Here [0-9] will match any number, like local_unnamed_addr # 0, local_unnamed_addr # 1 ... local_unnamed_addr # 9 will all be replaced (here deleted actually)
:%s/local_unnamed_addr #[0-9]//gc
Monday, 13 March 2017
LLVM test case update python script
Need to revisit this to understand it completely.
#!/usr/bin/env python2.7
"""A test case update script.
This script is a utility to update LLVM X86 'llc' based test cases with new
FileCheck patterns. It can either update all of the tests in the file or
a single test function.
"""
import argparse
import os # Used to advertise this file's name ("autogenerated_note").
import string
import subprocess
import sys
import re
# Invoke the tool that is being tested.
def llc(args, cmd_args, ir):
with open(ir) as ir_file:
stdout = subprocess.check_output(args.llc_binary + ' ' + cmd_args,
shell=True, stdin=ir_file)
# Fix line endings to unix CR style.
stdout = stdout.replace('\r\n', '\n')
return stdout
# RegEx: this is where the magic happens.
SCRUB_WHITESPACE_RE = re.compile(r'(?!^(| \w))[ \t]+', flags=re.M)
SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M)
SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n')
ASM_FUNCTION_X86_RE = re.compile(
r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n[^:]*?'
r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*'
r'^\s*(?:[^:\n]+?:\s*\n\s*\.size|\.cfi_endproc|\.globl|\.comm|\.(?:sub)?section)',
flags=(re.M | re.S))
SCRUB_X86_SHUFFLES_RE = (
re.compile(
r'^(\s*\w+) [^#\n]+#+ ((?:[xyz]mm\d+|mem)( \{%k\d+\}( \{z\})?)? = .*)$',
flags=re.M))
SCRUB_X86_SP_RE = re.compile(r'\d+\(%(esp|rsp)\)')
SCRUB_X86_RIP_RE = re.compile(r'[.\w]+\(%rip\)')
SCRUB_X86_LCP_RE = re.compile(r'\.LCPI[0-9]+_[0-9]+')
ASM_FUNCTION_ARM_RE = re.compile(
r'^(?P<func>[0-9a-zA-Z_]+):\n' # f: (name of function)
r'\s+\.fnstart\n' # .fnstart
r'(?P<body>.*?)\n' # (body of the function)
r'.Lfunc_end[0-9]+:\n', # .Lfunc_end0:
flags=(re.M | re.S))
RUN_LINE_RE = re.compile('^\s*;\s*RUN:\s*(.*)$')
TRIPLE_ARG_RE = re.compile(r'-mtriple=([^ ]+)')
TRIPLE_IR_RE = re.compile(r'^target\s+triple\s*=\s*"([^"]+)"$')
IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(')
CHECK_PREFIX_RE = re.compile('--?check-prefix(?:es)?=(\S+)')
CHECK_RE = re.compile(r'^\s*;\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:')
ASM_FUNCTION_PPC_RE = re.compile(
r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n'
r'\.Lfunc_begin[0-9]+:\n'
r'[ \t]+.cfi_startproc\n'
r'(?:\.Lfunc_[gl]ep[0-9]+:\n(?:[ \t]+.*?\n)*)*'
r'(?P<body>.*?)\n'
# This list is incomplete
r'(?:^[ \t]*(?:\.long[ \t]+[^\n]+|\.quad[ \t]+[^\n]+)\n)*'
r'.Lfunc_end[0-9]+:\n',
flags=(re.M | re.S))
def scrub_asm_x86(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Detect shuffle asm comments and hide the operands in favor of the comments.
asm = SCRUB_X86_SHUFFLES_RE.sub(r'\1 {{.*#+}} \2', asm)
# Generically match the stack offset of a memory operand.
asm = SCRUB_X86_SP_RE.sub(r'{{[0-9]+}}(%\1)', asm)
# Generically match a RIP-relative memory operand.
asm = SCRUB_X86_RIP_RE.sub(r'{{.*}}(%rip)', asm)
# Generically match a LCP symbol.
asm = SCRUB_X86_LCP_RE.sub(r'{{\.LCPI.*}}', asm)
# Strip kill operands inserted into the asm.
asm = SCRUB_KILL_COMMENT_RE.sub('', asm)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
def scrub_asm_arm_eabi(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Strip kill operands inserted into the asm.
