[Testing] Split up scan tests

[danhoeflinger]

This PR attempts to make the scan tests more manageable by breaking the current 2 tests into 8 tests.

The number of complicated kernels in individual tests is too large without this split causing issues with windows dpcpp debug tests. This also makes test results much more clear about what is failing when something does fail.

It should keep the same coverage with 3 caveats:

1. fixes a bug in what used to be scan_2.pass (now utils_scan.h) where keys_last + n was used as the end of the sequence for a serial call (should be keys_first + n).
2. Shifts the size of the "large" scan case down to 1M from 70M for debug cases. Its simply too large for our test suite.
3. _PSTL_TEST_INCLUSIVE/EXCLUSIVE_SCAN ifdefs removed (each file tests one variant)

Here is a map of how things moved around:

• scan.pass.cpp (deleted) -> split into:

  • test/support/utils_scan.h - helpers for both inclusive / exclusive tests
  • inclusive_scan.pass.cpp - test_with_plus (inclusive)
  • inclusive_scan_matrix.pass.cpp - test_matrix (inclusive binary_op)
  • inclusive_scan_multiplies.pass.cpp - test_with_multiplies (inclusive)
  • exclusive_scan.pass.cpp - test_with_plus (exclusive), added to existing file
  • exclusive_scan_matrix.pass.cpp - test_matrix (exclusive binary_op)
  • exclusive_scan_multiplies.pass.cpp - test_with_multiplies (exclusive)

• scan_2.pass.cpp (renamed) -> split into:

  • test/support/utils_scan.h - helpers for both inclusive / exclusive tests
  • exclusive_scan_usm.pass.cpp - exclusive USM in-place/non-inplace tests (kept)
  • inclusive_scan_usm.pass.cpp - inclusive USM in-place/non-inplace tests (new)

[Copilot]

Pull request overview

This PR refactors the numeric scan tests to be more manageable by splitting the prior combined coverage into multiple focused test files (inclusive/exclusive, plus/binary-op/multiplies, and USM), while also fixing a serial-range bug and reducing the debug “large” test size.

Changes:

• Split the former combined scan tests into separate inclusive/exclusive test executables (plus, matrix/binary-op, multiplies, and USM variants).
• Extracted shared USM scan test helpers into test/support/utils_scan.h and fixed the serial expected-range end iterator (keys_first + n).
• Reduced the debug large-case item count to 1M and removed inclusive/exclusive ifdef switching in favor of per-file variants.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
test/support/utils_scan.h	Convert former USM scan test into reusable helper header; fix expected-range end iterator.
test/parallel_api/numeric/numeric.ops/scan.pass.cpp	Delete old combined inclusive/exclusive scan test.
test/parallel_api/numeric/numeric.ops/inclusive_scan.pass.cpp	New inclusive “plus” scan coverage (host + hetero large case).
test/parallel_api/numeric/numeric.ops/inclusive_scan_matrix.pass.cpp	New inclusive binary-op “matrix” scan coverage (incl. in-place).
test/parallel_api/numeric/numeric.ops/inclusive_scan_multiplies.pass.cpp	New inclusive multiplies scan coverage (hetero).
test/parallel_api/numeric/numeric.ops/inclusive_scan_usm.pass.cpp	New inclusive USM scan coverage using shared helpers.
test/parallel_api/numeric/numeric.ops/exclusive_scan.pass.cpp	Add exclusive “plus” scan coverage alongside existing USM tests.
test/parallel_api/numeric/numeric.ops/exclusive_scan_matrix.pass.cpp	New exclusive binary-op “matrix” scan coverage (incl. in-place).
test/parallel_api/numeric/numeric.ops/exclusive_scan_multiplies.pass.cpp	New exclusive multiplies scan coverage (hetero).
test/parallel_api/numeric/numeric.ops/exclusive_scan_usm.pass.cpp	New exclusive USM scan coverage using shared helpers.

---

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

[SergeyKopienko]

@danhoeflinger can you confirm that all large submitters will continue using with the new data size in debug mode?

[danhoeflinger]

@danhoeflinger can you confirm that all large submitters will continue using with the new data size in debug mode?

@SergeyKopienko
Yes, for scan, the sizes for the large submitter are pretty small: either 2048 or 16k (depending on reduce_then_scan or not, or after my PRs ... CPU vs GPU). 1M elements will still hit the large submitters, as will some of the tests which are on the way to 100K in the main test search space.

[SergeyKopienko]

LGTM

[SergeyKopienko]

LGTM