变更摘要

该 PR 旨在修复 GEIR 示例（test_geir_indexer_quant_cache.cpp）成功路径中的 Session 内存泄漏（对应 issue #4792），并为 IndexerQuantCache kernel 补充基于真实字面值的数值 golden UT，替代原先仅判断输出非零的弱校验。示例侧在 RunGraph 成功后、调用 ge::GEFinalize() 之前新增 delete session，使成功路径与失败路径的资源释放行为一致；UT 侧新增 test_indexer_quant_cache_golden.cpp，针对 Normal 量化分支分别覆盖 roundScale=true/false 两种场景，golden 采用与被测 kernel 计算逻辑无关的固定字面值。

主要改动

• 修复 GEIR 示例 Session 泄漏：在 test_geir_indexer_quant_cache.cpp 的 ge::GEFinalize() 前新增 delete session;，确保 RunGraph 成功后释放 Session，与失败路径的资源回收行为保持一致，解决 #4792。
• 新增 Normal 量化分支 golden 测试文件：新增 test_indexer_quant_cache_golden.cpp，通过 RunAndCheckNormalGolden 以交替输入 1.0/-0.5 分别运行 roundScale=1 与 roundScale=0（TEST(IndexerQuantCacheNormalGolden, RoundScale) / NoRoundScale），断言输出与固定字面值 golden 完全一致。
• 校验量化输出的字节与位模式：逐字节校验完整 cache 输出（E4M3FN 字节 golden，如 0x78/0xF0、0x7E/0xF6），并以位模式校验 float32 scale（如 0x3B800000U、0x3B124925U），杜绝仅判断非零导致的数值错误漏检。
• 校验 slotMapping=-1 行的行为：验证当 slotMapping[row] < 0（如 SLOT_MAPPING = {0, -1, 2, 3}）时对应 cache 行与 scale 保持原值（初始 0）不变，并仅对非跳过的行应用量化结果。

Thanks for your pull-request.
The full list of commands accepted by me can be found at here.
You can get sig-info at here.
You can self-configure the PR merge rules for this repository. For more details, please refer to Here.
For more, you also can visit HICANN.

PR Approval Progress

✅ Congratulations! All modules have met the lgtm and approve requirements.

Module Approval Details

💡 Tip:

• Committer can comment /approve or /lgtm
• Commenting /approve implies both code review (lgtm) and intent to merge (approve)

CLA Signature Pass

qq_37913898, thanks for your pull request. All authors of the commits have signed the CLA. 👍

compile

/approve

/approve