port SpanHelpers.IndexOfAny(ref byte, byte, byte, int) to Vector128/256

[adamsitnik]

It's #73384 again

I've tried to repro #73474 locally and I've failed. Now I want to use the CI to verify if the issue is gone or not.

./build.sh --os browser -c Release
make -C src/mono/wasm provision-wasm
export EMSDK_PATH=/home/adam/projects/runtime/src/mono/wasm/emsdk
./dotnet.sh build /t:Test /p:TargetOS=Browser /p:TargetArchitecture=wasm /p:RunAOTCompilation=true /p:EnableAggressiveTrimming=true -c Release ./src/libraries/System.Runtime/tests

 ~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Runtime/tests
[1.0.0-prerelease.22375.5+d43453b9981bfd8705025250caff58d18e2abc7e] XHarness command issued: wasm test --app=. --output-directory=/home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output -s dotnet.js.symbols --symbol-patterns wasm-symbol-patterns.txt --symbolicator WasmSymbolicator.dll,Microsoft.WebAssembly.Internal.SymbolicatorWrapperForXHarness --engine=V8 --engine-arg=--stack-trace-limit=1000 --js-file=test-main.js -- --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
warn: Cannot find symbols file dotnet.js.symbols
info: info: Running v8 --expose_wasm --stack-trace-limit=1000 test-main.js -- --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: Incoming arguments: --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: Application arguments: --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: console.debug: mono_wasm_runtime_ready fe00e07a-5519-4dfe-b35a-f867dbaf2e28
info: console.info: Initializing.....
info: Discovering: System.Runtime.Tests.dll (method display = ClassAndMethod, method display options = None)
info: Discovered: System.Runtime.Tests.dll (found 8900 of 9043 test cases)
info: Using random seed for test cases: 1658957131
info: Using random seed for collections: 1658957131
info: Starting: System.Runtime.Tests.dll
info: Finished: System.Runtime.Tests.dll
info: info: === TEST EXECUTION SUMMARY ===
info: Total: 49567, Errors: 0, Failed: 0, Skipped: 85, Time: 42.670667s
info: info: Received expected 13414172 of /home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output/testResults.xml
info: Finished writing 13414172 bytes of RESULTXML
info: Xml file was written to the provided writer.
info: Tests run: 49567 Passed: 49482 Inconclusive: 0 Failed: 0 Ignored: 0 Skipped: 85
info: Process v8 exited with 0
info: Waiting to flush log messages with a timeout of 120 secs ..
info: Application has finished with exit code: 0
XHarness exit code: 0
~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Runtime/tests
----- end Mon Aug 8 11:33:56 CEST 2022 ----- exit code 0 ----------------------------------------------------------
XHarness artifacts: /home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output
Build succeeded.
0 Warning(s)
0 Error(s)
Time Elapsed 00:04:33.82

[ghost]

Tagging subscribers to this area: @dotnet/area-system-memory
See info in area-owners.md if you want to be subscribed.

Issue Details

---

It's #73384 again

I've tried to repro #73474 locally and I've failed. Now I want to use the CI to verify if the issue is gone or not.

./build.sh --os browser -c Release
make -C src/mono/wasm provision-wasm
export EMSDK_PATH=/home/adam/projects/runtime/src/mono/wasm/emsdk
./dotnet.sh build /t:Test /p:TargetOS=Browser /p:TargetArchitecture=wasm /p:RunAOTCompilation=true /p:EnableAggressiveTrimming=true -c Release ./src/libraries/System.Runtime/tests

