This PR addresses a few small bugfixes needed to make NanoGPT inference work, and also adds a new `--caching-precompile` argument to torchbench. With `--caching-precompile`, after every benchmark we save precompile artifacts to DynamoCache, allowing us to test caching precompile on all existing benchmarks.
The following bugfixes are in this PR to make all of this work:
- Fix global variables being pruned with DUPLICATE_INPUT guards. DUPLICATE_INPUT guards have additional vars from the second input, which we track with additional_local_vars, but we never tracked additional global variables. This fixes the issue. (See torch/_dynamo/guards.py changes)
- Return None from PRecompileContext.serialize() if no new dynamo compiles occurred. There's no reason to save artifacts (i.e. autotuning artifacts, etc) if no dynamo_compile occurred, so we return None early. We may later want to support editing existing dynamo artifacts as a TODO, but that's upcoming.
- log `dynamo_start` on CompilePackage.load: This is only needed so that tlparse doesn't ignore TORCH_TRACE logs generated when caching precompile hits. If there are no actual compiles, we never log a "dynamo_start" entry, which makes internal tlparse ignore the TORCH_TRACE file.
## Test Plan
After this PR, the following now works:
```
TORCH_LOGS=dynamo tlp python benchmarks/dynamo/torchbench.py --only nanogpt --performance --inference --backend inductor --caching-precompile --warm-start-latency
```
tlparse result (internal):
Cold Start (6 seconds):
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpAWe0zD/dedicated_log_torch_trace_vk9nkp4m.log/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
Warm Start (~1 s):
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpAWe0zD/dedicated_log_torch_trace_5l4iwrpm.log/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
The 1 second of warm start here can be improved: the costs here are mostly in starting up workers and triton and initializing CUDA, a lot of which should not be included in the compile time cost in real world scenarios where these are already loaded before training begins.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158847
Approved by: https://github.com/zhxchen17
This PR addresses a few small bugfixes needed to make NanoGPT inference work, and also adds a new `--caching-precompile` argument to torchbench. With `--caching-precompile`, after every benchmark we save precompile artifacts to DynamoCache, allowing us to test caching precompile on all existing benchmarks.
The following bugfixes are in this PR to make all of this work:
- Fix global variables being pruned with DUPLICATE_INPUT guards. DUPLICATE_INPUT guards have additional vars from the second input, which we track with additional_local_vars, but we never tracked additional global variables. This fixes the issue. (See torch/_dynamo/guards.py changes)
- Return None from PRecompileContext.serialize() if no new dynamo compiles occurred. There's no reason to save artifacts (i.e. autotuning artifacts, etc) if no dynamo_compile occurred, so we return None early. We may later want to support editing existing dynamo artifacts as a TODO, but that's upcoming.
- log `dynamo_start` on CompilePackage.load: This is only needed so that tlparse doesn't ignore TORCH_TRACE logs generated when caching precompile hits. If there are no actual compiles, we never log a "dynamo_start" entry, which makes internal tlparse ignore the TORCH_TRACE file.
## Test Plan
After this PR, the following now works:
```
TORCH_LOGS=dynamo tlp python benchmarks/dynamo/torchbench.py --only nanogpt --performance --inference --backend inductor --caching-precompile --warm-start-latency
```
tlparse result (internal):
Cold Start (6 seconds):
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpAWe0zD/dedicated_log_torch_trace_vk9nkp4m.log/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
Warm Start (~1 s):
https://manifold.edge.x2p.facebook.net/v0/read/tree/logs/.tmpAWe0zD/dedicated_log_torch_trace_5l4iwrpm.log/index.html?bucketName=tlparse_reports&apiKey=tlparse_reports-key&withPayload=1&timeoutMsec=10000
The 1 second of warm start here can be improved: the costs here are mostly in starting up workers and triton and initializing CUDA, a lot of which should not be included in the compile time cost in real world scenarios where these are already loaded before training begins.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158847
Approved by: https://github.com/zhxchen17
When loading a package and calling package.install(backends), we create a new frame and compile id for each package load, so that tlparse and chromium events still show compile times on warm start.
There is an argument for not doing this in AOT precompile, as no "compile" occurs. So for now, we put it in `package.install`, which hopefully won't be a thing for AOT precompile.
