osx-arm64 ABI changes. #41130

sdmaclea · 2020-08-21T00:08:00Z

osx-arm64 has a slightly different calling convention than the arm specified conventions

The obvious things that stood out to me was

R18/X18/W18 reserved
Removal of packing restrictions on arguments passed on the stack

Tasks from @sandreenko:

Support small stack arguments passing in managed <-> native calls;
Support small stack arguments passing via reflection (VM changes);
Support 16-byte struct passing starting with even register (like x1,x2);

category:correctness
theme:calling-convention
skill-level:expert
cost:large

The text was updated successfully, but these errors were encountered:

sdmaclea · 2020-08-21T00:09:36Z

I did a quick survey of R18 usage. It looks like it is not used in any of our manually written assembly. It also looks like the JIT doesn't generate code using that register (perhaps because Windows-Arm64 reserves it).

sandreenko · 2020-08-27T01:12:54Z

sandreenko@9abb9ad reproduces the issue:

Error in Function Marshal_InMany(Native Client)
Expected: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
Actual:1, 2, 3, 4, 5, 6, 7, 8, 9, 0
InMany return value is wrong

working on a fix.

sdmaclea · 2020-09-02T20:11:17Z

These failing tests are also related to this issue

// Assertion failed: (false && "null argument."), function SystemNative_ForkAndExecProcess, 
// file /Users/steve/git/runtime/src/libraries/Native/Unix/System.Native/pal_process.c, line 246.
baseservices/exceptions/stackoverflow/stackoverflowtester/stackoverflowtester.sh
Loader/binding/tracing/BinderTracingTest.ResolutionFlow/BinderTracingTest.ResolutionFlow.sh
Loader/binding/tracing/BinderTracingTest.Basic/BinderTracingTest.Basic.sh
profiler/unittest/metadatagetdispenser/metadatagetdispenser.sh
profiler/unittest/getappdomainstaticaddress/getappdomainstaticaddress.sh
profiler/eventpipe/eventpipe/eventpipe.sh
profiler/eventpipe/eventpipe_readevents/eventpipe_readevents.sh
profiler/gc/gc/gc.sh
profiler/gc/gcbasic/gcbasic.sh
profiler/elt/slowpatheltenter/slowpatheltenter.sh
profiler/elt/slowpatheltleave/slowpatheltleave.sh
ilasm/PortablePdb/IlasmPortablePdbTests/IlasmPortablePdbTests.sh
ilasm/System/Runtime/CompilerServices/MethodImplOptionsTests/MethodImplOptionsTests.sh

sdmaclea · 2020-09-03T16:24:44Z

These failures are also related

JIT/SIMD/Vector3Interop_r/Vector3Interop_r.sh
JIT/SIMD/Vector3Interop_ro/Vector3Interop_ro.sh

sdmaclea · 2020-09-06T22:33:05Z

These p1 interop test failures are also related

JIT/jit64/hfa/main/testG/hfa_nd2G_r/hfa_nd2G_r.sh
JIT/jit64/hfa/main/testG/hfa_sd2G_d/hfa_sd2G_d.sh
JIT/jit64/hfa/main/testG/hfa_sf2G_d/hfa_sf2G_d.sh
JIT/jit64/hfa/main/testG/hfa_nf2G_r/hfa_nf2G_r.sh
JIT/jit64/hfa/main/testG/hfa_sd2G_r/hfa_sd2G_r.sh
JIT/jit64/hfa/main/testG/hfa_nd2G_d/hfa_nd2G_d.sh
JIT/jit64/hfa/main/testG/hfa_nf2G_d/hfa_nf2G_d.sh
JIT/jit64/hfa/main/testG/hfa_sf2G_r/hfa_sf2G_r.sh
JIT/jit64/hfa/main/testA/hfa_nf2A_d/hfa_nf2A_d.sh
JIT/jit64/hfa/main/testA/hfa_sf2A_r/hfa_sf2A_r.sh
JIT/jit64/hfa/main/testA/hfa_sf2A_d/hfa_sf2A_d.sh
JIT/jit64/hfa/main/testA/hfa_nf2A_r/hfa_nf2A_r.sh
JIT/jit64/hfa/main/testC/hfa_nf2C_r/hfa_nf2C_r.sh
JIT/jit64/hfa/main/testC/hfa_sf2C_d/hfa_sf2C_d.sh
JIT/jit64/hfa/main/testC/hfa_sf2C_r/hfa_sf2C_r.sh
JIT/jit64/hfa/main/testC/hfa_nf2C_d/hfa_nf2C_d.sh
JIT/jit64/hfa/main/testB/hfa_nf2B_d/hfa_nf2B_d.sh
JIT/jit64/hfa/main/testB/hfa_sf2B_r/hfa_sf2B_r.sh
JIT/jit64/hfa/main/testB/hfa_sf2B_d/hfa_sf2B_d.sh
JIT/jit64/hfa/main/testB/hfa_nf2B_r/hfa_nf2B_r.sh