 ~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Runtime/tests
[1.0.0-prerelease.22375.5+d43453b9981bfd8705025250caff58d18e2abc7e] XHarness command issued: wasm test --app=. --output-directory=/home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output -s dotnet.js.symbols --symbol-patterns wasm-symbol-patterns.txt --symbolicator WasmSymbolicator.dll,Microsoft.WebAssembly.Internal.SymbolicatorWrapperForXHarness --engine=V8 --engine-arg=--stack-trace-limit=1000 --js-file=test-main.js -- --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
warn: Cannot find symbols file dotnet.js.symbols
info: info: Running v8 --expose_wasm --stack-trace-limit=1000 test-main.js -- --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: Incoming arguments: --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: Application arguments: --run WasmTestRunner.dll System.Runtime.Tests.dll -notrait category=AdditionalTimezoneChecks -notrait category=OuterLoop -notrait category=failing
info: console.debug: mono_wasm_runtime_ready fe00e07a-5519-4dfe-b35a-f867dbaf2e28
info: console.info: Initializing.....
info: Discovering: System.Runtime.Tests.dll (method display = ClassAndMethod, method display options = None)
info: Discovered: System.Runtime.Tests.dll (found 8900 of 9043 test cases)
info: Using random seed for test cases: 1658957131
info: Using random seed for collections: 1658957131
info: Starting: System.Runtime.Tests.dll
info: Finished: System.Runtime.Tests.dll
info: info: === TEST EXECUTION SUMMARY ===
info: Total: 49567, Errors: 0, Failed: 0, Skipped: 85, Time: 42.670667s
info: info: Received expected 13414172 of /home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output/testResults.xml
info: Finished writing 13414172 bytes of RESULTXML
info: Xml file was written to the provided writer.
info: Tests run: 49567 Passed: 49482 Inconclusive: 0 Failed: 0 Ignored: 0 Skipped: 85
info: Process v8 exited with 0
info: Waiting to flush log messages with a timeout of 120 secs ..
info: Application has finished with exit code: 0
XHarness exit code: 0
~/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Runtime/tests
----- end Mon Aug 8 11:33:56 CEST 2022 ----- exit code 0 ----------------------------------------------------------
XHarness artifacts: /home/adam/projects/runtime/artifacts/bin/System.Runtime.Tests/Release/net7.0-browser/browser-wasm/AppBundle/xharness-output
Build succeeded.
0 Warning(s)
0 Error(s)
Time Elapsed 00:04:33.82

Author:	adamsitnik
Assignees:	-
Labels:	area-System.Memory
Milestone:	-

[adamsitnik]

@jkotas the Build Browser wasm Linux Release LibraryTests_AOT build is green

[jkotas]

What fixed the underlying issue? The crash was caused by reading beyond end of the buffer. The CI may be green by luck - the memory after the end of the buffer might have been initialized just right to avoid the crash in the given run.

@vargaz or @lambdageek should inspect the IL to confirm that the problem is gone. Also, we should find the change that fixed it.

[jkotas]

Ok, #73493 fixed the out of bound read, replaced it by rejecting the whole method with BadImageFormatException. Do we still have a functional issue at runtime?

[adamsitnik]

Do we still have a functional issue at runtime?

System.Memory tests exercise this code path and they are passing on machine with AVX2