## Recompiles
Recompiles get saved to the same frame and code entry, so on warm start, each recompile will get collapsed into the same entry. Therefore, dynamo compiles that have recompiles on cold start (0/0, 0/1, 0/2, etc) will all get collapsed into a single compile id (0/0), as warm start will load all of the entries properly.
## Graph breaks
Graph breaks get their own compile id, and therefore their own code entry. These are replicated on warm start, so if cold start you had 4 different graphs (and therefore 4 compile ids), you'll have 4 compile ids on warm start as well.
## Test plan
Added a frame counter check to existing unit tests for automatic dynamic, showing that old and new frame counter between old and new load is the same.
This is the chromium event for test_automatic_dynamo_graph_breaks_device_cuda:
```
python test/dynamo/test_package.py -k test_automatic_dynamo_graph_breaks_device_cuda
```
<img width="2216" height="508" alt="image" src="https://github.com/user-attachments/assets/f604ed33-5c31-464b-9320-d67b2e6f57a1" />
Pull Request resolved: https://github.com/pytorch/pytorch/pull/158028
Approved by: https://github.com/oulgen
Users may want compile-related but customized logging info to dynamo_compile. One example is to logging the current training iteration index when recompilation happens. In general, current training iteration index is not available to compiler, since the same compiled function may be called multiple times in the same training iteration. The user could provide the training iteration index in a user hook where torch.compile logs it when recompilation happens.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/157961
Approved by: https://github.com/masnesral
This PR adds a new config option, `caching_precompile`, and a `DynamoCache`, which loads and saves Dynamo Cache entries automatically. It also hooks up DynamoCache to PrecompileContext, so that we can save multiple cache entries.
When this configuration is turned on, we:
- Automatically create and initialize a CompilePackage on every torch.compile
- Automatically use BundledAutogradcache
- Automatically save the CompilePackage entry to DynamoCache after every compile
You can also use PrecompileContext.serialize() to manually serialize a full object.
I've added unit tests to exhibit this behavior.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155913
Approved by: https://github.com/zhxchen17
This PR adds a new config option, `caching_precompile`, and a `DynamoCache`, which loads and saves Dynamo Cache entries automatically. It also hooks up DynamoCache to PrecompileContext, so that we can save multiple cache entries.
When this configuration is turned on, we:
- Automatically create and initialize a CompilePackage on every torch.compile
- Automatically use BundledAutogradcache
- Automatically save the CompilePackage entry to DynamoCache after every compile
You can also use PrecompileContext.serialize() to manually serialize a full object.
I've added unit tests to exhibit this behavior.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155913
Approved by: https://github.com/zhxchen17
`_torchdynamo_orig_callable` was being used in two distinct places:
- to get the original user function from nested eval_frame.py decorators
- to get the original backend from nested convert_frame.py callbacks
We rename ~the first usage to `_torchdynamo_orig_fn`~ and the second to `_torchdynamo_orig_backend` in order to distinguish these cases.
UPDATE: seems like both internal and OSS users depend on `_torchdynamo_orig_callable`, but it only seems in the first context. We should thus keep the original name for the first case then.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156901
Approved by: https://github.com/StrongerXi, https://github.com/jansel
`_torchdynamo_orig_callable` was being used in two distinct places:
- to get the original user function from nested eval_frame.py decorators
- to get the original backend from nested convert_frame.py callbacks
We rename the first usage to `_torchdynamo_orig_fn` and the second to `_torchdynamo_orig_backend` in order to distinguish these cases.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/156901
Approved by: https://github.com/StrongerXi, https://github.com/jansel
ghstack dependencies: #156527
This should prevent bad resume function prologues from slipping by. In particular, graph breaks in resume function prologues will now hard error.
Implementation details:
- The resume function prologue is surrounded by `LOAD_CONST arg, STORE_FAST __is_tracing_resume_prologue` instructions. The first sequence has `arg=True` and the second sequence has `arg=False`.
- InstructionTranslator will know when it is tracing a resume function prologue when it detects `STORE_FAST __is_tracing_resume_prologue`. The top of stack will be True to mark the start of the prologue, False to mark the end.
