Allow implicit widenings when tailcalling #54864

jakobbotsch · 2021-06-28T22:40:19Z

The managed calling convention dictates that the callee is responsible
for widening up to 4 bytes. We can use this to allow some more
tailcalls.

Fix #51957

I couldn't use SPMI as the new tailcalls change the jitinterface calls (thanks for explaining @sandreenko), so I used PMI. A few diffs in frameworks:

Analyzing CodeSize diffs...
Found 10 files with textual diffs.

PMI CodeSize Diffs for System.Private.CoreLib.dll, framework assemblies for  default jit

Summary of Code Size diffs:
(Lower is better)

Total bytes of base: 55385375
Total bytes of diff: 55385274
Total bytes of delta: -101 (-0,00% of base)
    diff is an improvement.


Top file improvements (bytes):
         -24 : Microsoft.CodeAnalysis.dasm (-0,00% of base)
         -20 : System.Private.DataContractSerialization.dasm (-0,00% of base)
         -15 : System.Reflection.Metadata.dasm (-0,00% of base)
         -14 : System.Drawing.Common.dasm (-0,00% of base)
         -10 : System.Diagnostics.EventLog.dasm (-0,01% of base)
          -6 : Microsoft.CodeAnalysis.CSharp.dasm (-0,00% of base)
          -6 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0,00% of base)
          -4 : System.Net.Quic.dasm (-0,00% of base)
          -2 : System.Reflection.MetadataLoadContext.dasm (-0,00% of base)

9 total files with Code Size differences (9 improved, 0 regressed), 263 unchanged.

Top method improvements (bytes):
         -14 (-33,33% of base) : System.Drawing.Common.dasm - System.Drawing.Printing.PrinterSettings:get_Duplex():int:this
         -10 (-47,62% of base) : System.Private.DataContractSerialization.dasm - System.Xml.XmlBinaryReader:GetNodeType():int:this
         -10 (-66,67% of base) : System.Private.DataContractSerialization.dasm - System.Xml.XmlBufferReader:GetNodeType():int:this
         -10 (-37,04% of base) : System.Diagnostics.EventLog.dasm - System.Diagnostics.EventLogEntry:get_EntryType():int:this
          -6 (-30,00% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Microsoft.CodeAnalysis.CSharp.Syntax.InternalSyntax.SyntaxToken:get_RawContextualKind():int:this
          -6 (-31,58% of base) : Microsoft.CodeAnalysis.dasm - Microsoft.CodeAnalysis.ConstantValue:get_Int16Value():short:this
          -6 (-30,00% of base) : Microsoft.CodeAnalysis.dasm - Microsoft.CodeAnalysis.ConstantValue:get_UInt16Value():ushort:this
          -6 (-30,00% of base) : Microsoft.CodeAnalysis.dasm - Microsoft.CodeAnalysis.ConstantValue:get_Int32Value():int:this
          -6 (-30,00% of base) : Microsoft.CodeAnalysis.dasm - Microsoft.CodeAnalysis.ConstantValue:get_UInt32Value():int:this
          -6 (-30,00% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Microsoft.CodeAnalysis.VisualBasic.Syntax.InternalSyntax.IdentifierTokenSyntax:get_RawContextualKind():int:this
          -5 (-12,20% of base) : System.Reflection.Metadata.dasm - System.Reflection.Metadata.GenericParameter:get_Index():int:this
          -5 (-12,20% of base) : System.Reflection.Metadata.dasm - System.Reflection.Metadata.Parameter:get_SequenceNumber():int:this
          -5 (-12,20% of base) : System.Reflection.Metadata.dasm - System.Reflection.Metadata.LocalVariable:get_Index():int:this
          -2 (-2,74% of base) : System.Net.Quic.dasm - System.Net.Quic.Implementations.MsQuic.MsQuicConnection:GetRemoteAvailableUnidirectionalStreamCount():int:this
          -2 (-2,74% of base) : System.Net.Quic.dasm - System.Net.Quic.Implementations.MsQuic.MsQuicConnection:GetRemoteAvailableBidirectionalStreamCount():int:this
          -2 (-3,45% of base) : System.Reflection.MetadataLoadContext.dasm - System.Reflection.TypeLoading.Ecma.EcmaGenericParameterType:ComputePosition():int:this

16 total methods with Code Size differences (16 improved, 0 regressed), 369818 unchanged.

