Files
Matt Godbolt (bot acct) 20505fba5a Filter NVCC fatbinData label from assembly output (#8506)
*(I'm Molty, an AI assistant acting on behalf of @mattgodbolt)*

Closes #5178

## Problem

NVCC embeds the CUDA fat binary blob in the host-side x86 assembly
inside a `#APP`/`#NO_APP` inline-assembly block, under a label called
`fatbinData`. In a realistic kernel this can be 100+ lines of `.quad`
hex values — a wall of noise before any user-readable code.

Example: https://godbolt.org/z/W3YMcq8oY

## Fix

Per-compiler pre-processing step in `NvccCompiler.processAsm()`: before
the host assembly reaches the ASM parser, any `#APP`/`#NO_APP` block
containing a `.nv_fatbin` section is stripped out entirely.

- Only `.nv_fatbin` blocks are removed; genuine user inline-assembly
blocks (which also use `#APP`/`#NO_APP` but without `.nv_fatbin`) are
left intact.
- Intentionally NVCC-specific — no changes to the base `AsmParser`, no
false-positive risk for other compilers.
- Stripping happens before `findUsedLabels` runs, so `fatbinData`
naturally disappears as unreferenced without any special-casing in the
parser's label-filtering logic.
- Gated on the existing Labels filter: with no filters active everything
remains visible; with Labels on the blob disappears.

## Testing

**New compiler unit tests** (`test/compilers/nvcc-tests.ts`):
- Strips `#APP`/`#NO_APP` blocks containing `.nv_fatbin`
- Preserves `#APP`/`#NO_APP` blocks without `.nv_fatbin` (user inline
asm)
- Handles multiple mixed blocks correctly
- No-op when no `#APP` blocks present
- Gracefully handles malformed unclosed blocks

**New parser filter-case**
(`test/filters-cases/nvcc-x86-host-example.asm`): representative NVCC
12.0 host assembly (real 15-line fat binary, boilerplate functions,
`.nvFatBinSegment` section) with nine filter-combination snapshots
documenting parser behaviour in isolation. These correctly show that the
**base parser itself does not filter `fatbinData`** — that's the
compiler pre-processor's job.

All 767 tests pass.

Co-authored-by: mattgodbolt-molty <mattgodbolt-molty@users.noreply.github.com>
2026-02-22 19:46:36 -06:00
..