**Summary**
Per the discussion in https://github.com/pytorch/pytorch/pull/123444, the `decomposed quant/dequant` patterns changed after https://github.com/pytorch/pytorch/pull/123445, we can move the optimization of `decomposed quant/dequant` from inductor decomposition into lowering phase to avoid the changes. In this way, we can:
- Avoid the pattern matcher failure introduced in https://github.com/pytorch/pytorch/pull/123445
- Make the quantization pattern clearer in the pattern matcher phase, since the `quant/dequant` nodes have not been decomposed.
**Changes in this PR**
- Move optimization of `decomposed quant/dequant` from inductor decomposition into lowering phase.
- Corresponding changes in the quantization pattern matcher to ensure no bc-breaking.
**TestPlan**
```
python -u -m pytest -s -v test/inductor/test_mkldnn_pattern_matcher.py -k test_q
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124041
Approved by: https://github.com/peterbell10, https://github.com/jgong5
**Summary**
Per the discussion in https://github.com/pytorch/pytorch/pull/123444, the `decomposed quant/dequant` patterns changed after https://github.com/pytorch/pytorch/pull/123445, we can move the optimization of `decomposed quant/dequant` from inductor decomposition into lowering phase to avoid the changes. In this way, we can:
- Avoid the pattern matcher failure introduced in https://github.com/pytorch/pytorch/pull/123445
- Make the quantization pattern clearer in the pattern matcher phase, since the `quant/dequant` nodes have not been decomposed.
**Changes in this PR**
- Move optimization of `decomposed quant/dequant` from inductor decomposition into lowering phase.
- Corresponding changes in the quantization pattern matcher to ensure no bc-breaking.
**TestPlan**
```
python -u -m pytest -s -v test/inductor/test_mkldnn_pattern_matcher.py -k test_q
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124041
Approved by: https://github.com/peterbell10, https://github.com/jgong5
Summary:
Added support for quantized linear on CPU with fbgemm.
Specifically, for torch.ops.quantized.linear_unpacked_dynamic_fp16, we
decompose it into two steps, pack weight, and fbgemm's qlinear with
packed weight.
Test Plan:
Included in commit.
test_aot_inductor::test_quantized_linear
Reviewers:
Subscribers:
Tasks:
Tags:
Differential Revision: [D55577959](https://our.internmc.facebook.com/intern/diff/D55577959)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/123069
Approved by: https://github.com/hl475
**Summary:**
This commit simplifies the existing decomposition hierarchy
of batch norm ops by adding a single, backend agnostic op:
`batch_norm_with_update`. The existing hierarchy looks like:
```
aten.batch_norm ->
aten._batch_norm_impl_index ->
[
aten.native_batch_norm ->
aten._native_batch_norm_legit (export only) ->
_batch_norm_legit_cpu/cuda (kernels, export only) ->
_batch_norm_cpu/cuda (kernels)
] OR
[ aten.cudnn_batch_norm ] OR
[ aten.miopen_batch_norm ]
```
Aside from complexity, an important problem with the
above decomposition hierarchy is cuda numerics in
export flows. We observed significantly worse convergence
when training a mobilenetv2-like model when using the
`_batch_norm_cuda` kernel instead of the `cudnn_batch_norm`
kernel. This means users who export their models on CPU
first then move the models to cuda later may silently
see worse accuracies even when cudnn is installed,
because they are using the worse kernel. This issue is
summarized in https://github.com/pytorch/pytorch/issues/111384.
Instead, the new hierarchy proposed by consolidating
existing batch norm ops will look like:
```
aten.batch_norm ->
aten.batch_norm_with_update ->
[ _batch_norm_cpu (kernel) ] OR
[ _batch_norm_cuda (kernel) ] OR
[ cudnn_batch_norm (kernel) ] OR
[ miopen_batch_norm (kernel) ]
```
The new op `batch_norm_with_update` hides backend
implementation details and automatically picks the right
kernel based on what is installed. This commit also adds
the following variants to this op:
```
batch_norm_with_update_functional
batch_norm_with_update.out
batch_norm_no_update
batch_norm_no_update.out
batch_norm_backward
```
Note that this commit only adds this op and its variants,
but does not actually change the decomps to produce these
ops in the graph. This will be done after the 2 week FC
window, and the ops used in the old stack is planned to
be removed after the 6 month BC window.
