JIT: enable tail calls and copy omission for implicit byref structs - #33004

Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged

JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom
AndyAyersMS:ImplicitByRefTailCallOpt

Conversation

@AndyAyersMS

@AndyAyersMSAndyAyersMS commented Feb 29, 2020

Copy link
Copy Markdown
Member

Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.

This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.

@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS

AndyAyersMS commented Feb 29, 2020

Copy link
Copy Markdown
MemberAuthor

cc @dotnet/jit-contrib

Addresses issues seen in #13824 (see comment) and #32843 (see comment).

This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.

Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.

Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:

;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:

  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    (and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice

@AndyAyersMSAndyAyersMS added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Feb 29, 2020
@AndyAyersMS

AndyAyersMS commented Feb 29, 2020

Copy link
Copy Markdown
MemberAuthor

Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.

@jashook
jashook self-requested a review February 29, 2020 04:14
jashook
jashook approved these changes Feb 29, 2020
View reviewed changes

@jashookjashook left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit

Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
//
// Otherwise, see how many appearances there are. We keep two early ref counts: total
// number of references to the struct or some field, and how many of these are
// arguments to calls. We undo promotion unless we see enougn non-call uses.

@jashookjashookFeb 29, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough

@benaadamsbenaadams mentioned this pull request Feb 29, 2020
Closed
@benaadams

benaadams commented Feb 29, 2020

Copy link
Copy Markdown
Member

Raised issue for "Linux arm checked" #33012

@AndyAyersMS

AndyAyersMS commented Feb 29, 2020•
edited
Loading

Copy link
Copy Markdown
MemberAuthor

As with anything aliasing related, checking is costly... still working on it.

@jaykrell

jaykrell commented Feb 29, 2020

Copy link
Copy Markdown
Contributor

It’d be nice to test & implement similar for mono. /cc @lambdageek

@AndyAyersMS

AndyAyersMS commented Feb 29, 2020

Copy link
Copy Markdown
MemberAuthor

Note for coreclr, this is windows-x64 and windows-arm64 only.

More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.

@AndyAyersMS

AndyAyersMS commented Mar 1, 2020•
edited
Loading

Copy link
Copy Markdown
MemberAuthor

A first cut at an alias analysis is in, not sure if its complexity matches its value, though.

Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.

Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.

This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32

runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757 in 7c7e8ed

publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);

;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool

Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.

@AndyAyersMSAndyAyersMS mentioned this pull request Mar 1, 2020
Closed
@GrabYourPitchforks

GrabYourPitchforks commented Mar 1, 2020•
edited
Loading

Copy link
Copy Markdown
Member

@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄

MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02

The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)

00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)

For reference, here's the C# that was used for this codegen:

https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158

@AndyAyersMS

AndyAyersMS commented Mar 2, 2020

Copy link
Copy Markdown
MemberAuthor

@GrabYourPitchforks thanks.

For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch

G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)

and emit something like:

G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:

and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.

SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.

G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool

@GrabYourPitchforksGrabYourPitchforks mentioned this pull request Mar 2, 2020
Closed
@AndyAyersMS

AndyAyersMS commented Mar 9, 2020

Copy link
Copy Markdown
MemberAuthor

@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...

@erozenfeld

erozenfeld commented Mar 10, 2020

Copy link
Copy Markdown
Contributor

I'll review this tomorrow.

erozenfeld
erozenfeld suggested changes Mar 10, 2020
View reviewed changes

@erozenfelderozenfeld left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.

Comment threadsrc/coreclr/src/jit/lclmorph.cpp Outdated
Comment threadsrc/coreclr/src/jit/lclmorph.cpp Outdated
Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
{
JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.

Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs Outdated
return result;
}

// Will return different answers if x and y refer to same struct

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs Outdated
Comment threadsrc/coreclr/src/jit/gentree.cpp
lcl = addr->AsLclVarCommon();
}
else if (addr->OperIs(GT_ADDR))
{

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.

See above?

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
// Here the call to ZZZ would need to be a slow tail call,
// so we won't tail call at all. But the only
// reference to x is at the call, and it's not in
// a loop, so we can still can avoid making a copy.

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.

Comment threadsrc/coreclr/src/jit/morph.cpp
// make a local copy of the struct for the callee.
hasByrefParameter = true;
break;

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.

sandreenko
sandreenko reviewed Mar 10, 2020
View reviewed changes

@sandreenkosandreenko left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.

Comment threadsrc/coreclr/src/jit/gentree.cpp Outdated

if (addr->OperIs(GT_LCL_VAR))
{
lcl = addr->AsLclVarCommon();

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.

The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.
This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.

Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.

Comment threadsrc/coreclr/src/jit/gentree.cpp Outdated
// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.

Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
{
assert(!call->IsTailCall());
unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.

Comment threadsrc/coreclr/src/jit/morph.cpp
assert(!call->IsTailCall());
unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
LclVarDsc* varDsc = lvaGetDesc(varNum);
const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.

Comment threadsrc/coreclr/src/jit/morph.cpp
{
// First, see if this arg is a byref param.
GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?

Comment threadsrc/coreclr/src/jit/morph.cpp Outdated
// a local struct assembled from the promoted fields.
if (lcl != nullptr)
{
JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",

@sandreenkosandreenkoMar 10, 2020•
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "
Can we check varDsc->lvPromoted before we print that it was not promoted?

Comment threadsrc/coreclr/src/jit/morph.cpp
JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
hasByrefParameter = false;
}
// If no other call arg refers to this byref, and no other arg is

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.

