mirror/rea
1
0
Fork 0
mirror of https://github.com/morluto/rea synced 2026-10-25 22:08:11 -04:00
Code Issues Packages Projects Releases Wiki Activity

fix(gpu): own CCOB boundaries and payload representations

Browse Source 
Use completed codec input to establish legacy CCOB extents, preserving declaration/integrity failures independently. Reuse bounded input windows, charge actual reads and decoding attempts, and qualify recovery after incomplete decoding. Publish only established or qualified terminal extents so interruption retains completed declarations. Recognize bitcode representation in HIP bundles while leaving inner facts unsupported.

Fixes #6887
Fixes #6880
This commit is contained in:
morluto
2026-10-24 02:32:54 +08:00
parent e737e6c1c6
commit 635f673a8c
9 changed files with 553 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/gpu-code-objects.md
+18 -3
View File
@@ -32,6 +32,12 @@ unexamined payloads remain `unexamined`. An exact AMDGPU target is normalized at
its note owner for comparison with a Clang bundle declaration; raw values remain
intact. Unresolved feature comparisons stay `unknown`.
A HIP target ID identifies a toolchain target, not its payload representation.
Recognized raw or wrapped LLVM bitcode stays `unsupported` for kernel metadata
inspection, with the bundle ID, range and selected payload digest retained.
Bitcode validity and inner targets remain unexamined; malformed recognized ELF
still reports `malformed`. Standalone bitcode is outside GPU container coverage.
HIP bundle and metadata note recognition accept both the legacy `amdgcn` triple
and LLVM's `amdgpu` subarchitecture triples, including declared generic processors
such as `gfx11-generic` and `gfx9-4-generic`. Notes with the empty or explicit
@@ -40,9 +46,18 @@ 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
and accepted as a bundle. CCOB v1 has no stored total size, so a following
legacy bundle is located by the next `CCOB` signature and zero-byte alignment
padding remains part of the first container extent.
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.
The declared decoded length bounds decompression. An understated length can
stop decoding before codec completion, so any subsequent recovery extent stays
unknown; the decoder does not expand its allowance to prove the bad declaration.
Interrupted observations omit unfinished v1 rows. A terminal capacity row
qualifies its retained range as the containing tail and does not infer siblings.
`cuda_target_header` retains the OS ABI, normalized raw `e_flags` and exact
header location. CUDA OS ABI `0x33` and the observed `0x41` flags layout use