fix(gpu): preserve inventory for oversized decoded declarations (#12319)

* fix(gpu): preserve CCOB extents beyond decoded size capacity

* fix(gpu): retain CUDA entries beyond decoded size capacity (#12326)

* fix(gpu): retain CUDA entries beyond decoded size capacity

* refactor(gpu): share declared decoded size projection
This commit is contained in:
morluto authored and GitHub committed 2026-11-19 12:44:46 +00:00
1 parent ff6f4f4e31
commit 067ebd7349
6 files changed
+165 -39

No files matched your search

+11 -2
View File
@@ -44,14 +44,23 @@ such as `gfx11-generic` and `gfx9-4-generic`. Notes with the empty or explicit
`unknown` environment use the same processor and feature comparison while
retaining their raw triple spelling. CUDA's legacy fatbinary compression flag is
retained as an unsupported codec with its encoded and decoded extents;
compressed bytes are not treated as PTX text. CCOB is recognized by the exact
four signature bytes `CCOB`; high-bit-corrupted bytes are not decoded as ASCII
compressed bytes are not treated as PTX text. A CUDA entry whose declared
decoded length exceeds the exact integer range of `decoded_size` retains
its encoded range and reports the exact length in a limitation. Kernel decoding
reports `capacity` for a selected entry with a supported codec.
CCOB is recognized by the exact four signature bytes `CCOB`;
high-bit-corrupted bytes are not decoded as ASCII
and accepted as a bundle. CCOB v1 has no stored total size: the completed zlib
or Zstd stream determines its encoded extent. Embedded `CCOB` text is ordinary
compressed data; zero-byte alignment after the stream belongs to neither
payload. If decoding fails or its codec is unsupported, a later `CCOB` marker
is only an uncertain recovery candidate. The affected row retains that
limitation and discovery remains partial.
CCOB v2 and v3 declare the full encoded extent in their headers. If a decode
exceeds the output budget, the row reports `capacity` and inspection can
continue at the next bundle. A v3 decoded length outside the exact integer
range of `decoded_size` leaves that field null and is recorded in the row's
limitation.
The declared decoded length bounds decompression. An understated length can
stop decoding before codec completion, so any subsequent recovery extent stays