1from __future__ import print_function
2import re
3import sys
4
5from . import common
6
7if sys.version_info[0] > 2:
8  class string:
9    expandtabs = str.expandtabs
10else:
11  import string
12
13# RegEx: this is where the magic happens.
14
15##### Assembly parser
16
17ASM_FUNCTION_X86_RE = re.compile(
18    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n(?:\s*\.?Lfunc_begin[^:\n]*:\n)?[^:]*?'
19    r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*'
20    r'^\s*(?:[^:\n]+?:\s*\n\s*\.size|\.cfi_endproc|\.globl|\.comm|\.(?:sub)?section|#+ -- End function)',
21    flags=(re.M | re.S))
22
23ASM_FUNCTION_ARM_RE = re.compile(
24        r'^(?P<func>[0-9a-zA-Z_]+):\n' # f: (name of function)
25        r'\s+\.fnstart\n' # .fnstart
26        r'(?P<body>.*?)\n' # (body of the function)
27        r'.Lfunc_end[0-9]+:', # .Lfunc_end0: or # -- End function
28        flags=(re.M | re.S))
29
30ASM_FUNCTION_AARCH64_RE = re.compile(
31     r'^_?(?P<func>[^:]+):[ \t]*\/\/[ \t]*@(?P=func)\n'
32     r'(?:[ \t]+.cfi_startproc\n)?'  # drop optional cfi noise
33     r'(?P<body>.*?)\n'
34     # This list is incomplete
35     r'.Lfunc_end[0-9]+:\n',
36     flags=(re.M | re.S))
37
38ASM_FUNCTION_AMDGPU_RE = re.compile(
39    r'^_?(?P<func>[^:]+):[ \t]*;+[ \t]*@(?P=func)\n[^:]*?'
40    r'(?P<body>.*?)\n' # (body of the function)
41    # This list is incomplete
42    r'^\s*(\.Lfunc_end[0-9]+:\n|\.section)',
43    flags=(re.M | re.S))
44
45ASM_FUNCTION_HEXAGON_RE = re.compile(
46    r'^_?(?P<func>[^:]+):[ \t]*//[ \t]*@(?P=func)\n[^:]*?'
47    r'(?P<body>.*?)\n' # (body of the function)
48    # This list is incomplete
49    r'.Lfunc_end[0-9]+:\n',
50    flags=(re.M | re.S))
51
52ASM_FUNCTION_MIPS_RE = re.compile(
53    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n[^:]*?' # f: (name of func)
54    r'(?:^[ \t]+\.(frame|f?mask|set).*?\n)+'  # Mips+LLVM standard asm prologue
55    r'(?P<body>.*?)\n'                        # (body of the function)
56    r'(?:^[ \t]+\.(set|end).*?\n)+'           # Mips+LLVM standard asm epilogue
57    r'(\$|\.L)func_end[0-9]+:\n',             # $func_end0: (mips32 - O32) or
58                                              # .Lfunc_end0: (mips64 - NewABI)
59    flags=(re.M | re.S))
60
61ASM_FUNCTION_PPC_RE = re.compile(
62    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n'
63    r'.*?'
64    r'\.Lfunc_begin[0-9]+:\n'
65    r'(?:[ \t]+.cfi_startproc\n)?'
66    r'(?:\.Lfunc_[gl]ep[0-9]+:\n(?:[ \t]+.*?\n)*)*'
67    r'(?P<body>.*?)\n'
68    # This list is incomplete
69    r'(?:^[ \t]*(?:\.long[ \t]+[^\n]+|\.quad[ \t]+[^\n]+)\n)*'
70    r'.Lfunc_end[0-9]+:\n',
71    flags=(re.M | re.S))
72
73ASM_FUNCTION_RISCV_RE = re.compile(
74    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n(?:\s*\.?Lfunc_begin[^:\n]*:\n)?[^:]*?'
