Test that the AOT autograd cache persists across separate processes by
running two subprocesses with a shared cache directory and verifying the
second one gets a cache hit.

ghstack-source-id: d69ed07
Pull Request resolved: #177397

Test that the AOT autograd cache persists across separate processes by
running two subprocesses with a shared cache directory and verifying the
second one gets a cache hit.

ghstack-source-id: dda4c48
Pull Request resolved: #177397

pre_grad_custom_pass was the only custom pass config without UUID-based
cache key integration. It was excluded from config serialization but not
handled specially via UUID extraction, so its effect was only captured
indirectly through the resulting FX graph. This meant two different
passes producing the same graph could incorrectly share a cache entry.

Align pre_grad_custom_pass with post-grad and joint passes: change its
type to CustomGraphPassType, add it to _cache_config_ignore_prefix (so
the UUID is extracted explicitly), include it in FxGraphHashDetails, and
validate it in _check_can_cache.

Pull Request resolved: #177403
Approved by: https://github.com/aorenste, https://github.com/zou3519
ghstack dependencies: #177397

Pull Request resolved: #177428
Approved by: https://github.com/mlazos, https://github.com/zou3519
ghstack dependencies: #177397, #177403

…77429)

Add a pre_grad_pass_timing config ("early", "late", or "default") that
controls when pre-grad passes run relative to the AOT autograd cache lookup.

- "early": passes run before cache lookup, so they execute on every compile
(including cache hits) and the cache key reflects the already-transformed
graph.
- "late": passes run after cache lookup (only on cache miss); requires
custom passes to provide a UUID for the cache key.
- "default": automatically resolves to "late" when possible (no custom pass,
or a custom pass with a UUID), and falls back to "early" when the custom
pass has no UUID.

Explicitly setting "late" with a UUID-less custom pass now raises a
RuntimeError instead of silently bypassing the cache. The existing
test_pre_grad_passes_called_on_cache_miss_only test is renamed and
pinned to "late" timing, and new tests cover early timing, both default
timing branches, and the error case.

Pull Request resolved: #177429
Approved by: https://github.com/aorenste, https://github.com/zou3519
ghstack dependencies: #177397, #177403, #177428

Test that the AOT autograd cache persists across separate processes by
running two subprocesses with a shared cache directory and verifying the
second one gets a cache hit.

Pull Request resolved: pytorch#177397
Approved by: https://github.com/aorenste, https://github.com/zou3519

…177403)

pre_grad_custom_pass was the only custom pass config without UUID-based
cache key integration. It was excluded from config serialization but not
handled specially via UUID extraction, so its effect was only captured
indirectly through the resulting FX graph. This meant two different
passes producing the same graph could incorrectly share a cache entry.

Align pre_grad_custom_pass with post-grad and joint passes: change its
type to CustomGraphPassType, add it to _cache_config_ignore_prefix (so
the UUID is extracted explicitly), include it in FxGraphHashDetails, and
validate it in _check_can_cache.

Pull Request resolved: pytorch#177403
Approved by: https://github.com/aorenste, https://github.com/zou3519
ghstack dependencies: pytorch#177397

Pull Request resolved: pytorch#177428
Approved by: https://github.com/mlazos, https://github.com/zou3519
ghstack dependencies: pytorch#177397, pytorch#177403

…torch#177429)

Add a pre_grad_pass_timing config ("early", "late", or "default") that
controls when pre-grad passes run relative to the AOT autograd cache lookup.

- "early": passes run before cache lookup, so they execute on every compile
(including cache hits) and the cache key reflects the already-transformed
graph.
- "late": passes run after cache lookup (only on cache miss); requires
custom passes to provide a UUID for the cache key.
- "default": automatically resolves to "late" when possible (no custom pass,
or a custom pass with a UUID), and falls back to "early" when the custom
pass has no UUID.

Explicitly setting "late" with a UUID-less custom pass now raises a
RuntimeError instead of silently bypassing the cache. The existing
test_pre_grad_passes_called_on_cache_miss_only test is renamed and
pinned to "late" timing, and new tests cover early timing, both default
timing branches, and the error case.

Pull Request resolved: pytorch#177429
Approved by: https://github.com/aorenste, https://github.com/zou3519
ghstack dependencies: pytorch#177397, pytorch#177403, pytorch#177428