Test Plan: `OpInfo` tests for `batch_norm_with_update`.
Reviewers: albanD, bdhirsh
Subscribers: albanD, bdhirsh, supriyar
Tasks: https://github.com/pytorch/pytorch/issues/111384
Differential Revision: [D54805279](https://our.internmc.facebook.com/intern/diff/D54805279)
Co-authored-by: Tugsbayasgalan Manlaibaatar <tmanlaibaatar@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/116092
Approved by: https://github.com/bdhirsh, https://github.com/albanD
**Summary:**
This commit simplifies the existing decomposition hierarchy
of batch norm ops by adding a single, backend agnostic op:
`batch_norm_with_update`. The existing hierarchy looks like:
```
aten.batch_norm ->
aten._batch_norm_impl_index ->
[
aten.native_batch_norm ->
aten._native_batch_norm_legit (export only) ->
_batch_norm_legit_cpu/cuda (kernels, export only) ->
_batch_norm_cpu/cuda (kernels)
] OR
[ aten.cudnn_batch_norm ] OR
[ aten.miopen_batch_norm ]
```
Aside from complexity, an important problem with the
above decomposition hierarchy is cuda numerics in
export flows. We observed significantly worse convergence
when training a mobilenetv2-like model when using the
`_batch_norm_cuda` kernel instead of the `cudnn_batch_norm`
kernel. This means users who export their models on CPU
first then move the models to cuda later may silently
see worse accuracies even when cudnn is installed,
because they are using the worse kernel. This issue is
summarized in https://github.com/pytorch/pytorch/issues/111384.
Instead, the new hierarchy proposed by consolidating
existing batch norm ops will look like:
```
aten.batch_norm ->
aten.batch_norm_with_update ->
[ _batch_norm_cpu (kernel) ] OR
[ _batch_norm_cuda (kernel) ] OR
[ cudnn_batch_norm (kernel) ] OR
[ miopen_batch_norm (kernel) ]
```
The new op `batch_norm_with_update` hides backend
implementation details and automatically picks the right
kernel based on what is installed. This commit also adds
the following variants to this op:
```
batch_norm_with_update_functional
batch_norm_with_update.out
batch_norm_no_update
batch_norm_no_update.out
batch_norm_backward
```
Note that this commit only adds this op and its variants,
but does not actually change the decomps to produce these
ops in the graph. This will be done after the 2 week FC
window, and the ops used in the old stack is planned to
be removed after the 6 month BC window.
Test Plan: `OpInfo` tests for `batch_norm_with_update`.
Reviewers: albanD, bdhirsh
Subscribers: albanD, bdhirsh, supriyar
Tasks: https://github.com/pytorch/pytorch/issues/111384
Co-authored-by: Tugsbayasgalan Manlaibaatar <tmanlaibaatar@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/116092
Approved by: https://github.com/bdhirsh, https://github.com/albanD
**Summary:**
This commit simplifies the existing decomposition hierarchy
of batch norm ops by adding a single, backend agnostic op:
`batch_norm_with_update`. The existing hierarchy looks like:
```
aten.batch_norm ->
aten._batch_norm_impl_index ->
[
aten.native_batch_norm ->
aten._native_batch_norm_legit (export only) ->
_batch_norm_legit_cpu/cuda (kernels, export only) ->
_batch_norm_cpu/cuda (kernels)
] OR
[ aten.cudnn_batch_norm ] OR
[ aten.miopen_batch_norm ]
```
Aside from complexity, an important problem with the
above decomposition hierarchy is cuda numerics in
export flows. We observed significantly worse convergence
when training a mobilenetv2-like model when using the
`_batch_norm_cuda` kernel instead of the `cudnn_batch_norm`
kernel. This means users who export their models on CPU
first then move the models to cuda later may silently
see worse accuracies even when cudnn is installed,
because they are using the worse kernel. This issue is
summarized in https://github.com/pytorch/pytorch/issues/111384.