 ~/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle ~/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Memory/tests
[1.0.0-prerelease.22375.5+d43453b9981bfd8705025250caff58d18e2abc7e] XHarness command issued: wasm test --app=. --output-directory=/home/adam/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle/xharness-output -s dotnet.js.symbols --symbol-patterns wasm-symbol-patterns.txt --symbolicator WasmSymbolicator.dll,Microsoft.WebAssembly.Internal.SymbolicatorWrapperForXHarness --engine=V8 --engine-arg=--stack-trace-limit=1000 --js-file=test-main.js -- --run WasmTestRunner.dll System.Memory.Tests.dll -notrait category=OuterLoop -notrait category=failing
warn: Cannot find symbols file dotnet.js.symbols
info: info: Running v8 --expose_wasm --stack-trace-limit=1000 test-main.js -- --run WasmTestRunner.dll System.Memory.Tests.dll -notrait category=OuterLoop -notrait category=failing
info: Incoming arguments: --run WasmTestRunner.dll System.Memory.Tests.dll -notrait category=OuterLoop -notrait category=failing
info: Application arguments: --run WasmTestRunner.dll System.Memory.Tests.dll -notrait category=OuterLoop -notrait category=failing
info: console.debug: mono_wasm_runtime_ready fe00e07a-5519-4dfe-b35a-f867dbaf2e28
info: console.info: Initializing.....
info: Discovering: System.Memory.Tests.dll (method display = ClassAndMethod, method display options = None)
info: Discovered: System.Memory.Tests.dll (found 2408 of 2430 test cases)
info: Using random seed for test cases: 75981927
info: Using random seed for collections: 75981927
info: Starting: System.Memory.Tests.dll
info: Finished: System.Memory.Tests.dll
info: info: === TEST EXECUTION SUMMARY ===
info: Total: 47871, Errors: 0, Failed: 0, Skipped: 0, Time: 28.899475s
info: info: Process v8 exited with 0
info: Received expected 12787906 of /home/adam/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle/xharness-output/testResults.xml
info: Waiting to flush log messages with a timeout of 120 secs ..
info: Finished writing 12787906 bytes of RESULTXML
info: Xml file was written to the provided writer.
info: Tests run: 47871 Passed: 47871 Inconclusive: 0 Failed: 0 Ignored: 0 Skipped: 0
info: Application has finished with exit code: 0
XHarness exit code: 0
~/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle ~/projects/runtime/src/libraries/System.Memory/tests
----- end Mon Aug 8 14:50:58 CEST 2022 ----- exit code 0 ----------------------------------------------------------
XHarness artifacts: /home/adam/projects/runtime/artifacts/bin/System.Memory.Tests/Release/net7.0/browser-wasm/AppBundle/xharness-output
Build succeeded.
0 Warning(s)
0 Error(s)

[jkotas]

Do the tests exercise the AOT code of method, or are they falling back to the interpreter silently? If it is the latter, we are likely going to see perf regressions.

[adamsitnik]

Do the tests exercise the AOT code of method, or are they falling back to the interpreter silently? If it is the latter, we are likely going to see perf regressions.

I am using instructions provided by @lambdageek in #73474 (comment) to run these tests

/p:RunAOTCompilation=true /p:EnableAggressiveTrimming=true 

I would expect that they exercise the AOT code of method. How can I check if they don't fall back to interpreter silently?

[jkotas]

I don’t know how to check that. Somebody from the Mono team needs to help you with at that.

[EgorBo]

I guess MONO_VERBOSE_METHOD=methodName might help (it's not expected to print anything when you run already AOT'd binary)

[adamsitnik]

I've tried to benchmark WASM AOT but I don't think that it's currently possible: dotnet/BenchmarkDotNet#1818 (comment)

[lambdageek]

/azp run runtime-wasm

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[lambdageek]

I'm trying the System.Memory tests locally. I would expect any call to IndexOfAny to throw BIFE on wasm AOT. there shouldn't be a fallback to the interpreter, AFAIK. Perhaps we're skipping some of these tests already for other reasons

[lambdageek]

Ah interesting.
Apparently in this configuration the AOT compiler does just skip the method and relies on the interpreter to throw.
But the interpreter doesn't throw - it doesn't have a check that the fallthru on a conditional branch is within the bounds of a method. and dynamically the branch is always taken.

So this PR is good; i'm working, separately, on adding a fallthru check to the interpreter

[jkotas]

this PR is good

Do you agree that this PR is going to introduce perf regression since the method won’t be AOT compiled anymore?

[lambdageek]

this PR is good

Do you agree that this PR is going to introduce perf regression since the method won’t be AOT compiled anymore?

Yes, I should qualify "good" - this PR wont' crash the AOT compiler or the runtime anymore. The perf won't be good.

[lambdageek]

dynamically the branch is always taken.

actually dynamically this code is unreachable

[adamsitnik]

@jkotas@lambdageek@vitek-karas I am using similar patterns in few other PRs and aiming to port all SpanHelpers code to Vector128/256 before we ship 7 (the PRs are ready). The thing I don't understand is what exactly is wrong here. Based on #73474 (comment) I know that linker is producing bad IL.

But why? Is it because I am using Vector128/256 types?
Is my understanding correct that NativeAOT is also using the linker and it's going to face same issues?