asm = SCRUB_KILL_COMMENT_RE.sub('', asm)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
def scrub_asm_powerpc64le(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
# Build up a dictionary of all the function bodies.
def build_function_body_dictionary(raw_tool_output, triple, prefixes, func_dict,
verbose):
target_handlers = {
'x86_64': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'i686': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'x86': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'i386': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'arm-eabi': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_RE),
'powerpc64le': (scrub_asm_powerpc64le, ASM_FUNCTION_PPC_RE),
}
handlers = None
for prefix, s in target_handlers.items():
if triple.startswith(prefix):
handlers = s
break
else:
raise KeyError('Triple %r is not supported' % (triple))
scrubber, function_re = handlers
for m in function_re.finditer(raw_tool_output):
if not m:
continue
func = m.group('func')
scrubbed_body = scrubber(m.group('body'))
if func.startswith('stress'):
# We only use the last line of the function body for stress tests.
scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
if verbose:
print >>sys.stderr, 'Processing function: ' + func
for l in scrubbed_body.splitlines():
print >>sys.stderr, ' ' + l
for prefix in prefixes:
if func in func_dict[prefix] and func_dict[prefix][func] != scrubbed_body:
if prefix == prefixes[-1]:
print >>sys.stderr, ('WARNING: Found conflicting asm under the '
'same prefix: %r!' % (prefix,))
else:
func_dict[prefix][func] = None
continue
func_dict[prefix][func] = scrubbed_body
def add_checks(output_lines, run_list, func_dict, func_name):
printed_prefixes = []
for p in run_list:
checkprefixes = p[0]
for checkprefix in checkprefixes:
if checkprefix in printed_prefixes:
break
if not func_dict[checkprefix][func_name]:
continue
# Add some space between different check prefixes.
if len(printed_prefixes) != 0:
output_lines.append(';')
printed_prefixes.append(checkprefix)
output_lines.append('; %s-LABEL: %s:' % (checkprefix, func_name))
func_body = func_dict[checkprefix][func_name].splitlines()
output_lines.append('; %s: %s' % (checkprefix, func_body[0]))
for func_line in func_body[1:]:
output_lines.append('; %s-NEXT: %s' % (checkprefix, func_line))
# Add space between different check prefixes and the first line of code.
# output_lines.append(';')
break
return output_lines
def should_add_line_to_output(input_line, prefix_set):
# Skip any blank comment lines in the IR.
if input_line.strip() == ';':
return False
# Skip any blank lines in the IR.
#if input_line.strip() == '':
# return False
# And skip any CHECK lines. We're building our own.
m = CHECK_RE.match(input_line)
if m and m.group(1) in prefix_set:
return False
return True
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('-v', '--verbose', action='store_true',
help='Show verbose output')
parser.add_argument('--llc-binary', default='llc',
help='The "llc" binary to use to generate the test case')
parser.add_argument(
'--function', help='The function in the test file to update')
parser.add_argument('tests', nargs='+')
args = parser.parse_args()
autogenerated_note = ('; NOTE: Assertions have been autogenerated by '
'utils/' + os.path.basename(__file__))
for test in args.tests:
if args.verbose:
print >>sys.stderr, 'Scanning for RUN lines in test file: %s' % (test,)
with open(test) as f:
input_lines = [l.rstrip() for l in f]
triple_in_ir = None
for l in input_lines:
m = TRIPLE_IR_RE.match(l)
if m:
triple_in_ir = m.groups()[0]
break
raw_lines = [m.group(1)
for m in [RUN_LINE_RE.match(l) for l in input_lines] if m]
run_lines = [raw_lines[0]] if len(raw_lines) > 0 else []
for l in raw_lines[1:]:
if run_lines[-1].endswith("\\"):
run_lines[-1] = run_lines[-1].rstrip("\\") + " " + l
else:
run_lines.append(l)
if args.verbose:
print >>sys.stderr, 'Found %d RUN lines:' % (len(run_lines),)
for l in run_lines:
print >>sys.stderr, ' RUN: ' + l
run_list = []
for l in run_lines:
commands = [cmd.strip() for cmd in l.split('|', 1)]
llc_cmd = commands[0]
triple_in_cmd = None
m = TRIPLE_ARG_RE.search(llc_cmd)
if m:
triple_in_cmd = m.groups()[0]
filecheck_cmd = ''
if len(commands) > 1:
filecheck_cmd = commands[1]
if not llc_cmd.startswith('llc '):
print >>sys.stderr, 'WARNING: Skipping non-llc RUN line: ' + l
continue
if not filecheck_cmd.startswith('FileCheck '):
print >>sys.stderr, 'WARNING: Skipping non-FileChecked RUN line: ' + l
continue
llc_cmd_args = llc_cmd[len('llc'):].strip()
llc_cmd_args = llc_cmd_args.replace('< %s', '').replace('%s', '').strip()
check_prefixes = [item for m in CHECK_PREFIX_RE.finditer(filecheck_cmd)
for item in m.group(1).split(',')]
if not check_prefixes:
check_prefixes = ['CHECK']