Instead, the new hierarchy proposed by consolidating
existing batch norm ops will look like:
```
aten.batch_norm ->
aten.batch_norm_with_update ->
[ _batch_norm_cpu (kernel) ] OR
[ _batch_norm_cuda (kernel) ] OR
[ cudnn_batch_norm (kernel) ] OR
[ miopen_batch_norm (kernel) ]
```
The new op `batch_norm_with_update` hides backend
implementation details and automatically picks the right
kernel based on what is installed. This commit also adds
the following variants to this op:
```
batch_norm_with_update_functional
batch_norm_with_update.out
batch_norm_no_update
batch_norm_no_update.out
batch_norm_backward
```
Note that this commit only adds this op and its variants,
but does not actually change the decomps to produce these
ops in the graph. This will be done after the 2 week FC
window, and the ops used in the old stack is planned to
be removed after the 6 month BC window.
Test Plan: `OpInfo` tests for `batch_norm_with_update`.
Reviewers: albanD, bdhirsh
Subscribers: albanD, bdhirsh, supriyar
Tasks: https://github.com/pytorch/pytorch/issues/111384
Co-authored-by: Tugsbayasgalan Manlaibaatar <tmanlaibaatar@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/116092
Approved by: https://github.com/bdhirsh, https://github.com/albanD
**description**
Enable lowering of dynamic qlinear for X86Inductor. The pattern is `choose_qparams -> getitem -> q -> dq -> linear`. We only fuse `dq -> linear` and get `choose_qparams -> getitem -> q -> onednn.qlinear_pointwise`. So, we treat it as dynamic quantization of activation + static quantized linear.
The previous implementation of `onednn.qlinear_pointwise` is for the case where `x_scale` and `x_zp` are scalars. Since `choose_qparams` returns tensors, we added a variation `onednn.qlinear_pointwise.tensor` to support the case.
This feature is targeting PyTorch 2.3 release.
**Test plan**
```
python inductor/test_mkldnn_pattern_matcher.py -k test_dynamic_qlinear_cpu
python inductor/test_mkldnn_pattern_matcher.py -k test_dynamic_qlinear_qat_cpu
python inductor/test_cpu_cpp_wrapper.py -k test_dynamic_qlinear
```
**Performance before and after lowering `choose_qparam` to Inductor**
Before
- latency for shape (32, 32) = 0.151 ms
latency for shape (128, 128) = 0.153 ms
latency for shape (1024, 1024) = 0.247 ms
After
- latency for shape (32, 32) = 0.049 ms
- latency for shape (128, 128) = 0.052 ms
- latency for shape (1024, 1024) = 0.133 ms
Test method: A module with a single Linear layer, dynamic-quantize, lower to X86Inductor
Test env & config: Intel(R) Xeon(R) Platinum 8358 CPU @ 2.60GHz, single instance, single core, using Intel OpenMP and Tcmalloc
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120605
Approved by: https://github.com/leslie-fang-intel, https://github.com/jgong5, https://github.com/jerryzh168
Inductor codegen for `_assert_async` is currently disabled because we don't really understand how to codegen `scalar_to_tensor` on a Sympy expression. I initially tried to see if I could get this to work, but I got into some weird problem involving stride sorting, so I decided to fix it properly by not going through a tensor.
So we introduce an `_assert_scalar` which takes a scalar as an argument, avoiding needing to turn a SymBool into a tensor before asserting on it. I also add `_functional_assert_scalar` for good luck, although this doesn't do anything right now because https://github.com/pytorch/pytorch/pull/104203 still hasn't been landed.
I need to customize the codegen for this operator, so I decide to directly implement it in Inductor, rather than trying to treat it as a generic ExternKernel. This leads to the new AssertScalar IR node. This is written carefully so that it doesn't get DCE'd by Inductor.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114148
Approved by: https://github.com/jansel
ghstack dependencies: #120800
By changing runtime symbolic asserts to using assert_scalar, the asserts can call into `expect_true` and modify the shape env so that we can run through the traced graph module with fake tensors. With assert_async, the asserts only get hit during runtime, but that means if we run the graph module with fake tensors, the asserts will not affect the shape env, so later data dependent calls to the fake tensors may result in GuardOnDataDependentSymNode errors.
