#### change 1: if compute_strides stride fail for reshape just clone.
Lets consider the most general case, if torch compile is asked to reshape [u0, u1][u3, u4] -> [u5, u6] what shall it do?
The shape is general enough to represent both contiguous and non contiguous tensors, tensors where a clone free reshape can happen and other where a clone free cant happen. The current algorithm will fail due to data dependent errors.
The general idea is if its impossible to tell if the reshape can happen in place, (because for some concrete inputs
it will and other not) then its ok to take the general path and clone, instead of failing or asking the user to give hints.
**Because the user want a single graph (single compilations)** and this is the only way it can be done.
Had this been a view? then the user is explicitly asking for a copy-free reshape, we would fail asking for more
information (hints in torch.checks form).
with this change reshape works as the following:
1. if we know the input is contiguous we will convert the reshape to view.
2. if compute_strides succeed we will use view. (compute_strides was changed to not fail when when unbacked presented instead it will just return nullptr if it cant compute the strides meaning we shall use a clone).
3. if neither 1, 2 works clone and use a view.
Side note: having a view does not mean that inductor will not clone, for inductor there is a pass that converts all views back to reshapes and inductor has its logic dealing with those.
#### change 2 : skip _reshape_view_helper and fall back to simpler logic if it fail.
We trace _reshape_view_helper when doing fake tensor tracing , but not during proxy tracing. hence such tracing wont effect the graph (only compute output shapes of several operations). We should not fail there, because it should always be possible for us to pass it in case of reshape.
i.e. when reshape_symint was called we would have either cloned, or compute_strides succeeded so the view should pass. What I did is the following: we run _reshape_view_helper, if we fail due to unbacked we call _view_simple which will succeed always for reshapes, (might fail for views when its impossible to do the view, in such case we throw the dde that was thrown by the original algorithm).
Ideally I would want to register _view_simple as the meta for view and avoid calling _reshape_view_helper completely but I am running some issues with the dispatcher with subclasses and I do not have time to debug it. Namely one test
would end up calling some c++ view function that does not support symints during meta dispatch when i register a
python meta decompositions
```python test/dynamo/test_subclasses.py SubclassTests.test_subclass_views_dynamic_True ```
https://github.com/pytorch/pytorch/issues/153303.I will follow up with that change in a separate PR. cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @chenyang78 @kadeng @muchulee8 @amjames @chauhang @aakhundov @bdhirsh
Two other alternatives for registering _view_simple as meta and the try catch approach in this PR is:
1. call _view_simple if any input is dynamic see #153521
2. if we make is_compiling works for framework code tracing (does not work rn) we can call _view_simple
is if is_compiling.
#### Note:
Reshape can still fail when is_contiguous is called, Next PR will handle that by calling is_known_contiguous.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153198
Approved by: https://github.com/etaf, https://github.com/bobrenjc93
This PR adds support for SimpleFSDP's composability with Tensor Parallel + torch.compile.
`_StridedShard` is used in SimpleFSDP/FSDP2 to support correct distributed checkpointing when FSDP+TP is applied. Previously, `_StridedShard` is not guarded by torch.compile. This PR adds `_StridedShard` as an additional placement type to be guarded by torch.compile.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152286
Approved by: https://github.com/bdhirsh
During our debug session, @wdvr and I found out that the benchmark database is growing much faster than we expect. After taking a closer look, the majority of them coming from TorchInductor benchmark and the top 3 are all debug information not used by any dashboard atm. In the period of 7 days, there are close to 6 millions records ([query](https://paste.sh/GUVCBa0v#UzszFCZaWQxh7oSVsZtfZdVE))
```
Benchmark,Metric,Count
"TorchInductor","user_stack","1926014"
"TorchInductor","reason","1926014"
"TorchInductor","model","1926014"
```
Let's skip uploading them to avoid bloating the database.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153769
Approved by: https://github.com/malfet
Summary: There should be no reason to check for existence of this GNU C++ header here in this file. It doesn't include it. Removing this condition to make it build under libc++.
Differential Revision: D75179136
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154080
Approved by: https://github.com/soumith
Fixes#153777
CSE is an optimization and shouldn't block a compile if it hits recursion depth limits. Unfortunately we can't write this iteratively due to a dependency on `ast.unparse` which necessarily needs to do recursion. This PR catches opts out of CSE when we hit recursion depth errors.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154039
Approved by: https://github.com/Microve
Summary:
Context:
Recently we've added a couple more kernel types support other than inductor generated triton kernels,
such as cpu cpp kernels, extern kernels.
