- 24 9月, 2022 13 次提交
-
-
由 Will Smith 提交于
* Add test * Add CHECKs * Build scripts and test settings * COMPlus -> DOTNET * Fix filenames * Missing > * Missing quotes * use %scriptPath% * Rework properties, start two examples * Arch demo * Cleanup * Remove tab * Easier environment variables. Undo precommand changes. * undo blank line * clean CHECKs * Draft of bash * Bash, conditionals * More variables, start on run.cmd/sh/py * another output * Support in cmd/bash for RunningDisasmChecks * copy, factor, formatting * Initial work to include FileCheck. Added SuperFileCheck. * Able to build SuperFileCheck * Do not DisasmCheck TypeEquality_r for now. Update some FileChecks to follow SuperFileCheck rules. * Partially wiring up SuperFileCheck to tests * Piping list of method names from SuperFileCheck to JitDisasm * Handling bash a little bit * Moving SuperFileCheck to tests/Common * Few tweaks * Building SuperFileCheck as part of the test build * Tweaking a few things * Fixed a bug * Moving SuperFileCheck back to src\coreclr\tools. Removed checks from TypeEquality_r. * Restore original logic in Runtime_73681 * Trying to add CI leg for disasmchecks * Use x64 package if x86 platform detected for JIT tools package * Remove innerloop for disasmchecks * Trying to fix build. Only run in Windows for now. * Update Runtime_73681.cs Trying to fail test * Trying to fix build * Update Runtime_73681.cs * Update Runtime_73681.cs * Fixing a few issues * Trying to run disasmchecks as part of CI * Trying to run disasmchecks * Trying to run disasmchecks * Trying to run disasmchecks * Revert a change * Trying to run disasmchecks * Trying to run disasmchecks * build SuperFileCheck on non-windows * few tweaks * Trying to fix CI * Including SuperFileCheck for tests * Cleanup * More cleanup * Cleanup * Changed SuperFileCheck to not publish everything. Changed SuperFileCheck's lookup for FileCheck. * Invoking SuperFileCheck using dotnet * Making the test pass * Only run disasm checks for coreclr and not mono * Using HasBatchDisasmCheck and HasBashDisasmCheck to determine to run the check * Enabling filecheck on linux and osx * Added more comments * Added ARM64 specific test. Do not run SuperFileCheck if no methods were found. * Added documentation. Changed disasm-output. * Minor doc tweak * Minor doc tweak * Minor doc tweak * Minor doc tweak * Minor doc tweak * Cleanup. Trying to fix linux * Fixing test * Add information on additional functionality * cleanup * Add FileCheck snippet * Undo environment variable changes * Feedback from Mark * Cleanup * Trying to fix linux test run * Trying to fix linux test run * A few missing changes from the original branch * Enable OSX for disasm checks * cleanup / comment * Force test failure * Update Runtime_73681.cs * Set env vars after errorlevel check * Reverting back on setting environment variables in test. Added new FileCheck test for mod optimization * Force a failure by changing the register * Ignore native binaries for superpmi collect * Update Runtime_34937.cs * Force the correct failure * Update Runtime_34937.cs * Update Runtime_34937.cs * Adding specific OS check prefixes. Changed dump-input context amount * Added getting fully qualified method names with wildcards for SuperFileCheck * More tests. Fixed a few issues with generics. * Disabling generic support * Error if it cannot find enclosing type declaration * Fixing build * Remove namespac * Bring generics back, but in a limited form Co-authored-by: NMark Plesko <markples@microsoft.com>
-
由 dotnet-maestro[bot] 提交于
* Update dependencies from https://github.com/dotnet/hotreload-utils build 20220920.1 Microsoft.DotNet.HotReload.Utils.Generator.BuildTool From Version 1.1.0-alpha.0.22469.1 -> To Version 1.1.0-alpha.0.22470.1 * Update dependencies from https://github.com/dotnet/roslyn-analyzers build 20220921.2 Microsoft.CodeAnalysis.NetAnalyzers From Version 7.0.0-preview1.22466.1 -> To Version 7.0.0-preview1.22471.2 Co-authored-by: Ndotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com>
-
由 Vlad Brezae 提交于
Existing since its prototype
-
由 Ankit Jain 提交于
* CI: trigger library tests builds, when building runtime tests .. as they depend on it. * Mono product build jobs only build the runtime, so don't trigger on library changes * CI: Exclude some eng/pipelines files from triggering pipelines ```yml # src/workloads is only used in runtime-official builds # where evaluate-paths is not used _always_exclude: [ eng/pipelines/common/evaluate-default-paths.yml eng/pipelines/coreclr/* eng/pipelines/installer/* eng/pipelines/libraries/* eng/pipelines/mono/* eng/pipelines/runtime* src/workloads/* ] ``` * wip * Revert "wip" This reverts commit 0b08b8d170fbd9168f9bff3470a54f6f40daa783.