https://github.com/pytorch/pytorch/issues/119587
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119608
Approved by: https://github.com/ezyang
By changing runtime symbolic asserts to using assert_scalar, the asserts can call into `expect_true` and modify the shape env so that we can run through the traced graph module with fake tensors. With assert_async, the asserts only get hit during runtime, but that means if we run the graph module with fake tensors, the asserts will not affect the shape env, so later data dependent calls to the fake tensors may result in GuardOnDataDependentSymNode errors.
https://github.com/pytorch/pytorch/issues/119587
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119608
Approved by: https://github.com/ezyang
Fixes https://github.com/pytorch/pytorch/issues/117361
The implementation here slightly diverges from what was proposed in the issue, so I will recap what this PR is doing here. Today, when doing computations involving size-like unbacked SymInts, we assume for all operations that the compile time range of the integer is `[2, inf]`, even though at runtime we also accept zero and one.
This PR removes the carte blanche assumption, and instead does the analysis in a much more limited and controlled fashion: only for guards which we have designated as "size oblivious" are we willing to do the analysis under the assumption that the range of all size-like unbacked SymInts is `[2, inf]`; otherwise, we will faithfully only do analysis with `[0, inf]` (or whatever the user provided) bounds.
The infra pieces of this PR are:
* Remove runtime_var_to_range from torch/fx/experimental/symbolic_shapes.py; modify `_constrain_range_for_size` to refine the range without clamping min to 2, and instead add the symbol to a `size_like` set in the ShapeEnv
* When evaluating an expression, if the expression is requested to be evaluated in a `size_oblivious` way, we attempt to statically compute the value of the expression with the assumption that all symbols in `size_like` are updated to assume that they are `>= 2`.
* Add Python and C++ APIs for guarding on a SymBool in a size-oblivious way. In C++, I also need to add some helpers for performing symbolic comparisons, since the stock comparisons immediately specialize in the "normal" way.
The rest of the changes of the PR are marking various spots in PyTorch framework code as size oblivious, based on what our current test suite exercises.
As you review the places where we have marked things as size oblivious, it may become clear why I ended up not opting for the "designate a branch as the default branch when it's not statically obvious which way to go": for some of the conditions, this answer is rather non-obvious. I think potentially there is another refinement on top of this PR, which is something like "I don't care if you can't figure it out with ValueRange analysis, go down this path anyway if there are unbacked sizes involved." But even if we add this API, I think we are obligated to attempt the ValueRange analysis first, since it can lead to better outcomes sometimes (e.g., we are able to figure out that something is contiguous no matter what the unbacked size is.)
When is it permissible to mark something as size oblivious? Heuristically, it is OK anywhere in framework code if it gets you past a guard on unbacked SymInt problem. It is somewhat difficult to provide a true semantic answer, however. In particular, these annotations don't have any observational equivalence guarantee; for example, if I have `torch.empty(u0, 1).squeeze()`, we will always produce a `[u0]` size tensor, even though if `u0 == 1` PyTorch will actually produce a `[]` size tensor. The argument that I gave to Lezcano is that we are in fact defining an alternate semantics for a "special" size = 0, 1, for which we have these alternate eager mode semantics. In particular, suppose that we have a constant `special1` which semantically denotes 1, but triggers alternate handling rules. We would define `torch.empty(special1, 1).squeeze()` to always produce a `[special1]` size tensor, making its semantics coincide with unbacked SymInt semantics. In this model, the decision to designate guards as size oblivious is simply a user API question: you put them where ever you need some handling for special1! As we conservatively error out whenever it is not obvious what `special1` semantics should be, it is always valid to expand these semantics to cover more cases (although you can always choose the wrong semantics!)
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118579
Approved by: https://github.com/eellison, https://github.com/lezcano
Summary:
A follow up for #117097. In that PR I didn't add
`_scaled_dot_product_attention_for_cpu` into the core_aten_decomposition
table. This PR does that and also add a unit test.
Test Plan: python test/export/test_export.py -k
test_scaled_dot_product_attention
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/117390
Approved by: https://github.com/drisspg
Summary: Previously, when the first tensor argument to `aten.cat` was empty and there was only one non-empty tensor argument, the first (empty) tensor was erroneously returned by the `aten.cat` decomposition. Here we fix the bug.