[jkotas]

Yes, something in the refactored methods exposed a bug in the IL linker.

Native AOT does not have the bug. It is not sharing the relevant part of the linker.

[lambdageek]

But why? Is it because I am using Vector128/256 types?

I'm not sure, but I think I have a workaround:

if I change the last else if (Vector.IsHardwareAccelerated) to:

else{Debug.Assert(Vector.IsHardwareAccelerated);

then I can also delete

Debug.Fail("Unreachable");gotoNotFound;

and then in the IL that comes out of the linker, I don't see a conditional jump at the end of the method anymore.

(I'm not sure if this has any undesired impact on CoreCLR performance)

[jkotas]

My guess is that the bug is triggered by certain shape of the method flowgraph. It is not about types used.

[vitek-karas]

Sorry - I didn't get to this yet (first on my list after emails/management)...
My theory is that:
The method ends with an if which looks like:

Early1:Early2:if(something){gotoEarly1;}else{gotoEarly2;}

Basically the method ends with an if/else where both branches go backwards. In that case it's invalid to remove the "false" branch, because there's nothing left in the method - and even though that branch is technically not reachable, it needs to exist for validity.

[EgorBo]

Can it still remove it and fold ldc.0 + brfalse to just br ? or there can be other similar shapes (e.g. a local instead of constant)

[vitek-karas]

I'll have to go through the code - but in general the code in the linker is very simplistic and tries to avoid changing the number of instructions (it's problematic to move code around in that setup) - so it tries to keep the same number of instructions. It's also relatively problematic to fully remove the ldc.0 and so on - if the condition is over a call then removing the entire call is relatively complicated - and we don't do that. Which means we need to "pop" somehow.

I'm considering either disabling the optimization in this case or to force insert a dummy ret at the end of the method.

[jkotas]

force insert a dummy ret at the end of the method.

ldnull + throw may be better. Ret requires exactly zero or one items on the evaluation stack.

[adamsitnik]

@jkotas@lambdageek@vitek-karas Thanks for the explanation! I've removed all the gotos, similified the code and ensured that there are no regressions for configs that I can benchmark (CLR x64 AVX2 AVX & arm64 AdvSIMD).
It was a good excercise and I am defnitely not going to recommend using goto in the guidelines that I am currently working on ;)

The perf has not regressed, it's on par. When reading the Ratio columns please keep in mind that for small number of elements we are dealing with nanobenchmarks and 3.4 ns vs 3.2 ns is a 6% difference.

AVX2 (x64)

Details

BenchmarkDotNet=v0.13.1.1845-nightly, OS=Windows 11 (10.0.22000.795/21H2)
AMD Ryzen Threadripper PRO 3945WX 12-Cores, 1 CPU, 24 logical and 12 physical cores
.NET SDK=7.0.100-preview.7.22377.5
[Host] : .NET 7.0.0 (7.0.22.37506), X64 RyuJIT AVX2
Job-FHASLE : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT AVX2
Job-JRHZCS : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT AVX2
LaunchCount=9

Method	Toolchain	Size	Mean	Ratio
IndexOfAnyTwoValues	\PR\corerun.exe	1	1.922 ns	0.82
IndexOfAnyTwoValues	\baseline\corerun.exe	1	2.357 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	4	2.513 ns	0.91
IndexOfAnyTwoValues	\baseline\corerun.exe	4	2.768 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	8	3.459 ns	1.08
IndexOfAnyTwoValues	\baseline\corerun.exe	8	3.216 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	32	2.202 ns	0.68
IndexOfAnyTwoValues	\baseline\corerun.exe	32	3.223 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	33	2.200 ns	0.69
IndexOfAnyTwoValues	\baseline\corerun.exe	33	3.202 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	63	2.224 ns	0.69
IndexOfAnyTwoValues	\baseline\corerun.exe	63	3.231 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	64	2.773 ns	0.75
IndexOfAnyTwoValues	\baseline\corerun.exe	64	3.689 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	512	6.235 ns	0.76
IndexOfAnyTwoValues	\baseline\corerun.exe	512	8.181 ns	1.00