sdmaclea · 2020-09-08T21:23:26Z

This is also related

Interop/PInvoke/Generics/GenericsTest/GenericsTest.sh

sdmaclea · 2020-09-09T18:28:51Z

This is also related... seg faults calling SystemNative_ForkAndExecProcess

JIT/superpmi/superpmicollect/superpmicollect.sh

sandreenko · 2020-10-15T09:29:42Z

A small update before my vacation: master...sandreenko:Arm64OSX is a combination of Jit changes that are on reviews right now and initial VM work. It passes tests where we don't need marshaling (when we call only with primitive types, no pointers or structs) but fails with marshaling because VM keeps calculating sizes in stack slots and corrupts arguments.

To see the issue run lldb -- Tests/Core_Root/corerun Interop/PInvoke/Generics/GenericsTest/GenericsTest.dll with export COMPlus_JitHashDump=7798599a (ILStubClass:IL_STUB_PInvoke(bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,long) (MethodHash=7798599a)) and export COMPlus_JitHalt=TestVector128B stop at the call to IL_STUB_PInvoke, you will see that arguments are passed correctly:

->  0x2819c6848: bl     0x281785068
    0x2819c684c: nop
    0x2819c6850: mov    x17, #0x198
    0x2819c6854: ldr    q16, [x29, x17]
Target 0: (corerun) stopped.
(lldb) memory read -s1 -fu -c20 $sp
0x16fdfe410: 1 <- true bool
0x16fdfe411: 0 <- false bool
0x16fdfe412: 1 <- true bool etc.
0x16fdfe413: 0 
0x16fdfe414: 1
0x16fdfe415: 0
0x16fdfe416: 1
0x16fdfe417: 0
0x16fdfe418: 184 <- our pointer1 
0x16fdfe419: 229 <- our pointer2
0x16fdfe41a: 223 <- our pointer3
0x16fdfe41b: 111 <- out pointer4

but then ILStubClass:IL_STUB_PInvoke(bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,bool,long) mixes them and does:


               [000144] --CXG-------              *  CALL ind unman void  
               [000118] ------------ arg0         +--*  LCL_VAR   int    V18 loc1         
               [000119] ------------ arg1         +--*  LCL_VAR   int    V19 loc2         
               [000120] ------------ arg2         +--*  LCL_VAR   int    V20 loc3         
               [000121] ------------ arg3         +--*  LCL_VAR   int    V21 loc4         
               [000122] ------------ arg4         +--*  LCL_VAR   int    V22 loc5         
               [000123] ------------ arg5         +--*  LCL_VAR   int    V23 loc6         
               [000124] ------------ arg6         +--*  LCL_VAR   int    V24 loc7         
               [000125] ------------ arg7         +--*  LCL_VAR   int    V25 loc8         
               [000126] ------------ arg8         +--*  LCL_VAR   int    V26 loc9         
               [000127] ------------ arg9         +--*  LCL_VAR   int    V27 loc10        
               [000128] ------------ arg10        +--*  LCL_VAR   int    V28 loc11        
               [000129] ------------ arg11        +--*  LCL_VAR   int    V29 loc12        
               [000130] ------------ arg12        +--*  LCL_VAR   int    V30 loc13        
               [000131] ------------ arg13        +--*  LCL_VAR   int    V31 loc14        
               [000132] ------------ arg14        +--*  LCL_VAR   int    V32 loc15        
               [000133] ------------ arg15        +--*  LCL_VAR   int    V33 loc16        
               [000134] ------------ arg16        +--*  LCL_VAR   long   V34 loc17        
               [000143] ------------ calli tgt    \--*  LCL_VAR   long   V37 tmp2

when instead of int for the first 17 arguments we should see byte, the type is returned from info.compCompHnd->getArgType(sig, sigArgs, &classHnd).