75    r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*'
76    r'.Lfunc_end[0-9]+:\n',
77    flags=(re.M | re.S))
78
79ASM_FUNCTION_LANAI_RE = re.compile(
80    r'^_?(?P<func>[^:]+):[ \t]*!+[ \t]*@(?P=func)\n'
81    r'(?:[ \t]+.cfi_startproc\n)?'  # drop optional cfi noise
82    r'(?P<body>.*?)\s*'
83    r'.Lfunc_end[0-9]+:\n',
84    flags=(re.M | re.S))
85
86ASM_FUNCTION_SPARC_RE = re.compile(
87    r'^_?(?P<func>[^:]+):[ \t]*!+[ \t]*@(?P=func)\n'
88    r'(?P<body>.*?)\s*'
89    r'.Lfunc_end[0-9]+:\n',
90    flags=(re.M | re.S))
91
92ASM_FUNCTION_SYSTEMZ_RE = re.compile(
93    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n'
94    r'[ \t]+.cfi_startproc\n'
95    r'(?P<body>.*?)\n'
96    r'.Lfunc_end[0-9]+:\n',
97    flags=(re.M | re.S))
98
99ASM_FUNCTION_AARCH64_DARWIN_RE = re.compile(
100     r'^_(?P<func>[^:]+):[ \t]*;[ \t]@(?P=func)\n'
101     r'([ \t]*.cfi_startproc\n[\s]*)?'
102     r'(?P<body>.*?)'
103     r'([ \t]*.cfi_endproc\n[\s]*)?'
104     r'^[ \t]*;[ \t]--[ \t]End[ \t]function',
105     flags=(re.M | re.S))
106
107ASM_FUNCTION_ARM_DARWIN_RE = re.compile(
108     r'^[ \t]*\.globl[ \t]*_(?P<func>[^ \t])[ \t]*@[ \t]--[ \t]Begin[ \t]function[ \t](?P=func)'
109     r'(?P<directives>.*?)'
110     r'^_(?P=func):\n[ \t]*'
111     r'(?P<body>.*?)'
112     r'^[ \t]*@[ \t]--[ \t]End[ \t]function',
113     flags=(re.M | re.S ))
114
115ASM_FUNCTION_ARM_MACHO_RE = re.compile(
116     r'^_(?P<func>[^:]+):[ \t]*\n'
117     r'([ \t]*.cfi_startproc\n[ \t]*)?'
118     r'(?P<body>.*?)\n'
119     r'[ \t]*\.cfi_endproc\n',
120     flags=(re.M | re.S))
121
122ASM_FUNCTION_ARM_IOS_RE = re.compile(
123     r'^_(?P<func>[^:]+):[ \t]*\n'
124     r'^Lfunc_begin(?P<id>[0-9][1-9]*):\n'
125     r'(?P<body>.*?)'
126     r'^Lfunc_end(?P=id):\n'
127     r'^[ \t]*@[ \t]--[ \t]End[ \t]function',
128     flags=(re.M | re.S))
129
130ASM_FUNCTION_WASM32_RE = re.compile(
131    r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n'
132    r'(?P<body>.*?)\n'
133    r'^\s*(\.Lfunc_end[0-9]+:\n|end_function)',
134    flags=(re.M | re.S))
135
136
137SCRUB_LOOP_COMMENT_RE = re.compile(
138    r'# =>This Inner Loop Header:.*|# in Loop:.*', flags=re.M)
139
140SCRUB_X86_SHUFFLES_RE = (
141    re.compile(
142        r'^(\s*\w+) [^#\n]+#+ ((?:[xyz]mm\d+|mem)( \{%k\d+\}( \{z\})?)? = .*)$',
143        flags=re.M))
144SCRUB_X86_SPILL_RELOAD_RE = (
145    re.compile(
146        r'-?\d+\(%([er])[sb]p\)(.*(?:Spill|Reload))$',
147        flags=re.M))
148SCRUB_X86_SP_RE = re.compile(r'\d+\(%(esp|rsp)\)')
149SCRUB_X86_RIP_RE = re.compile(r'[.\w]+\(%rip\)')
150SCRUB_X86_LCP_RE = re.compile(r'\.LCPI[0-9]+_[0-9]+')
151SCRUB_X86_RET_RE = re.compile(r'ret[l|q]')
152
153def scrub_asm_x86(asm, args):
154  # Scrub runs of whitespace out of the assembly, but leave the leading
155  # whitespace in place.
156  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
157  # Expand the tabs used for indentation.
158  asm = string.expandtabs(asm, 2)
159  # Detect shuffle asm comments and hide the operands in favor of the comments.