Comment threadsrc/coreclr/src/jit/morph.cpp
// Do we pass 'lcl' more than once to the callee?
if (arg2->isStruct && arg2->passedByRef)
{
GenTreeLclVarCommon* const lcl2 =

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.
Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.

@erozenfelderozenfeldMar 10, 2020•
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.

No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.

I think that's fine, there is no real address exposure here.

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?

@erozenfelderozenfeldMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?

@sandreenkosandreenkoMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:

runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809 in 73268c7

if (notStruct && notLastInstr && notDebugCode &&
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd))
{
// We can skip the addrtaken, as next IL instruction consumes
// the address.
}
else
{
lvaTable[varNum].lvHasLdAddrOp = 1;
if (!info.compIsStatic && (varNum == 0))
{
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.

If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.

AndyAyersMSand others added 4 commits March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS

AndyAyersMS commented Mar 11, 2020

Copy link
Copy Markdown
MemberAuthor

OSX builds failing with

Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook

jashook commented Mar 11, 2020

Copy link
Copy Markdown
Contributor

Fixed this morning with #33481

jashook
jashook reviewed Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)

@jashookjashookMar 10, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.

We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.

runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446 in 5ea5b57

#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
#endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)

Perhaps we need a define just for this and a cleanup of all references?

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.

@jashookjashookMar 11, 2020

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.

@jashookjashookMar 11, 2020•
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.

@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)

If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.

So the right condition is windows x64, and all arm64.

sandreenko
sandreenko approved these changes Mar 11, 2020
View reviewed changes

@sandreenkosandreenko left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

erozenfeld
erozenfeld approved these changes Mar 11, 2020
View reviewed changes

@erozenfelderozenfeld left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.

@AndyAyersMS
AndyAyersMS merged commit 514db28 into dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS deleted the ImplicitByRefTailCallOpt branch March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks mentioned this pull request Mar 12, 2020
Merged
@AndyAyersMSAndyAyersMS mentioned this pull request Jul 22, 2020
Closed
@AndyAyersMSAndyAyersMS mentioned this pull request Aug 9, 2020
Closed
@ghostghost locked as resolved and limited conversation to collaborators Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub. Already have an account? Sign in.

Reviewers

3 more reviewers
@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko
, 'i'); if (__m === '*' || __re.test(location.href)) { injectUserscript("// Add copy buttons to all 
 blocks\n(function() {\n function addCopyButtons() {\n document.querySelectorAll('pre code').forEach(function(codeBlock) {\n if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;\n codeBlock.parentElement.setAttribute('data-copy-added', 'true');\n \n var btn = document.createElement('button');\n btn.textContent = 'Copy';\n btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';\n btn.onmouseover = function() { this.style.opacity = '1'; };\n btn.onmouseout = function() { this.style.opacity = '0.7'; };\n btn.onclick = function() {\n navigator.clipboard.writeText(codeBlock.textContent).then(function() {\n btn.textContent = 'Copied!';\n setTimeout(function() { btn.textContent = 'Copy'; }, 1500);\n });\n };\n codeBlock.parentElement.style.position = 'relative';\n codeBlock.parentElement.appendChild(btn);\n });\n }\n \n addCopyButtons();\n \n // Re-run on dynamic content\n var observer = new MutationObserver(addCopyButtons);\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Add Copy Buttons to Code Blocks");
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Force GitHub README to respect dark mode\n(function() {\n var style = document.createElement('style');\n style.textContent = '\n .markdown-body {\n color-scheme: dark light;\n }\n .markdown-body pre { background: #161b22 !important; }\n .markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }\n .markdown-body table th, .markdown-body table td { border-color: #30363d !important; }\n .markdown-body img { background: #0d1117; }\n .markdown-body blockquote { border-left-color: #8b949e; }\n .markdown-body hr { border-color: #30363d; }\n ';\n document.head.appendChild(style);\n})();", "GitHub Dark Mode README Fix");
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Highlight search terms from Google/DuckDuckGo/Bing referrer\n(function() {\n var ref = document.referrer;\n var terms = [];\n \n if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {\n var url = new URL(ref);\n var q = url.searchParams.get('q') || url.searchParams.get('p');\n if (q) {\n terms = q.split(/\\s+/).filter(function(t) { return t.length > 2; });\n }\n }\n \n if (terms.length === 0) return;\n \n var style = document.createElement('style');\n style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';\n document.head.appendChild(style);\n \n function highlight(node) {\n if (node.nodeType === 3) { // text node\n var text = node.textContent;\n var found = false;\n terms.forEach(function(term) {\n var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\\]\\\\]/g, '\\\\') + ')', 'gi');\n if (regex.test(text)) {\n found = true;\n var frag = document.createDocumentFragment();\n var parts = text.split(regex);\n parts.forEach(function(part, i) {\n if (i % 2 === 0) {\n frag.appendChild(document.createTextNode(part));\n } else {\n var span = document.createElement('span');\n span.className = 'userscript-highlight';\n span.textContent = part;\n frag.appendChild(span);\n }\n });\n node.parentNode.replaceChild(frag, node);\n }\n });\n } else if (node.nodeType === 1 && node.childNodes) { // element\n var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];\n if (!skipTags.includes(node.tagName)) {\n Array.from(node.childNodes).forEach(highlight);\n }\n }\n }\n \n highlight(document.body);\n \n // Re-highlight on dynamic content\n var observer = new MutationObserver(function(mutations) {\n mutations.forEach(function(m) {\n m.addedNodes.forEach(function(node) {\n if (node.nodeType === 1 || node.nodeType === 3) highlight(node);\n });\n });\n });\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Highlight Search Terms");
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Strip utm_, fbclid, gclid, etc. from all links on page\n(function() {\n var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',\n 'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',\n 'ref', 'ref_src', 'source', 'medium', 'campaign'];\n \n function cleanUrl(url) {\n try {\n var u = new URL(url, window.location.origin);\n var changed = false;\n trackingParams.forEach(function(p) {\n if (u.searchParams.has(p)) {\n u.searchParams.delete(p);\n changed = true;\n }\n });\n return changed ? u.toString() : url;\n } catch (e) {\n return url;\n }\n }\n \n function cleanLinks() {\n document.querySelectorAll('a[href]').forEach(function(a) {\n var clean = cleanUrl(a.href);\n if (clean !== a.href) a.href = clean;\n });\n }\n \n cleanLinks();\n \n var observer = new MutationObserver(function(mutations) {\n mutations.forEach(function(m) {\n m.addedNodes.forEach(function(node) {\n if (node.nodeType === 1) {\n if (node.tagName === 'A') cleanLinks();\n node.querySelectorAll('a[href]').forEach(function(a) {\n var clean = cleanUrl(a.href);\n if (clean !== a.href) a.href = clean;\n });\n }\n });\n });\n });\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "Remove Tracking Parameters from Links");
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Auto-enable theater mode on YouTube\n(function() {\n function tryTheater() {\n var btn = document.querySelector('button[aria-label=\"Theater mode\"], ytd-player #player button[title=\"Theater mode\"]');\n if (btn && !btn.classList.contains('activated')) {\n btn.click();\n }\n }\n \n // Try immediately\n tryTheater();\n \n // Try after navigation (SPA)\n var lastUrl = location.href;\n setInterval(function() {\n if (location.href !== lastUrl) {\n lastUrl = location.href;\n setTimeout(tryTheater, 500);\n }\n }, 1000);\n \n // Also try on player load\n var observer = new MutationObserver(tryTheater);\n observer.observe(document.body, { childList: true, subtree: true });\n})();", "YouTube Theater Mode Default");
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Remove or un-stick sticky/fixed headers that block content\n(function() {\n function unstick() {\n document.querySelectorAll('header, nav, [role=\"banner\"], .header, .navbar, .sticky, .fixed-top, [style*=\"position: fixed\"], [style*=\"position:sticky\"]').forEach(function(el) {\n if (el.style.position === 'fixed' || el.style.position === 'sticky' || \n getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {\n el.style.position = 'static';\n el.style.top = 'auto';\n el.style.zIndex = 'auto';\n }\n });\n }\n \n unstick();\n \n var observer = new MutationObserver(unstick);\n observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });\n})();", "Kill Sticky Headers");
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko

, 'i');
if (__m === '*' || __re.test(location.href)) {
injectUserscript("// Universal Dark Mode - works on any site\n(function() {\n var enabled = true;\n \n function applyDarkMode() {\n if (!enabled) return;\n \n // Create style element if it doesn't exist\n var style = document.getElementById('universal-dark-mode-style');\n if (!style) {\n style = document.createElement('style');\n style.id = 'universal-dark-mode-style';\n document.head.appendChild(style);\n }\n \n // Dark mode CSS - inverts colors but preserves images/video\n style.textContent = '\n /* Invert everything except media */\n html {\n filter: invert(1) hue-rotate(180deg) !important;\n background: #1a1a2e !important;\n }\n \n /* Restore images, videos, iframes, canvas */\n img, video, iframe, canvas, svg, picture, [style*=\"background-image\"] {\n filter: invert(1) hue-rotate(180deg) !important;\n }\n \n /* Preserve specific elements that should not be inverted */\n .no-dark-mode, .no-dark-mode *,\n [data-theme=\"light\"], [data-theme=\"light\"],\n .ace_editor, .ace_editor *,\n .CodeMirror, .CodeMirror *,\n .monaco-editor, .monaco-editor *,\n .markdown-body pre, .markdown-body pre *,\n .highlight, .highlight *,\n pre code, pre code * {\n filter: none !important;\n }\n \n /* Fix common UI elements */\n .modal, .popup, .dropdown-menu, .tooltip, .popover {\n filter: invert(1) hue-rotate(180deg) !important;\n background: #2d2d44 !important;\n border-color: #444 !important;\n }\n \n /* Scrollbars */\n ::-webkit-scrollbar { background: #1a1a2e !important; }\n ::-webkit-scrollbar-thumb { background: #444 !important; }\n ::-webkit-scrollbar-thumb:hover { background: #555 !important; }\n \n /* Selection */\n ::selection { background: #4ecdc4 !important; color: #1a1a2e !important; }\n ::-moz-selection { background: #4ecdc4 !important; color: #1a1a2e !important; }\n ';\n }\n \n function removeDarkMode() {\n var style = document.getElementById('universal-dark-mode-style');\n if (style) style.remove();\n }\n \n // Toggle with Alt+Shift+D\n document.addEventListener('keydown', function(e) {\n if (e.altKey && e.shiftKey && e.key === 'D') {\n e.preventDefault();\n enabled = !enabled;\n if (enabled) {\n applyDarkMode();\n console.log('[Universal Dark Mode] Enabled');\n } else {\n removeDarkMode();\n console.log('[Universal Dark Mode] Disabled');\n }\n }\n });\n \n // Apply on load\n applyDarkMode();\n \n // Re-apply on dynamic content\n var observer = new MutationObserver(function(mutations) {\n if (enabled && !document.getElementById('universal-dark-mode-style')) {\n applyDarkMode();\n }\n });\n observer.observe(document.head, { childList: true });\n \n console.log('[Universal Dark Mode] Loaded - Press Alt+Shift+D to toggle');\n})();", "Universal Dark Mode");
}
} catch(__e) { console.warn('[Userscript:Universal Dark Mode]', __e); }
})();
})();

