-
Notifications
You must be signed in to change notification settings - Fork 4.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Update mono to support shuffle for constant inputs #105299
Conversation
Tagging subscribers to this area: @lambdageek, @steveisok |
This comment was marked as resolved.
This comment was marked as resolved.
This comment was marked as resolved.
This comment was marked as resolved.
No diffs for x64 is expected above since the x64 changes are MonoLLVM only. For arm64 it should be both MonoJIT and MonoLLVM |
This comment was marked as resolved.
This comment was marked as resolved.
@EgorBot -arm64 -mono --envvars "MONO_VERBOSE_METHOD:ShuffleBench" using System.Numerics;
using BenchmarkDotNet.Attributes;
using System.Runtime.Intrinsics;
public class StrengthReduction
{
Vector128<int> srcVec = Vector128.Create(1, 2, 3, 4);
Vector128<int> dstVec;
[Benchmark]
public void ShuffleBench() =>
dstVec = Vector128.Shuffle(srcVec, Vector128.Create(2, 2, 2, 2));
[Benchmark]
public Quaternion ConcatenateBenchmark() =>
Quaternion.Concatenate(Quaternion.Identity, Quaternion.Identity);
} |
Benchmark results on Arm64
|
Need to fix a MonoLLVM assert, but... Before:
After:
|
…tch the necessary parameter count
return ins; | ||
} | ||
|
||
if (is_xconst (ins) && is_const (element)) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@matouskozak or @fanyang-mono, is there an easy way to remove MonoInst *
that are known to be dead or is it perhaps handled elsewhere?
In a case like this, element
will be dead because we're just folding it directly into the constant. In the case that ins
is xzero
or xones
we will likewise be replacing it with an xconst
that includes the underlying constant from element
.
So in the example given above, the following IR is "dead":
iconst R36 <- [2]
iconst R37 <- [2]
iconst R38 <- [2]
iconst R39 <- [2]
xzero R40 <- [System.Runtime.Intrinsics.Vector128`1<int>]
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I haven't tried it but I think MONO_REMOVE_INS
(or MONO_DELETE_INS
) could work for this.
762d241
to
2995754
Compare
d05673f
to
b35b1e3
Compare
This should now be ready for review. -- Please feel free to call out anywhere I messed up the styling. Tried to follow the conventions, but it's entirely possible I made some mistakes. |
@EgorBot -arm64 -mono --envvars "MONO_VERBOSE_METHOD:ShuffleBench" using System.Numerics;
using BenchmarkDotNet.Attributes;
using System.Runtime.Intrinsics;
public class StrengthReduction
{
Vector128<int> srcVec = Vector128.Create(1, 2, 3, 4);
Vector128<int> dstVec;
[Benchmark]
public void ShuffleBench() =>
dstVec = Vector128.Shuffle(srcVec, Vector128.Create(2, 2, 2, 2));
[Benchmark]
public Quaternion ConcatenateBenchmark() =>
Quaternion.Concatenate(Quaternion.Identity, Quaternion.Identity);
} |
Benchmark results on Arm64
|
Both of those numbers for MonoJIT are very good improvements:
I had done some local testing for a few other scenarios and seen similar improvements to APIs using |
For Before 0: a9ba7bfd stp x29, x30, [sp, #-96]!
