Add new_empty (with dtype argument only) to torch::stable #159508

mikaylagawarecki · 2025-07-30T20:47:49Z

Stack from ghstack (oldest at bottom):

…(pending header-onlyness) [ghstack-poisoned]

pytorch-bot · 2025-07-30T20:47:53Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/159508

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 188f656 with merge base a44a0d3 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

…(pending header-onlyness) ghstack-source-id: 6cde171 Pull Request resolved: #159508

…le::Tensor (pending header-onlyness)" needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

…(pending header-onlyness) ghstack-source-id: b06a267 Pull Request resolved: #159508

…le::Tensor (pending header-onlyness)" needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

ghstack-source-id: 4c53a24 Pull Request resolved: #159508

mikaylagawarecki · 2025-08-11T21:16:11Z

torchgen/gen_aoti_c_shim.py

+            if not has_function_variant:
+                # Functions with both function and method variants can use the at::{*}_symint version
+                # (e.g., narrow -> at::narrow_symint), BUT
+                # Method-only functions with symint parameters should use at::symint:: namespace
+                # Remove the _symint suffix since at::symint:: namespace uses the base name
+                # (e.g., new_empty -> at::symint::new_empty<c10::SymInt>)
+                base_name = cpp_sig.name()
+                base_name = base_name.removesuffix("_symint")  # Remove "_symint" suffix
+                return f"at::symint::{base_name}<c10::SymInt>"
+


This is the relevant part here

pytorch/torchgen/gen.py

Lines 717 to 739 in 6d91d6d

if Variant.function in f.variants:

result += f"""

// aten::{f.func}

inline {sig.decl()} {{

return at::_ops::{f.func.name.unambiguous_name()}::call({exprs_str});

}}"""

# The template function can be used from template situations

# where you want to switch between the symint or not version

# depending on a template argument

#

# NB: we ALWAYS generate this even for methods. But we put it in

# this header so it can take advantage of per-op headers

if has_symint:

result += f"""

namespace symint {{

template <typename T, typename = std::enable_if_t<std::is_same_v<T, {intlike_t}>>>

{sig.decl(suppress_symint_suffix=True)} {{

return at::_ops::{f.func.name.unambiguous_name()}::call({exprs_str});

}}

}}

"""

return result

needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

ghstack-source-id: ac037b9 Pull Request resolved: #159508

janeyx99 · 2025-08-12T18:16:04Z

torch/csrc/stable/ops.h

+  // Handle dtype - use input tensor's dtype if not specified
+  int32_t target_dtype;
+  if (dtype.has_value()) {
+    target_dtype = static_cast<int32_t>(dtype.value());


hmmm lol is this actually stable...this looks to expose a detail on how dtype is encapsulated. Should we/I write that translation layer from the headeronly ScalarType to their corresponding shim aoti_torch_get_dtype... maybe there's no other way to hide this

Yes, it would be nice if you could write that!

Can you describe what you mean in more detail of what the translation layer would do

like a pair of utils that take in and return ScalarType::BFloat`6 -> aoti_torch_dtype_bfloat16 and vice versa. I'll prioritize writing this (i think it;d be helpful for my Tensor scalar_type PR too)

torch/csrc/stable/ops.h

janeyx99 · 2025-08-12T18:23:49Z

torchgen/gen_aoti_c_shim.py

    cpp_sig = gen_static_dispatch_backend_call_signature(sig, f)
+
    if backend_index is None:
+        # Check if this is a symint function and if the function only has method variants


Are there any other ops we expect to go through this branch? Maybe list new_empty as an example here?

Hm it's listed below (when all the if checks pass) on 412

janeyx99

Looks fine to me, this is making me think i need to land my scalartype liaisoning asap so we aren't exposing weird details by accident

torchgen/gen_aoti_c_shim.py

test/cpp_extensions/libtorch_agnostic_extension/test/test_libtorch_agnostic.py

test/cpp_extensions/libtorch_agnostic_extension/libtorch_agnostic/ops.py

test/cpp_extensions/libtorch_agnostic_extension/libtorch_agnostic/csrc/kernel.cpp

[ghstack-poisoned]

ghstack-source-id: 4c8f1d6 Pull Request resolved: #159508

[ghstack-poisoned]

ghstack-source-id: d64e073 Pull Request resolved: #159508

mikaylagawarecki · 2025-08-19T22:29:49Z

@pytorchbot merge

pytorchmergebot · 2025-08-19T22:33:02Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

…9508) Pull Request resolved: pytorch#159508 Approved by: https://github.com/janeyx99 ghstack dependencies: pytorch#160557

Add new_empty to torch::stable + dtype/scalar_type to stable::Tensor …

cbcdcfe

…(pending header-onlyness) [ghstack-poisoned]