The full log for the problem stub is attached.
wrongStubCompilation.arm64.osx.txt

sdmaclea · 2020-11-19T21:03:21Z

@sandreenko Is the Arm64OSX branch representative of your current status. I would like to pull your changes into the SDK and see if it fixes any of the issues we are seeing.

sandreenko · 2020-11-23T12:07:42Z

@sandreenko Is the Arm64OSX branch representative of your current status. I would like to pull your changes into the SDK and see if it fixes any of the issues we are seeing.

I am working in the same branch master...sandreenko:Arm64OSX, it passes tests with small arguments, like

ilasm/System/Runtime/CompilerServices/MethodImplOptionsTests/MethodImplOptionsTests.dll
Loader/binding/tracing/BinderTracingTest.Basic/BinderTracingTest.Basic.dll

and profiler tests:

/Users/seandree/git/runtime/artifacts/tests/coreclr/OSX.arm64.Debug/Tests/Core_Root/corerun  /Users/seandree/git/runtime/artifacts/tests/coreclr/OSX.arm64.Debug/profiler/unittest/metadatagetdispenser/metadatagetdispenser.dll

Profiler path: /Users/seandree/git/runtime/artifacts/tests/coreclr/OSX.arm64.Debug/profiler/unittest/metadatagetdispenser/libProfiler.dylib
Profiler.dll!DllGetClassObject
Profiler.dll!Profiler::Initialize
Initialize started
Got IMetaDataDispenserEx
ModuleLoadStarted exiting
Got IMetaDataDispenserEx
ModuleLoadStarted exiting
Got IMetaDataDispenserEx
ModuleLoadStarted exiting
Got IMetaDataDispenserEx
ModuleLoadStarted exiting
Profiler.dll!Profiler::Shutdown
PROFILER TEST PASSES

but it misses 16-byte struct passed on registers changes, so it is not ready.

sandreenko · 2020-12-02T23:58:03Z

With #45467 being merged we have small stack arguments passing support on Apple Arm64, it fixes the following tests from pri0:

Interop/PInvoke/Primitives/Int/PInvokeIntTest/PInvokeIntTest.sh
Loader/binding/tracing/BinderTracingTest.Basic/BinderTracingTest.Basic.sh
JIT/Methodical/explicit/rotate/_opt_relrotarg_objref/_opt_relrotarg_objref.sh
JIT/Methodical/explicit/rotate/_il_relrotarg_double/_il_relrotarg_double.sh
JIT/superpmi/superpmicollect/superpmicollect.sh
profiler/unittest/releaseondetach/releaseondetach.sh
profiler/unittest/metadatagetdispenser/metadatagetdispenser.sh
profiler/unittest/getappdomainstaticaddress/getappdomainstaticaddress.sh
profiler/eventpipe/eventpipe/eventpipe.sh
profiler/eventpipe/eventpipe_readevents/eventpipe_readevents.sh
profiler/gc/gc/gc.sh
profiler/gc/gcbasic/gcbasic.sh
profiler/elt/slowpatheltenter/slowpatheltenter.sh
profiler/elt/slowpatheltleave/slowpatheltleave.sh
ilasm/PortablePdb/IlasmPortablePdbTests/IlasmPortablePdbTests.sh
ilasm/System/Runtime/CompilerServices/MethodImplOptionsTests/MethodImplOptionsTests.sh

I will publish results for pri1 later, I have not seen any regressions, but I suppose there could be some missed cases.

