This change improves the AArch64 implementation of String.equals by introducing
SIMD-based fast paths using SVE and NEON.
SVE implementation:
- Uses predicated loads and comparisons for short lengths (len < VL)
- Uses a full predicated loop for longer inputs
- Handles the tail via an overlapped compare at (base + len - VL)
NEON implementation:
- Uses an 8-byte pre-read to simplify tail handling and eliminate 4/2/1-byte
scalar branches
- Processes 16-byte chunks using LDP pair loads
- Uses CMP/CCMP to collapse comparisons into a single branch on mismatch
These changes reduce branch pressure and improve throughput for both short and
long strings.
Correctness:
- The implementation preserves existing semantics and matches behavior for all
lengths
Testing:
- Updated and extended intrinsic tests to cover boundary conditions and
mismatch positions
Benchmark:
Across evaluated macrobenchmarks (DaCapo and Renaissance), most workloads spend
<0.5% of CPU time in String.equals. DaCapo biojava is a notable exception
(~8–9%). In biojava, most String.equals calls are on very short strings (1–2
bytes), where SVE shows ~1% end-to-end improvement, while NEON is largely
neutral or shows a small regression (~1%).
Measured using JMH on AArch64 (Arm Neoverse V2 CPU). Values are relative (%) vs
baseline. Negative values indicate regressions. Mismatch results are reported
across first(DF), middle(DM), and last(DL) difference positions.
SVE results:
Length | L1_EQ L1_DF L1_DM L1_DL | U16_EQ U16_DF U16_DM U16_DL | Avg
-------+----------------------------+-----------------------------+------
0 | 19.63 | 20.05 | 19.84
1 | 16.59 17.81 16.57 18.34 | 16.02 0.71 0.42 1.39 | 10.98
2 | 16.44 1.32 0.30 -0.16 | 15.90 -5.17 -4.55 -1.09 | 2.87
3 | 26.58 1.60 1.43 27.07 | 30.34 -8.86 -7.06 14.08 | 10.65
7 | 41.47 -2.94 -3.37 39.82 | 24.02 -8.82 -6.27 20.48 | 13.05
8 | 19.08 -1.16 -3.50 -0.90 | 22.49 -9.75 17.50 13.13 | 7.11
9 | 20.17 -4.12 -5.17 19.03 | 9.25 -2.24 21.35 3.39 | 7.71
15 | 19.48 -3.83 -4.50 19.01 | 29.26 -10.06 11.76 17.07 | 9.77
16 | 19.04 -3.15 16.41 16.85 | 38.37 -11.12 13.18 27.70 | 14.66
17 | 8.95 -2.40 5.68 6.38 | 16.32 -1.61 7.49 11.44 | 6.53
31 | 28.87 -0.01 19.79 23.37 | 41.43 -7.57 23.85 35.89 | 20.70
32 | 32.58 3.38 12.39 26.90 | 46.01 -10.99 20.53 44.15 | 21.87
33 | 11.62 -15.20 6.04 13.27 | 32.27 -9.38 20.33 32.28 | 11.40
63 | 44.66 -11.59 37.20 42.56 | 55.41 -10.57 43.19 55.90 | 32.10
64 | 53.99 -2.19 27.04 51.79 | 59.36 -8.72 35.41 60.32 | 34.63
65 | 33.79 -14.01 23.95 29.15 | 48.91 -11.58 36.54 50.03 | 24.60
127 | 62.10 -3.79 47.51 62.79 | 58.13 -8.89 60.68 60.90 | 42.43
128 | 67.38 -2.47 38.62 67.09 | 62.83 -0.38 51.72 61.87 | 43.33
129 | 52.02 -1.42 39.17 49.20 | 55.04 -9.52 53.23 52.81 | 36.32
256 | 66.11 -1.38 56.12 64.93 | 70.67 -3.68 53.67 74.54 | 47.62
Average:
33.03 -2.40 17.46 30.34 | 37.60 -7.27 23.84 33.49 | 20.91
NEON results:
Length | L1_EQ L1_DF L1_DM L1_DL | U16_EQ U16_DF U16_DM U16_DL | Avg
-------+----------------------------+-----------------------------+------
0 | 9.22 | 8.69 | 8.95
1 | 3.07 3.59 1.34 5.42 | 6.36 -6.20 -6.71 -10.59 | -0.47
2 | 3.23 -4.79 -5.67 -4.09 | 8.06 -8.43 -9.89 -9.20 | -3.85
3 | 12.80 -4.16 -3.95 11.28 | 11.94 -14.50 -14.41 11.83 | 1.36
7 | 31.00 -7.21 -12.76 33.59 | 4.73 -17.67 -17.38 1.65 | 1.99
8 | 4.43 -7.20 -4.70 -6.73 | 2.71 -18.05 -3.17 -4.05 | -4.59
9 | -9.33 -19.90 -16.27 -1.80 | 16.65 -23.72 4.26 8.78 | -5.17
15 | -6.96 -16.17 -15.60 -4.01 | 7.46 -24.60 -3.19 77.82 | 1.84
16 | 2.48 -16.38 -2.56 -3.62 | 9.08 -19.29 -5.45 77.93 | 5.27
17 | 4.88 -18.85 -0.18 19.35 | 18.43 -19.80 -8.37 84.96 | 10.05
31 | 6.92 -21.13 -4.62 60.71 | 24.42 -21.81 9.48 188.59 | 30.32
32 | 7.75 -24.20 -5.29 68.23 | 25.33 -20.57 4.17 183.65 | 29.88
33 | 20.23 -20.42 -11.33 98.60 | 23.76 -24.76 5.97 188.57 | 35.08
63 | 30.25 -22.30 14.29 152.37 | 25.02 -28.37 21.43 419.68 | 76.55
64 | 28.99 -22.91 9.03 185.51 | 38.20 -22.82 19.76 446.60 | 85.29
65 | 16.13 -21.77 1.45 211.38 | 27.94 -24.79 17.50 446.80 | 84.33
127 | 33.69 -28.94 28.75 429.23 | 41.75 -24.86 37.35 832.68 |168.71
128 | 26.28 -29.03 24.13 432.87 | 43.48 -18.53 26.44 810.20 |164.48
129 | 27.73 -20.30 20.84 439.01 | 44.09 -22.35 30.09 827.38 |168.31
256 | 53.30 -20.27 26.09 841.37 | 56.66 -21.07 47.41 1604.98|323.56
Average:
15.30 -16.97 2.26 156.24 | 22.24 -20.12 8.17 325.70 | 59.10
Observations:
- SVE shows consistent improvements across all tested lengths, with gains
increasing as input size grows
- NEON improves equal-string performance across all lengths
- NEON shows regressions for short mismatched inputs due to the loss of the
scalar tbz-based early-exit sequence, which efficiently detects mismatches at
small sizes and at early positions
- The scalar implementation relies on a branchy 4/2/1 tbz ladder, which is
efficient for early mismatches but suboptimal for equal strings
- The NEON implementation replaces this with a branchless SIMD approach and
performs upfront comparisons of the first and last 8 bytes, improving
throughput and late-mismatch detection
---------
- [x] I confirm that I make this contribution in accordance with the [OpenJDK
Interim AI Policy](https://openjdk.org/legal/ai).
-------------
Commit messages:
- 8381560: AArch64: Optimize String.equals intrinsic
Changes: https://git.openjdk.org/jdk/pull/31400/files
Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=31400&range=00
Issue: https://bugs.openjdk.org/browse/JDK-8381560
Stats: 229 lines in 6 files changed: 156 ins; 1 del; 72 mod
Patch: https://git.openjdk.org/jdk/pull/31400.diff
Fetch: git fetch https://git.openjdk.org/jdk.git pull/31400/head:pull/31400
PR: https://git.openjdk.org/jdk/pull/31400