4: 910003fd mov x29, sp
8: f9000bba str x26, [x29, #16]
c: aa0003fa mov x26, x0
10: d29c1000 mov x0, #0xe080 // #57472
14: f2b6d620 movk x0, #0xb6b1, lsl #16
18: f2defa00 movk x0, #0xf7d0, lsl #32
1c: f9400011 ldr x17, [x0]
20: b4000051 cbz x17, 0x28
24: 954b13d1 bl 0x52c4f68
28: 3dc00740 ldr q0, [x26, #16]
2c: 3d800fa0 str q0, [x29, #48]
30: d2800044 mov x4, #0x2 // #2
34: d2800043 mov x3, #0x2 // #2
38: d2800042 mov x2, #0x2 // #2
3c: d2800041 mov x1, #0x2 // #2
40: 4f00e400 movi v0.16b, #0x0
44: 4e041c80 mov v0.s[0], w4
48: 4e0c1c60 mov v0.s[1], w3
4c: 4e141c40 mov v0.s[2], w2
50: 4e1c1c20 mov v0.s[3], w1
54: 3d8013a0 str q0, [x29, #64]
58: 910083a0 add x0, x29, #0x20
5c: f9002ba0 str x0, [x29, #80]
60: f9401ba0 ldr x0, [x29, #48]
64: f9401fa1 ldr x1, [x29, #56]
68: f94023a2 ldr x2, [x29, #64]
6c: f94027a3 ldr x3, [x29, #72]
70: 9400001c bl 0xe0
74: f9402bbe ldr x30, [x29, #80]
78: a90007c0 stp x0, x1, [x30]
7c: eb1f035f cmp x26, xzr
80: 10000011 adr x17, 0x80
84: 540000e0 b.eq 0xa0 // b.none
88: 3dc00ba0 ldr q0, [x29, #32]
8c: 3d800b40 str q0, [x26, #32]
90: f9400bba ldr x26, [x29, #16]
94: 910003bf mov sp, x29
98: a8c67bfd ldp x29, x30, [sp], #96
9c: d65f03c0 ret
a0: d28027c0 mov x0, #0x13e // #318
a4: aa1103e1 mov x1, x17
a8: 954b50ac bl 0x52d4358 After: 0: a9be7bfd stp x29, x30, [sp, #-32]!
4: 910003fd mov x29, sp
8: f9000bba str x26, [x29, #16]
c: aa0003fa mov x26, x0
10: 3dc00740 ldr q0, [x26, #16]
14: 9c0001e1 ldr q1, 0x50
18: 4e010000 tbl v0.16b, {v0.16b}, v1.16b
1c: eb1f035f cmp x26, xzr
20: 10000011 adr x17, 0x20
24: 540000c0 b.eq 0x3c // b.none
28: 3d800b40 str q0, [x26, #32]
2c: f9400bba ldr x26, [x29, #16]
30: 910003bf mov sp, x29
34: a8c27bfd ldp x29, x30, [sp], #32
38: d65f03c0 ret
3c: d28027c0 mov x0, #0x13e // #318
40: aa1103e1 mov x1, x17
44: 94000007 bl 0x60
48: b9006740 str w0, [x26, #100]
4c: b9806740 ldrsw x0, [x26, #100]
50: 0b0a0908 add w8, w8, w10, lsl #2
54: 0b0a0908 add w8, w8, w10, lsl #2
58: 0b0a0908 add w8, w8, w10, lsl #2
5c: 0b0a0908 add w8, w8, w10, lsl #2 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! Just some minor format adjustment needed. Thanks for adding this for mono.