If we ever need to port this support back here are the PRs:
#43024 (lowering->codegen support)
#43130 (importation->morph support)
#42503 (morph->lowering support)
#45461 (a few additions to the previous changes)

#45467 (Enable Arm64 Mac OS ABI.)

sandreenko · 2020-12-08T19:26:28Z

I extracted HFA(float) support #45780 as a separate issue for a clearer discussion.

sdmaclea · 2020-12-09T17:28:31Z

Closing in favor of #45780

sdmaclea · 2020-12-09T17:38:36Z

Looking more closely, the 3 tasks above aren't fully covered by #45780

sandreenko · 2020-12-29T20:09:52Z

Closing this meta, the last item is covered by #46456

sdmaclea added arch-arm64 area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI os-macos-bigsur (macOS11) labels Aug 21, 2020

sdmaclea added this to the 6.0.0 milestone Aug 21, 2020

Dotnet-GitSync-Bot added the untriaged New issue has not been triaged by the area owner label Aug 21, 2020

BruceForstall removed the untriaged New issue has not been triaged by the area owner label Aug 24, 2020

BruceForstall assigned sandreenko Aug 24, 2020

sandreenko mentioned this issue Aug 27, 2020

Add initial CoreCLR compilation support for Apple Silicon #40435

Merged

JulieLeeMSFT added the Epic Groups multiple user stories. Can be grouped under a theme. label Oct 12, 2020

JulieLeeMSFT added Team Epic and removed Epic Groups multiple user stories. Can be grouped under a theme. labels Oct 19, 2020

BruceForstall added the JitUntriaged CLR JIT issues needing additional triage label Oct 28, 2020

sandreenko mentioned this issue Oct 30, 2020

arm64 osx: support byte sizes from lowering to codegen. #43024

Merged

JulieLeeMSFT removed the JitUntriaged CLR JIT issues needing additional triage label Nov 2, 2020

JulieLeeMSFT added User Story A single user-facing feature. Can be grouped under an epic. Bottom Up Work Not part of a theme, epic, or user story and removed Team Epic labels Nov 16, 2020

sdmaclea closed this as completed Dec 9, 2020

sdmaclea reopened this Dec 9, 2020

sdmaclea mentioned this issue Dec 26, 2020

Epic: Support Apple Silicon #43313

Closed

31 tasks

sandreenko mentioned this issue Dec 29, 2020

Fix stack arguments passing via reflection. #46456

Closed

sandreenko closed this as completed Dec 29, 2020

JulieLeeMSFT mentioned this issue Jan 28, 2021

What's new in .NET 6 Preview 1 dotnet/core#5853

Closed

ghost locked as resolved and limited conversation to collaborators Jan 28, 2021

JulieLeeMSFT added this to .NET Core CodeGen Jun 5, 2024

JulieLeeMSFT moved this to Done in .NET Core CodeGen Jun 5, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

osx-arm64 ABI changes. #41130

osx-arm64 ABI changes. #41130

sdmaclea commented Aug 21, 2020 •

edited by sandreenko

Loading

sdmaclea commented Aug 21, 2020

sandreenko commented Aug 27, 2020

sdmaclea commented Sep 2, 2020

sdmaclea commented Sep 3, 2020

sdmaclea commented Sep 6, 2020

sdmaclea commented Sep 8, 2020

sdmaclea commented Sep 9, 2020 •

edited

Loading

sandreenko commented Oct 15, 2020

sdmaclea commented Nov 19, 2020

sandreenko commented Nov 23, 2020

sandreenko commented Dec 2, 2020

sandreenko commented Dec 8, 2020

sdmaclea commented Dec 9, 2020

sdmaclea commented Dec 9, 2020

sandreenko commented Dec 29, 2020

osx-arm64 ABI changes. #41130

osx-arm64 ABI changes. #41130

Comments

sdmaclea commented Aug 21, 2020 • edited by sandreenko Loading

sdmaclea commented Aug 21, 2020

sandreenko commented Aug 27, 2020

sdmaclea commented Sep 2, 2020

sdmaclea commented Sep 3, 2020

sdmaclea commented Sep 6, 2020

sdmaclea commented Sep 8, 2020

sdmaclea commented Sep 9, 2020 • edited Loading

sandreenko commented Oct 15, 2020

sdmaclea commented Nov 19, 2020

sandreenko commented Nov 23, 2020

sandreenko commented Dec 2, 2020

sandreenko commented Dec 8, 2020

sdmaclea commented Dec 9, 2020

sdmaclea commented Dec 9, 2020

sandreenko commented Dec 29, 2020

sdmaclea commented Aug 21, 2020 •

edited by sandreenko

Loading

sdmaclea commented Sep 9, 2020 •

edited

Loading