On Fri, 21 Feb 2025 20:14:19 GMT, Chen Liang <li...@openjdk.org> wrote:
>> Simplify the layout access var handles to be direct in some common cases. >> Also made `VarHandle::isAccessModeSupported` report if an access mode is >> supported for a VH. >> >> Reduces the instructions to execute this code in a simple main by 47%: >> >> long[] arr = new long[8]; >> var ms = MemorySegment.ofArray(arr); >> ms.setAtIndex(ValueLayout.JAVA_BYTE, 12, (byte) 3); >> >> >> Main overheads in FFM are identified to be: >> 1. Eager initialization of direct MethodHandle; can be CDS archived >> 2. MH combinator forms via LambdaFormEditor, not cached right now and always >> have large overhead >> >> Still need other measures to deal with common user patterns of >> `MethodHandles.insertCoordinates(vh, 1, 0L)` which currently is still very >> slow. >> >> Tests: 2 unrelated failures on tier 1-3 > > Chen Liang has updated the pull request incrementally with one additional > commit since the last revision: > > Review remarks, dates, some more simplifications Benchmark results for the latest revision appears performance neutral. bytestacks same as last revision. 11 jobs left in tier 1-3, no failure so far. Also created CSR for this minor behavioral change. Benchmark (polluteProfile) Mode Cnt Score Error Units LoopOverNonConstantHeap.BB_get false avgt 30 0.934 ± 0.027 ns/op LoopOverNonConstantHeap.BB_get true avgt 30 0.946 ± 0.028 ns/op LoopOverNonConstantHeap.BB_loop false avgt 30 0.208 ± 0.004 ms/op LoopOverNonConstantHeap.BB_loop true avgt 30 0.211 ± 0.003 ms/op LoopOverNonConstantHeap.segment_get false avgt 30 1.123 ± 0.040 ns/op LoopOverNonConstantHeap.segment_get true avgt 30 1.120 ± 0.040 ns/op LoopOverNonConstantHeap.segment_loop false avgt 30 0.205 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop true avgt 30 0.202 ± 0.003 ms/op LoopOverNonConstantHeap.segment_loop_instance false avgt 30 0.209 ± 0.005 ms/op LoopOverNonConstantHeap.segment_loop_instance true avgt 30 0.202 ± 0.003 ms/op LoopOverNonConstantHeap.segment_loop_instance_unaligned false avgt 30 0.209 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop_instance_unaligned true avgt 30 0.210 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop_readonly false avgt 30 0.206 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop_readonly true avgt 30 0.206 ± 0.005 ms/op LoopOverNonConstantHeap.segment_loop_slice false avgt 30 0.203 ± 0.002 ms/op LoopOverNonConstantHeap.segment_loop_slice true avgt 30 0.207 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop_unaligned false avgt 30 0.206 ± 0.004 ms/op LoopOverNonConstantHeap.segment_loop_unaligned true avgt 30 0.209 ± 0.003 ms/op LoopOverNonConstantHeap.unsafe_get false avgt 30 0.386 ± 0.017 ns/op LoopOverNonConstantHeap.unsafe_get true avgt 30 0.381 ± 0.017 ns/op LoopOverNonConstantHeap.unsafe_loop false avgt 30 0.205 ± 0.004 ms/op LoopOverNonConstantHeap.unsafe_loop true avgt 30 0.204 ± 0.004 ms/op ------------- PR Comment: https://git.openjdk.org/jdk/pull/23720#issuecomment-2675734830