The name appeared in tlparse chrome link can be confusing to users.
Rename from
`inductor_triton_kernel_to_post_grad_nodes.json`
to `inductor_generated_kernel_to_post_grad_nodes.json`
Test Plan: CI
Differential Revision: D75159042
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154046
Approved by: https://github.com/yushangdi
Summary:
This is a reland of D74910193.
We change the dtype to torch.float8_e5m2 in unit test since it is not supported.
Test Plan:
```
buck2 test 'fbcode//mode/dev-nosan' fbcode//caffe2/test/inductor:quantization
```
Differential Revision: D75169792
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154057
Approved by: https://github.com/Mingming-Ding
`isin_Tensor_Scalar_out` is just a redispatch to eq/neq
`isin_Scalar_Tensor_out` redispatches back to generic `isin` op, but needs a small tweak to handle float scalars
Make sure that `out` is resized to an expected value in `isin_Tensor_Tensor_out_mps`
Add unittests to validate that, but skip them on MacOS-13, where MPS op just returns garbage
Before this change both of those failed
```python
>>> import torch
>>> t = torch.tensor([0, 1, 2], device='mps')
>>> torch.isin(t, 1)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NotImplementedError: The operator 'aten::isin.Tensor_Scalar_out' is not currently implemented for the MPS device. If you want this op to be considered for addition please comment on https://github.com/pytorch/pytorch/issues/141287 and mention use-case, that resulted in missing op as well as commit hash 3b875c25ea6d8802a0c53af9eb961ddf2f058188. As a temporary fix, you can set the environment variable `PYTORCH_ENABLE_MPS_FALLBACK=1` to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.
>>> torch.isin(1, t)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
NotImplementedError: The operator 'aten::isin.Scalar_Tensor_out' is not currently implemented for the MPS device. If you want this op to be considered for addition please comment on https://github.com/pytorch/pytorch/issues/141287 and mention use-case, that resulted in missing op as well as commit hash 3b875c25ea6d8802a0c53af9eb961ddf2f058188. As a temporary fix, you can set the environment variable `PYTORCH_ENABLE_MPS_FALLBACK=1` to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154010
Approved by: https://github.com/Skylion007, https://github.com/dcci, https://github.com/manuelcandales
ghstack dependencies: #153970, #153971, #153997
After https://github.com/pytorch/pytorch/pull/154004, one of the model `phlippe_resnet` needs higher tolerance for fp16 on CUDA 12.8. I can reproduce it locally with:
```
python benchmarks/dynamo/torchbench.py --accuracy --timing --explain --print-compilation-time --inductor --device cuda --training --amp --only phlippe_resnet
E0522 02:47:12.392000 2130213 site-packages/torch/_dynamo/utils.py:2949] RMSE (res-fp64): 0.00144, (ref-fp64): 0.00036 and shape=torch.Size([]). res.dtype: torch.float32, multiplier: 3.000000, tol: 0.001000, use_larger_multiplier_for_smaller_tensor: 0
```
I'm not sure what exactly happens behind the scene, but this should help fix the CI failure.
Also remove some left over expected accuracy results for CUDA 12.4 which we are not using anymore on CI for benchmark jobs.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154109
Approved by: https://github.com/Skylion007, https://github.com/malfet
https://github.com/pytorch/pytorch/issues/148222
Goal:
At the moment autograd saved tensors hooks are run in eager after compiled forward.
They are executed at the same time for all saved tensors.
Hooks can be used to reduce amout of memory used for saved tensors, doing quantization or offloading to cpu.
This is suboptimal for optimization of peak memory.
Better solution will be to put the hooks in the graph, as close as possible to the last usage of the tensor.
To get user specified autograd saved tensors hooks in the graph.
Logic:
UX:
If user specifies with torch.autograd.graph.saved_tensors_hooks(pack_gm, unpack_gm).
Where pack_gm and unpack_gm are torch.fx.GraphModule.
Then AotAutograd will retrace those graph modules, doing decompositions and functionalization in aot_autograd, inlining the result graphs in forward epilogue and backward prologue.
User may want to use control logic in the hooks, for example applying quantization only for specific dtypes and sizes.
This is also possible, user can put it into torch.fx.wrap function and use symbolic trace to make a GraphModule.
In that case AotAutograd cahing will work only in case when user explicitly set to the torch.fx.wrap call_function node "user_cache_hash" metadata.
If this metadata set - then aot_autograd cache can use saved cache artifact.
If metadata is not set - then cache is bypassed.