I'm not sure whether interp does this kind of optimization, but jiterpreter does, just FYI. If it becomes important we could probably move this optimization from the jiterpreter up into the interpreter, but I think the only target where interpreter performance really matters right now is WASM. |
* Support mono creating xconst in a few more places * Update mono to support shuffle for constant inputs * Ensure that arm64 also accelerates shuffle for non-constant inputs * Ensure OP_XZERO and OP_XONES are recognized as being constant * Ensure shuffle creates a correct instruction when the fsig doesn't match the necessary parameter count * Ensure that getting the index for floating-point shuffle is possible * Ensure the right class handle is passed down to LLVM so overload resolution can function * Make sure we update the original xconst if we mutate it * Return a new constant and instead of mutating the existing one * Insert relevant xcast nodes * Add some asserts around the ecount * Ensure we get the right element type * Ensure we don't create nodes unnecessarily for create_elementwise * Ensure that create_elementwise still works for other vector sizes * Ensure indentation of switch cases is correct for Mono
…05110) * Remove support for older LLVM versions, and re-order linker flags We have generally tried to support linking multiple versions of LLVM within our git tree. Every new LLVM version moves symbols around between libraries, and as a result, every new version of LLVM requires different linker flags to build. The command line tool `llvm-config` should tell you the exact flags you need, but it is a problem for us when cross-compiling to rely on this, and as a result, we transcribe the result of various llvm-config outputs directly into Mono's CMakeLists.txt. In an effort to support multiple versions of LLVM, flags common between all supported versions were kept in one place, then the version-specific flags appended afterwards. And this has worked fine for years. However: 1. Whilst we only link with `lld`, it is common for contributors and source-build to link with `gold`, `bfd`, or some other GNU-flavoured linker, where library order is essential 2. The list of common libraries to link has remained unchanged for years, but the symbol intra-dependencies may have changed a long time ago, so common symbol order cannot be assumed to remain valid between LLVM versions This has resulted in a long-standing problem for people using e.g. Debian or Ubuntu or GitHub CodeSpaces, rather than always building with one of our dockerfiles representing our "real" build environment, when targeting platforms which use Mono and link LLVM. * Bumping clang and llvm - make docs less ambiguous. (#105401) * Bump main to RC1 (#105338) * Update SDK to preview 6 (#104696) * Update SDK to preview 6 * Update Shared.csproj Fix `error NU1903: Package 'System.Text.Json' 8.0.0 has a known high severity vulnerability` * Fix with existing version. --------- Co-authored-by: Ilona Tomkowicz <32700855+ilonatommy@users.noreply.github.com> Co-authored-by: Viktor Hofer <viktor.hofer@microsoft.com> * Change DefaultMaximumErrorResponseLength to KB from Byte (#105396) * Change DefaultMaximumErrorResponseLength to KB from Byte * Handle overflow * Review feedback * Fix warning for MakeGenericType annotation mismatch (#104921) Fixes warning code when a generic type whose type parameters have DAM annotations is used with MakeGenericType, over a type that doesn't have matching annotations. The code IL2070 used to mention the 'this' argument. Instead it should have been IL2071 which mentions the generic argument as the cause of the mismatch. Similar for MakeGenericMethod with IL2090 and IL2091. * Set GCStressIncompatible on GenericContext tests (#104686) Co-authored-by: Vladimir Sadov <vsadov@microsoft.com> * Add runtime config parameter to force ijwhost to load assemblies in an isolated context (#105337) * Add support for isolated load context in LoadInMemoryAssemblyInContext by passing -1 as loadContext * Have ijwhost check a runtime config parameter to determine if it should run in an isolated load context * Added test for ijwhost isolated load context runtime