# FIXME: We should use multiple check prefixes to common check lines. For
# now, we just ignore all but the last.
run_list.append((check_prefixes, llc_cmd_args, triple_in_cmd))
func_dict = {}
for p in run_list:
prefixes = p[0]
for prefix in prefixes:
func_dict.update({prefix: dict()})
for prefixes, llc_args, triple_in_cmd in run_list:
if args.verbose:
print >>sys.stderr, 'Extracted LLC cmd: llc ' + llc_args
print >>sys.stderr, 'Extracted FileCheck prefixes: ' + str(prefixes)
raw_tool_output = llc(args, llc_args, test)
if not (triple_in_cmd or triple_in_ir):
print >>sys.stderr, "Cannot find a triple. Assume 'x86'"
build_function_body_dictionary(raw_tool_output,
triple_in_cmd or triple_in_ir or 'x86', prefixes, func_dict, args.verbose)
is_in_function = False
is_in_function_start = False
func_name = None
prefix_set = set([prefix for p in run_list for prefix in p[0]])
if args.verbose:
print >>sys.stderr, 'Rewriting FileCheck prefixes: %s' % (prefix_set,)
output_lines = []
output_lines.append(autogenerated_note)
for input_line in input_lines:
if is_in_function_start:
if input_line == '':
continue
if input_line.lstrip().startswith(';'):
m = CHECK_RE.match(input_line)
if not m or m.group(1) not in prefix_set:
output_lines.append(input_line)
continue
# Print out the various check lines here.
output_lines = add_checks(output_lines, run_list, func_dict, func_name)
is_in_function_start = False
if is_in_function:
if should_add_line_to_output(input_line, prefix_set) == True:
# This input line of the function body will go as-is into the output.
output_lines.append(input_line)
else:
continue
if input_line.strip() == '}':
is_in_function = False
continue
if input_line == autogenerated_note:
continue
# If it's outside a function, it just gets copied to the output.
output_lines.append(input_line)
m = IR_FUNCTION_RE.match(input_line)
if not m:
continue
func_name = m.group(1)
if args.function is not None and func_name != args.function:
# When filtering on a specific function, skip all others.
continue
is_in_function = is_in_function_start = True
if args.verbose:
print>>sys.stderr, 'Writing %d lines to %s...' % (len(output_lines), test)
with open(test, 'wb') as f:
f.writelines([l + '\n' for l in output_lines])
if __name__ == '__main__':
main()
#!/usr/bin/env python2.7
"""A test case update script.
This script is a utility to update LLVM X86 'llc' based test cases with new
FileCheck patterns. It can either update all of the tests in the file or
a single test function.
"""
import argparse
import os # Used to advertise this file's name ("autogenerated_note").
import string
import subprocess
import sys
import re
# Invoke the tool that is being tested.
def llc(args, cmd_args, ir):
with open(ir) as ir_file:
stdout = subprocess.check_output(args.llc_binary + ' ' + cmd_args,
shell=True, stdin=ir_file)
# Fix line endings to unix CR style.
stdout = stdout.replace('\r\n', '\n')
return stdout
# RegEx: this is where the magic happens.
SCRUB_WHITESPACE_RE = re.compile(r'(?!^(| \w))[ \t]+', flags=re.M)
SCRUB_TRAILING_WHITESPACE_RE = re.compile(r'[ \t]+$', flags=re.M)
SCRUB_KILL_COMMENT_RE = re.compile(r'^ *#+ +kill:.*\n')
ASM_FUNCTION_X86_RE = re.compile(
r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n[^:]*?'
r'(?P<body>^##?[ \t]+[^:]+:.*?)\s*'
r'^\s*(?:[^:\n]+?:\s*\n\s*\.size|\.cfi_endproc|\.globl|\.comm|\.(?:sub)?section)',
flags=(re.M | re.S))
SCRUB_X86_SHUFFLES_RE = (
re.compile(
r'^(\s*\w+) [^#\n]+#+ ((?:[xyz]mm\d+|mem)( \{%k\d+\}( \{z\})?)? = .*)$',
flags=re.M))
SCRUB_X86_SP_RE = re.compile(r'\d+\(%(esp|rsp)\)')
SCRUB_X86_RIP_RE = re.compile(r'[.\w]+\(%rip\)')
SCRUB_X86_LCP_RE = re.compile(r'\.LCPI[0-9]+_[0-9]+')
ASM_FUNCTION_ARM_RE = re.compile(
r'^(?P<func>[0-9a-zA-Z_]+):\n' # f: (name of function)
r'\s+\.fnstart\n' # .fnstart
r'(?P<body>.*?)\n' # (body of the function)
r'.Lfunc_end[0-9]+:\n', # .Lfunc_end0:
flags=(re.M | re.S))
RUN_LINE_RE = re.compile('^\s*;\s*RUN:\s*(.*)$')
TRIPLE_ARG_RE = re.compile(r'-mtriple=([^ ]+)')
TRIPLE_IR_RE = re.compile(r'^target\s+triple\s*=\s*"([^"]+)"$')
IR_FUNCTION_RE = re.compile('^\s*define\s+(?:internal\s+)?[^@]*@(\w+)\s*\(')
CHECK_PREFIX_RE = re.compile('--?check-prefix(?:es)?=(\S+)')
CHECK_RE = re.compile(r'^\s*;\s*([^:]+?)(?:-NEXT|-NOT|-DAG|-LABEL)?:')
ASM_FUNCTION_PPC_RE = re.compile(
r'^_?(?P<func>[^:]+):[ \t]*#+[ \t]*@(?P=func)\n'
r'\.Lfunc_begin[0-9]+:\n'
r'[ \t]+.cfi_startproc\n'
r'(?:\.Lfunc_[gl]ep[0-9]+:\n(?:[ \t]+.*?\n)*)*'
r'(?P<body>.*?)\n'
# This list is incomplete
r'(?:^[ \t]*(?:\.long[ \t]+[^\n]+|\.quad[ \t]+[^\n]+)\n)*'
r'.Lfunc_end[0-9]+:\n',
flags=(re.M | re.S))
def scrub_asm_x86(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Detect shuffle asm comments and hide the operands in favor of the comments.