Dynamo:
Dynamo traces pack and unpack hooks and installs them as subgraph and explicitly adds to the output_graph. (As those subgraphs are not used and will not be copied in the result by default).
The complexity here is that at this moment we do not have example of inputs for the hooks.
We trace pack_hook with some Tensor from the inputs.
The result subgraphs are added to the hashing of AotAutograd Cache.
In AotAutograd we retrace the graph with the true saved tensors coming from partitioner.
Backwards Compatibility:
As current hooks are executed in eager mode and not all of them will be traceable - we only try to put in the graph hooks, explicitly marked by user with annotation (@_inlineable_saved_tensors_hooks).
For other hooks or if compiled autograd is enabled - keep the same logic.
Recompilations:
Hooks are guarded with lambda guard matching function id to cause recompilation if user reruns compiled function.
Aot_autograd:
After partitioner prepared forward and backward module - we trace prepared at Dynamo graphs for pack and unpack hooks and inline them in epilogue of forward and prologue of backward. Forward outputs and backward inputs are changed, transparently for user.
We do not try to put it close the last usage etc., relying on inductor to do this optimization.
```
INFO: TRACED GRAPH
===== Forward graph pre saved_tensors_hooks inlining 3 =====
/data/users/ivankobzarev/a/pytorch/torch/fx/_lazy_graph_module.py class GraphModule(torch.nn.Module):
def forward(self, primals_1: "Sym(s0)", primals_2: "Sym(s1)", primals_3: "f32[s0, s1][s1, 1]cuda:0"):
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6660 in simple_fn, code: x = x + 1
add: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.add.Tensor(primals_3, 1); primals_3 = None
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6661 in simple_fn, code: x = SAF.apply(x)
view: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.view.default(add, [primals_1, primals_2])
return (view, add, primals_1, primals_2)
INFO: TRACED GRAPH
===== Backward graph pre saved_tensors_hooks inlining 3 =====
/data/users/ivankobzarev/a/pytorch/torch/fx/_lazy_graph_module.py class GraphModule(torch.nn.Module):
def forward(self, primals_1: "Sym(s0)", primals_2: "Sym(s1)", primals_3: "f32[s0, s1][s1, 1]cuda:0"):
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6660 in simple_fn, code: x = x + 1
add: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.add.Tensor(primals_3, 1); primals_3 = None
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6661 in simple_fn, code: x = SAF.apply(x)
view: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.view.default(add, [primals_1, primals_2])
return (view, add, primals_1, primals_2)
INFO: TRACED GRAPH
===== saved_tensors_pack_hook add 3 =====
/data/users/ivankobzarev/a/pytorch/torch/fx/_lazy_graph_module.py class pack_float8(torch.nn.Module):
def forward(self, x_1: "f32[s0, s1][s1, 1]cuda:0"):
# No stacktrace found for following nodes
_to_copy: "f8e4m3fn[s0, s1][s1, 1]cuda:0" = torch.ops.aten._to_copy.default(x_1, dtype = torch.float8_e4m3fn); x_1 = None
return (torch.float32, _to_copy)
INFO: TRACED GRAPH
===== saved_tensors_unpack_hook add 3 =====
<eval_with_key>.22 from /data/users/ivankobzarev/a/pytorch/torch/fx/experimental/proxy_tensor.py:1225 in wrapped class pack_float8(torch.nn.Module):
def forward(self, x_1: "f32[s0, s1][s1, 1]cuda:0"):
# No stacktrace found for following nodes
_to_copy: "f8e4m3fn[s0, s1][s1, 1]cuda:0" = torch.ops.aten._to_copy.default(x_1, dtype = torch.float8_e4m3fn); x_1 = None
return (torch.float32, _to_copy)
INFO: TRACED GRAPH
===== Forward graph 3 =====
/data/users/ivankobzarev/a/pytorch/torch/fx/_lazy_graph_module.py class GraphModule(torch.nn.Module):
def forward(self, primals_1: "Sym(s0)", primals_2: "Sym(s1)", primals_3: "f32[s0, s1][s1, 1]cuda:0"):
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6660 in simple_fn, code: x = x + 1
add: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.add.Tensor(primals_3, 1); primals_3 = None
# No stacktrace found for following nodes
_to_copy: "f8e4m3fn[s0, s1][s1, 1]cuda:0" = torch.ops.aten._to_copy.default(add, dtype = torch.float8_e4m3fn)
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6661 in simple_fn, code: x = SAF.apply(x)
view: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.view.default(add, [primals_1, primals_2]); add = None
return (view, _to_copy, primals_1, primals_2)
INFO: TRACED GRAPH
===== Backward graph 3 =====
<eval_with_key>.21 class GraphModule(torch.nn.Module):
def forward(self, primals_1: "Sym(s0)", primals_2: "Sym(s1)", add_packed_2: "f8e4m3fn[s0, s1][s1, 1]cuda:0", tangents_1: "f32[s0, s1][s1, 1]cuda:0"):
# No stacktrace found for following nodes
_to_copy: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten._to_copy.default(add_packed_2, dtype = torch.float32); add_packed_2 = None
# File: /data/users/ivankobzarev/a/pytorch/test/functorch/test_aotdispatch.py:6661 in simple_fn, code: x = SAF.apply(x)
add_7: "f32[s0, s1][s1, 1]cuda:0" = torch.ops.aten.add.Tensor(tangents_1, _to_copy); tangents_1 = _to_copy = None
return (None, None, add_7)
```
Differential Revision: [D72187044](https://our.internmc.facebook.com/intern/diff/D72187044)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150032
Approved by: https://github.com/bdhirsh
Add ninja-build for pytorch tests.
