pc:~/.asciidoc3/doc$ python3 -m cProfile ../asciidoc3.py userguide.txt
8979863 function calls (8952700 primitive calls) in 13.708 seconds
Ordered by: standard name
ncalls  tottime  percall  cumtime  percall filename:lineno(function)
41    0.000    0.000    0.000    0.000 frozen importlib._bootstrap:103(release)
41    0.000    0.000    0.000    0.000 frozen importlib._bootstrap:143(__init__)
41    0.000    0.000    0.000    0.000 frozen importlib._bootstrap:147(__enter__)
41    0.000    0.000    0.000    0.000 frozen importlib._bootstrap:151(__exit__)
41    0.000    0.000    0.000    0.000 frozen importlib._bootstrap:157(_get_module_lock)
...
46    0.000    0.000    0.000    0.000 abc.py:9(abstractmethod)
3961    0.009    0.000    0.020    0.000 asciidoc3.py:101(items)
2122    0.024    0.000    0.372    0.000 asciidoc3.py:4253(subs)
1599    0.006    0.000    0.096    0.000 asciidoc3.py:4263(isnext)
2648    0.008    0.000    0.027    0.000 asciidoc3.py:427(dovetail)
910823    1.635    0.000    2.364    0.000 asciidoc3.py:4276(match)
1924    0.022    0.000    0.110    0.000 asciidoc3.py:4295(extract_passthroughs)
1924    0.003    0.000    0.004    0.000 asciidoc3.py:4304(restore_passthroughs)
...

We are interested only in functions of asciidoc3.py (see column "filename"), not in system calls like frozen importlib._bootstrap. I tried out some of the shown functions, but in the end I got stuck with this candidate in "class Macros, def match(...)"

910823    1.635    0.000    2.364    0.000 asciidoc3.py:4276(match)

We're talking about more than 900,000 calls and 2.3 seconds cumtime - about 18% of the entire run-time: "def match(...)" deserves some acceleration. Let's start (asciidoc3.py - edited lines are bold #####):

...
from decimal import Decimal, ROUND_HALF_UP
from functools import lru_cache  #####
from io import StringIO as io_StringIO
...

class Macros:
    ...
    def isnext(self):
        """Return matching macro if block macro is next on reader."""
        reader.skip_blank_lines()
        line = reader.read_next()
        if line:
            for m in self.macros:
                if m.prefix == '#':
                    if m.reo.match(line):
                        self.current = m
                        return m
        return False

    @lru_cache()  #####
    def match(self, prefix, name, text):
        """Return re match object matching 'text' with macro type 'prefix',
        macro name 'name'."""
        for m in self.macros:
            if m.prefix == prefix:
                mo = m.reo.match(text)
                if mo:
                    if m.name == name:
                        return mo
                    if re.match(name, mo.group('name')):
                        return mo
        return None

    def extract_passthroughs(self, text, prefix=''):
    ...

    ...

def asciidoc3(backend, doctype, confiles, infile, outfile, options):
    ...
        else:
            if document.attributes.get('DoNotUse') == 'AndDoNotEdit':
                Ad3Codec.fixed_encoding = False
            else:
                Ad3Codec.fixed_encoding = True
            # to ensure encoding in subprocess and conf-files:
            Ad3Codec.add_encoding_for_compatibility()
            if 'quirks' in document.attributes:
                message.verbose('No quirks-mode in AsciiDoc3 any more!')
            writer.newline = config.newline
            try:
                writer.open(outfile, reader.bom)
                try:
                    document.translate(has_header) # Generate the output.
                finally:
                    writer.close()
            finally:
                reader.closefile()
        # inspect cache use               #####
        print(macros.match.cache_info())  #####
    except KeyboardInterrupt:
        raise
    except Exception as e:
        # Cleanup
        ...

The output looks very interesting:

CacheInfo(hits=885758, misses=25065, maxsize=128, currsize=128)
7205035 function calls (7177872 primitive calls) in 10.923 seconds
...
25065 0.046 0.000 0.075 0.000 asciidoc3.py:4275(match)

Trying other values for maxsize in steps of 2**x

@lru_cache(131072)
2**17
CacheInfo(hits=889588, misses=21235, maxsize=131072, currsize=21235)
7196631 function calls (7169468 primitive calls) in 10.912 seconds

@lru_cache(65536)
2**16
CacheInfo(hits=889588, misses=21235, maxsize=65536, currsize=21235)
7196631 function calls (7169468 primitive calls) in 10.915 seconds

BTW, the output-file "userguide.html" is binary identical in all cases.

Wow! We see (13.708 - 10.915) / 0.13708 = ~20% less time.

But beware, we have to keep in mind, AsciiDoc3 should run with any version of Python3, and @functools.lru_cache was implemented in Python 3.2. So we have to edit asciidoc3.py avoiding uncatched exceptions:

from decimal import Decimal, ROUND_HALF_UP
try:                                                   #####
    from functools import lru_cache                    #####
except (NameError, ModuleNotFoundError, ImportError):  #####
    pass                                               #####
from io import StringIO as io_StringIO

...

    def match(self, prefix, name, text):
        """Return re match object matching 'text' with macro type 'prefix',
        macro name 'name'."""
        #print("y")
        for m in self.macros:
            if m.prefix == prefix:
                mo = m.reo.match(text)
                if mo:
                    if m.name == name:
                        return mo
                    if re.match(name, mo.group('name')):
                        return mo
        return None
    try:                                                  #####
        #lru_cache not implemented in Python < 3.3        #####
        match = lru_cache(65536)(match) # 2**16 = 65536   #####
    except (NameError, ModuleNotFoundError, ImportError): #####
        pass                                              #####

...