asm = SCRUB_X86_SHUFFLES_RE.sub(r'\1 {{.*#+}} \2', asm)
# Generically match the stack offset of a memory operand.
asm = SCRUB_X86_SP_RE.sub(r'{{[0-9]+}}(%\1)', asm)
# Generically match a RIP-relative memory operand.
asm = SCRUB_X86_RIP_RE.sub(r'{{.*}}(%rip)', asm)
# Generically match a LCP symbol.
asm = SCRUB_X86_LCP_RE.sub(r'{{\.LCPI.*}}', asm)
# Strip kill operands inserted into the asm.
asm = SCRUB_KILL_COMMENT_RE.sub('', asm)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
def scrub_asm_arm_eabi(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Strip kill operands inserted into the asm.
asm = SCRUB_KILL_COMMENT_RE.sub('', asm)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
def scrub_asm_powerpc64le(asm):
# Scrub runs of whitespace out of the assembly, but leave the leading
# whitespace in place.
asm = SCRUB_WHITESPACE_RE.sub(r' ', asm)
# Expand the tabs used for indentation.
asm = string.expandtabs(asm, 2)
# Strip trailing whitespace.
asm = SCRUB_TRAILING_WHITESPACE_RE.sub(r'', asm)
return asm
# Build up a dictionary of all the function bodies.
def build_function_body_dictionary(raw_tool_output, triple, prefixes, func_dict,
verbose):
target_handlers = {
'x86_64': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'i686': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'x86': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'i386': (scrub_asm_x86, ASM_FUNCTION_X86_RE),
'arm-eabi': (scrub_asm_arm_eabi, ASM_FUNCTION_ARM_RE),
'powerpc64le': (scrub_asm_powerpc64le, ASM_FUNCTION_PPC_RE),
}
handlers = None
for prefix, s in target_handlers.items():
if triple.startswith(prefix):
handlers = s
break
else:
raise KeyError('Triple %r is not supported' % (triple))
scrubber, function_re = handlers
for m in function_re.finditer(raw_tool_output):
if not m:
continue
func = m.group('func')
scrubbed_body = scrubber(m.group('body'))
if func.startswith('stress'):
# We only use the last line of the function body for stress tests.
scrubbed_body = '\n'.join(scrubbed_body.splitlines()[-1:])
if verbose:
print >>sys.stderr, 'Processing function: ' + func
for l in scrubbed_body.splitlines():
print >>sys.stderr, ' ' + l
for prefix in prefixes:
if func in func_dict[prefix] and func_dict[prefix][func] != scrubbed_body:
if prefix == prefixes[-1]:
print >>sys.stderr, ('WARNING: Found conflicting asm under the '
'same prefix: %r!' % (prefix,))
else:
func_dict[prefix][func] = None
continue
func_dict[prefix][func] = scrubbed_body
def add_checks(output_lines, run_list, func_dict, func_name):
printed_prefixes = []
for p in run_list:
checkprefixes = p[0]
for checkprefix in checkprefixes:
if checkprefix in printed_prefixes:
break
if not func_dict[checkprefix][func_name]:
continue
# Add some space between different check prefixes.
if len(printed_prefixes) != 0:
output_lines.append(';')
printed_prefixes.append(checkprefix)
output_lines.append('; %s-LABEL: %s:' % (checkprefix, func_name))
func_body = func_dict[checkprefix][func_name].splitlines()
output_lines.append('; %s: %s' % (checkprefix, func_body[0]))
for func_line in func_body[1:]:
output_lines.append('; %s-NEXT: %s' % (checkprefix, func_line))
# Add space between different check prefixes and the first line of code.
# output_lines.append(';')
break
return output_lines
def should_add_line_to_output(input_line, prefix_set):
# Skip any blank comment lines in the IR.
if input_line.strip() == ';':