Switch to gcc 14 due to fix for precompiled headers and s390x vectorization interaction.
Disable -Werror when building onnxruntime.
Pin onnx version.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153619
Approved by: https://github.com/huydhn
This avoid replaying load_input on a cache hit on the generate_code_cache.
the idea is that if a template have prologue_loads_all_inputs = True, it means that
all all inputs are loaded and hence no need to replay
Effect on the current benchmark on a local run on dev server.
18549985383 -> 15072230073
25697270062 -> 20738613297
Pull Request resolved: https://github.com/pytorch/pytorch/pull/150869
Approved by: https://github.com/eellison
Usage:
```python
from torch._higher_order_ops.wrap import dynamo_bypassing_wrapper
# Your ordinary function wrapper
def my_hop_fn_impl(fn, *args, k=1, **kwargs):
def wrapper(*args, **kwargs):
out = fn(*args, **kwargs)
if isinstance(out, tuple):
return (out[0] + k,)
return out + k
return wrapper
# Calling `my_hop_fn` instead of the impl directly captures a HOP into the dynamo graph
def my_hop_fn(fn, *args, k=1, **kwargs):
return dynamo_bypassing_wrapper(
functools.partial(my_hop_fn_impl, k=k), fn, *args, **kwargs
)
```
Notes:
- The dynamo captured graph now stashes arbitrary callable objects (the wrapper_fn) - this is equivalent to what SAC does today with policy_fn.
- The `wrapper_fn` passed to `dynamo_bypassing_wrapper ` should have signature `Callable -> Callable`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153487
Approved by: https://github.com/ydwu4
Graph partition relies on `read_writes` to collect partition inputs and outputs. There are three edge cases:
1. `NoneLayout` is not allocated so it cannot become a partition input or output.
2. Codegen may decide a buffer to be internal to a kernel (e.g., triton kernel). One example is some buffers internal to a FusedSchedulerNode. These buffers are never actually allocated as `buf_id`.
3. We should use mutation_real_name for graph partition inputs and outputs to match the behavior of other codegen.
This PR supports these 3 cases.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153899
Approved by: https://github.com/eellison
Previously, we didn't track the unbacked symbols leaked out of true_branch and false_branch if they have the same shape expr. This cause the the fake output of cond operator itself doesn't set up its unbacked_bindings meta properly (because they're ignored).
In this PR, we also check whether there're leaked out unbacked symbols and create new unbacked symbols for it and track it as output of cond.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/148206
Approved by: https://github.com/zou3519
## PR
There are a few cases that my previous PR (#153220) didn't cover.
1. The LHS/RHS matters. Today, if you do `torch._check(lhs == rhs)` then it will show up as a deferred runtime assert with `Eq(lhs, rhs)`.
2. There can be transitive replacements. For example, expr1 -> expr2 -> u0. `test_size_with_unbacked_add_expr_transitive` tests for this.
3. An unbacked symint expr may not have a replacement that's purely a symbol, for instance, it could be another expression. `test_size_with_unbacked_add_and_mul_expr` tests for this.
## Device assertion msg
```
/tmp/tmp07mu50tx/6y/c6ym2jzadwfigu3yexredb7qofviusz3p7ozcdjywvayhxgcqxkp.py:40: unknown: block: [8681,0,0], thread: [4,0,0] Assertion `index out of bounds: 0 <= tl.broadcast_to(tmp13, [XBLOCK]) < ks0` failed.