- When `convert_frame.py` detects that an error occurred while the InstructionTranslator was tracing a resume function prologue, we will wrap the exception and hard error
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154564
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289, #154782, #156762, #155166
- Make the fullgraph argument of set_fullgraph a positional argument
- Fix behavior on nested calls by updating `tracer.error_on_graph_break` in more places. In particular, a tracer's error_on_graph_break is set to the inlined tracer's error_on_graph_break upon the latter's exit. We also track error_on_graph_break in the speculation log now, since if we encounter a nested graph break, we will restart analysis and we need to somehow remember the error_on_graph_break setting after attempting to run the nested function (but we don't actually trace into it in the restart analysis).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154782
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289
Implements https://github.com/pytorch/pytorch/issues/144908.
Implementation notes:
- `set_fullgraph` is implemented using `patch_config`, which changes config correctly during runtime and tracing.
- Moved setting `config.error_on_graph_break` from convert_frame.py to eval_frame.py. This is because this should only be done at the top-level decorated function. If we kept this in convert_frame.py, we would be changing `config.error_on_graph_break` on every top-level frame, which causes confusing behavior (see added test for example).
- InstructionTranslator reads from `config.error_on_graph_break` every `step()`. This is to determine the value of `config.error_on_graph_break` at the time of the graph break, because tracer cleanup will restore the value of `config.error_on_graph_break` .
- `convert_frame.py` determines whether we should abort tracing (fullgraph=True) or continue (fullgraph=False) by reading the value of the tracer's `error_on_graph_break`. If there is no tracer (failed to initialize), then default to reading `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154289
Approved by: https://github.com/jansel, https://github.com/zou3519
ghstack dependencies: #154283
`torch.compile` now always goes through `torch._dynamo._optimize`. fullgraph is now implemented in `torch.compile` by looking at `config.error_on_graph_break`. Export still goes through `torch._dynamo._optimize_assert`, which uses `tx.one_graph` instead of `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154283
Approved by: https://github.com/jansel, https://github.com/anijain2305
Based on the [conversation](https://github.com/pytorch/pytorch/issues/121791), we plan to drop the "highest, high, medium" to represent fp32 internal computation data types . Instead, we will directly use the algorithm to represent it.
### Design Choice: Directly use algorithms name like "TF32", "BF16".
#### Pros
- The names are more informative. 'tf32' is more informative than a simple "high".
- Easier to extend new algorithm like `tf32x3`
#### Cons
- "HIGHEST, HIGH, MEDIUM" indicated the relative precision between different algorithms. However, we can have more documents to discuss them.
### We provide a layered structure for backends/operators.
('f32' is short for 'fp32_precision')
### We provide 3 fp32 compute precision can be set:
- **"ieee"**: Not allowed to use any other internal computation data types .
- **"tf32"**: Allowed to use tf32 as internal computation data types.
- **"bf16"**: Allowed to use bf16 as internal computation data types.
- **"none"**: Precision's are not set. Can be override by its father node.
### Overriding Precision Settings
Child node can be override by its father node if it is set to default.
For current default settings:
```
backend = generic, op = all, precision setting = none
backend = cuda, op = all, precision setting = none
backend = cuda, op = conv, precision setting = tf32
backend = cuda, op = rnn, precision setting = tf32
backend = cuda, op = matmul, precision setting = none
backend = matmul, op = all, precision setting = none
backend = matmul, op = conv, precision setting = none
backend = matmul, op = rnn, precision setting = none
backend = matmul, op = matmul, precision setting = none
```
- If the user set `torch.backends.mkldnn.fp32_precision="bf16"`, his child nodes `torch.backends.mkldnn.matmul.fp32_precision` / `torch.backends.mkldnn.conv.fp32_precision` / `torch.backends.mkldnn.rnn.fp32_precision` will also be override to "bf16".
- If the user set `torch.backends.fp32_precision="bf16"`, `torch.backends.mkldnn.fp32_precision` and his child nodes will also we override to "bf16".
### Backward Compatible
Since new API allow user to have more fine-grained control. There will be some conflict. For example, previous `torch.backends.cudnn.allow_tf32` are not enough to represent the status for `torch.backends.cudnn.rnn.fp32_precision="ieee"` and `torch.backends.cudnn.conv.fp32_precision="tf32"`. Therefore, our goal for backward compatible is
- If the user only uses previous APIs, it will work as previous expectations.