return False
# Skip any blank lines in the IR.
#if input_line.strip() == '':
# return False
# And skip any CHECK lines. We're building our own.
m = CHECK_RE.match(input_line)
if m and m.group(1) in prefix_set:
return False
return True
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('-v', '--verbose', action='store_true',
help='Show verbose output')
parser.add_argument('--llc-binary', default='llc',
help='The "llc" binary to use to generate the test case')
parser.add_argument(
'--function', help='The function in the test file to update')
parser.add_argument('tests', nargs='+')
args = parser.parse_args()
autogenerated_note = ('; NOTE: Assertions have been autogenerated by '
'utils/' + os.path.basename(__file__))
for test in args.tests:
if args.verbose:
print >>sys.stderr, 'Scanning for RUN lines in test file: %s' % (test,)
with open(test) as f:
input_lines = [l.rstrip() for l in f]
triple_in_ir = None
for l in input_lines:
m = TRIPLE_IR_RE.match(l)
if m:
triple_in_ir = m.groups()[0]
break
raw_lines = [m.group(1)
for m in [RUN_LINE_RE.match(l) for l in input_lines] if m]
run_lines = [raw_lines[0]] if len(raw_lines) > 0 else []
for l in raw_lines[1:]:
if run_lines[-1].endswith("\\"):
run_lines[-1] = run_lines[-1].rstrip("\\") + " " + l
else:
run_lines.append(l)
if args.verbose:
print >>sys.stderr, 'Found %d RUN lines:' % (len(run_lines),)
for l in run_lines:
print >>sys.stderr, ' RUN: ' + l
run_list = []
for l in run_lines:
commands = [cmd.strip() for cmd in l.split('|', 1)]
llc_cmd = commands[0]
triple_in_cmd = None
m = TRIPLE_ARG_RE.search(llc_cmd)
if m:
triple_in_cmd = m.groups()[0]
filecheck_cmd = ''
if len(commands) > 1:
filecheck_cmd = commands[1]
if not llc_cmd.startswith('llc '):
print >>sys.stderr, 'WARNING: Skipping non-llc RUN line: ' + l
continue
if not filecheck_cmd.startswith('FileCheck '):
print >>sys.stderr, 'WARNING: Skipping non-FileChecked RUN line: ' + l
continue
llc_cmd_args = llc_cmd[len('llc'):].strip()
llc_cmd_args = llc_cmd_args.replace('< %s', '').replace('%s', '').strip()
check_prefixes = [item for m in CHECK_PREFIX_RE.finditer(filecheck_cmd)
for item in m.group(1).split(',')]
if not check_prefixes:
check_prefixes = ['CHECK']
# FIXME: We should use multiple check prefixes to common check lines. For
# now, we just ignore all but the last.
run_list.append((check_prefixes, llc_cmd_args, triple_in_cmd))
func_dict = {}
for p in run_list:
prefixes = p[0]
for prefix in prefixes:
func_dict.update({prefix: dict()})
for prefixes, llc_args, triple_in_cmd in run_list:
if args.verbose:
print >>sys.stderr, 'Extracted LLC cmd: llc ' + llc_args
print >>sys.stderr, 'Extracted FileCheck prefixes: ' + str(prefixes)
raw_tool_output = llc(args, llc_args, test)
if not (triple_in_cmd or triple_in_ir):
print >>sys.stderr, "Cannot find a triple. Assume 'x86'"
build_function_body_dictionary(raw_tool_output,
triple_in_cmd or triple_in_ir or 'x86', prefixes, func_dict, args.verbose)
is_in_function = False
is_in_function_start = False
func_name = None
prefix_set = set([prefix for p in run_list for prefix in p[0]])
if args.verbose:
print >>sys.stderr, 'Rewriting FileCheck prefixes: %s' % (prefix_set,)
output_lines = []
output_lines.append(autogenerated_note)
for input_line in input_lines:
if is_in_function_start:
if input_line == '':
continue
if input_line.lstrip().startswith(';'):
m = CHECK_RE.match(input_line)
if not m or m.group(1) not in prefix_set:
output_lines.append(input_line)
continue
# Print out the various check lines here.
output_lines = add_checks(output_lines, run_list, func_dict, func_name)
is_in_function_start = False
if is_in_function:
if should_add_line_to_output(input_line, prefix_set) == True:
# This input line of the function body will go as-is into the output.
output_lines.append(input_line)
else:
continue
if input_line.strip() == '}':
is_in_function = False
continue
if input_line == autogenerated_note:
continue
# If it's outside a function, it just gets copied to the output.
output_lines.append(input_line)
m = IR_FUNCTION_RE.match(input_line)
if not m:
continue
func_name = m.group(1)
if args.function is not None and func_name != args.function:
# When filtering on a specific function, skip all others.
continue
is_in_function = is_in_function_start = True
if args.verbose:
print>>sys.stderr, 'Writing %d lines to %s...' % (len(output_lines), test)
with open(test, 'wb') as f:
f.writelines([l + '\n' for l in output_lines])
if __name__ == '__main__':
main()
Sunday, 12 March 2017
ORC/orc (OR with Complement)
OR with Complement X-form
orc RA,RS,RB (Rc=0)
orc. RA,RS,RB (Rc=1)
RA <= (RS) | ¬(RB)
The contents of register RS are ORed with the complement
of the contents of register RB and the result is
placed into register RA.
Special Registers Altered:
CR0 (if Rc=1)
orc RA,RS,RB (Rc=0)
orc. RA,RS,RB (Rc=1)
RA <= (RS) | ¬(RB)
The contents of register RS are ORed with the complement
of the contents of register RB and the result is
placed into register RA.
