JIT: revise inlinee block numbering, enable synthesis for inlinees #83610
Conversation
Start numbering inlinee blocks from 1 instead of 1 + the root compiler's max BB num. Update inlinee block bbNums when they are inserted into the root compiler's graph. Adjust computations in various places that knew about the old approach and looked from inlinee compiler to root compiler for bitset, epochs and the like. Enable synthesis for inlinees, now that regular bitsets on inlinee compiler instances behave sensibly. There is still some messiness around inlinees inheriting root compiler EH info which requires special checks. I will clean this up separately. Fixes dotnet#82755. Contributes to dotnet#82964. enable synthesis for inlinees
ghost
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch, @kunalspathak Issue DetailsStart numbering inlinee blocks from 1 instead of 1 + the root compiler's max BB num. Update inlinee block bbNums when they are inserted into the root compiler's graph. Adjust computations in various places that knew about the old approach and looked from inlinee compiler to root compiler for bitset, epochs and the like. Enable synthesis for inlinees, now that regular bitsets on inlinee compiler instances behave sensibly. There is still some messiness around inlinees inheriting root compiler EH info which requires special checks. I will clean this up separately. Fixes #82755. enable synthesis for inlinees
|
|
@jakobbotsch PTAL Should be no diff with default behavior. There are diffs with synthesis enabled. |
|
No diffs, but shows a TP win, which is a bit unexpected. Perhaps better block density in the |
I was curious so I did a detailed TP run: Base: 230604816272, Diff: 230161636846, -0.1922%
?compInit@Compiler@@QEAAXPEAVArenaAllocator@@PEAUCORINFO_METHOD_STRUCT_@@PEAVICorJitInfo@@PEAUCORINFO_METHOD_INFO@@PEAUInlineInfo@@@Z : 144304290 : +47.32% : 18.68% : +0.0626%
?fgInsertInlineeBlocks@Compiler@@AEAAXPEAUInlineInfo@@@Z : 15361212 : +5.08% : 1.99% : +0.0067%
?GetDescriptorForSwitch@Compiler@@QEAA?AUSwitchUniqueSuccSet@1@PEAUBasicBlock@@@Z : 4476052 : +5.51% : 0.58% : +0.0019%
?impImport@Compiler@@QEAAXXZ : -2753802 : -0.77% : 0.36% : -0.0012%
?WalkSpanningTree@Compiler@@QEAAXPEAVSpanningTreeVisitor@@@Z : -7157865 : -6.20% : 0.93% : -0.0031%
?fgRenumberBlocks@Compiler@@QEAA_NXZ : -10770841 : -3.76% : 1.39% : -0.0047%
memcpy : -15466063 : -9.29% : 2.00% : -0.0067%
?allocateMemory@ArenaAllocator@@QEAAPEAX_K@Z : -17243298 : -0.35% : 2.23% : -0.0075%
?bbNewBasicBlock@Compiler@@QEAAPEAUBasicBlock@@W4BBjumpKinds@@@Z : -26240701 : -7.04% : 3.40% : -0.0114%
?EnsureCoversInd@?$JitExpandArray@E@@IEAAXI@Z : -35773458 : -17.74% : 4.63% : -0.0155%
memset : -109241007 : -5.80% : 14.14% : -0.0474%
?fgInit@Compiler@@QEAAXXZ : -176797854 : -100.00% : 22.88% : -0.0767%
?InitializeRange@?$JitExpandArray@E@@IEAAXII@Z : -206210940 : -31.89% : 26.69% : -0.0894%The importer sizes |
|
/azp run runtime-coreclr jitstress |
|
Azure Pipelines successfully started running 1 pipeline(s). |
Start numbering inlinee blocks from 1 instead of 1 + the root compiler's max BB num. Update inlinee block bbNums when they are inserted into the root compiler's graph.
Adjust computations in various places that knew about the old approach and looked from inlinee compiler to root compiler for bitset, epochs and the like.
Enable synthesis for inlinees, now that regular bitsets on inlinee compiler instances behave sensibly.
There is still some messiness around inlinees inheriting root compiler EH info which requires special checks. I will clean this up separately.
Fixes #82755.
Contributes to #82964.
enable synthesis for inlinees