Re: JIT compiling with LLVM v9.0
Tomas Vondra <tomas.vondra@2ndquadrant.com>
Attachments
- bench-10gb-duration.png (image/png)
- bench-10gb-pct.png (image/png)
- bench-50gb-duration.png (image/png)
- bench-50gb-pct.png (image/png)
- bench2-10gb-duration.png (image/png)
- bench2-10gb-pct.png (image/png)
- bench2-50gb-duration.png (image/png)
- bench2-50gb-pct.png (image/png)
- dbt3-jit-parallel.ods (application/vnd.oasis.opendocument.spreadsheet)
Hi, On 02/03/2018 01:05 PM, Tomas Vondra wrote: > Hi, > > ... > > In this round of tests I've disabled parallelism. Based on > discussion with Andres I've decided to repeat the tests with parallel > queries enabled - that's running now, and will take some time to > complete. > And here are the results with parallelism enabled - same machines, but with max_parallel_workers_per_gather > 0. Based on discussions and Andres' FOSDEM talk I somehow expected more significant JIT benefits in the parallel case, but the results are pretty much exactly the same (modulo speedup thanks to parallelism, of course). In fact, the JIT impact is much noisier with parallelism enabled, for some reason, with regressions where there were no measurable regressions before (particularly for the 10GB case). That is not to say we shouldn't be doing JIT, or that Andres did not observe the speedups/benefits he mentioned during the talk - I have no trouble believing it depends on queries, and DBT-3 may not match that. I don't plan doing any further benchmarks on this patch series unless someone requests that (possibly with ideas what to focus on). I'll keep looking at the patch, of course. I've seen some build issues, so I'll try finding more details. regards -- Tomas Vondra http://www.2ndQuadrant.com PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
Commits
-
Improve JIT docs.
- fb604780114c 11.0 landed
-
Add documentation for the JIT feature.
- e6c039d13e16 11.0 landed
-
Add EXPLAIN support for JIT.
- 1f0c6a9e7dca 11.0 landed
-
Add inlining support to LLVM JIT provider.
- 9370462e9a79 11.0 landed
-
JIT tuple deforming in LLVM JIT provider.
- 32af96b2b118 11.0 landed
-
Add FIELDNO_* macro designating offset into structs required for JIT.
- 7ced1d124728 11.0 landed
-
Add expression compilation support to LLVM JIT provider.
- 2a0faed9d702 11.0 landed
-
Expand list of synchronized types and functions in LLVM JIT provider.
- fb46ac26fe49 11.0 landed
-
Add helpers for emitting LLVM IR.
- 7ec0d80c0508 11.0 landed
-
Basic planner and executor integration for JIT.
- cc415a56d09a 11.0 landed
-
Debugging and profiling support for LLVM JIT provider.
- 250bca7fc145 11.0 landed
-
Support for optimizing and emitting code in LLVM JIT provider.
- b96d550eb03c 11.0 landed
-
Add file containing extensions of the LLVM C API.
- 31bc604e0b74 11.0 landed
-
Basic JIT provider and error handling infrastructure.
- 432bb9e04da4 11.0 landed
-
Add configure infrastructure (--with-llvm) to enable LLVM support.
- 5b2526c83832 11.0 landed
-
Add C++ support to configure.
- 6869b4f25847 11.0 landed
-
Add PGAC_PROG_VARCC_VARFLAGS_OPT autoconf macro.
- 3de04e4ed12d 11.0 landed
-
Fix VM buffer pin management in heap_lock_updated_tuple_rec().
- 0b1d1a038bab 11.0 cited
-
Allow tupleslots to have a fixed tupledesc, use in executor nodes.
- ad7dbee368a7 11.0 landed
-
Expression evaluation based aggregate transition invocation.
- 69c3936a1499 11.0 landed
-
Perform slot validity checks in a separate pass over expression.
- b40933101ca6 11.0 landed
-
Rely on executor utils to build targetlist for DML RETURNING.
- 4717fdb14cf0 11.0 landed
-
Refer to OS X as "macOS", except for the port name which is still "darwin".
- da6c4f6ca88d 10.0 cited