Test Plan:
```
$ python test/inductor/test_torchinductor.py -k test_cat_empty
...
----------------------------------------------------------------------
Ran 2 tests in 5.760s
OK
```
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/113514
Approved by: https://github.com/jansel
Tracks https://github.com/pytorch/pytorch/issues/98161
Complex number support in Pytorch isn't ideal today as complex operations will mostly end up taken care of by the aten runtime, except for `torch.angle` which is handled in [105609](https://github.com/pytorch/pytorch/pull/105609). In general a better way to handle that could be to decompose complex operations first so that more opportunities for fusion could be unveiled, and then to have Triton take care of non-continuous (strided) tensor operations more efficiently. This change adds support to decompose complex addtions.
```
@triton.jit
def triton_(in_ptr0, in_ptr1, out_ptr0, xnumel, XBLOCK : tl.constexpr):
xnumel = 6
xoffset = tl.program_id(0) * XBLOCK
xindex = xoffset + tl.arange(0, XBLOCK)[:]
xmask = xindex < xnumel
x0 = xindex
tmp0 = tl.load(in_ptr0 + (x0), xmask)
tmp1 = tl.load(in_ptr1 + (x0), xmask)
tmp2 = tmp0 + tmp1
tl.store(out_ptr0 + (x0), tmp2, xmask)
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110740
Approved by: https://github.com/jansel
## Context
Add decompositions for `aten.max`, `aten.min`, and `aten.var_mean`. These operators follow a pattern of returning a tuple of outputs from two component operators:
```
aten.max(x) -> return aten.amax(x), aten.argmax(x)
aten.min(x) -> return aten.amin(x), aten.argmin(x)
aten.var_mean(x) -> return aten.var(x), aten.mean(x)
```
For `var_mean`, the `refs` implementation was doing something similar, so I changed it to call `torch.` ops instead like was done for other `refs` implementations previously. cc: @peterbell10 @lezcano
Note that Inductor lowers all these directly, so they are excluded from the Inductor decomp table.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110906
Approved by: https://github.com/manuelcandales
Summary:
## Context
Both `aten.sum` and `aten.squeeze` have a "most generic" variant in the form of `aten.sum.dim_IntList` and `aten.squeeze.dims` respectively. Add decompositions for other non generic variants of these operators to express them using the most generic variant.
Note that to register these decomps, the reference implementation under `_refs` had to be removed as registered decompositions. cc: @lezcano @peterbell10
Test Plan: Github CI + Meta Internal CI
Differential Revision: D49965952
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110645
Approved by: https://github.com/peterbell10, https://github.com/digantdesai, https://github.com/manuelcandales
Partially fixes `test_memory_format_factory_like_functions_preserve` with PYTORCH_TEST_WITH_INDUCTOR. Inductor preserves memory layouts for user-visible outputs as annotated on the fx graph that it is passed in. That graph is generated from running aot_autograd with decompositions. If the decompositions give incorrect strides, so will inductor.
This preserves the layout of `_like` operators when it corresponds to a `torch.memory_format`. It doesnt fix a) arbitrary permutations, b) striding of non-dense outputs. Both of these are lower-pri compared to preserving channels last. We would need either https://github.com/pytorch/pytorch/issues/92920 or a `to` variant that takes in a physical layout arbitrary permutations. I converted the output of rand to the correct layout instead of passing the layout in so that this would compose with the `replace_random` pass, and because the two pointwise ops will get fused anyway.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110242
Approved by: https://github.com/int3
Example of when the `evict_first` heuristic helps.
```
@torch.compile
def f(a, b):
return (a * b).sum(dim=-1)
N = 512
inps = (torch.randn(N, N, N).permute(2, 1, 0), torch.randn(N, N, N).permute(1, 2, 0))
from torch._inductor.utils import do_bench
print(do_bench(lambda: f(*inps)))
```
This generates code like this: http://ix.io/4HFs
```
Original: 3.8 ms
This PR: 3.54 ms
Always `evict_first: 5.4ms
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108841
Approved by: https://github.com/lezcano, https://github.com/jansel