Diffs look as expected, we get some cases where we were relying on the widening anyway.

 ; Emitting BLENDED_CODE for X64 CPU with AVX - Windows
 ; optimized code
 ; rsp based frame
-; partially interruptible
+; fully interruptible
 ; No PGO data
 ; Final local variable assignments
 ;
 ;  V00 this         [V00,T00] (  4,  4   )     ref  ->  rcx         this class-hnd
-;  V01 OutArgs      [V01    ] (  1,  1   )  lclBlk (32) [rsp+00H]   "OutgoingArgSpace"
+;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
 ;
-; Lcl frame size = 40
+; Lcl frame size = 0
 
 G_M58155_IG01:
-       sub      rsp, 40
-						;; bbWeight=1    PerfScore 0.25
+						;; bbWeight=1    PerfScore 0.00
 G_M58155_IG02:
        mov      rax, qword ptr [rcx]
        mov      rax, qword ptr [rax+120]
-       call     qword ptr [rax+8]Microsoft.CodeAnalysis.VisualBasic.Syntax.InternalSyntax.IdentifierTokenSyntax:get_PossibleKeywordKind():ushort:this
-       nop      
-						;; bbWeight=1    PerfScore 7.25
+       mov      rax, qword ptr [rax+8]
+						;; bbWeight=1    PerfScore 6.00
 G_M58155_IG03:
-       add      rsp, 40
-       ret      
-						;; bbWeight=1    PerfScore 1.25
+       rex.jmp  rax
+						;; bbWeight=1    PerfScore 2.00
 
-; Total bytes of code 20, prolog size 4, PerfScore 10.75, instruction count 7, allocated bytes for code 20 (MethodHash=91bc1cd4) for method Microsoft.CodeAnalysis.VisualBasic.Syntax.InternalSyntax.IdentifierTokenSyntax:get_RawContextualKind():int:this
+; Total bytes of code 14, prolog size 0, PerfScore 9.40, instruction count 4, allocated bytes for code 14 (MethodHash=91bc1cd4) for method Microsoft.CodeAnalysis.VisualBasic.Syntax.InternalSyntax.IdentifierTokenSyntax:get_RawContextualKind():int:this

 ; Emitting BLENDED_CODE for X64 CPU with AVX - Windows
 ; optimized code
 ; rsp based frame
-; partially interruptible
+; fully interruptible
 ; No PGO data
 ; 0 inlinees with PGO data; 1 single block inlinees; 0 inlinees without PGO data
 ; Final local variable assignments
 ;
 ;  V00 this         [V00,T00] (  4,  3.50)     ref  ->  rcx         this class-hnd
-;  V01 OutArgs      [V01    ] (  1,  1   )  lclBlk (32) [rsp+00H]   "OutgoingArgSpace"
+;# V01 OutArgs      [V01    ] (  1,  1   )  lclBlk ( 0) [rsp+00H]   "OutgoingArgSpace"
 ;  V02 cse0         [V02,T01] (  3,  2.50)     int  ->  rax         "CSE - aggressive"
 ;
-; Lcl frame size = 40
+; Lcl frame size = 0
 
 G_M35284_IG01:
-       sub      rsp, 40
-						;; bbWeight=1    PerfScore 0.25
+						;; bbWeight=1    PerfScore 0.00
 G_M35284_IG02:
        mov      eax, dword ptr [rcx+56]
        cmp      eax, -1
        je       SHORT G_M35284_IG04
 						;; bbWeight=1    PerfScore 3.25
 G_M35284_IG03:
-       add      rsp, 40
        ret      
-						;; bbWeight=0.50 PerfScore 0.63
+						;; bbWeight=0.50 PerfScore 0.50
 G_M35284_IG04:
        mov      edx, 8
        mov      r8d, 1
        xor      r9d, r9d
-       call     System.Drawing.Printing.PrinterSettings:GetModeField(int,short,long):short:this
-       nop      
-						;; bbWeight=0.50 PerfScore 1.00
+						;; bbWeight=0.50 PerfScore 0.38
 G_M35284_IG05:
-       add      rsp, 40
-       ret      
-						;; bbWeight=0.50 PerfScore 0.63
+       jmp      System.Drawing.Printing.PrinterSettings:GetModeField(int,short,long):short:this
+						;; bbWeight=0.50 PerfScore 1.00
 