-
由 Larry Ewing 提交于
* Work around main still targeting net7.0
-
由 Vladimir Sadov 提交于
-
由 Adeel Mujahid 提交于
* Switch ilc to use System.CommandLine * Switch cg2 to use System.CommandLine * Fix release build * Fix test build * ilc: --insructionset, crossgen2: --instruction-set 🥲 * Fix a couple of defaults * Fix repro package arguments for cg2 * ilc: rename --instructionset to --instruction-set * Remove 'Microsoft (R)' from tools' banner * Merge conflict typo fix * Add -v alias for version (to match old behavior)
-
由 SingleAccretion 提交于
There is no need for this.
-
由 SingleAccretion 提交于
On ARM64, we turn the "EQ(CAST<byte>(x), 0)" into "TEST_EQ(x, 0xFF)", to avoid materializing the cast. However, this breaks recognition of the "cb[n]z" idiom. If "x" ends up in a register, that's ok - we're essentially exchanging relop for a cast. If, however, it was contained, it is better to leave it alone: ``` - ldr w0, [fp,#0x18] - tst w0, #255 - bne G_M25131_IG10 - ;; size=12 bbWeight=1 PerfScore 3.50 + ldrb w0, [fp,#0x18] + cbnz w0, G_M25131_IG10 + ;; size=8 bbWeight=1 PerfScore 3.00 ```
-
由 Jakob Botsch Nielsen 提交于
We may get here for any invariant dividend/divisor but these can be 'complex' address-of trees that gtClone does not handle. Fix #76051
-
由 SingleAccretion 提交于
* Delete IND<struct> handling from local morph * Fix raw handle intrinsics
-
由 Jakob Botsch Nielsen 提交于
Also change it to check for any local, not just GT_LCL_VAR.
-
由 Andy Ayers 提交于
Under jit stress, `optOptimizeBoolsGcStress` will modify trees. Since this now happens in a phase that runs after `fgSetBlockOrder`, it must also rethread the trees.
-
- 23 9月, 2022 14 次提交
-
-
由 Jo Shields 提交于
Fixes build failures in official build. Tested locally
-
由 Nathan Ricci 提交于
-
由 Michal Strehovský 提交于
This was disabled by default in #76000. I started filling out the backport template when it hit me that we want to keep it here. We'll want to backport these in tandem.
-
由 Nathan Ricci 提交于
Better descriptions for why these tests are disabled.
-
由 Stephen Toub 提交于
* Fix Dictionary perf regression for non-string ref type keys Several releases ago, some performance improvements were made to Dictionary that were particularly valuable for value type keys, enabling equality comparisons to be devirtualized and possibly inlined. For non-string reference type keys, however, this can end up being a regression, as every dictionary access then needs to access `EqualityComparer<TKey>.Default`, which can be more expensive with shared generics. The access is hoisted out of a loop, but at least one per call is still needed. This fixes the regression by ensuring we always store a comparer in the dictionary's constructor if TKey is a reference type. The same fix is applied to `HashSet<T>`. This also means we can delete some now unreachable code. * Address PR feedback * Fix one more existing TODO * Revert previous PR feedback changes * Address PR feedback
-
由 Michal Strehovský 提交于
`DiagnosticSource` is currently not AOT compatible. If a machine-wide DiagnosticSource-related event listener is enabled (such as PerfView, or possibly even a managed VS debugging session) it activates `DiagnosticSource` code paths within the executable and basically injects a runtime failure into NativeAOT processes due to the AOT-incompatibility of the code. E.g. trying to do a `HttpClient` web request with PerfView collecting in the background causes a runtime exception to be thrown. This uses the documented switch to disable `EventSource` support (unless the user specified a different value). Indirectly, it disables `DiagnosticSource` as well. As a side effect, disabling `EventSource` drops the size of a NativeAOT-compiled Hello World from 3.48 MB to 2.85 MB 🥳.