Special Registers Altered:
CR0 (if Rc=1)
Saturday, 11 March 2017
NOT Rx, ry <=> nor Rx,Ry,Ry
NOR X-form
nor RA,RS,RB (Rc=0)
nor. RA,RS,RB (Rc=1)
RA <= ¬((RS) | (RB))
The contents of register RS are ORed with the contents
of register RB and the complemented result is placed
into register RA.
Special Registers Altered:
CR0 (if Rc=1)
Extended Mnemonics:
Example of extended mnemonics for NOR:
Extended: Equivalent to:
not Rx,Ry nor Rx,Ry,Ry
nor RA,RS,RB (Rc=0)
nor. RA,RS,RB (Rc=1)
RA <= ¬((RS) | (RB))
The contents of register RS are ORed with the contents
of register RB and the complemented result is placed
into register RA.
Special Registers Altered:
CR0 (if Rc=1)
Extended Mnemonics:
Example of extended mnemonics for NOR:
Extended: Equivalent to:
not Rx,Ry nor Rx,Ry,Ry
Friday, 10 March 2017
Spec Reducer (sreducer)
66. How to reduce a spec test case?
Let us suppose there is a failure in h264ref benchmark.
Step1: create a build, link and run script
basically you need to create a script that just lists the compile commands,
the link command and the run command:
For example: I have a script named /home/jtony/tools/spec_reducer/buildLinkRun.sh
The build and link commands can be found by running the build.dry.sh script, this should
be run in your spec directory (like /home/jtony/scrum/s2/114/spec/cpu2006),
1(a): Here is my build command (part example):
cd /home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/build/build_base_none.0000
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o annexb.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char
annexb.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o biariencode.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char
biariencode.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o block.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char b
lock.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o cabac.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char c
abac.c &
...
1(b): Here is my link command:
/home/jtony/git-llvm/build/team-llvm//bin/clang -O2 -DSPEC_CPU_LINUX -fsigned-char annexb.o biariencode.o block.o cabac.o configfile.o context_ini.o decoder.o explicit_gop.o fast_me.o filehandle.o fmo.o header.o image.o intrarefresh.o leaky_bucket.o lencod.o loopFilter.o macroblock.o mb_access.o mbuffer.o memalloc.o mv-search.o nal.o nalu.o nalucommon.o output.o parset.o parsetcommon.o q_matrix.o q_offsets.o ratectl.o rdopt.o rdopt_coding_state.o rdpicdecision.o refbuf.o rtp.o sei.o slice.o transform8x8.o vlc.o weighted_prediction.o specrand.o -lm -m64 -Wl,-q -Wl,-rpath=/home/jtony/git-llvm/build/team-llvm//lib64 -o h264ref
1(c): the run command is a little bit harder to find, but you should be able to piece together the run
command from the speccmds.cmd file eg:
/home/jtony/perf-runs/temp/benchspec/CPU2006/464.h264ref/run/run_base_test_none.0000/speccmds.cmd
Here is the run command for my case:
cd /home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/run/run_base_test_none.0000
/home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/build/build_base_none.0000/h264ref -d foreman_test_encoder_baseline.cfg > tony.out
specperl /home/jtony/scrum/s2/114/spec/cpu2006/bin/specdiff -m -l 10 --cw --floatcompare /home/jtony/scrum/s2/114/spec/cpu2006/benchspec/CPU2006/464.h264ref/data/test/output/foreman_test_baseline_encodelog.out > tony.out
Step2, you need a good (no bug) and a bad (contains the bug) compiler. In my case, I have
The bad one /home/jtony/git-llvm/build/team-llvm/bin/clang++
The good one /home/jtony/git-llvm/build/good/bin/clang++
run this:
produceBinReduceScript.sh good_compiler bad_compiler, like:
/home/nemanjai/llvm/Git/llvm-on-power/utils/produceBinReduceScript.sh tony.sh /home/jtony/git-llvm/build/good/bin/clang /home/jtony/git-llvm/build/team-llvm/bin/clang
that produced `ReduceScript.sh`
Step3, once you have the ReduceScript, you can use binary search to find which file is miscompiled. In my case I have 43, you can do the following:
like
./ReduceScript.sh 1 43
./ReduceScript.sh 1 22
./ReduceScript.sh 11 22
./ReduceScript.sh 15 22
./ReduceScript.sh 18 22
./ReduceScript.sh 18 20
./ReduceScript.sh 18 19
./ReduceScript.sh 19 19
Step 4:
go into PPCISelDAGToDAG.cpp
and add:
`"#include "EnvDecider.hpp"`
and add:
#if 1
static EnvDecider RunMe("INSTANCE");
if (!(++RunMe))
return false;
#endif
Step 5:
then you can invoke the script like this (once you know how many there are):
`INSTANCE=1 ./ReduceScript.sh`
and so on and so forth
Let us suppose there is a failure in h264ref benchmark.