-; Total bytes of code 42, prolog size 4, PerfScore 9.95, instruction count 13, allocated bytes for code 42 (MethodHash=b14c762b) for method System.Drawing.Printing.PrinterSettings:get_Duplex():int:this
+; Total bytes of code 28, prolog size 0, PerfScore 7.93, instruction count 8, allocated bytes for code 28 (MethodHash=b14c762b) for method System.Drawing.Printing.PrinterSettings:get_Duplex():int:this

cc @dotnet/jit-contrib

The managed calling convention dictates that the callee is responsible for widening up to 4 bytes. We can use this to allow some more tailcalls. Fix dotnet#51957

jakobbotsch · 2021-06-28T22:41:16Z

src/coreclr/jit/morph.cpp

-//    is tracked with https://github.com/dotnet/runtime/issues/8413.
-//
-//    3) If the callee has a 9 to 16 byte struct argument and the callee has
+//    If the callee has a 9 to 16 byte struct argument and the callee has


This comment was outdated, both of the issues have been fixed.

jakobbotsch · 2021-07-01T11:39:47Z

/azp run runtime-coreclr jitstress, runtime-coreclr outerloop

azure-pipelines · 2021-07-01T11:40:20Z

Azure Pipelines successfully started running 2 pipeline(s).

jakobbotsch · 2021-07-01T19:36:35Z

PTAL @dotnet/jit-contrib

EgorBo · 2021-07-05T09:19:08Z

Are failed jitstress jobs related?

EgorBo · 2021-07-05T09:23:56Z

Are we losing method name here? 🙂 I'd be nice to keep it for indirect calls where possible for better readability but it's a separate issue.

jakobbotsch · 2021-07-05T10:21:58Z

Are failed jitstress jobs related?

I believe the failing test is #54200

Are we losing method name here? 🙂 I'd be nice to keep it for indirect calls where possible for better readability but it's a separate issue.

Hmm, looks like it. I can try to take a look at that separately.

Allow implicit widenings when tailcalling

99701db

The managed calling convention dictates that the callee is responsible for widening up to 4 bytes. We can use this to allow some more tailcalls. Fix dotnet#51957

dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jun 28, 2021

jakobbotsch commented Jun 28, 2021

View reviewed changes

Formatting

0c3b721

EgorBo approved these changes Jul 5, 2021

View reviewed changes

jakobbotsch merged commit 66c339b into dotnet:main Jul 5, 2021

jakobbotsch deleted the allow-widening-tailcalls branch July 5, 2021 14:02

This was referenced Jul 15, 2021

Test Failure:Microsoft.Extensions.Logging.Generators.Tests.LoggerMessageGeneratorEmitterTests.TestBaseline_TestWithTwoParams_Success #55253

Closed

Widen return values in CallTailCallTarget stubs #55740

Closed

ghost locked as resolved and limited conversation to collaborators Aug 4, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow implicit widenings when tailcalling #54864

Allow implicit widenings when tailcalling #54864

jakobbotsch commented Jun 28, 2021

jakobbotsch Jun 28, 2021

jakobbotsch commented Jul 1, 2021

azure-pipelines bot commented Jul 1, 2021

jakobbotsch commented Jul 1, 2021

EgorBo commented Jul 5, 2021

EgorBo commented Jul 5, 2021 •

edited

Loading

jakobbotsch commented Jul 5, 2021

Allow implicit widenings when tailcalling #54864

Allow implicit widenings when tailcalling #54864

Conversation

jakobbotsch commented Jun 28, 2021

jakobbotsch Jun 28, 2021

Choose a reason for hiding this comment

jakobbotsch commented Jul 1, 2021

azure-pipelines bot commented Jul 1, 2021

jakobbotsch commented Jul 1, 2021

EgorBo commented Jul 5, 2021

EgorBo commented Jul 5, 2021 • edited Loading

jakobbotsch commented Jul 5, 2021

EgorBo commented Jul 5, 2021 •

edited

Loading