Summary:
Add a new path in `post_grad.py` for replacing addmm + ReLU / GELU activation with the corresponding `_addmm_activation` call (with `use_gelu=False` or `True`, respectively). The replacement is done only on `max_autotune_gemm=False` and when the activation is fusible.
Test Plan:
$ python test/inductor/test_pattern_matcher.py -k test_addmm_activation -v
(__main__.TestPaternMatcher.test_addmm_activation) ... /data/users/aakhundov/pytorch/torch/_inductor/compile_fx.py:128: UserWarning: TensorFloat32 tensor cores for float32 matrix multiplication available but not enabled. Consider setting `torch.set_float32_matmul_precision('high')` for better performance.
warnings.warn(
Using FallbackKernel: aten._addmm_activation.default
Using FallbackKernel: aten._addmm_activation.default
/data/users/aakhundov/pytorch/torch/_dynamo/eval_frame.py:373: UserWarning: changing options to `torch.compile()` may require calling `torch._dynamo.reset()` to take effect
warnings.warn(
frames [('total', 1), ('ok', 1)]
stats [('calls_captured', 2), ('unique_graphs', 1)]
aot_autograd [('total', 1), ('ok', 1)]
inductor []
ok
----------------------------------------------------------------------
Ran 1 test in 13.415s
OK
Reviewers: @eellison
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104132
Approved by: https://github.com/eellison, https://github.com/jansel
Some notes:
* I now manually turn off `_generate` jobs from running with cudagraphs, as it is unrealistic to expect to cudagraph autoregressive generation up to max sequence length, this would imply compiling the entire unrolled sequence generation. Concretely, cm3leon_generate was timing out post this change, likely due to the compile time slowdown of dynamic shapes ON TOP OF accidentally unrolling all the loops
* A few torch._dynamo.reset tactically inserted to force recompiles on tests that expected it
* expectedFailureAutomaticDynamic flip into patching automatic_dynamic_shapes=False
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103623
Approved by: https://github.com/voznesenskym
This PR handles inference. Will do similar thing for training later.
Some manual testing results shows this can improve inference perf by 2-3% (absolute improvement not relative one).
- convmixer: 4.285x -> 4.309x
- resnet50: 2.170x -> 2.203x
The PR is built upon freezing. Since without freezing, the weight input for a conv node may not be a parameter directly but be the output of precision converting ops. It's so much easier to implement this PR after freezing.
Commands
```
TORCHINDUCTOR_FREEZING=1 python benchmarks/dynamo/timm_models.py --backend inductor --amp --performance --only convmixer_768_32 --inference
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103642
Approved by: https://github.com/eellison
Adds Conv-BN folding to inductor freezing. One thing that's a little awkward now is we'll want different decompositions to run depending on if we are in the inference compiler. For now, I require that you run with torch.no_grad() so we can detect if no gradients are required before calling aot_autograd.
Differential Revision: [](https://our.internmc.facebook.com/intern/diff/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100653
Approved by: https://github.com/jansel
Adds a freezing pass that will constant fold parameters in inductor `config.freezing`. This occurs post functionalization in aot autograd to capture both dispatching and allow passes to occur post functionalization. A few notes:
- There is an option to discard parameters `config.freezing_discard_parameters` which will take the current eager modules and wrap parameters to a Tensor subclass which will error if used.
- I needed to expose flat_params in aot_autograd in order to discard old references when we constant fold away parameters, like with amp. I also exposed `fw_metadata` to avoid constant folding mutated paraemters.
- Caching parameter transformations/constant folding across different inferences nyi
- Checking version_counter of constant folded params nyi
I'm not really sure what the actual naming should be. In jit there was both "freezing", which was platform agnostic, and "optimize for inference", which made device specific optimizations. We're doing the latter here but maybe freezing is a better name.
Differential Revision: [D46244033](https://our.internmc.facebook.com/intern/diff/D46244033)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100652
Approved by: https://github.com/jansel
Previously, cudagraphs and dynamic_shapes were incompatible and enabling
dynamic shapes would forcibly disable cudagraphs. This new strategy
I think is better. The idea is essentially that cudagraphs is an
"optimization" that happens to guard on every input. When cudagraphs
is on, we force everything static, and this automatically does the right
thing because we will force a recompile if sizes change.
This obsoletes https://github.com/pytorch/pytorch/pull/101813
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103290
Approved by: https://github.com/voznesenskym, https://github.com/eellison
Previously, cudagraphs and dynamic_shapes were incompatible and enabling
dynamic shapes would forcibly disable cudagraphs. This new strategy
I think is better. The idea is essentially that cudagraphs is an
"optimization" that happens to guard on every input. When cudagraphs
is on, we force everything static, and this automatically does the right
thing because we will force a recompile if sizes change.
