fix(simd): umasked AVX2 load

[ahuber21]

The previous code always performed a full width load on the provided data. In ragged-epilogue scenarios, where we request a masked load, this resulted in SEGV errors in certain runs with address sanitizer.

    if (i < count.size()) {
        auto mask = create_mask<simd_width>(count);
        s0 = op.accumulate(mask, s0, op.load_a(mask, a + i), op.load_b(mask, b + i));
    }

Why wasn't this caught sooner?

The OS only triggers a segmentation fault if a read accesses an unmapped memory page. Since memory protection (typically) operates at a 4KB page granularity, reading past the end of a buffer is "safe" from the OS's perspective unless the overflow happens to cross exactly into an unmapped page.

Why is ASan catching it sporadically?

Since our underlying object storage is std::vector, ASan detection requires two specific conditions to align:

• No Spare Capacity: The vector's size() must equal its capacity(). If there is spare capacity, the unsafe load simply reads valid (though uninitialized) memory owned by the vector.
• Alignment & Redzones: The underlying heap allocation must be sized and aligned such that the full-width SIMD read (e.g., 32 bytes) actually crosses the allocation boundary into the ASan redzone. If the allocator adds padding for alignment, the read might land in that valid padding instead.

[ahuber21]

@copilot summarize the changes in this PR

[Copilot]

@ahuber21 I've opened a new pull request, #242, to work on those changes. Once the pull request is ready, I'll request review from you.

[Copilot]

Pull request overview

This PR fixes a critical SIMD memory safety bug in AVX2 masked load operations that could cause segmentation faults when reading beyond allocated memory boundaries. The fix ensures masked loads respect buffer boundaries in ragged-epilogue scenarios.

Key Changes:

• Fixed AVX2 masked load implementation to prevent out-of-bounds reads
• Added comprehensive ASan-detected regression tests for distance computations
• Enhanced CI with dedicated AddressSanitizer build configuration

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
tests/svs/lib/avx_detection.cpp	Added test for runtime AVX flag patching mechanism
tests/svs/core/distance.cpp	Added ASan regression tests for distance computation with ragged epilogues
tests/CMakeLists.txt	Updated Catch2 to v3.11.0 and improved test discovery with tag-based labels
.github/workflows/build-linux.yml	Added clang++-18 ASan build configuration with leak detection disabled
tests/svs/index/vamana/multi.cpp	Tagged test as long-running to exclude from ASan builds
tests/svs/index/vamana/index.cpp	Tagged test as long-running to exclude from ASan builds
tests/svs/index/inverted/memory_based.cpp	Tagged test as long-running to exclude from ASan builds
tests/svs/index/inverted/clustering.cpp	Tagged test as long-running to exclude from ASan builds

---

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

[ibhati]

The new workflow is added in the existing workflows, please make sure the earlier flow and options did not change.

[ibhati]

What is this [long]? Why is this required?

[ahuber21]

Sorry, had a comment that I didn't post. I added a new tag [long] that marks long-running tests. They are skipped in the debug asan build.

ScalableVectorSearch/.github/workflows/build-linux.yml

Lines 63 to 64 in 63e58cd

	# skip longer-running tests
	ctest_args: "-LE long"

[ethanglaser]

Do we know how long the ASan run takes if completing successfully?

[ahuber21]

Do we know how long the ASan run takes if completing successfully?

@ethanglaser about 10 minutes total, < 5 mins of which is testing.

[ahuber21]

The new workflow is added in the existing workflows, please make sure the earlier flow and options did not change.

@ibhati the only change is that the existing steps are executing three more tests (with negligible runtime).