JIT: Don't run optSetBlockWeights when we have PGO data
#111764
Conversation
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
I would rather see us strive to get these flags set properly by each phase that manipulates flow. So yet another aspect of pgo post phase checks: if root method has PGO, all blocks must have |
That sounds like a better plan. I suppose a reasonable goal is to enable |
) Part of #107749. Prerequisite to #111764 (comment). Add a post-phase check that ensures BBF_PROF_WEIGHT is set for every reachable block when we have profile data. This check revealed only one site -- when incorporating an inlinee's blocks -- where we couldn't rely on normal flow propagation to set the flag.
|
Diffs are smaller this time around; some of these are probably from the |
* main: (22 commits) Clean up Stopwatch a bit (dotnet#111834) JIT: Fix embedded broadcast simd size (dotnet#111638) Revert potential UB due to aliasing + more WB removals (dotnet#111733) re-enable acceleration of Vector512<long>.op_Multiply (dotnet#111832) Handle OSSL 3.4 change to SAN:othername formatting JIT: Fix stack allocated arrays for NativeAOT (dotnet#111827) JIT: enhance RBO inference for similar compares to constants (dotnet#111766) JIT: Don't run optSetBlockWeights when we have PGO data (dotnet#111764) [Android] Make sure RuntimeFlavor=CoreCLR when clr subset is specified (dotnet#111821) Change empty subject test certificate to include a critical SAN. Fix reversed code offsets in GcInfo (dotnet#111792) Swap some libraries areas between leads (dotnet#111816) Add left-handed spherical and cylindrical billboards (dotnet#109605) JIT: revise `optRelopImpliesRelop` to always set `reverseSense` (dotnet#111803) Fix Zip64ExtraField handling (dotnet#111802) Add build support for Android+CoreCLR (dotnet#110471) arm64: Add bic(s) compact encoding (dotnet#111452) JIT: Ensure `BBF_PROF_WEIGHT` flag is set when we have PGO data (dotnet#111780) Add support for AVX10.2, Add AVX10.2 API surface and template tests (dotnet#111209) JIT: Preliminary for enabling inlining late devirted calls (dotnet#111782) ...
Part of #107749. For PGO scenarios, we plan to rely entirely on profile synthesis to propagate profile-derived weights throughout the flowgraph. This is the first step in transitioning away from the existing weight propagation heuristics. We want to keep these around for non-PGO scenarios for now, though ideally, we'll eventually be able to replace these entirely with profile synthesis (plus branch prediction heuristics when we don't have profile data), and consolidate how the profile is maintained for PGO and non-PGO scenarios.
This had few diffs locally, as
optSetBlockWeightsalready avoids modifying profile-derived weights. The diffs seem to stem fromBBF_PROF_WEIGHTnot being propagated to all blocks when we have profile data (such as during inlining). This doesn't seem concerning, since the next profile repair pass (which I've yet to add) will propagate the flag as needed.