config option * [RISC-V] Fix passing float and uint arguments in VM (#105021) * Add tests * Fix passing float and uint arguments in VM * Change test lib name so it doesn't clash with managed DLL on Windows * Fix platform analyzer attribute order for MacCatalyst (#105409) We need to make sure the attribute for MacCatalyst comes _after_ the iOS one due to how MacCatalyst is a superset of iOS: https://learn.microsoft.com/en-us/dotnet/standard/analyzers/platform-compat-analyzer#platform-inclusion This caused an error in aspnetcore in the latest dependency flow because the analyzer thought AesGcm is _only_ supported on MacCatalyst: > error CA1416: (NETCORE_ENGINEERING_TELEMETRY=Build) This call site is reachable on all platforms. 'AesGcm.Decrypt(ReadOnlySpan<byte>, ReadOnlySpan<byte>, ReadOnlySpan<byte>, Span<byte>, ReadOnlySpan<byte>)' is only supported on: 'maccatalyst' 13.0 and later. * Use correct `ExceptionArgument` value in `System.IO.Pipelines` (#105418) * Remove zlib from requirements script and instruction files (#105419) * Remove zlib from requirements instructions * Remove zlib from native requirements installation script * Revert "Remove zlib from requirements script and instruction files (#105419)" (#105449) This reverts commit 3ec6286. * Ensure that WaitForPendingFinalizers has seen the expected Full GC count (#105289) * Ensure that WaitForPendingFinalizers has seen the expected Full GC * NativeAOT and some renames * a testcase * make the test not unsafe and make OuterLoop * Use unsigned math when comparing collection ticks * cast the diff to int when comparing gc ticks * Migrate to zlib-ng, part 3: Remove zlib and zlib-intel source code and license mentions (second attempt) (#105371) * Remove zlib/ * Remove zlib-intel/ * Remove third party notice * Remove patches * Remove version txts * Remove cgmanifest.json entries * Remove installer third party notice * Update docs --------- Co-authored-by: Jan Kotas <jkotas@microsoft.com> * [browser] Trigger relink on `EmccMaximumHeapSize` change (#105027) * Edit test + trigger relink. * Remove logging to speed up the test + decrease loop runs to prevent "Browser has been disconnected" error. * Feedback - properties are not bool-only anymore. * Fix: workload needed when heap size set. --------- Co-authored-by: Larry Ewing <lewing@microsoft.com> * Delete erroneous Socket test (#105448) * Pull the python dependency from the EmsdkVersion where possible (#105437) * Use `BinaryPrimitives` more in the ILCompiler (#105404) * JIT: use VNVisitReachingVNs in IsVNNeverNegative (#105197) * Set xunit env var to not print output for passing tests (#105392) * Set xunit env var to not print output for passing tests Fixes #103445 * Update xunit.console.targets * Update xunit.console.targets * Some more automated C# modernization in corelib (#105151) * Fix IDE0056 on corelib (indexing can be simplified) * Fix IDE (null check can be simplified) * Fix IDE0078 (use pattern matching) * Fix IDE0019 (use pattern matching) * Fix IDE0066 (use switch expression) * Fix IDE0250 (struct can be made readonly) * Fix nullability warning and address PR feedback * Address PR feedback and revert a downlevel change * Wrap any `?? throw new`s that go beyond 120 characters * Fix ECMA 355 Partition download links (#105454) On https://github.com/dotnet/runtime/blob/main/docs/project/dotnet-standards.md the Partition with Notes download links using HTTP protocol fail to download in Chrome: >Mixed Content: The site at 'https://github.com/' was loaded over a secure connection, but the file at 'https://download.microsoft.com/download/7/3/3/733AD403-90B2-4064-A81E-01035A7FE13C/MS%20Partition%20I.pdf' was redirected through an insecure connection. This file should be served over HTTPS. See https://blog.chromium.org/2020/02/protecting-users-from-insecure.html for more details. (It might be reasonble for somebody else to followup fixing all domains on all pages with regex `\(http://([.a-z0-9-]+)` replacing with `(https://$1`, but I didn't test each blog site supports HTTPS.) * Add swiftcall signature check for `mono_class_try_get_swift_error_class` (#105408) * Add signature check for swiftcall * Handle null values for