Skip to content
JIT: enable tail calls and copy omission for implicit byref structs - #33004
Merged
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Mar 11, 2020
Merged
JIT: enable tail calls and copy omission for implicit byref structs#33004
AndyAyersMS merged 9 commits into
dotnet:masterfrom 
AndyAyersMS:ImplicitByRefTailCallOpt
Conversation
@AndyAyersMS
@AndyAyersMSAndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Handle cases where a caller passes an implicit byref argument struct to

a callee in tail position. There is no need to create a local copy or

to block tail calling.
This pattern comes up increasingly often as we write more layered code

operating on spans and similar structs.
@AndyAyersMS
JIT: enable tail calls and copy omission for implicit byref structs
13f0f6a
Handle cases where a caller passes an implicit byref argument struct to
a callee in tail position. There is no need to create a local copy or
to block tail calling.
This pattern comes up increasingly often as we write more layered code
operating on spans and similar structs.
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
cc @dotnet/jit-contrib
Addresses issues seen in #13824 (see comment) and #32843 (see comment).
This modifies the promotion heuristic for implicit byrefs to not consider uses at (some) call sites as significant for promotion, and modifies the legality checks for tail calls to allow implicit byref params to be passed as args in tail calls. To make this work we then must avoid copying the param to a local struct at the call.
Jit diffs shows a lot of hits in the json parser and other places where spans are passed through various layers.
Found 92 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -25235 (-0.05% of base)
diff is an improvement.
Top file regressions (bytes):
124 : System.Linq.Parallel.dasm (0.01% of base)
116 : System.Reflection.Metadata.dasm (0.03% of base)
14 : xunit.core.dasm (0.02% of base)
6 : System.Composition.Hosting.dasm (0.01% of base)
2 : System.CodeDom.dasm (0.00% of base)
Top file improvements (bytes):
-5553 : System.Private.CoreLib.dasm (-0.12% of base)
-2843 : System.Security.Cryptography.Pkcs.dasm (-0.66% of base)
-2260 : System.Text.Json.dasm (-0.47% of base)
-2253 : Microsoft.CodeAnalysis.VisualBasic.dasm (-0.04% of base)
-1963 : Microsoft.CodeAnalysis.CSharp.dasm (-0.05% of base)
-1586 : System.Security.Cryptography.Algorithms.dasm (-0.52% of base)
-1024 : System.Data.Common.dasm (-0.07% of base)
-813 : System.Collections.Immutable.dasm (-0.07% of base)
-793 : Microsoft.CodeAnalysis.dasm (-0.05% of base)
-625 : Newtonsoft.Json.dasm (-0.07% of base)
-603 : System.Threading.Tasks.Dataflow.dasm (-0.07% of base)
-476 : System.Memory.dasm (-0.19% of base)
-451 : System.Security.Cryptography.Cng.dasm (-0.30% of base)
-434 : System.Net.Http.dasm (-0.06% of base)
-297 : System.Reflection.MetadataLoadContext.dasm (-0.16% of base)
-257 : System.Runtime.Numerics.dasm (-0.36% of base)
-245 : System.Drawing.Common.dasm (-0.08% of base)
-212 : System.IO.Compression.dasm (-0.28% of base)
-203 : System.Collections.dasm (-0.04% of base)
-167 : System.Numerics.Tensors.dasm (-0.07% of base)
92 total files with Code Size differences (87 improved, 5 regressed), 139 unchanged.
Top method regressions (bytes):
596 ( 3.62% of base) : System.Linq.Parallel.dasm - UnionQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - ExceptQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
158 ( 1.06% of base) : System.Linq.Parallel.dasm - IntersectQueryOperator`1:WrapPartitionedStream(PartitionedStream`2,PartitionedStream`2,IPartitionedStreamRecipient`1,bool,QuerySettings):this (49 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
37 ( 4.91% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - Binder:PassArgumentByRef(bool,BoundExpression,KeyValuePair`2,KeyValuePair`2,TypeSymbol,String,DiagnosticBag):BoundExpression:this
35 ( 1.12% of base) : System.Numerics.Tensors.dasm - DenseTensor`1:Reshape(ReadOnlySpan`1):Tensor`1:this (7 methods)
30 ( 4.97% of base) : Microsoft.CodeAnalysis.CSharp.dasm - LocalRewriter:MakeIsOperator(BoundIsOperator,CSharpSyntaxNode,BoundExpression,BoundTypeExpression,Conversion,TypeSymbol):BoundExpression:this
28 ( 4.54% of base) : Microsoft.CodeAnalysis.dasm - ValueTuple`2:op_Equality(ValueTuple`2,ValueTuple`2):bool (7 methods)
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
26 ( 3.04% of base) : System.Private.CoreLib.dasm - GenericArraySortHelper`1:HeapSort(Span`1) (5 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTimeOffset):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,Decimal):this (2 methods)
21 ( 3.58% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,Guid):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,double):this (2 methods)
19 ( 3.79% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,float):this (2 methods)
17 ( 1.31% of base) : System.Runtime.Numerics.dasm - BigInteger:.ctor(ReadOnlySpan`1,bool,bool):this
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
16 ( 1.08% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
16 ( 2.71% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteBase64Escape(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
14 ( 4.81% of base) : System.Private.CoreLib.dasm - MemoryStream:Read(Span`1):int:this
Top method improvements (bytes):
-761 (-19.18% of base) : System.Private.CoreLib.dasm - Utf8Parser:TryParse(ReadOnlySpan`1,byref,byref,ushort):bool (16 methods)
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-324 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (4 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-306 (-3.05% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadEncryptedPkcs8(ref,ReadOnlyMemory`1,ReadOnlySpan`1,ReadOnlySpan`1,KeyReader`1,byref,byref) (7 methods)