This obsoletes https://github.com/pytorch/pytorch/pull/101813
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103290
Approved by: https://github.com/voznesenskym
Summary: bias_addmm is not backed up by a cpp funciton, so turn
autotune_cublasLt for cpp_wrapper + max_autotune. We can add a cpp
function implementation if there is a performance need.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103004
Approved by: https://github.com/jansel
Key change - seed, offset are the last 2 args in both the fwd and bwd graphs
Reason - The cudagraphs implementation in inductor currently relies on very simple ordering guarantees i.e. first n inputs are static for both fwd and bwd graphs. In the current implementation of functionalization of rng ops, this assumption is broken because the first 2 inputs are seed, offset.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/102344
Approved by: https://github.com/eellison
Summary:
Replace _dynamo.config with an object instead of module
Current usage patterns of setting and reading fields on config will work
unchanged.
Only changes needed going forward:
1. import torch._dynamo.config will not work. However, just doing
import torch._dynamo is sufficient to access dynamo config
as torch._dynamo.config.
2. Files inside of _dynamo folder need to access config via
from torch._dynamo.config_util import config instead of
from torch._dynamo import config. Because _dynamo/__init__.py
imports some of the files so it would be circular import.
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags:
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/96455
Approved by: https://github.com/williamwen42
As the comment for `get_expanded_dims` says:
```
# copy_ fails when trying to write to tensors with memory overlap,
# for expanded dimensions (a dimension which used to have size 1 -> ?)
# we can select one element from that dimension and write to it
# to achieve writing to all values of that dimension of the input tensor
```
We were doing this for the copy, for not for checking if we could copy. Update it so we index then check for memory overlap. This covers all of the `complex_striding` warnings I observed in TB.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98656
Approved by: https://github.com/ngimel, https://github.com/yf225
Patterns based on https://github.com/pytorch/pytorch/pull/94729 mainly as a forcing function for implementing joint graph replacements.
Up until now, we had two places to do pattern matching
1) Pre-grad has janky infra (graph not normalized or functional), but is
desirable for many types of passes where you want your change to
affect grad formulas.
2) Post-grad has good infra, but cant change grad formulas.
This PR adds a third place to do pattern matching: the joint
forward+backwards graph. The idea is to take the patterns and lower
them to a joint graph and replace both the forwards+backwards before
we partition them. This allows us to do something similar to pre-grad
transforms, but run after normalization and functionalization.
Note that we don't seem to have kernels for all of these patterns, some get decomposed in the dispatcher.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97741
Approved by: https://github.com/Chillee
Summary: This is a reland of #98264.
When _inductor.config.cpp_wrapper is specified, we run a
two-pass wrapper codegen to generate wrapper code in cpp which calls
cuLaunchKernel to launch pre-compiled cuda kernels, and then call
load_inline to load that generated wrapper back into the python world.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98534
Approved by: https://github.com/huydhn
Summary: when _inductor.config.cpp_wrapper is specified, we run a
two-pass wrapper codegen to generate wrapper code in cpp which calls
cuLaunchKernel to launch pre-compiled cuda kernels, and then call
load_inline to load that generated wrapper back into the python world.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98264
Approved by: https://github.com/ngimel
This replaces fake_mode_from_tensors but it preferentially looks for
fake_mode in TracingContext and also if there is an active fake mode
on the dispatch stack, before groveling in tensors to find it.
This advances PegasusForCausalLM, which was previously failing because
we generated a graph that had a parameter (non-fake) and a SymInt,
and thus previously we failed to detect the correct fake mode.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98321
Approved by: https://github.com/voznesenskym
Previously, when we would run a forward graph whose backward we never invoked it would prevent us from switching from warmup to recording. Now, refine the heuristic to allow incrementing the generation as soon as we invoke a backward graph. This still handles the
```
mod1 = torch.compile(...)
mod2 = torch.compile(...)
mod2(mod1(x)).sum().backward()
```
case while accounting for graphs which we may not run backward of.
It also now handles the case where we skip cudagraphify the backward of a forward.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98112
Approved by: https://github.com/jansel
Summary:
This is a copy of https://github.com/pytorch/pytorch/pull/97152 to make
the landing easier.
This PR implements a two-pass wrapper codegen for the Triton
backend to achieve ahead-of-time compilation. In the first pass, the
regular python wrapper code will be generated, and then the generated
code will be executed to perform Triton compilation and autotuning.
After that, the second pass wrapper codegen will generate C++ wrapper
with proper CUDA API to load and launch Triton-generated CUDA kernels.
Like the AOT mode for the cpp backend, the next step would be to provide
a more complete API for AOT.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98214
Approved by: https://github.com/eellison
Summary: to prepare for further AOT Inductor changes
<!--
copilot:summary
-->
### <samp>🤖 Generated by Copilot at 7dff885</samp>
This pull request adds support for AOT compilation and C++ wrapper code generation for inductor models. It modifies the `GraphLowering` class in `torch/_inductor/graph.py` and the `compile_fx` function in `torch/_inductor/compile_fx.py` to enable this feature.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97709
Approved by: https://github.com/jansel