swift_error_ptr * Enable NativeAOT runtime tests on MacCatalyst (#102882) This PR updates the CLRTest.Execute.Bash.targets file to set the apple run command for MacCatalyst. The command apple just-run used on Apple mobile is not permitted, and apple test requires the a test runner. Additionally, it is necessary to locate Info.plist in the Contents/ directory and the binary in Contents/MacOS/ within the bundle. --------- Co-authored-by: Ivan Povazan <ivan.povazan@gmail.com> * Resolving an antigen failure (#105260) * Resolving an antigen failure * Fix method accessibility so xunit doesn't complain * Support field access on GetType() of T constrained to be Enum (#105351) Adds trimming support for instance.GetType().GetFields(), where instance is a variable of type `T` that is constrained to System.Enum. This includes a change to have ILLink's TypeProxy track a TypeReference instead of TypeDefinition, which was necessary to allow TypeProxy to represent a generic parameter. Note that this only supports the specific case where `GetType()` is called on a variable of type `T` that is constrained to `Enum`. A variable of type `Enum` is not supported, so the following will still warn: ```csharp static void M(Enum v) { v.GetType().GetFields(); } ``` * Update mono to support shuffle for constant inputs (#105299) * Support mono creating xconst in a few more places * Update mono to support shuffle for constant inputs * Ensure that arm64 also accelerates shuffle for non-constant inputs * Ensure OP_XZERO and OP_XONES are recognized as being constant * Ensure shuffle creates a correct instruction when the fsig doesn't match the necessary parameter count * Ensure that getting the index for floating-point shuffle is possible * Ensure the right class handle is passed down to LLVM so overload resolution can function * Make sure we update the original xconst if we mutate it * Return a new constant and instead of mutating the existing one * Insert relevant xcast nodes * Add some asserts around the ecount * Ensure we get the right element type * Ensure we don't create nodes unnecessarily for create_elementwise * Ensure that create_elementwise still works for other vector sizes * Ensure indentation of switch cases is correct for Mono * Make TooDeepJsonDocument test more consistent across platforms (#105445) * Make TooDeepJsonDocument test more consistent across platforms Run the test on a thread with as consistent a stack size as possible so that we don't inadvertently succeed due to having a really large stack. * Disable test on mono * Update the TypeLib embedding and add comments on API use (#105416) There is an undocumented semantic of Win32 Resource APIs. The missing semantic is that all resource type/name strings are transparently converted to uppercase when calling any of the Win32 Resource APIs. We don't want to apply this undocumented semantic to the reader/writer API so we document it instead. We are avoiding applying the behavior since ReadyToRun scenarios are designed to be a byte for byte copy of the resource, including name as it was written by other tooling. * Update docs for ByRefLike with generics for work in .NET 10 (#103318) Co-authored-by: Jan Kotas <jkotas@microsoft.com> * Fix double printing in StressLog and simplify stresslog macros (#105420) * Add support for nested types in the `corelib.h` parser for rooting descriptors (#105432) * Fixed for assertion failure due to not checking if we are processing Eph samples (#105164) * [PERF] Use correct python executable on windows in venv (#105451) * Fix up Fuzzlyn CI scripts for new hardware intrinsics support (#105470) 1) Strip out the extensions in the seed name when using it for file/directory names, since the list of extensions is quite long 2) Limit the number of unreduced/uncategorized example seeds we show * Try to re-enable DeepEquals_TooDeepJsonDocument_ThrowsInsufficientExecutionStackException test on mono (#105509) * zlib-ng: avoid suppressing WD4242 and WD4244 (#105433) WD4242 and WD4244 are compiler warnings that should not be suppressed because the warn about possible loss of data. WD4242 shows up in zlib-ng/arch/*/slide_hash*.c files and comes from the arguments