Code gen before:

Details

; System.SpanHelpers.IndexOfAny(Byte ByRef, Byte, Byte, Int32)pushrsivzerouppermovzxedx,dlmoveax,edxmovzxr8d,r8bmovr10d,r8dxorr11d,r11dmovesi,r9dmovsxdr9,r9daddr9,0FFFFFFFFFFFFFFF0js near ptr M00_L01movrsi,r9jmp near ptr M00_L13M00_L00:addrsi,0FFFFFFFFFFFFFFF8movzxedx,byte ptr [rcx+r11]cmpeax,edxje near ptr M00_L05cmpr10d,edxje near ptr M00_L05movzxedx,byte ptr [rcx+r11+1]cmpeax,edxje near ptr M00_L06cmpr10d,edxje near ptr M00_L06movzxedx,byte ptr [rcx+r11+2]cmpeax,edxje near ptr M00_L07cmpr10d,edxje near ptr M00_L07movzxedx,byte ptr [rcx+r11+3]cmpeax,edxje near ptr M00_L08cmpr10d,edxje near ptr M00_L08movzxedx,byte ptr [rcx+r11+4]cmpeax,edxje near ptr M00_L09cmpr10d,edxje near ptr M00_L09movzxedx,byte ptr [rcx+r11+5]cmpeax,edxje near ptr M00_L10cmpr10d,edxje near ptr M00_L10movzxedx,byte ptr [rcx+r11+6]cmpeax,edxje near ptr M00_L11cmpr10d,edxje near ptr M00_L11movzxedx,byte ptr [rcx+r11+7]cmpeax,edxje near ptr M00_L12cmpr10d,edxje near ptr M00_L12addr11,8M00_L01:cmprsi,8jae near ptr M00_L00cmprsi,4jb short M00_L02addrsi,0FFFFFFFFFFFFFFFCmovzxedx,byte ptr [rcx+r11]cmpeax,edxje short M00_L05cmpr10d,edxje short M00_L05movzxedx,byte ptr [rcx+r11+1]cmpeax,edxje short M00_L06cmpr10d,edxje short M00_L06movzxedx,byte ptr [rcx+r11+2]cmpeax,edxje short M00_L07cmpr10d,edxje short M00_L07movzxedx,byte ptr [rcx+r11+3]cmpeax,edxje short M00_L08cmpr10d,edxje short M00_L08addr11,4M00_L02:testrsi,rsije short M00_L04M00_L03:movzxedx,byte ptr [rcx+r11]cmpeax,edxje short M00_L05cmpr10d,edxje short M00_L05incr11decrsijne short M00_L03M00_L04:moveax,0FFFFFFFFvzeroupperpoprsiretM00_L05:moveax,r11djmp near ptr M00_L20M00_L06:leaeax,[r11+1]jmp near ptr M00_L20M00_L07:leaeax,[r11+2]jmp near ptr M00_L20M00_L08:leaeax,[r11+3]jmp near ptr M00_L20M00_L09:leaeax,[r11+4]jmp near ptr M00_L20M00_L10:leaeax,[r11+5]jmp near ptr M00_L20M00_L11:leaeax,[r11+6]jmp near ptr M00_L20M00_L12:leaeax,[r11+7]jmp near ptr M00_L20M00_L13:cmprsi,10jb short M00_L16 vmovd xmm0,edxvpbroadcastbymm0,xmm0 vmovd xmm1,r8dvpbroadcastbymm1,xmm1addrsi,0FFFFFFFFFFFFFFF0je short M00_L15M00_L14:vmovupdymm2,[rcx+r11] vpcmpeqb ymm3,ymm0,ymm2 vpcmpeqb ymm2,ymm1,ymm2vporymm2,ymm3,ymm2 vpmovmskb eax,ymm2testeax,eaxjne near ptr M00_L19addr11,20cmprsi,r11ja short M00_L14M00_L15:vmovupdymm2,[rcx+rsi]movr11,rsi vpcmpeqb ymm0,ymm0,ymm2 vpcmpeqb ymm1,ymm1,ymm2vporymm0,ymm0,ymm1 vpmovmskb eax,ymm0testeax,eaxjne short M00_L19jmp near ptr M00_L04M00_L16: vmovd xmm0,edxvpbroadcastbxmm0,xmm0 vmovd xmm1,r8dvpbroadcastbxmm1,xmm1testrsi,rsije short M00_L18M00_L17:vmovupdxmm2,[rcx+r11] vpcmpeqb xmm3,xmm0,xmm2 vpcmpeqb xmm2,xmm1,xmm2vporxmm2,xmm3,xmm2 vpmovmskb eax,xmm2testeax,eaxjne short M00_L19addr11,10cmprsi,r11ja short M00_L17M00_L18:vmovupdxmm2,[rcx+rsi]movr11,rsi vpcmpeqb xmm0,xmm0,xmm2 vpcmpeqb xmm1,xmm1,xmm2vporxmm0,xmm0,xmm1 vpmovmskb eax,xmm0testeax,eaxje near ptr M00_L04M00_L19:tzcnteax,eaxaddr11,raxjmp near ptr M00_L05M00_L20:vzeroupperpoprsiretint3int3int3int3int3int3int3int3int3int3int3int3int3int3int3int3int3int3int3int3sbb[rcx],eaxadd[rax],eaxadd[rax],esp; Total bytes of code 663