Step1: create a build, link and run script
basically you need to create a script that just lists the compile commands,
the link command and the run command:
For example: I have a script named /home/jtony/tools/spec_reducer/buildLinkRun.sh
The build and link commands can be found by running the build.dry.sh script, this should
be run in your spec directory (like /home/jtony/scrum/s2/114/spec/cpu2006),
1(a): Here is my build command (part example):
cd /home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/build/build_base_none.0000
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o annexb.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char
annexb.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o biariencode.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char
biariencode.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o block.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char b
lock.c &
/home/jtony/git-llvm/build/team-llvm//bin/clang -c -o cabac.o -DSPEC_CPU -DNDEBUG -O2 -DSPEC_CPU_LINUX -fsigned-char c
abac.c &
...
1(b): Here is my link command:
/home/jtony/git-llvm/build/team-llvm//bin/clang -O2 -DSPEC_CPU_LINUX -fsigned-char annexb.o biariencode.o block.o cabac.o configfile.o context_ini.o decoder.o explicit_gop.o fast_me.o filehandle.o fmo.o header.o image.o intrarefresh.o leaky_bucket.o lencod.o loopFilter.o macroblock.o mb_access.o mbuffer.o memalloc.o mv-search.o nal.o nalu.o nalucommon.o output.o parset.o parsetcommon.o q_matrix.o q_offsets.o ratectl.o rdopt.o rdopt_coding_state.o rdpicdecision.o refbuf.o rtp.o sei.o slice.o transform8x8.o vlc.o weighted_prediction.o specrand.o -lm -m64 -Wl,-q -Wl,-rpath=/home/jtony/git-llvm/build/team-llvm//lib64 -o h264ref
1(c): the run command is a little bit harder to find, but you should be able to piece together the run
command from the speccmds.cmd file eg:
/home/jtony/perf-runs/temp/benchspec/CPU2006/464.h264ref/run/run_base_test_none.0000/speccmds.cmd
Here is the run command for my case:
cd /home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/run/run_base_test_none.0000
/home/jtony/perf-runs/benchspec/CPU2006/464.h264ref/build/build_base_none.0000/h264ref -d foreman_test_encoder_baseline.cfg > tony.out
specperl /home/jtony/scrum/s2/114/spec/cpu2006/bin/specdiff -m -l 10 --cw --floatcompare /home/jtony/scrum/s2/114/spec/cpu2006/benchspec/CPU2006/464.h264ref/data/test/output/foreman_test_baseline_encodelog.out > tony.out
Step2, you need a good (no bug) and a bad (contains the bug) compiler. In my case, I have
The bad one /home/jtony/git-llvm/build/team-llvm/bin/clang++
The good one /home/jtony/git-llvm/build/good/bin/clang++
run this:
produceBinReduceScript.sh good_compiler bad_compiler, like:
/home/nemanjai/llvm/Git/llvm-on-power/utils/produceBinReduceScript.sh tony.sh /home/jtony/git-llvm/build/good/bin/clang /home/jtony/git-llvm/build/team-llvm/bin/clang
that produced `ReduceScript.sh`
Step3, once you have the ReduceScript, you can use binary search to find which file is miscompiled. In my case I have 43, you can do the following:
like
./ReduceScript.sh 1 43
./ReduceScript.sh 1 22
./ReduceScript.sh 11 22
./ReduceScript.sh 15 22
./ReduceScript.sh 18 22
./ReduceScript.sh 18 20
./ReduceScript.sh 18 19
./ReduceScript.sh 19 19
Step 4:
go into PPCISelDAGToDAG.cpp
and add:
`"#include "EnvDecider.hpp"`
and add:
#if 1
static EnvDecider RunMe("INSTANCE");
if (!(++RunMe))
return false;
#endif
Step 5:
then you can invoke the script like this (once you know how many there are):
`INSTANCE=1 ./ReduceScript.sh`
and so on and so forth
Thursday, 9 March 2017
findbadobjectfile script (learning)
#!/usr/bin/perl
@PURPOSE="Find bad compilation unit";
@SYNTAX="findbadobjectfile";
@NOTE="Uses scripts compileme, link.sh, and executeme. Also uses file object_names.";
