WIP: feat(probe-subop): normalized-key probe AArch64 SVE on HashAgg

[LeiRui]

What problem does this PR solve?

HashAgg with normalized-key hash tables can spend significant time in groupNormalizedKeyProbe on AArch64. This PR adds an opt-in SVE vectorized probe/insert path (17-byte slot layout + tag region), wired through the aggregation hash-table factory, with scalar parity tests and non-aarch64 link stubs. Join hash tables (createForJoin / joinProbe) are unchanged and stay on native Bolt.

Issue Number: N/A

Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)
• ✨ New feature (non-breaking change which adds functionality)
• 🚀 Performance improvement (optimization)
• ⚠️ Breaking change (fix or feature that would cause existing functionality to change)
• 🔨 Refactoring (no logic changes)
• 🔧 Build/CI or Infrastructure changes
• 📝 Documentation only

Description

What

• Add SVE normalized-key probe for HashAgg: groupNormalizedKeyProbeSVE, insertForGroupBySve, 17B slot+tag table layout; scalar path uses BaseHashTable::NormalizedKeySlot (16B key+pointer, in HashTable.h).
• Split SVE TU HashTableNormalizedKeyProbeSve.cpp (aarch64, -march=armv8-a+sve); stubs HashTableNormalizedKeyProbeStubs.cpp on other platforms for link safety.
• HashTableSveRuntime.h: boltHashAggSveNormalizedKeyProbeEnabledFromEnv() + linuxAarch64RuntimeHasSve() (AT_HWCAP / HWCAP_SVE).
• Extension hooks (default stock behavior): HashAggregation::createGroupingSetForHashAggregation, GroupingSet::createAggregationHashTable — override point for custom BaseHashTable / env-driven table type without copying HashAggregation.
• NormalizedKeyMode: kNativeBolt (default) / kScalar (16B slots) / kSve (17B+tag); Join build forces kNativeBolt via createForJoin(..., sve=false) and isJoinBuild_.

Env opt-in (BOLT_HASH_AGG_SVE_NORMALIZED_KEY_PROBE)

Env value	Aggregation hash table
unset, empty, or unknown	off → kNativeBolt (same as upstream default)
1	enable SVE path when Linux aarch64 + runtime SVE
true, yes, or on (ASCII, case-insensitive)	same as 1
0	off
false, no, or off (ASCII, case-insensitive)	same as 0

Executor example (Gluten / Spark): spark.executorEnv.BOLT_HASH_AGG_SVE_NORMALIZED_KEY_PROBE=1.

Registration / runtime chain (HashAgg only)

createForAggregation does not call updateNormalizedKeyModeForAggregation() directly; it only sets sveNormalizedKeyProbeRequested_ in the ctor. Mode is chosen later when the table switches to HashMode::kNormalizedKey via setHashMode (from checkSize / analyze).

GroupingSet::createAggregationHashTable()  [override hook]
  └─ sveFlag = boltHashAggSveNormalizedKeyProbeEnabledFromEnv()
        └─ HashTable::createForAggregation(..., sveFlag)
              └─ ctor: sveNormalizedKeyProbeRequested_ = sveFlag
                  (normalizedKeyMode_ stays kNativeBolt until setHashMode)

addInput → checkSize / analyze → setHashMode(kNormalizedKey)   [when applicable]
  └─ updateNormalizedKeyModeForAggregation()
        ├─ isJoinBuild_ || !sveNormalizedKeyProbeRequested_ → kNativeBolt
        ├─ linuxAarch64RuntimeHasSve() → kSve
        └─ else → kScalar

groupProbe (hashMode_ == kNormalizedKey)
  └─ groupNormalizedKeyProbe()
        ├─ kScalar  → groupNormalizedKeyProbeScalar
        ├─ kSve     → groupNormalizedKeyProbeSVE  (SVE TU)
        └─ kNativeBolt → ProbeState / fullProbe (existing)

Join (out of scope for SVE)

HashTable::createForJoin(..., sveNormalizedKeyProbeEnabled=false, isJoinBuild=true)
  → normalizedKeyMode_ = kNativeBolt always
joinProbe → joinNormalizedKeyProbe   (never groupNormalizedKeyProbeSVE)