160  asm = SCRUB_X86_SHUFFLES_RE.sub(r'\1 {{.*#+}} \2', asm)
161  # Detect stack spills and reloads and hide their exact offset and whether
162  # they used the stack pointer or frame pointer.
163  asm = SCRUB_X86_SPILL_RELOAD_RE.sub(r'{{[-0-9]+}}(%\1{{[sb]}}p)\2', asm)
164  # Generically match the stack offset of a memory operand.
165  asm = SCRUB_X86_SP_RE.sub(r'{{[0-9]+}}(%\1)', asm)
166  if getattr(args, 'x86_scrub_rip', False):
167    # Generically match a RIP-relative memory operand.
168    asm = SCRUB_X86_RIP_RE.sub(r'{{.*}}(%rip)', asm)
169  # Generically match a LCP symbol.
170  asm = SCRUB_X86_LCP_RE.sub(r'{{\.LCPI.*}}', asm)
171  if getattr(args, 'extra_scrub', False):
172    # Avoid generating different checks for 32- and 64-bit because of 'retl' vs 'retq'.
173    asm = SCRUB_X86_RET_RE.sub(r'ret{{[l|q]}}', asm)
174  # Strip kill operands inserted into the asm.
175  asm = common.SCRUB_KILL_COMMENT_RE.sub('', asm)
176  # Strip trailing whitespace.
177  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
178  return asm
179
180def scrub_asm_amdgpu(asm, args):
181  # Scrub runs of whitespace out of the assembly, but leave the leading
182  # whitespace in place.
183  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
184  # Expand the tabs used for indentation.
185  asm = string.expandtabs(asm, 2)
186  # Strip trailing whitespace.
187  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
188  return asm
189
190def scrub_asm_arm_eabi(asm, args):
191  # Scrub runs of whitespace out of the assembly, but leave the leading
192  # whitespace in place.
193  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
194  # Expand the tabs used for indentation.
195  asm = string.expandtabs(asm, 2)
196  # Strip kill operands inserted into the asm.
197  asm = common.SCRUB_KILL_COMMENT_RE.sub('', asm)
198  # Strip trailing whitespace.
199  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
200  return asm
201
202def scrub_asm_hexagon(asm, args):
203  # Scrub runs of whitespace out of the assembly, but leave the leading
204  # whitespace in place.
205  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
206  # Expand the tabs used for indentation.
207  asm = string.expandtabs(asm, 2)
208  # Strip trailing whitespace.
209  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
210  return asm
211
212def scrub_asm_powerpc(asm, args):
213  # Scrub runs of whitespace out of the assembly, but leave the leading
214  # whitespace in place.
215  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
216  # Expand the tabs used for indentation.
217  asm = string.expandtabs(asm, 2)
218  # Stripe unimportant comments, but leave the token '#' in place.
219  asm = SCRUB_LOOP_COMMENT_RE.sub(r'#', asm)
220  # Strip trailing whitespace.
221  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
222  return asm
223
224def scrub_asm_mips(asm, args):
225  # Scrub runs of whitespace out of the assembly, but leave the leading
226  # whitespace in place.
227  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
228  # Expand the tabs used for indentation.
229  asm = string.expandtabs(asm, 2)
230  # Strip trailing whitespace.
231  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
232  return asm
233
234def scrub_asm_riscv(asm, args):
235  # Scrub runs of whitespace out of the assembly, but leave the leading
236  # whitespace in place.
237  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
238  # Expand the tabs used for indentation.
239  asm = string.expandtabs(asm, 2)
240  # Strip trailing whitespace.
241  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
242  return asm
243
244def scrub_asm_lanai(asm, args):
245  # Scrub runs of whitespace out of the assembly, but leave the leading
246  # whitespace in place.
247  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
248  # Expand the tabs used for indentation.
249  asm = string.expandtabs(asm, 2)
250  # Strip trailing whitespace.
251  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
252  return asm
253
254def scrub_asm_sparc(asm, args):
255  # Scrub runs of whitespace out of the assembly, but leave the leading
256  # whitespace in place.
257  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
258  # Expand the tabs used for indentation.
259  asm = string.expandtabs(asm, 2)
260  # Strip trailing whitespace.
261  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
262  return asm
263
264def scrub_asm_systemz(asm, args):
265  # Scrub runs of whitespace out of the assembly, but leave the leading
266  # whitespace in place.