- If the user use **new** API to change the status to an **un-representable** status for old API, and try to access the status by **old** API. We will raise Runtime Error and point the document for user.
### Test Plan
```
python test/test_cuda.py -k test_fp32_precision_with_tf32
python test/test_cuda.py -k test_fp32_precision_with_float32_matmul_precision
python test/test_cuda.py -k test_invalid_status_for_legacy_api
python test/test_mkldnn.py -k test_mlkdnn_get_set
python test/test_mkldnn.py -k test_generic_precision
python test/test_mkldnn.py -k test_invalid
python test/test_mkldnn.py -k test_default_use_parent
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/125888
Approved by: https://github.com/jgong5, https://github.com/albanD
Co-authored-by: Jiang, Yanbing <yanbing.jiang@intel.com>
Summary:
In D75617963, we started logging dynamic whitelist suggestions to PT2 Compile Events. The whitelists were aggregated across all frames, intending to avoid manual work for the user (e.g. if frame 0/1 saw L['x'] turn dynamic, and later 1/1 saw L['y'], we'd log "L['x'],L['y']" on frame 1/1).
This switches to frame-specific whitelists, as attributing dynamism changes to certain frames was difficult, and suggestions are sometimes polluted by problematic frames (e.g. optimizer states).
The globally aggregated whitelist is still available in tlparse, by looking at the final `put_local_code_state_*` entry.
Test Plan:
loggercli codegen GeneratedPt2CompileEventsLoggerConfig
Rollback Plan:
Differential Revision: D76628834
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155959
Approved by: https://github.com/bobrenjc93
This should prevent bad resume function prologues from slipping by. In particular, graph breaks in resume function prologues will now hard error.
Implementation details:
- The resume function prologue is surrounded by `LOAD_CONST arg, STORE_FAST __is_tracing_resume_prologue` instructions. The first sequence has `arg=True` and the second sequence has `arg=False`.
- InstructionTranslator will know when it is tracing a resume function prologue when it detects `STORE_FAST __is_tracing_resume_prologue`. The top of stack will be True to mark the start of the prologue, False to mark the end.
- When `convert_frame.py` detects that an error occurred while the InstructionTranslator was tracing a resume function prologue, we will wrap the exception and hard error
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154564
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289, #154782, #155166
- Make the fullgraph argument of set_fullgraph a positional argument
- Fix behavior on nested calls by updating `tracer.error_on_graph_break` in more places. In particular, a tracer's error_on_graph_break is set to the inlined tracer's error_on_graph_break upon the latter's exit. We also track error_on_graph_break in the speculation log now, since if we encounter a nested graph break, we will restart analysis and we need to somehow remember the error_on_graph_break setting after attempting to run the nested function (but we don't actually trace into it in the restart analysis).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154782
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289
Implements https://github.com/pytorch/pytorch/issues/144908.
Implementation notes:
- `set_fullgraph` is implemented using `patch_config`, which changes config correctly during runtime and tracing.
- Moved setting `config.error_on_graph_break` from convert_frame.py to eval_frame.py. This is because this should only be done at the top-level decorated function. If we kept this in convert_frame.py, we would be changing `config.error_on_graph_break` on every top-level frame, which causes confusing behavior (see added test for example).
- InstructionTranslator reads from `config.error_on_graph_break` every `step()`. This is to determine the value of `config.error_on_graph_break` at the time of the graph break, because tracer cleanup will restore the value of `config.error_on_graph_break` .
- `convert_frame.py` determines whether we should abort tracing (fullgraph=True) or continue (fullgraph=False) by reading the value of the tracer's `error_on_graph_break`. If there is no tracer (failed to initialize), then default to reading `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154289
Approved by: https://github.com/jansel, https://github.com/zou3519
ghstack dependencies: #154283
`torch.compile` now always goes through `torch._dynamo._optimize`. fullgraph is now implemented in `torch.compile` by looking at `config.error_on_graph_break`. Export still goes through `torch._dynamo._optimize_assert`, which uses `tx.one_graph` instead of `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154283
Approved by: https://github.com/jansel, https://github.com/anijain2305
Recursive function collect_temp_source with closure in PyCodegen caused cycle reference issue when torch.compile is used.
This issue may cause major tensors will not freed timely even there are no user references to these tensors.