...
/tmp/tmp07mu50tx/6y/c6ym2jzadwfigu3yexredb7qofviusz3p7ozcdjywvayhxgcqxkp.py:40: unknown: block: [8681,0,0], thread: [6,0,0] Assertion `index out of bounds: 0 <= tl.broadcast_to(tmp13, [XBLOCK]) < ks0` failed.
```
## Autotuning code setup
This is the autotuning code for a concat kernel which takes input tensors (`in_buf`) and writes them to the (`out_buf`).
It's important to note the size of `in_buf0` is the same as `in_buf1` don't match along dim=0. This is bad because all concat inputs must share the same size for each dim except for the concat dim (here that's dim=1).
```
in_buf0 = generate_example_value(size=(u1 + s0, 256)) # concrete size is (17900, 256)
in_buf1 = generate_example_value(size=(u0, 10)) # concrete size is (8192, 10)
...
out_buf = generate_example_value(size=(u1 + s0, 266)) # concrete size is (17900, 256+10)
triton_poi_fused_cat_1.run(in_buf0, in_buf1, ..., out_buf, xnumel=(u1 + s0) * 266 ...)
```
If we look into the kernel code, you'll see that `tmp9` loads `in_buf1` (our incorrectly shaped input tensor). There is also a mask to prevent OOB loads.
- `tmp6` makes sure we're only loading with the `xindex` from 256 to 264.
- `xmask` makes sure we're only loading with the `xindex` within `xnumel`.
- `tmp6 & xmask` together is essentially checking `0 ≤ x0 < u1 + s0` and `256 ≤ x1 < 264`.
The mask logic is correct, however, `in_buf1` has the shape `[8192, 10]` this means any load where `8192 ≤ x0 < u1 + s0` will be an OOB load.
```
def triton_poi_fused_cat_1(in_buf0, in_buf1, ... out_buf, xnumel, XBLOCK):
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)
xmask = xindex < xnumel
x0 = (xindex % 264)
x1 = xindex // 264
...
tmp6 = x0 >= tl.full([1], value=256)
tmp9 = tl.load(in_buf1 + (x1), tmp6 & xmask)
# device assertion is thrown here
tl.device_assert(((0 <= tl.broadcast_to(tmp13, [XBLOCK])) & (tl.broadcast_to(tmp13, [XBLOCK]) < ks0)) | ~(xmask & tmp6), "index out of bounds: 0 <= tl.broadcast_to(tmp13, [XBLOCK]) < ks0")
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153768
Approved by: https://github.com/jingsh
https://github.com/pytorch/pytorch/pull/152708 expanded support of `get_estimated_runtime` to many more types of `SchedulerNodes`. This caused an increase in compile time because we're always calling `get_estimated_runtime` to populate the metrics table. This PR adds a flag for this logging, which reduces the instruction count by 8%. Long term, we should probably merge metrics.py with TORCH_LOGS/tlparse (suggestion from @xmfan).
Update: added support for TORCH_LOGS for the metrics logging.
Test Plan:
mm_loop.py and many existing tests cover.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/153506
Approved by: https://github.com/eellison
This PR adds code generation for CK-tile based universal gemm kernels to the CK backend for Inductor, and adds these kernels to autotune choices.
Unlike legacy-CK based kernels (which are generated by parsing the CK instances from CK library), we generate the set of instances by manually specifying the tuning parameters.
This PR introduces a new template for code generation, and compilation/autotuning is handled by the existing infrastructure.
Points of discussion:
* For simplicity and reduced coupling with CK, the instance filter checks only data type and layout, and doesn't check the alignment requirement - meaning that more instances will be compiled than necessary - while keeping the code generation independent from internal CK logic which checks the alignment validity at runtime
* CK-tile instances are enabled whenever legacy-CK instances are enabled. A config knob could be introduced to differentiate between the instance types if that's needed
* Whether gemm problem size K is ever dynamic, since whenever it's not a compile-time constant, we need to perform a runtime dispatch between several kernels
** Testing **
Use the existing tests in `test/inductor/test_ck_backend.py`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/152341
Approved by: https://github.com/chenyang78
Work around issues like #153960, #152623
NCCL 2.26 seems to introduce random hang in non-blocking API mode. This PR opts out of non-blocking mode to work around it. Previously torch turned it on by default in eager init (i.e. `device_id` passed) to avoid init overhead.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/154055
Approved by: https://github.com/atalman