This was referenced Jul 30, 2025

Update torch::stable::Tensor() default constructor #159507

Closed

Add pad and narrow to torch/csrc/stable/ops.h #159328

Closed

mikaylagawarecki added a commit that referenced this pull request Jul 30, 2025

Add new_empty to torch::stable + dtype/scalar_type to stable::Tensor …

2ec4dcd

…(pending header-onlyness) ghstack-source-id: 6cde171 Pull Request resolved: #159508

mikaylagawarecki mentioned this pull request Jul 31, 2025

ABI stable fa3 Dao-AILab/flash-attention#1791

Merged

Update on "Add new_empty to torch::stable + dtype/scalar_type to stab…

6fced91

…le::Tensor (pending header-onlyness)" needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Aug 4, 2025

Add new_empty to torch::stable + dtype/scalar_type to stable::Tensor …

90be175

…(pending header-onlyness) ghstack-source-id: b06a267 Pull Request resolved: #159508

Update on "Add new_empty to torch::stable + dtype/scalar_type to stab…

648adaa

…le::Tensor (pending header-onlyness)" needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Aug 11, 2025

Add new_empty to torch::stable

0a80862

ghstack-source-id: 4c53a24 Pull Request resolved: #159508

pytorch-bot bot added ciflow/inductor release notes: inductor (aoti) labels Aug 11, 2025

mikaylagawarecki commented Aug 11, 2025

View reviewed changes

mikaylagawarecki changed the title ~~Add new_empty to torch::stable + dtype/scalar_type to stable::Tensor (pending header-onlyness)~~ Add new_empty (with dtype argument only) to torch::stable Aug 11, 2025

Update on "Add new_empty (with dtype argument only) to torch::stable"

ae40182

needs ScalarType.h, which needs to be header-only [ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Aug 11, 2025

Add new_empty to torch::stable

96d99d6

ghstack-source-id: ac037b9 Pull Request resolved: #159508

mikaylagawarecki marked this pull request as ready for review August 11, 2025 21:31

mikaylagawarecki requested a review from janeyx99 as a code owner August 11, 2025 21:31

janeyx99 reviewed Aug 12, 2025

View reviewed changes

torch/csrc/stable/ops.h Outdated Show resolved Hide resolved

janeyx99 reviewed Aug 12, 2025

View reviewed changes

Update on "Add new_empty (with dtype argument only) to torch::stable"

3d7d239

[ghstack-poisoned]

mikaylagawarecki mentioned this pull request Aug 19, 2025

Add ScalarType -> shim conversion, add stable::Tensor.scalar_type #160557

Closed

mikaylagawarecki added a commit that referenced this pull request Aug 19, 2025

Add new_empty to torch::stable

a5f8cf1

ghstack-source-id: 4c8f1d6 Pull Request resolved: #159508

mikaylagawarecki requested a review from janeyx99 August 19, 2025 20:51

Update on "Add new_empty (with dtype argument only) to torch::stable"

188f656

[ghstack-poisoned]

mikaylagawarecki added a commit that referenced this pull request Aug 19, 2025

Add new_empty to torch::stable

d3a40ad

ghstack-source-id: d64e073 Pull Request resolved: #159508

mikaylagawarecki added the ciflow/trunk Trigger trunk jobs on your pull request label Aug 19, 2025

janeyx99 approved these changes Aug 19, 2025

View reviewed changes

mikaylagawarecki added the release notes: build release notes category label Aug 19, 2025

pytorchmergebot added the merging label Aug 19, 2025

pytorchmergebot closed this in 78a8e6a Aug 20, 2025

pytorchmergebot added Merged and removed merging labels Aug 20, 2025

github-actions bot deleted the gh/mikaylagawarecki/331/head branch September 19, 2025 02:11

	if Variant.function in f.variants:
	result += f"""
	// aten::{f.func}
	inline {sig.decl()} {{
	return at::_ops::{f.func.name.unambiguous_name()}::call({exprs_str});
	}}"""

	# The template function can be used from template situations
	# where you want to switch between the symint or not version
	# depending on a template argument
	#
	# NB: we ALWAYS generate this even for methods. But we put it in
	# this header so it can take advantage of per-op headers
	if has_symint:
	result += f"""
	namespace symint {{
	template <typename T, typename = std::enable_if_t<std::is_same_v<T, {intlike_t}>>>
	{sig.decl(suppress_symint_suffix=True)} {{
	return at::_ops::{f.func.name.unambiguous_name()}::call({exprs_str});
	}}
	}}
	"""
	return result

Add new_empty (with dtype argument only) to torch::stable #159508

Add new_empty (with dtype argument only) to torch::stable #159508

Uh oh!

Conversation

mikaylagawarecki commented Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Jul 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/159508

✅ No Failures

Uh oh!

mikaylagawarecki Aug 11, 2025

Choose a reason for hiding this comment

Uh oh!

janeyx99 Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

mikaylagawarecki Aug 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

janeyx99 Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

janeyx99 Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

mikaylagawarecki Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

janeyx99 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mikaylagawarecki commented Aug 19, 2025

Uh oh!

pytorchmergebot commented Aug 19, 2025

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

mikaylagawarecki commented Jul 30, 2025 •

edited

Loading

pytorch-bot bot commented Jul 30, 2025 •

edited

Loading

mikaylagawarecki Aug 12, 2025 •

edited

Loading