We saw OOM issues because of this problem in many cases including training and inference using torch.compile.
The fix is to use iterative function implementation to replace the recursive function implementation.
Fixes#155778
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155791
Approved by: https://github.com/ezyang
This should prevent bad resume function prologues from slipping by. In particular, graph breaks in resume function prologues will now hard error.
Implementation details:
- The resume function prologue is surrounded by `LOAD_CONST arg, STORE_FAST __is_tracing_resume_prologue` instructions. The first sequence has `arg=True` and the second sequence has `arg=False`.
- InstructionTranslator will know when it is tracing a resume function prologue when it detects `STORE_FAST __is_tracing_resume_prologue`. The top of stack will be True to mark the start of the prologue, False to mark the end.
- When `convert_frame.py` detects that an error occurred while the InstructionTranslator was tracing a resume function prologue, we will wrap the exception and hard error
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154564
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289, #154782, #155166
- Make the fullgraph argument of set_fullgraph a positional argument
- Fix behavior on nested calls by updating `tracer.error_on_graph_break` in more places. In particular, a tracer's error_on_graph_break is set to the inlined tracer's error_on_graph_break upon the latter's exit. We also track error_on_graph_break in the speculation log now, since if we encounter a nested graph break, we will restart analysis and we need to somehow remember the error_on_graph_break setting after attempting to run the nested function (but we don't actually trace into it in the restart analysis).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154782
Approved by: https://github.com/jansel
ghstack dependencies: #154283, #154289
Implements https://github.com/pytorch/pytorch/issues/144908.
Implementation notes:
- `set_fullgraph` is implemented using `patch_config`, which changes config correctly during runtime and tracing.
- Moved setting `config.error_on_graph_break` from convert_frame.py to eval_frame.py. This is because this should only be done at the top-level decorated function. If we kept this in convert_frame.py, we would be changing `config.error_on_graph_break` on every top-level frame, which causes confusing behavior (see added test for example).
- InstructionTranslator reads from `config.error_on_graph_break` every `step()`. This is to determine the value of `config.error_on_graph_break` at the time of the graph break, because tracer cleanup will restore the value of `config.error_on_graph_break` .
- `convert_frame.py` determines whether we should abort tracing (fullgraph=True) or continue (fullgraph=False) by reading the value of the tracer's `error_on_graph_break`. If there is no tracer (failed to initialize), then default to reading `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154289
Approved by: https://github.com/jansel, https://github.com/zou3519
ghstack dependencies: #154283
`torch.compile` now always goes through `torch._dynamo._optimize`. fullgraph is now implemented in `torch.compile` by looking at `config.error_on_graph_break`. Export still goes through `torch._dynamo._optimize_assert`, which uses `tx.one_graph` instead of `config.error_on_graph_break`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154283
Approved by: https://github.com/jansel, https://github.com/anijain2305
Adding a per torch.compile() object CompilePackage which tracks dynamo artifact. CompilePackage is considered a low level component and should not be directly exposed to end users. It has the following interface:
1. `CompilePackage.__init__()` which optionally takes previously serialized dynamo states.
a. when `dynamo` argument is None, it will contruct a brand new CompilePackage object.
b. when `dynamo` argument is not None, it will load a pre-compiled dynamo state.
2. `package.save()` which dumps the dynamo states into _DynamoCacheEntry.
3. `package.install(backends)` which will handle all the side-effectful global scope updates with compiled functions and resume functions.
This diff focus on making the low level mechanism for precompile. It will be left to upper level interface to use these API to build more user-facing frontend.
Differential Revision: [D75956538](https://our.internmc.facebook.com/intern/diff/D75956538/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155118
Approved by: https://github.com/jamesjwu
Co-authored-by: James Wu <jjwu@meta.com>
#153622 introduced a hook for getting the relevant code objects after frame tracing. The idea is to have vLLM use this instead of monkey-patching `inline_call_()` to determine the source code files to hash. Unfortunately, the hook runs too late; the vLLM backend needs access to the set of source code filenames while it's running.
This PR replaces the newly-added hook with a utility function that a backend can call to get this information. I've made the change in vLLM and can verify that this allows the information to be queried at the right time.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/155249
Approved by: https://github.com/zou3519