-262 (-14.24% of base) : Newtonsoft.Json.dasm - DictionaryWrapper`2:Contains(KeyValuePair`2):bool:this (7 methods)
-243 (-9.14% of base) : System.Collections.Immutable.dasm - ImmutableHashSet`1:IsProperSubsetOf(IEnumerable`1,MutationInput):bool (7 methods)
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-238 (-9.06% of base) : System.Reflection.MetadataLoadContext.dasm - EcmaModule:GetTypeCoreNoCache(ReadOnlySpan`1,ReadOnlySpan`1,byref):RoDefinitionType:this
-216 (-69.23% of base) : System.Text.Json.dasm - JsonWriterHelper:ValidatePropertyAndValue(ReadOnlySpan`1,ReadOnlySpan`1) (4 methods)
-190 (-6.61% of base) : System.Security.Cryptography.Algorithms.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-190 (-6.61% of base) : System.Security.Cryptography.Cng.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-187 (-6.50% of base) : System.Security.Cryptography.Pkcs.dasm - PasswordBasedEncryption:Decrypt(byref,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,Span`1):int
-176 (-38.60% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberByOptions(ReadOnlySpan`1,long):this (4 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Algorithms.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.22% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadSubjectPublicKeyInfo(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-174 (-4.23% of base) : System.Security.Cryptography.Pkcs.dasm - KeyFormatHelper:ReadPkcs8(ref,ReadOnlyMemory`1,KeyReader`1,byref,byref) (7 methods)
-167 (-59.01% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpAttributeData:CreateMetadataExpression(TypedConstant,EmitContext):IMetadataExpression:this
-162 (-53.64% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteLiteralByOptions(ReadOnlySpan`1,ReadOnlySpan`1):this (2 methods)
Top method regressions (percentages):
9 (12.50% of base) : Microsoft.CodeAnalysis.dasm - SyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.CSharp.dasm - CSharpSyntaxWalker:VisitTrivia(SyntaxTrivia):this
9 ( 8.74% of base) : Microsoft.CodeAnalysis.VisualBasic.dasm - VisualBasicSyntaxWalker:VisitTrivia(SyntaxTrivia):this
12 ( 8.28% of base) : CommandLine.dasm - <>c__DisplayClass0_1:<MapValues>b__1(KeyValuePair`2):bool:this
13 ( 7.93% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCurrentChanged(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 7.69% of base) : System.Reflection.Metadata.dasm - StateMachineMethodTableReader:.ctor(int,bool,int,MemoryBlock,int):this
4 ( 7.41% of base) : System.Net.Sockets.dasm - UdpReceiveResult:op_Equality(UdpReceiveResult,UdpReceiveResult):bool
7 ( 6.48% of base) : Microsoft.CodeAnalysis.CSharp.dasm - Binder:ReportDiagnosticsIfObsolete(DiagnosticBag,Conversion,SyntaxNodeOrToken,bool):this
10 ( 6.41% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionPromoted(TransactionTraceIdentifier,TransactionTraceIdentifier):this
11 ( 6.15% of base) : System.Reflection.Metadata.dasm - CustomDebugInformationTableReader:.ctor(int,bool,int,int,int,MemoryBlock,int):this
27 ( 5.68% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteStringEscape(ReadOnlySpan`1,DateTime):this (2 methods)
54 ( 5.61% of base) : System.Text.Json.dasm - Utf8JsonWriter:WriteNumberEscape(ReadOnlySpan`1,long):this (4 methods)
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - FieldRVATableReader:.ctor(int,bool,int,MemoryBlock,int):this
9 ( 5.56% of base) : System.Reflection.Metadata.dasm - NestedClassTableReader:.ctor(int,bool,int,MemoryBlock,int):this
2 ( 5.26% of base) : Microsoft.CodeAnalysis.dasm - <>c:<ReportAnalyzerExecutionTime>b__42_2(KeyValuePair`2):Assembly:this
2 ( 5.26% of base) : System.Private.DataContractSerialization.dasm - XmlUTF8NodeWriter:WriteGuidText(Guid):this
9 ( 5.26% of base) : System.Reflection.Metadata.dasm - ClassLayoutTableReader:.ctor(int,bool,int,MemoryBlock,int):this
6 ( 5.17% of base) : System.Transactions.Local.dasm - TransactionsEtwProvider:TransactionScopeCreated(TransactionTraceIdentifier,int):this
9 ( 5.14% of base) : System.Reflection.Metadata.dasm - InterfaceImplTableReader:.ctor(int,bool,int,int,MemoryBlock,int):this
Top method improvements (percentages):
-240 (-83.33% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-336 (-82.76% of base) : System.Private.CoreLib.dasm - Path:Join(ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1,ReadOnlySpan`1):String
-20 (-80.00% of base) : Microsoft.VisualBasic.Core.dasm - Conversion:Int(Decimal):Decimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Multiply(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDecimal:Divide(SqlDecimal,SqlDecimal):SqlDecimal
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Add(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Subtract(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Multiply(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlDouble:Divide(SqlDouble,SqlDouble):SqlDouble
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:OnesComplement(SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Add(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Subtract(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Multiply(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Divide(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Mod(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Modulus(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseAnd(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:BitwiseOr(SqlInt64,SqlInt64):SqlInt64
-20 (-80.00% of base) : System.Data.Common.dasm - SqlInt64:Xor(SqlInt64,SqlInt64):SqlInt64
1996 total methods with Code Size differences (1811 improved, 185 regressed), 238046 unchanged.