passed to the slide_hash_chain method. WD4244 happens in Windows when building in Debug configuration, in various zlib-ng/deflate*.c files, and comes from the arguments passed to the check_match method. Fixed by: - Adding asserts to verify the values are below the maximum allowed for their type. - Casting them the proper type before passing them as arguments to their methods. - Removing the WD suppressions, which unfortunately also propagated to other unrelated cmake files. - Fixed a similar loss of data error in an unrelated mono file where the warning suppression was propagated due to this. * Delete outdated comments (#105519) * Arm64/Sve: Add FFR register liveness tracking (#105348) * Add tracking of FFR register somewhat workable code cleanup Remove FFR Add all the GetFfr* wip Work with MskCns() model Use physReg approach Remove commented prototypes working Remove bunch of unnecessary code Remove SpecialImport from GetFFR/SetFFR/LoadFirstFaulting some more code cleanup some fixup * Change condition for PhysReg * jit format * review feedback * unspill for LoadVectorFirstFaulting as well * Use the right opReg * skip spilling tracking * review feedback * Use non-existent REG_FFR * Do not reload from FFR for GetFfr() * review feedback * Make just GrabTemp * fix build and formatting * missed another build failure for arm * Fix throwing exception when calling RunClassConstructor on a generic type with a static constructor (#105513) * Fix throwing exception when calling RunClassConstructor on a generic type with a static constructor #99183 seems to have done away with assuming that a generic type's static constructor was always "initialized". As a result, if you call RunClassConstructor on it, the runtime would throw an exception. Fixes #103891 * Apply suggestions from code review --------- Co-authored-by: Jan Kotas <jkotas@microsoft.com> * [RISC-V][LoongArch64] New passing info for floating-point structs (#103945) * Replace StructFloatFieldInfoFlags with FpStructInRegistersInfo which carries also exact field sizes and offsets * Replace StructFloatFieldInfoFlags with FpStruct::Flags in profiler * Remove FpStructInRegistersInfo::FromOldFlags() * Fix duplicating types in HandleInlineArray * Remove signedness from FpStruct::IntKind because most probably we won't need it * Remove old StructFloatFieldInfoFlags calculating routine * Typo in TARGET_LOONGARCH64 * Remove m_returnedFpFieldOffsets from ArgIterator * Add missing ENREGISTERED_PARAMTYPE_MAXSIZE condition to C# version of FpStruct info calculation * Rename RISCV64PassStructInRegister to match settled casing for RiscV in class names * Update hardcoded flags for float and double in ArgIteratorTemplate::ComputeReturnFlags() This fixes JIT/HardwareIntrinsics/General/Vector* tests. * Fix build on other platforms * Update LoongArch to use FpStructInRegistersInfo * Remove unused old flag masks * LoongArch64 typo Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> * Missing FpStruct namespace Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> * Missing FpStruct namespace Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> * Missing FpStruct namespace Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> * Use FpStruct namespace everywhere in JIT * JIT review * Update StructFloatFieldInfoFlags description * Revert to hitherto instruction set order as it's not the point of this PR * Unify get{LoongArch,RiscV}64PassFpStructInRegistersInfo JIT interfaces * Use JIT_TO_EE_TRANSITION instead of _LEAF because MethodTable::GetFpStructInRegistersInfo may throw * Remove FpStruct::IntKind, we should have similar info in ClassLayout in JIT * Change JIT interface to return a struct similar to CORINFO_SWIFT_LOWERING to facilitate code unification in the future * Change JIT to use new Swift-like getFpStructLowering * Cache CORINFO_FPSTRUCT_LOWERING * Update LoongArch classifier to use CORINFO_FPSTRUCT_LOWERING * Update StructFloatInfoFlags doc comment on C# * Move StructFloatFieldInfoFlags and FpStructInRegistersInfo out of the JIT interface * Merge LoongArch and RISC-V AOT calculation of FpStructInRegistersInfo because they were identical. Move it to Common\Internal/Runtime because