Codegen after:

Details

.NET 7.0.0 (42.42.42.42424), X64 RyuJIT AVX2

; System.SpanHelpers.IndexOfAny(Byte ByRef, Byte, Byte, Int32)vzeroupperxorr10d,r10dcmpr9d,20jl short M00_L02nop dword ptr [rax]moveax,r9dandeax,0FFFFFFE0movzxr11d,dl vmovd xmm0,r11dvpbroadcastbymm0,xmm0movzxr11d,r8b vmovd xmm1,r11dvpbroadcastbymm1,xmm1M00_L00: vmovdqu ymm2,ymmword ptr [rcx+r10] vpcmpeqb ymm3,ymm0,ymm2 vpcmpeqb ymm2,ymm1,ymm2vporymm2,ymm3,ymm2vptestymm2,ymm2jne short M00_L01addr10,20cmpr10,raxjb short M00_L00jmp short M00_L05nop dword ptr [rax]M00_L01: vpmovmskb r9d,ymm2xoreax,eaxtzcnteax,r9daddeax,r10djmp near ptr M00_L09M00_L02:cmpr9d,10jl short M00_L05moveax,r9dandeax,0FFFFFFF0movzxr11d,dl vmovd xmm0,r11dvpbroadcastbxmm0,xmm0movzxr11d,r8b vmovd xmm1,r11dvpbroadcastbxmm1,xmm1M00_L03: vmovdqu xmm2,xmmword ptr [rcx+r10] vpcmpeqb xmm3,xmm0,xmm2 vpcmpeqb xmm2,xmm1,xmm2vporxmm2,xmm3,xmm2vptestxmm2,xmm2jne short M00_L04addr10,10cmpr10,raxjb short M00_L03jmp short M00_L05xchgax,axM00_L04: vpmovmskb edx,xmm2xoreax,eaxtzcnteax,edxaddeax,r10djmp short M00_L09M00_L05:movzxeax,dlmovzxedx,r8bmovr8d,r9dcmpr10,r8jae short M00_L07M00_L06:movzxr9d,byte ptr [rcx+r10]cmpeax,r9dje short M00_L08cmpedx,r9dje short M00_L08incr10cmpr10,r8jb short M00_L06M00_L07:moveax,0FFFFFFFFvzeroupperretM00_L08:moveax,r10dM00_L09:vzeroupperretadd[rcx],bladd[rax],aladd[rax],aladd[rax],aladd[rax],aladd[rax],aladd[rax],aladd[rax],aladd[rax-55C021C],bhjg short M00_L10M00_L10:add[rbp+48],dl (bad); Total bytes of code 291

AVX (x64)

