Make reduction bound temporaries use PRIVATE address space

[Copilot]

Scalar temporaries generated to store non-affine reduction bounds were allocated in GLOBAL address space, causing race conditions when parallel threads share those locations.

Changes

• get_loopy_temporary: Add address_space parameter (default lp.AddressSpace.GLOBAL preserves existing behavior).
• add_store: Add address_space parameter, forwarded to get_loopy_temporary when output_to_temporary=True.
• Reduction bounds call site: Pass address_space=lp.AddressSpace.PRIVATE when storing non-affine reduction bound scalars, giving each thread its own copy.

# Before: bound temps were GLOBAL — shared across threadsbound_result=StoredResult(bound_name, 0, frozenset([
add_store(..., output_to_temporary=True, ...)]))
# After: bound temps are PRIVATE — thread-localbound_result=StoredResult(bound_name, 0, frozenset([
add_store(..., output_to_temporary=True, ...,
address_space=lp.AddressSpace.PRIVATE)]))

• Test: test_reduction_bound_temps_are_private uses a CSR matrix multiply (which triggers non-affine bounds) and asserts all generated *_lbound/*_ubound temporaries have PRIVATE address space.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>Temporaries for bounds should not be global</issue_title>
<issue_description>In loopy code generation for reduction bounds (see pytato/target/loopy/codegen.py around line 570), bounds get stored in what turn out to be global temporaries. That's likely to create race conditions; those temporaries should be private.

cc @majosm

x-ref: inducer/arraycontext#349 (comment)</issue_description>

<agent_instructions>Please create a plan for how this could be fixed. Likely, add_store will need an argument indicating the address space.</agent_instructions>

Comments on the Issue (you are @copilot in this section)

• FixesTemporaries for bounds should not be global #648

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.

[Copilot]

Pull request overview

This PR adjusts Pytato’s Loopy code generation so that scalar temporaries created for non-affine reduction bounds are placed in PRIVATE address space (instead of GLOBAL), preventing races in parallel kernels. It also adds a regression test and updates Sphinx reference aliasing for the new Loopy symbol.

Changes:

• Generate reduction-bound temporaries with lp.AddressSpace.PRIVATE to avoid cross-thread races.
• Extend add_store/get_loopy_temporary to accept an address_space parameter (defaulting to GLOBAL for existing behavior).
• Add a regression test asserting reduction-bound temps use PRIVATE address space; add a Sphinx missing-reference alias for lp.AddressSpace.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
pytato/target/loopy/codegen.py	Adds address_space plumbing and uses PRIVATE for reduction bound temporaries.
test/test_codegen.py	Adds regression test checking address space of reduction-bound temporaries.
doc/conf.py	Adds Sphinx missing-reference alias for lp.AddressSpace.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.