267  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
268  # Expand the tabs used for indentation.
269  asm = string.expandtabs(asm, 2)
270  # Strip trailing whitespace.
271  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
272  return asm
273
274def scrub_asm_wasm32(asm, args):
275  # Scrub runs of whitespace out of the assembly, but leave the leading
276  # whitespace in place.
277  asm = common.SCRUB_WHITESPACE_RE.sub(r' ', asm)
278  # Expand the tabs used for indentation.
279  asm = string.expandtabs(asm, 2)
280  # Strip trailing whitespace.
281  asm = common.SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
282  return asm
283
284def get_triple_from_march(march):
285  triples = {
286      'amdgcn': 'amdgcn',
287      'r600': 'r600',
288      'mips': 'mips',
289      'sparc': 'sparc',
290      'hexagon': 'hexagon',
291  }
292  for prefix, triple in triples.items():
293    if march.startswith(prefix):
294      return triple
295  print("Cannot find a triple. Assume 'x86'", file=sys.stderr)
296  return 'x86'
297
298def build_function_body_dictionary_for_triple(args, raw_tool_output, triple, prefixes, func_dict):
299  target_handlers = {
300      'i686': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
301      'x86': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
302      'i386': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
303      'aarch64': (scrub_asm_arm_eabi, ASM_FUNCTION_AARCH64_RE),
304      'aarch64-apple-darwin': (scrub_asm_arm_eabi, ASM_FUNCTION_AARCH64_DARWIN_RE),
305      'hexagon': (scrub_asm_hexagon, ASM_FUNCTION_HEXAGON_RE),
306      'r600': (scrub_asm_amdgpu, ASM_FUNCTION_AMDGPU_RE),
307      'amdgcn': (scrub_asm_amdgpu, ASM_FUNCTION_AMDGPU_RE),
308      'arm': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_RE),
309      'arm64': (scrub_asm_arm_eabi, ASM_FUNCTION_AARCH64_RE),
310      'arm64-apple-ios': (scrub_asm_arm_eabi, ASM_FUNCTION_AARCH64_DARWIN_RE),
311      'armv7-apple-ios' : (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_IOS_RE),
312      'armv7-apple-darwin': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_DARWIN_RE),
313      'thumb': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_RE),
314      'thumb-macho': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_MACHO_RE),
315      'thumbv5-macho': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_MACHO_RE),
316      'thumbv7-apple-ios' : (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_IOS_RE),
317      'mips': (scrub_asm_mips, ASM_FUNCTION_MIPS_RE),
318      'ppc32': (scrub_asm_powerpc, ASM_FUNCTION_PPC_RE),
319      'powerpc': (scrub_asm_powerpc, ASM_FUNCTION_PPC_RE),
320      'riscv32': (scrub_asm_riscv, ASM_FUNCTION_RISCV_RE),
321      'riscv64': (scrub_asm_riscv, ASM_FUNCTION_RISCV_RE),
322      'lanai': (scrub_asm_lanai, ASM_FUNCTION_LANAI_RE),
323      'sparc': (scrub_asm_sparc, ASM_FUNCTION_SPARC_RE),
324      's390x': (scrub_asm_systemz, ASM_FUNCTION_SYSTEMZ_RE),
325      'wasm32': (scrub_asm_wasm32, ASM_FUNCTION_WASM32_RE),
326  }
327  handler = None
328  best_prefix = ''
329  for prefix, s in target_handlers.items():
330    if triple.startswith(prefix) and len(prefix) > len(best_prefix):
331      handler = s
332      best_prefix = prefix
333
334  if handler is None:
335    raise KeyError('Triple %r is not supported' % (triple))
336
337  scrubber, function_re = handler
338  common.build_function_body_dictionary(
339          function_re, scrubber, [args], raw_tool_output, prefixes,
340          func_dict, args.verbose)
341
342##### Generator of assembly CHECK lines
343
344def add_asm_checks(output_lines, comment_marker, prefix_list, func_dict, func_name):
345  # Label format is based on ASM string.
346  check_label_format = '{} %s-LABEL: %s:'.format(comment_marker)
347  common.add_checks(output_lines, comment_marker, prefix_list, func_dict, func_name, check_label_format, True, False)
348