Details

BenchmarkDotNet=v0.13.1.1845-nightly, OS=Windows 11 (10.0.22000.795/21H2)
AMD Ryzen Threadripper PRO 3945WX 12-Cores, 1 CPU, 24 logical and 12 physical cores
.NET SDK=7.0.100-preview.7.22377.5
[Host] : .NET 7.0.0 (7.0.22.37506), X64 RyuJIT AVX2
Job-YLMNVS : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT AVX
Job-DFRNLR : .NET 7.0.0 (42.42.42.42424), X64 RyuJIT AVX
EnvironmentVariables=COMPlus_EnableAVX2=0 LaunchCount=9

Method	Toolchain	Size	Mean	Ratio
IndexOfAnyTwoValues	\PR\corerun.exe	1	1.913 ns	0.80
IndexOfAnyTwoValues	\baseline\corerun.exe	1	2.399 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	4	3.005 ns	1.17
IndexOfAnyTwoValues	\baseline\corerun.exe	4	2.566 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	8	3.457 ns	1.12
IndexOfAnyTwoValues	\baseline\corerun.exe	8	3.043 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	32	3.045 ns	0.81
IndexOfAnyTwoValues	\baseline\corerun.exe	32	3.749 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	33	3.037 ns	0.72
IndexOfAnyTwoValues	\baseline\corerun.exe	33	4.188 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	63	3.010 ns	0.72
IndexOfAnyTwoValues	\baseline\corerun.exe	63	4.152 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	64	3.674 ns	0.77
IndexOfAnyTwoValues	\baseline\corerun.exe	64	4.787 ns	1.00
IndexOfAnyTwoValues	\PR\corerun.exe	512	11.500 ns	0.94
IndexOfAnyTwoValues	\baseline\corerun.exe	512	12.203 ns	1.00

AdvSIMD(arm64)

Details

BenchmarkDotNet=v0.13.1.1845-nightly, OS=ubuntu 20.04
Unknown processor
.NET SDK=7.0.100-rc.1.22408.5
[Host] : .NET 7.0.0 (7.0.22.40308), Arm64 RyuJIT AdvSIMD
Job-NZXMHK : .NET 7.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD
Job-LQCIXE : .NET 7.0.0 (42.42.42.42424), Arm64 RyuJIT AdvSIMD
LaunchCount=9

Method	Toolchain	Size	Mean	Ratio
IndexOfAnyTwoValues	/PR/corerun	1	6.295 ns	0.98
IndexOfAnyTwoValues	/main/corerun	1	6.841 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	4	6.846 ns	1.00
IndexOfAnyTwoValues	/main/corerun	4	7.454 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	8	8.182 ns	1.05
IndexOfAnyTwoValues	/main/corerun	8	7.793 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	32	6.823 ns	1.03
IndexOfAnyTwoValues	/main/corerun	32	6.936 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	33	7.015 ns	1.07
IndexOfAnyTwoValues	/main/corerun	33	6.920 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	63	7.337 ns	1.05
IndexOfAnyTwoValues	/main/corerun	63	6.988 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	64	11.086 ns	1.00
IndexOfAnyTwoValues	/main/corerun	64	12.297 ns	1.00
IndexOfAnyTwoValues	/PR/corerun	512	39.362 ns	1.03
IndexOfAnyTwoValues	/main/corerun	512	38.550 ns	1.00

[vitek-karas]

I think I have a tentative fix for the linker bug (with the gotos - I couldn't find any other way to get to that state 😉 ). I'll clean it up and if possible we should try it on your previous change... anyway, thanks a lot for finding it!

[SwapnilGaikwad]

Hi @adamsitnik, I'm looking at the char equivalent of IndexOfAny. It's half done (blocked by #73777) but can add as a draft PR if you are going to find it useful.

[SwapnilGaikwad]

Hi @adamsitnik, I'm looking at the char equivalent of IndexOfAny. It's half done (blocked by #73777) but can add as a draft PR if you are going to find it useful.

#73788

[adamsitnik]

We don't need it anymore due to #73556