Lots of interesting diffs, here's simple one:
;; SqlDecimal:Add(SqlDecimal,SqlDecimal):SqlDecimal
;; before
G_M20502_IG01:
push rsi
sub rsp, 32
mov rsi, rcx
G_M20502_IG02:
mov rcx, rsi
call SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal
mov rax, rsi
G_M20502_IG03:
add rsp, 32
pop rsi
ret ;; after
G_M20502_IG02:
jmp SqlDecimal:op_Addition(SqlDecimal,SqlDecimal):SqlDecimal

More is possible here:
  • can do this for any call in tail position, not just tail calls
  • can handle passing a nested sub-struct to callee
  • can handle promoted case by writing back promoted fields to the original struct before tail calling
    
(and avoid write-back if original struct and promoted fields haven't diverged)
  • true liveness for robust last-use detection would be nice
@AndyAyersMSAndyAyersMS
added
the 
area-CodeGen-coreclr
CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
label
Feb 29, 2020
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Needs an extra safety check to make sure we're not introducing aliasing -- eg same byref passed to callee in two param slots, and callee writes to one.... hopefully doesn't happen often.
@jashook
jashook
self-requested a review
February 29, 2020 04:14
jashook
jashook
approved these changes
Feb 29, 2020
View reviewed changes
@jashookjashook
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm minus a nit
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 //
 // Otherwise, see how many appearances there are. We keep two early ref counts: total
 // number of references to the struct or some field, and how many of these are
 // arguments to calls. We undo promotion unless we see enougn non-call uses.
@jashookjashookFeb 29, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: enougn -> enough
@benaadamsbenaadams
mentioned this pull request
Feb 29, 2020

Closed

@benaadams
benaadams
commented
Feb 29, 2020

Copy link


Copy Markdown

Member
Raised issue for "Linux arm checked" #33012
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
As with anything aliasing related, checking is costly... still working on it.
@jaykrell
jaykrell
commented
Feb 29, 2020

Copy link


Copy Markdown

Contributor
It’d be nice to test & implement similar for mono. /cc @lambdageek
@AndyAyersMS
AndyAyersMS
commented
Feb 29, 2020

Copy link


Copy Markdown

MemberAuthor
Note for coreclr, this is windows-x64 and windows-arm64 only.
More register-intensive ABIs like SysV should (ideally) already be able to optimize some of these cases -- that is, if we keep passed-via-registers structs in registers, they can just flow through to callees. But if we materialize the struct on the frame (as has been our habit) then we will miss these cases on SysV too. The fix for that is quite different than the fix in this PR.
@AndyAyersMS
AndyAyersMS
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

MemberAuthor
A first cut at an alias analysis is in, not sure if its complexity matches its value, though.
Belive there is a bug in morph that I'm working around for now; will chase this down separately, and if/when fixed we can get rid of the check for TYP_I_IMPL args aliasing our byref arg; anything that can alias the struct should be TYP_BYREF.
Have changed the "do not return" case to require that there be just one reference to the implicit byref in the method. We're only doing the alias analysis for tail calls, so those are the only cases where we can safely deal with multiple references.
This version has basically the same diffs as before (we lose 9 cases); we don't expect to see true aliases very often. Magic arg limit of 6 chosen so we still get the copy elimination and tail calls in SPC's Number:TryFormatInt32
runtime/src/libraries/System.Private.CoreLib/src/System/Number.Formatting.cs