-
由 Filip Navara 提交于
* Ensure that frozen objects respect the minimum object size * Handle FrozenStringNode too
-
由 Andy Ayers 提交于
Update `fgGetDomSpeculatively` to return nullptr if the idom of a block is unreachable, even if other preds have refs. Fixes #72767.
-
由 Adeel Mujahid 提交于
* Remove usages of Ubuntu 16.04 from pipelines * Delete old jenkins scripts * Improve handling of /p:CrossBuild=true
-
由 Parker Bibus 提交于
* Updated the Maui Perf net6.0 to get the latest daily 6.0 version. * Remove TODO comment. * Setup for testing. * Revert "Setup for testing." This reverts commit 903014e269670595b05e08eca4d6de1b8edeab9f.
-
由 Egor Bogatov 提交于
-
由 Aaron Robinson 提交于
* Remove locks from COM events delegate management. This removes locks and instead assumes the collection is immutable. * Use array instead of List<T> * Remove usage of `Delegate.Combine`. Upon deeper inspection there doesn't seem to be any value to using that mechanism.
-
由 Andy Ayers 提交于
Use a bit vector to track the visited blocks. This scales much better than using the per-block visited flags. Fixes #44341.
-
由 Egor Bogatov 提交于
Co-authored-by: NMichal Strehovský <MichalStrehovsky@users.noreply.github.com> Co-authored-by: NJan Kotas <jkotas@microsoft.com> Co-authored-by: NJakob Botsch Nielsen <Jakob.botsch.nielsen@gmail.com>
-
- 22 9月, 2022 13 次提交
-
-
由 Shay Rojansky 提交于
-
由 Dan Moseley 提交于
-
由 Eric Erhardt 提交于
* Configuring Console Json Formatting with JsonWriterOptions doesn't work with Trimming Ensure the correct members are preserved when trimming and add a trimming test. Fix #73822 * Enable NativeAOT test that should work now Fix #73436
-
由 Marek Safar 提交于
-
由 SingleAccretion 提交于
* Delete dead verification code * Untangle verCheckTailCallConstraint We only use it for tail call stress.
-
由 SingleAccretion 提交于
It is unused. Tighten an assert in the ref counter.
-
由 Viktor Hofer 提交于
* Update slngen version to work with .NET 7 SDKs * Batch update all solution files under src/libs ... by invoking the following command: "dotnet.cmd build src/libraries/slngen.proj".
-
由 Viktor Hofer 提交于
* Clean-up InteropServices.JavaScript project file
-
由 Matous Kozak 提交于
* TODO entry point for SN_sum * Sum for Vector128 AMD64 in progress * Vector128 Sum intrinsic (not supporting byte type) * remove semicolon * space in function def * defined macro * code style fix * SIMD intrinsics check, code style fixes * move ISA check after i64 code * Fast log -1 illegal entries
-
由 Anton Lapounov 提交于
-
由 Hyungju Lee 提交于
* Set names to runtime internal threads * Fix MAC build * Shorten ThreadPool thread names * Fix not to return SetThreadDescription() * Change SetThreadDescription to SetThreadName * Set thread names inside thread functions * Add missing name for MAC * Move name change info thread function Co-authored-by: NJan Kotas <jkotas@microsoft.com>
-
由 Andy Ayers 提交于
In stress modes (and in normal uses, see test case) the jit may insert patchpoint helper calls in regions where last error is live. So the helpers need to preserve last error. Because some invocations of the helpers may transition to OSR methods instead of returning, we can't use the normal macros for this. Fixes #75828.
-
由 Adam Sitnik 提交于
* bring back the old code... * bring back more old code * Use an ifdef around clr code instead of a separate file * Delete SpanHelpers.Clr.cs * Remove a remaining INumber<T> helper from mono Co-authored-by: NJeff Handley <jeffhandley@users.noreply.github.com>
-