it's no longer exposed in JIT interface. * Don't zero-initialize CORINFO_FPSTRUCT_LOWERING * Add note for CORINFO_FPSTRUCT_LOWERING::loweredElements type --------- Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> * Ensure we don't reuse temps when calling fgMorphArgs on LIR nodes (#105508) * Ensure constant evaluation of shifts on xarch broadcast the operand to the correct size (#105487) * Ensure constant evaluation of shifts on xarch broadcast the operand to the correct size * Ensure we don't try to execute AVX2 code on unsupported platforms * Use ConcurrentDictionary in runtimecounters test (#105520) * Use ConcurrentDictionary in runtimecounters test Fixes #105443 * Fix build break * Fix issue 98506 - Excessive exceptions generated in StackTraceSymbols (#105530) * Fix issue 98506 - Excessive exceptions generated in StackTraceSymbols * Code review feedback * Clean up some usages of LowLevelList<T> (#105407) * Fix ShuffleThunk cache heap (#105480) There was a problem with using heap from the related LoaderAllocator for shuffle thunk cache heap. I have tested it again and it seems that the issue is gone. So I am removing the workaround, making the cache use LoaderAllocator local heap. Close #55697 * [browser] Fix computing destination sub path and publish extension target path in Wasm SDK (#105458) * Bump flags to LLVM 19, not 16 --------- Co-authored-by: Larry Ewing <lewing@microsoft.com> Co-authored-by: Alexander Köplinger <alex.koeplinger@outlook.com> Co-authored-by: Ilona Tomkowicz <32700855+ilonatommy@users.noreply.github.com> Co-authored-by: Carlos Sánchez López <1175054+carlossanlop@users.noreply.github.com> Co-authored-by: Adeel Mujahid <3840695+am11@users.noreply.github.com> Co-authored-by: Viktor Hofer <viktor.hofer@microsoft.com> Co-authored-by: Ahmet Ibrahim Aksoy <aaksoy@microsoft.com> Co-authored-by: Sven Boemer <sbomer@gmail.com> Co-authored-by: Steve Pfister <steveisok@users.noreply.github.com> Co-authored-by: Vladimir Sadov <vsadov@microsoft.com> Co-authored-by: Mike Oliphant <oliphant@nostatic.org> Co-authored-by: Tomasz Sowiński <tomeksowi@gmail.com> Co-authored-by: xtqqczze <45661989+xtqqczze@users.noreply.github.com> Co-authored-by: Jan Kotas <jkotas@microsoft.com> Co-authored-by: Stephen Toub <stoub@microsoft.com> Co-authored-by: Paulus Pärssinen <paulus.parssinen@gmail.com> Co-authored-by: Egor Bogatov <egorbo@gmail.com> Co-authored-by: Carl Walsh <darthwalsh@gmail.com> Co-authored-by: Milos Kotlar <kotlarmilos@gmail.com> Co-authored-by: Ivan Povazan <ivan.povazan@gmail.com> Co-authored-by: Tanner Gooding <tagoo@outlook.com> Co-authored-by: Aaron Robinson <arobins@microsoft.com> Co-authored-by: Jeremy Koritzinsky <jekoritz@microsoft.com> Co-authored-by: Mukund Raghav Sharma (Moko) <68247673+mrsharm@users.noreply.github.com> Co-authored-by: Cameron Aavik <99771732+caaavik-msft@users.noreply.github.com> Co-authored-by: Jakob Botsch Nielsen <Jakob.botsch.nielsen@gmail.com> Co-authored-by: Kunal Pathak <Kunal.Pathak@microsoft.com> Co-authored-by: Qiao Pengcheng <qiaopengcheng@loongson.cn> Co-authored-by: Mike McLaughlin <mikem@microsoft.com> Co-authored-by: Huo Yaoyuan <huoyaoyuan@hotmail.com> Co-authored-by: Jan Vorlicek <janvorli@microsoft.com> Co-authored-by: Marek Fišera <mara@neptuo.com>
Mono wasn't accelerating
shuffle
for anything other than byte/sbyte on x64. This was in part because doing so required detecting inputs as constant which didn't exist when that support was first brought online.However, I had since added basic support for constants last year in #81902. This PR adds support that mirrors most of the RyuJIT
Shuffle
importation logic and so we can now do the right thing. As a side effect, the general support for generating such constants forCreate
calls when all inputs are constants was also expanded which should have some indirect benefits in other places.Shuffle
having not being accelerated is very impactful to the entire BCL and I would expect this to have a broad impact across many core benchmarks includingHexConverter
,XxHash
,TensorPrimitives
,ReverseEndianness
,Base64Decoder
,Guid
,Matrix4x4
,Quaternion
,IndexOfAny
,ProbablisticMap
, variousSpanHelpers
APIs and more.