# To use this script, create two directories: good_objects and bad_objects.
# Build the testcase twice. First build with good driver and move all the object files to good_objects directory.
# Build again with bad driver, and move all the object files to bad_objects directory.
# Create a script called link.sh which has the original link command. Remove all the object files from the link command.
# Create a file called object_names which lists all the object files, one per line.
# Create a script called executeme which runs the testcase and returns 0 for success.
sub link_program
{
print "***Running link cmd\n";
system("./temp_link.sh");
}
sub run_program
{
print "****Running program\n";
system("./executeme");
}
@obj_array = `cat object_names`;
use integer;
$lower = 0;
$size = @obj_array;
$upper = $size;
$lastgood = $size+1;
open (my $input, '<', 'link.sh');
$link = <$input>;
chomp ($link);
$j=0;
while( $lower < $upper-1 ) {
$link_cmd = $link;
my $link_file = "temp_link.sh";
open (my $fh, '>',$link_file) or die "Could not open $link_file";
system ("chmod +x temp_link.sh");
$middle = ($lower + $upper) / 2;
if($lower > 0){
for($i=0;$i<$lower;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
}
for($i=$lower;$i<$middle;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
for($i=$middle;$i<$upper;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " bad_objects/".@obj_array[$i];
}
if($upper < $size){
for($i=$upper;$i<$size;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
}
$link_cmd = $link_cmd."\n";
# print $link_cmd;
print $fh $link_cmd;
close $fh;
link_program;
run_program;
if($?==0){
print "Passed round $j\n";
$upper = $middle;
}else{
print "Failed round $j\n";
$lower = $middle;
}
$j=$j+1;
system("rm temp_link.sh");
}
print "Bad compilation unit: @obj_array[$lower]\n";
@PURPOSE="Find bad compilation unit";
@SYNTAX="findbadobjectfile";
@NOTE="Uses scripts compileme, link.sh, and executeme. Also uses file object_names.";
# To use this script, create two directories: good_objects and bad_objects.
# Build the testcase twice. First build with good driver and move all the object files to good_objects directory.
# Build again with bad driver, and move all the object files to bad_objects directory.
# Create a script called link.sh which has the original link command. Remove all the object files from the link command.
# Create a file called object_names which lists all the object files, one per line.
# Create a script called executeme which runs the testcase and returns 0 for success.
sub link_program
{
print "***Running link cmd\n";
system("./temp_link.sh");
}
sub run_program
{
print "****Running program\n";
system("./executeme");
}
@obj_array = `cat object_names`;
use integer;
$lower = 0;
$size = @obj_array;
$upper = $size;
$lastgood = $size+1;
open (my $input, '<', 'link.sh');
$link = <$input>;
chomp ($link);
$j=0;
while( $lower < $upper-1 ) {
$link_cmd = $link;
my $link_file = "temp_link.sh";
open (my $fh, '>',$link_file) or die "Could not open $link_file";
system ("chmod +x temp_link.sh");
$middle = ($lower + $upper) / 2;
if($lower > 0){
for($i=0;$i<$lower;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
}
for($i=$lower;$i<$middle;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
for($i=$middle;$i<$upper;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " bad_objects/".@obj_array[$i];
}
if($upper < $size){
for($i=$upper;$i<$size;$i++){
chomp(@obj_array[$i]);
$link_cmd = $link_cmd . " good_objects/".@obj_array[$i];
}
}
$link_cmd = $link_cmd."\n";
# print $link_cmd;
print $fh $link_cmd;
close $fh;
link_program;
run_program;
if($?==0){
print "Passed round $j\n";
$upper = $middle;
}else{
print "Failed round $j\n";
$lower = $middle;
}
$j=$j+1;
system("rm temp_link.sh");
}
print "Bad compilation unit: @obj_array[$lower]\n";
Wednesday, 8 March 2017
RLDCR/rldcr (Rotate Left Doubleword then Clear Right)
Rotate Left Doubleword then Clear Right
MDS-form
rldcr RA,RS,RB,ME (Rc=0)
rldcr. RA,RS,RB,ME (Rc=1)
n <= (RB)58:63
r <= ROTL64((RS), n)
e <= me5 || me0:4
m <= MASK(0, e)
RA <= r & m
The contents of register RS are rotated64 left the number
of bits specified by (RB)58:63. A mask is generated
having 1-bits from bit 0 through bit ME and 0-bits elsewhere.
The rotated data are ANDed with the generated
mask and the result is placed into register RA.
Special Registers Altered:
CR0 (if Rc=1)
Programming Note:
rldcr can be used to extract an n-bit field that starts
at variable bit position b in register RS, left-justified
into register RA (clearing the remaining 64-n bits
of RA), by setting RB58:63=b and ME=n-1. It can
be used to rotate the contents of a register left
(right) by variable n bits, by setting RB58:63=n
(64-n) and ME=63.
Extended mnemonics are provided for some of
these uses (some devolve to rldcl); see
Appendix C, "assembler Extended Mnemonics" on
page 795.
MDS-form
rldcr RA,RS,RB,ME (Rc=0)
rldcr. RA,RS,RB,ME (Rc=1)
n <= (RB)58:63
r <= ROTL64((RS), n)
e <= me5 || me0:4
m <= MASK(0, e)
RA <= r & m
The contents of register RS are rotated64 left the number
of bits specified by (RB)58:63. A mask is generated
having 1-bits from bit 0 through bit ME and 0-bits elsewhere.
The rotated data are ANDed with the generated
mask and the result is placed into register RA.
Special Registers Altered:
CR0 (if Rc=1)
Programming Note:
rldcr can be used to extract an n-bit field that starts
at variable bit position b in register RS, left-justified
into register RA (clearing the remaining 64-n bits
of RA), by setting RB58:63=b and ME=n-1. It can
be used to rotate the contents of a register left
(right) by variable n bits, by setting RB58:63=n
(64-n) and ME=63.
Extended mnemonics are provided for some of
these uses (some devolve to rldcl); see
Appendix C, "assembler Extended Mnemonics" on
page 795.
Subscribe to:
Posts (Atom)