diff options
author | Koundinya Veluri <kouvel@users.noreply.github.com> | 2018-01-25 12:01:32 -0800 |
---|---|---|
committer | GitHub <noreply@github.com> | 2018-01-25 12:01:32 -0800 |
commit | 209415618ca5d1a5d1d9e39ca78d643d0935534e (patch) | |
tree | d83c946783390afbb52e3e0f968018c38dfd2560 /src/vm/codeversion.cpp | |
parent | e9985126acb0f1efd7c780faac4e66bc798b73c0 (diff) | |
download | coreclr-209415618ca5d1a5d1d9e39ca78d643d0935534e.tar.gz coreclr-209415618ca5d1a5d1d9e39ca78d643d0935534e.tar.bz2 coreclr-209415618ca5d1a5d1d9e39ca78d643d0935534e.zip |
Enable tiered jitting for R2R methods (#15967)
Enable tiered jitting for R2R methods
- Included R2R methods and generics over value types in CoreLib for tiered jitting. Tier 0 for R2R methods is the precompiled code if available, and tier 1 is selectively scheduled based on call counting.
- Added a delay before starting to count calls for tier 1 promotion. The delay is a short duration after frequent tier 0 jitting stops (current heuristic for identifying startup).
- Startup time and steady-state performance have improved on JitBench. There is a regression shortly following startup due to call counting and tier 1 jitting, for a short duration before steady-state performance stabilizes.
- Added two new config values, one for configuring the call count threshold for promoting to tier 1, and another for specifying the delay from the last tier 0 JIT invocation before starting to count calls
Diffstat (limited to 'src/vm/codeversion.cpp')
-rw-r--r-- | src/vm/codeversion.cpp | 15 |
1 files changed, 12 insertions, 3 deletions
diff --git a/src/vm/codeversion.cpp b/src/vm/codeversion.cpp index 10d3013f35..da808e8839 100644 --- a/src/vm/codeversion.cpp +++ b/src/vm/codeversion.cpp @@ -2177,12 +2177,14 @@ PCODE CodeVersionManager::PublishVersionableCodeIfNecessary(MethodDesc* pMethodD // attempt to publish the active version still under the lock if (FAILED(hr = PublishNativeCodeVersion(pMethodDesc, activeVersion, fEESuspend))) { - // if we need an EESuspend to publish then start over. We have to leave the lock in order to suspend, - // and when we leave the lock the active version might change again. However now we know that suspend + // If we need an EESuspend to publish then start over. We have to leave the lock in order to suspend, + // and when we leave the lock the active version might change again. However now we know that suspend is + // necessary. if (hr == CORPROF_E_RUNTIME_SUSPEND_REQUIRED) { _ASSERTE(!fEESuspend); fEESuspend = true; + continue; // skip RestartEE() below since SuspendEE() has not been called yet } else { @@ -2215,6 +2217,8 @@ PCODE CodeVersionManager::PublishVersionableCodeIfNecessary(MethodDesc* pMethodD HRESULT CodeVersionManager::PublishNativeCodeVersion(MethodDesc* pMethod, NativeCodeVersion nativeCodeVersion, BOOL fEESuspended) { + // TODO: This function needs to make sure it does not change the precode's target if call counting is in progress. Track + // whether call counting is currently being done for the method, and use a lock to ensure the expected precode target. LIMITED_METHOD_CONTRACT; _ASSERTE(LockOwnedByCurrentThread()); _ASSERTE(pMethod->IsVersionable()); @@ -2236,7 +2240,12 @@ HRESULT CodeVersionManager::PublishNativeCodeVersion(MethodDesc* pMethod, Native { EX_TRY { - hr = pPrecode->SetTargetInterlocked(pCode, FALSE) ? S_OK : E_FAIL; + pPrecode->SetTargetInterlocked(pCode, FALSE); + + // SetTargetInterlocked() would return false if it lost the race with another thread. That is fine, this thread + // can continue assuming it was successful, similarly to it successfully updating the target and another thread + // updating the target again shortly afterwards. + hr = S_OK; } EX_CATCH_HRESULT(hr); return hr; |