SOLARIS_THREADS) || defined(WIN32_THREADS) || defined(BEOS_THREADS)'
142 print '# define THREADS'
147 print '#ifdef GLX_USE_TLS'
149 print '\t.globl _x86_64_get_get_dispatch; HIDDEN(_x86_64_get_get_dispatch)'
150 print '_x86_64_get_get_dispatch:'
151 print '\tlea\t_x86_64_get_dispatch(%rip), %rax'
154 print '\t.p2align\t4,,15'
155 print '_x86_64_get_dispatch:'
156 print '\tmovq\t_glapi_tls_Dispatch@GOTTPOFF(%rip), %rax'
157 print '\tmovq\t%fs:(%rax), %rax'
159 print '\t.size\t_x86_64_get_dispatch, .-_x86_64_get_dispatch'
161 print '#elif defined(PTHREADS)'
163 print '\t.extern\t_glapi_Dispatch'
164 print '\t.extern\t_gl_DispatchTSD'
165 print '\t.extern\tpthread_getspecific'
167 print '\t.p2align\t4,,15'
168 print '_x86_64_get_dispatch:'
169 print '\tmovq\t_gl_DispatchTSD(%rip), %rdi'
170 print '\tjmp\tpthread_getspecific@PLT'
172 print '#elif defined(THREADS)'
174 print '\t.extern\t_glapi_get_dispatch'
181 def printRealFooter(self
):
183 print '#if defined(GLX_USE_TLS) && defined(__linux__)'
184 print ' .section ".note.ABI-tag", "a"'
186 print ' .long 1f - 0f /* name length */'
187 print ' .long 3f - 2f /* data length */'
188 print ' .long 1 /* note length */'
189 print '0: .asciz "GNU" /* vendor name */'
190 print '1: .p2align 2'
191 print '2: .long 0 /* note data: the ABI tag */'
192 print ' .long 2,4,20 /* Minimum kernel version w/TLS */'
193 print '3: .p2align 2 /* pad out section */'
194 print '#endif /* GLX_USE_TLS */'
196 print '#if defined (__ELF__) && defined (__linux__)'
197 print ' .section .note.GNU-stack,"",%progbits'
202 def printFunction(self
, f
):
204 # The x86-64 ABI divides function parameters into a couple
205 # classes. For the OpenGL interface, the only ones that are
206 # relevent are INTEGER and SSE. Basically, the first 8
207 # GLfloat or GLdouble parameters are placed in %xmm0 - %xmm7,
208 # the first 6 non-GLfloat / non-GLdouble parameters are placed
209 # in registers listed in int_parameters.
211 # If more parameters than that are required, they are passed
212 # on the stack. Therefore, we just have to make sure that
213 # %esp hasn't changed when we jump to the actual function.
214 # Since we're jumping to the function (and not calling it), we
215 # have to make sure of that anyway!
217 int_parameters
= ["%rdi", "%rsi", "%rdx", "%rcx", "%r8", "%r9"]
223 for p
in f
.parameterIterator():
224 type_name
= p
.get_base_type_string()
226 if p
.is_pointer() or (type_name
!= "GLfloat" and type_name
!= "GLdouble"):
228 registers
.append( [int_parameters
[int_class
], stack_offset
] )
233 registers
.append( ["%%xmm%u" % (sse_class
), stack_offset
] )
237 if ((int_class
& 1) == 0) and (sse_class
== 0):
238 registers
.append( ["%rbp", 0] )
241 name
= f
.dispatch_name()
243 print '\t.p2align\t4,,15'
244 print '\t.globl\tGL_PREFIX(%s)' % (name
)
245 print '\t.type\tGL_PREFIX(%s), @function' % (name
)
246 if not f
.is_static_entry_point(f
.name
):
247 print '\tHIDDEN(GL_PREFIX(%s))' % (name
)
248 print 'GL_PREFIX(%s):' % (name
)
249 print '#if defined(GLX_USE_TLS)'
250 print '\tcall\t_x86_64_get_dispatch@PLT'
251 print '\tmovq\t%u(%%rax), %%r11' % (f
.offset
* 8)
253 print '#elif defined(PTHREADS)'
255 save_all_regs(registers
)
256 print '\tcall\t_x86_64_get_dispatch@PLT'
257 restore_all_regs(registers
)
260 print '\tmovq\t(%rax), %r11'
262 print '\tmovq\t%u(%%rax), %%r11' % (f
.offset
* 8)
267 print '\tmovq\t_glapi_Dispatch(%rip), %rax'
268 print '\ttestq\t%rax, %rax'
270 print '\tmovq\t%u(%%rax), %%r11' % (f
.offset
* 8)
274 save_all_regs(registers
)
275 print '\tcall\t_glapi_get_dispatch'
276 restore_all_regs(registers
)
278 print '\tmovq\t%u(%%rax), %%r11' % (f
.offset
* 8)
280 print '#endif /* defined(GLX_USE_TLS) */'
282 print '\t.size\tGL_PREFIX(%s), .-GL_PREFIX(%s)' % (name
, name
)
287 def printBody(self
, api
):
288 for f
in api
.functionIterateByOffset():
289 self
.printFunction(f
)
292 for f
in api
.functionIterateByOffset():
293 dispatch
= f
.dispatch_name()
294 for n
in f
.entry_points
:
296 if f
.is_static_entry_point(n
):
297 text
= '\t.globl GL_PREFIX(%s) ; .set GL_PREFIX(%s), GL_PREFIX(%s)' % (n
, n
, dispatch
)
299 if f
.has_different_protocol(n
):
300 print '#ifndef GLX_INDIRECT_RENDERING'
309 print "Usage: %s [-f input_file_name] [-m output_mode]" % sys
.argv
[0]
312 if __name__
== '__main__':
313 file_name
= "gl_API.xml"
317 (args
, trail
) = getopt
.getopt(sys
.argv
[1:], "m:f:")
321 for (arg
,val
) in args
:
327 if mode
== "generic":
328 printer
= PrintGenericStubs()
330 print "ERROR: Invalid mode \"%s\" specified." % mode
333 api
= gl_XML
.parse_GL_API(file_name
, glX_XML
.glx_item_factory())