199 create a set of cvgswarp versions that do not get an input gpumat

[morousg]

No description provided.

[Copilot]

Pull request overview

Adds input-independent warp operation descriptors for flexible pipeline composition.

Changes:

• Adds standalone and batch no-input warp overloads.
• Extracts reusable GpuMat reader construction.
• Adds tests for no-input affine and perspective warping.

Reviewed changes

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

File	Description
include/cvGPUSpeedup.cuh	Adds no-input warp overloads.
include/cvGPUSpeedupHelpers.cuh	Adds the reusable reader helper.
tests/warping/test_warping_opencv.cu	Exercises the new operation composition paths.

Suppressed comments (2)

tests/warping/test_warping_opencv.cu:421

• Every per-image comparison is discarded and the test returns success unconditionally. Aggregate the comparison results and return them (using either the existing cvGS overload as the oracle or an explicit tolerance), otherwise batch-output regressions in the new overload remain undetected.

    return true;

tests/warping/test_warping_opencv.cu:492

• This test also returns success regardless of every comparison. After switching it to the no-input overload, aggregate checks for the active planes and verify inactive planes against defaultValue; do not compare inactive OpenCV buffers, because indices usedPlanes..NUM_IMGS2 were never initialized by warpPerspective.

    return true;

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

Suppressed comments (4)

include/cvGPUSpeedup.cuh:484

• usedPlanes is caller-controlled but is used as an array bound before any validation. Values greater than BATCH index past transform_matrices/dstSize, while negative values are forwarded to FKL. Reject values outside [0, BATCH] before entering the loop.

                           const int& usedPlanes, const cv::Scalar& defaultValue) {
    for (int i = 0; i < usedPlanes; ++i) {

tests/warping/test_warping_opencv.cu:306

• This test always returns success even when the new no-input perspective pipeline produces an incorrect result; correct only affects the log. Since exact border equality is known not to hold, use a border-aware/tolerant assertion and return that predicate so this overload has effective regression coverage.

    return true;

tests/warping/test_warping_opencv.cu:421

• Every per-plane comparison is discarded and the full-batch no-input test returns true unconditionally. Aggregate the comparison results (with the intended perspective-border tolerance) and return the aggregate so regressions in this new overload fail CTest.

    return true;

tests/warping/test_warping_opencv.cu:491

• Only the first usedPlanes OpenCV outputs are written, but this loop compares all ten planes, including uninitialized reference buffers, and then the function returns success regardless. Check active planes against OpenCV, explicitly validate the expected default for inactive planes, and return the aggregate result so the partial-batch overload is actually tested.

    for (int i = 0; i < NUM_IMGS2; ++i) {
        cv::Mat resultcv(d_resultscv[i]);
        cv::Mat resultcvGS(d_resultscvGS[i]);
        const bool correct = compareAndCheck<CV_8UC3>(resultcv, resultcvGS);
        std::cout << "Perspective transformation batch " << i << ": " << (correct ? "PASS" : "EXPECTED_FAIL") << std::endl;