Fixed issue where row reductions added an extra unnecessary wrapper

[cliffburdick]

No description provided.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[cliffburdick]

/build

[greptile-apps]

Greptile Summary

This PR corrects the CUB version guard from the broken CUB_MAJOR_VERSION >= 3 && CUB_MINOR_VERSION >= 2 (already flagged in a prior review thread) to the semantically correct CUB_MAJOR_VERSION > 3 || (CUB_MAJOR_VERSION == 3 && CUB_MINOR_VERSION >= 2) across all four affected reduction functions. It also adds the contiguous-tensor fast path (direct pointer, fixed segment size) to ExecMin and ExecMax, matching the pattern already present in ExecReduce and ExecSum, and backs the new paths with comprehensive segmented reduction tests.

Confidence Score: 5/5

Safe to merge — the version guard fix is correct and the new fast paths mirror the already-reviewed ExecReduce/ExecSum pattern.

All previously flagged P1 issues (incorrect CUB version guard in ExecMin and ExecMax) are now resolved with the correct > 3 || (== 3 && >= 2) expression. No new logic or correctness issues were found. The new tests cover the relevant shapes and validate expected values correctly.

No files require special attention.

Vulnerabilities

No security concerns identified.

Important Files Changed

Filename	Overview
include/matx/transforms/cub.h	Fixes the CUB version guard from the incorrect >= 3 && >= 2 to the correct `> 3
test/00_operators/ReductionTests.cu	Adds SegmentedMin and SegmentedMax typed tests covering 2D→1D, 3D→2D, 3D→1D, and full-reduction-to-scalar cases with correct expected values.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["ExecMin / ExecMax called"] --> B{"OutputTensor::Rank() > 0?"}
    B -- No --> C["cub::DeviceReduce::Min/Max\n(scalar output)"]
    B -- Yes --> D{"CUB >= 3.2?"}
    D -- No --> E["ReduceInput with iterator offsets\n(legacy path)"]
    D -- Yes --> F{"is_tensor_view AND contiguous?"}
    F -- Yes --> G["cub::DeviceSegmentedReduce::Min/Max\nwith fixed seg_size\n(fast path — NEW)"]
    F -- No --> H["ReduceInput with iterator offsets\n(iterator path)"]

Loading

_{Reviews (2): Last reviewed commit: "Fixing CUB macro for when CUB 4.0 comes ..." | Re-trigger Greptile}

[cliffburdick]

/build

[coveralls]

Coverage is 91.745% — cburdick/940 into main. No base build found for main.