mirror of
https://github.com/tangerineturtle1/NX2navi.git
synced 2026-11-15 17:09:57 -04:00
fc6d84aa63
The shader reader now reads Switch 2 (NVN2) programs, whose 0xe0-byte control sections and task and mesh stages the runtime keys as it does, and runs on Python 3.12 through the zstandard package. The kit's new last step writes the 36,031 shaders of the player's game files as a seed and runs ocarina.exe --prepare-shaders, which translates them and has the GPU driver build each one's library, about 5 minutes. On a cold walk into play this took the stalls from 14 seconds of pipeline builds and 9 of translations to none. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
48 lines
2.2 KiB
Python
48 lines
2.2 KiB
Python
"""The part of Python 3.14's compression.zstd the shader tools use, on older Pythons through the zstandard package.
|
|
|
|
decompress(data, zstd_dict=None), ZstdDict(raw) with .dict_id, and get_frame_info(data) with .decompressed_size (None
|
|
when the frame does not say) and .dictionary_id (0 for none). On 3.14 the standard module itself is used.
|
|
"""
|
|
try:
|
|
from compression.zstd import ZstdDict, decompress, get_frame_info # noqa: F401
|
|
except ImportError:
|
|
from collections import namedtuple
|
|
|
|
import zstandard
|
|
|
|
FrameInfo = namedtuple('FrameInfo', 'decompressed_size dictionary_id')
|
|
LIMIT = 1 << 30 # the largest output a frame without a content size may decompress to
|
|
|
|
class ZstdDict:
|
|
def __init__(self, raw):
|
|
self._d = zstandard.ZstdCompressionDict(raw, dict_type=zstandard.DICT_TYPE_AUTO)
|
|
self.dict_id = self._d.dict_id()
|
|
|
|
def get_frame_info(data):
|
|
p = zstandard.get_frame_parameters(data)
|
|
size = p.content_size if p.content_size >= 0 and p.content_size != zstandard.CONTENTSIZE_UNKNOWN else None
|
|
return FrameInfo(size, p.dict_id)
|
|
|
|
def decompress(data, zstd_dict=None):
|
|
d = zstandard.ZstdDecompressor(dict_data=zstd_dict._d if zstd_dict else None)
|
|
p = zstandard.get_frame_parameters(data)
|
|
if 0 <= p.content_size != zstandard.CONTENTSIZE_UNKNOWN:
|
|
# one frame of a known size, the common case: decompressed in one call (the streaming path below is
|
|
# several times slower on the shader archives' hundreds of megabytes)
|
|
frame = zstandard.frame_content_size(data)
|
|
if frame == p.content_size:
|
|
try:
|
|
return d.decompress(data)
|
|
except zstandard.ZstdError:
|
|
pass # more than one frame: the streaming path
|
|
out, pos = [], 0
|
|
# One frame after another, as compression.zstd decompresses concatenated frames.
|
|
while pos < len(data):
|
|
obj = d.decompressobj()
|
|
out.append(obj.decompress(data[pos:]))
|
|
rest = obj.unused_data
|
|
pos = len(data) - len(rest)
|
|
if not rest:
|
|
break
|
|
return b''.join(out)
|