Lines 747 to 757
in
7c7e8ed
publicstaticboolTryFormatInt32(intvalue,inthexMask,ReadOnlySpan<char>format,IFormatProvider?provider,Span<char>destination,outintcharsWritten)
{
// Fast path for default format
if(format.Length==0)
{
returnvalue>=0?
TryUInt32ToDecStr((uint)value,digits:-1,destination,outcharsWritten):
TryNegativeInt32ToDecStr(value,digits:-1,NumberFormatInfo.GetInstance(provider).NegativeSign,destination,outcharsWritten);
}
returnTryFormatInt32Slow(value,hexMask,format,provider,destination,outcharsWritten);
;;; BEFORE; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; partially interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 2, 4 ) byref -> rcx ; V05 arg5 [V05,T16] ( 3, 1.50) byref -> rdi ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (48) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T17] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)"; V10 tmp4 [V10,T08] ( 4, 2.50) byref -> rbx V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)"; V11 tmp5 [V11,T09] ( 4, 2.50) int -> rbp V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 tmp8 [V14 ] ( 6, 6 ) struct (16) [rsp+0x48] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V15 tmp9 [V15,T05] ( 3, 3 ) byref -> r9 stack-byref "BlockOp address local"; V16 tmp10 [V16,T12] ( 2, 2 ) int -> rcx "argument with side effect"; V17 tmp11 [V17,T10] ( 2, 2 ) ref -> r8 "argument with side effect"; V18 tmp12 [V18,T06] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V19 tmp13 [V19,T13] ( 2, 2 ) int -> rcx "argument with side effect"; V20 tmp14 [V20 ] ( 2, 2 ) struct (16) [rsp+0x38] do-not-enreg[XSB] must-init addr-exposed "by-value struct argument"; V21 tmp15 [V21,T07] ( 3, 3 ) byref -> r8 stack-byref "BlockOp address local"; V22 tmp16 [V22,T14] ( 2, 2 ) int -> rcx "argument with side effect"; V23 tmp17 [V23,T15] ( 2, 2 ) int -> rdx "argument with side effect"; V24 tmp18 [V24,T11] ( 2, 2 ) ref -> r9 "argument with side effect";; Lcl frame size = 88G_M23990_IG01:pushrdipushrsipushrbppushrbxsubrsp,88vzeroupperxorrax,raxmov qword ptr [rsp+48H],raxmov qword ptr [rsp+38H],raxmovesi,ecxmovrcx, bword ptr [rsp+A0H]movrdi, bword ptr [rsp+A8H] ;; bbWeight=1 PerfScore 9.75G_M23990_IG02:movrbx, bword ptr [rcx]movebp, dword ptr [rcx+8] ;; bbWeight=1 PerfScore 4.00G_M23990_IG03:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG10 ;; bbWeight=1 PerfScore 3.00G_M23990_IG04:testesi,esijge SHORT G_M23990_IG08testr9,r9je SHORT G_M23990_IG05movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG06 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG05:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG06:movecx,esimovr8, gword ptr [rax+40]lear9, bword ptr [rsp+48H]mov bword ptr [r9],rbxmov dword ptr [r9+8],ebpmov bword ptr [rsp+20H],rdilear9, bword ptr [rsp+48H]movedx,-1call Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.88G_M23990_IG07:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG08:movecx,esilear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebplear8, bword ptr [rsp+48H]movr9,rdimovedx,-1call Number:TryUInt32ToDecStr(int,int,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 2.50G_M23990_IG09:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63G_M23990_IG10:movecx,esi ;; bbWeight=0.50 PerfScore 0.13G_M23990_IG11: vmovdqu xmm0, xmmword ptr [r8] vmovdqu xmmword ptr [rsp+38H],xmm0 ;; bbWeight=0.50 PerfScore 1.50G_M23990_IG12:lear8, bword ptr [rsp+48H]mov bword ptr [r8],rbxmov dword ptr [r8+8],ebpmov bword ptr [rsp+28H],rdilear8, bword ptr [rsp+38H]learax, bword ptr [rsp+48H]mov bword ptr [rsp+20H],raxcall Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):boolnop ;; bbWeight=0.50 PerfScore 3.38G_M23990_IG13:addrsp,88poprbxpoprbppoprsipoprdiret ;; bbWeight=0.50 PerfScore 1.63; Total bytes of code 228, prolog size 23, PerfScore 59.28, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool;;; AFTER; Assembly listing for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool; Emitting BLENDED_CODE for X64 CPU with AVX - Windows; optimized code; rsp based frame; fully interruptible; Final local variable assignments;; V00 arg0 [V00,T01] ( 6, 4 ) int -> rsi ; V01 arg1 [V01,T03] ( 3, 2.50) int -> rdx ; V02 arg2 [V02,T00] ( 4, 7 ) byref -> r8 ld-addr-op; V03 arg3 [V03,T02] ( 5, 3.50) ref -> r9 class-hnd; V04 arg4 [V04,T04] ( 3, 3 ) byref -> rdi ; V05 arg5 [V05,T06] ( 3, 1.50) byref -> rbx ; V06 OutArgs [V06 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"; V07 tmp1 [V07,T07] ( 3, 1.50) ref -> rax class-hnd "Inline return value spill temp";* V08 tmp2 [V08 ] ( 0, 0 ) byref -> zero-ref V12._pointer(offs=0x00) P-INDEP "field V02._pointer (fldOffset=0x0)";* V09 tmp3 [V09 ] ( 0, 0 ) int -> zero-ref V12._length(offs=0x08) P-INDEP "field V02._length (fldOffset=0x8)";* V10 tmp4 [V10 ] ( 0, 0 ) byref -> zero-ref V13._pointer(offs=0x00) P-INDEP "field V04._pointer (fldOffset=0x0)";* V11 tmp5 [V11 ] ( 0, 0 ) int -> zero-ref V13._length(offs=0x08) P-INDEP "field V04._length (fldOffset=0x8)";* V12 tmp6 [V12 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref";* V13 tmp7 [V13 ] ( 0, 0 ) struct (16) zero-ref "Promoted implicit byref"; V14 rat0 [V14,T05] ( 2, 2 ) byref -> r9 "Fast tail call lowering is creating a new local variable";; Lcl frame size = 32G_M23990_IG01:pushrdipushrsipushrbxsubrsp,32movesi,ecxmovrdi, bword ptr [rsp+60H]movrbx, bword ptr [rsp+68H] ;; bbWeight=1 PerfScore 5.50G_M23990_IG02:cmp dword ptr [r8+8],0jne SHORT G_M23990_IG09 ;; bbWeight=1 PerfScore 3.00G_M23990_IG03:testesi,esijge SHORT G_M23990_IG07testr9,r9je SHORT G_M23990_IG04movrcx,r9call NumberFormatInfo:<GetInstance>g__GetProviderNonNull|42_0(IFormatProvider):NumberFormatInfojmp SHORT G_M23990_IG05 ;; bbWeight=0.50 PerfScore 2.88G_M23990_IG04:call NumberFormatInfo:get_CurrentInfo():NumberFormatInfo ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG05:movr9,rdimov bword ptr [rsp+60H],rbxmovr8, gword ptr [rax+40]movecx,esimovedx,-1 ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG06:addrsp,32poprbxpoprsipoprdijmp Number:TryNegativeInt32ToDecStr(int,int,String,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG07:movecx,esimovr9,rbxmovr8,rdimovedx,-1 ;; bbWeight=0.50 PerfScore 0.50G_M23990_IG08:addrsp,32poprbxpoprsipoprdijmp Number:TryUInt32ToDecStr(int,int,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88G_M23990_IG09:mov bword ptr [rsp+60H],rdimov bword ptr [rsp+68H],rbxmovecx,esi ;; bbWeight=0.50 PerfScore 1.13G_M23990_IG10:addrsp,32poprbxpoprsipoprdijmp Number:<TryFormatInt32>g__TryFormatInt32Slow|33_0(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool ;; bbWeight=0.50 PerfScore 1.88; Total bytes of code 130, prolog size 19, PerfScore 34.00, (MethodHash=5b40a24c) for method Number:TryFormatInt32(int,int,ReadOnlySpan`1,IFormatProvider,Span`1,byref):bool
Note we could tighten up the latter a bit more, there's no need to copy the span in to RDI/RBX and back.
@AndyAyersMSAndyAyersMS
mentioned this pull request
Mar 1, 2020