Flowchart — HashAgg probe mode (mermaid — click to expand)

flowchart TD
  A[groupProbe + kNormalizedKey] --> B[groupNormalizedKeyProbe]
  B --> C{normalizedKeyMode_}
  C -->|kScalar| D[groupNormalizedKeyProbeScalar]
  C -->|kSve| E[groupNormalizedKeyProbeSVE]
  C -->|kNativeBolt| F[ProbeState / fullProbe]
  G[createForJoin / joinProbe] --> H[kNativeBolt + joinNormalizedKeyProbe]

Loading

File roles (HashTable + GroupingSet) (mermaid — click to expand)

flowchart TB
  subgraph hooks["Extension hooks"]
    HA["HashAggregation.h/cpp\ncreateGroupingSetForHashAggregation"]
    GS["GroupingSet.h/cpp\ncreateAggregationHashTable"]
  end

  subgraph hashtable["HashTable"]
    HT["HashTable.cpp / .h\ndispatch, allocateTables, insert"]
    RT["HashTableSveRuntime.h\nenv + HWCAP_SVE"]
    SVE["HashTableNormalizedKeyProbeSve.cpp\nSVE probe + insertForGroupBySve"]
    STUB["HashTableNormalizedKeyProbeStubs.cpp\nnon-aarch64"]
  end

  GS -->|env| RT
  GS --> HT
  HA --> GS
  HT --> SVE
  HT --> STUB
  classDef prNew fill:#d4edda,stroke:#28a745,stroke-width:2px,color:#000
  classDef prModified fill:#fff3cd,stroke:#ffc107,stroke-width:2px,color:#000
  class SVE,STUB,RT prNew
  class HT,HA,GS prModified

Loading

Legend: green = new; yellow = modified.

Step	File / symbol	Role
0	GroupingSet::createAggregationHashTable	Reads env; calls createForAggregation(..., sveFlag). Override for plugin/custom table.
1	HashTableSveRuntime.h	Env parser; linuxAarch64RuntimeHasSve() on Linux aarch64.
2	HashTable ctor	Stores sveNormalizedKeyProbeRequested_ from createForAggregation(..., sveFlag).
2b	setHashMode(kNormalizedKey) → updateNormalizedKeyModeForAggregation	Sets kSve / kScalar / kNativeBolt; Join build → always kNativeBolt.
3	allocateTables / clear / insertForGroupBy	16B vs 17B+tag vs F14 layout; SVE insert via insertForGroupBySve when kSve.
4	groupNormalizedKeyProbe	Dispatches scalar / SVE / native probe.
5	HashTableNormalizedKeyProbeSve.cpp	SVE intrinsics; not linked on non-aarch64.
6	HashTableNormalizedKeyProbeStubs.cpp	Non-aarch64: SVE probe → scalar fallback; insert → BOLT_FAIL if called.
7	createForJoin	Always sveNormalizedKeyProbeEnabled=false; joinProbe unchanged.

Tests

• HashTableTest.boltHashAggSveNormalizedKeyProbeEnv — env parser (setenv / unset / 0 / 1)
• HashTableTest.normalizedKeySveMatchesScalar — #if __aarch64__ && __linux__: helper sets kSve vs kScalar, compares probe hits; GTEST_SKIP if no runtime SVE

make unittest_release TEST=bbolt_exec_test \
  TEST_ARGS='--gtest_filter=HashTableTest.boltHashAggSveNormalizedKeyProbeEnv:HashTableTest.normalizedKeySveMatchesScalar*'

Performance Impact

• No Impact
• Positive Impact: I have run benchmarks.
• Negative Impact

Release Note

Please describe the changes in this PR

Release Note:

Release Note:
- HashAgg normalized-key probe: optional AArch64 SVE path (17B slot + tag layout).
- Opt-in: set BOLT_HASH_AGG_SVE_NORMALIZED_KEY_PROBE=1 (or true/yes/on) on executors; default off (native Bolt).
- Join hash tables (createForJoin / joinProbe) unchanged.
- Extension hooks on HashAggregation / GroupingSet for custom aggregation hash tables.

Checklist (For Author)

• I have added/updated unit tests (ctest).
• I have verified the code with local build (Release/Debug).
• I have run clang-format / linters.
• (Optional) I have run Sanitizers (ASAN/TSAN) locally for complex C++ changes.
• No need to test or manual test.

Breaking Changes

• No
• Yes (Description: ...)