gf2121 commented on PR #14361: URL: https://github.com/apache/lucene/pull/14361#issuecomment-2729250628
Thanks for feedback, > Should we floor to a multiple of 16 instead of 8 so that we have a perfect second loop with AVX-512 as well? That is what i thought initially. But my AVX-512 machine (hopefully it is) somehow only deals 256 bit once for its `vpand` and `vpslld` instructions so flooring to multiple of 8 works on it. Flooring to multiple of 16 makes benchmark a bit slower on my machine as more remaining longs need to be read but i'm fine to floor it to multiple of 16 there are machines requiring that. ``` Benchmark Mode Cnt Score Error Units Decode21Benchmark.decode21Scalar thrpt 5 28.114 ? 0.013 ops/ms Decode21Benchmark.decode21Scalar:asm thrpt NaN --- Decode21Benchmark.decode21Vector thrpt 5 49.160 ? 1.661 ops/ms Decode21Benchmark.decode21Vector:asm thrpt NaN --- Decode21Benchmark.decode21VectorFloorToMultipleOf16 thrpt 5 74.828 ? 0.463 ops/ms Decode21Benchmark.decode21VectorFloorToMultipleOf16:asm thrpt NaN --- Decode21Benchmark.decode21VectorFloorToMultipleOf8 thrpt 5 81.078 ? 0.397 ops/ms Decode21Benchmark.decode21VectorFloorToMultipleOf8:asm thrpt NaN --- ``` cpu flags: ``` fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology cpuid tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid mpx avx512f avx512dq rdseed adx smap clflushopt clwb avx512cd avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves arat umip pku ospke avx512_vnni arch_capabilities ``` > By the way, which of your machine produced the above benchmark results? The luceneutil results get on the intel chip (Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz (AVX 512)). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org