Closed

@GrabYourPitchforks
GrabYourPitchforks
commented
Mar 1, 2020•

edited

Loading

Copy link


Copy Markdown

Member
@AndyAyersMS I rebased #32843 on top of this change (and removed the in weirdness that was contentious in that PR) and I'm seeing significantly improved codegen. 😄
MethodToolchainMeanErrorStdDevRatioRatioSD
TryParseInt32master1,218.0 ns30.51 ns33.91 ns1.000.00
TryParseInt32utf8parser841.8 ns7.34 ns6.87 ns0.690.02
The only remaining nit is the "jump-to-jump" pattern that appears in the below codegen, but I don't know if that's related to this PR or if it's a separate opportunity for further cleanup. (Edit: maybe it's for predict-not-taken scenarios?)
00007ffe`449db320 4585c9 testr9d,r9d00007ffe`449db323 7505jne System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca905a (00007ffe`449db32a)00007ffe`449db325 e9dedaffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e08 (00007ffe`449d8e08)00007ffe`449db32a 4183c920 orr9d,20h00007ffe`449db32e 4183f967 cmpr9d,67h00007ffe`449db332 7f0e jg System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9072 (00007ffe`449db342)00007ffe`449db334 4183f964 cmpr9d,64h00007ffe`449db338 74eb je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db33a 4183f967 cmpr9d,67h00007ffe`449db33e 74e5 je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9055 (00007ffe`449db325)00007ffe`449db340 eb18 jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db342 4183f96e cmpr9d,6Eh00007ffe`449db346 7408je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9080 (00007ffe`449db350)00007ffe`449db348 4183f978 cmpr9d,78h00007ffe`449db34c 7407je System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca9085 (00007ffe`449db355)00007ffe`449db34e eb0a jmp System_Private_CoreLib!System.Buffers.Text.Utf8Parser.TryParse(System.ReadOnlySpan`1<Byte>, Int32 ByRef, Int32 ByRef, Int32)+0xffffffff`a0ca908a (00007ffe`449db35a)00007ffe`449db350 e9d3daffff jmp CLRStub[MethodDescPrestub]@7ffe449d8e28 (00007ffe`449d8e28)00007ffe`449db355 e94edbffff jmp CLRStub[MethodDescPrestub]@7ffe449d8ea8 (00007ffe`449d8ea8)00007ffe`449db35a e991f3ffff jmp CLRStub[MethodDescPrestub]@7ffe449da6f0 (00007ffe`449da6f0)
For reference, here's the C# that was used for this codegen:
https://github.com/GrabYourPitchforks/runtime/blob/ab36e65d9d1938a22e8c0baa6aa51e1797016817/src/libraries/System.Private.CoreLib/src/System/Buffers/Text/Utf8Parser/Utf8Parser.Integer.Signed.cs#L129-L158
@AndyAyersMS
AndyAyersMS
commented
Mar 2, 2020

Copy link


Copy Markdown

MemberAuthor
@GrabYourPitchforks thanks.
For the jump intensive TryParse example -- we don't have a concept of a "conditional tail jump" so what you see is about the best we can do for now. Ideally we could invert the first branch
G_M5679_IG02: 4585C9 testr9d,r9d7505jne SHORT G_M5679_IG04G_M5679_IG03: E9D613A5FF jmp TryParseInt32D(System.ReadOnlySpan`1[Byte],...)
and emit something like:
G_M5679_IG02: 4585C9 testr9d,r9d xxxxxxxx je TryParseInt32D(System.ReadOnlySpan`1[Byte],...)G_M5679_IG04:
and similarly elsewhere, but this sort of construct would be a new concept for the jit. And I'm not sure how often we see tail calls with no need for argument setup. Will keep it on my list of things to watch for.
SysV codegen looks like it needs some TLC here too. On SysV spans are passed in registers, and here the jit can't resist spilling them to the stack.
G_M51636_IG01:pushrbpsubrsp,16learbp,[rsp+10H]mov bword ptr [rbp-10H],rdimov qword ptr [rbp-08H],rsiG_M51636_IG02:testr8d,r8djne SHORT G_M51636_IG05G_M51636_IG03:movrdi, bword ptr [rbp-10H]movrsi, qword ptr [rbp-08H]G_M51636_IG04:learsp,[rbp]poprbpjmp TryParseInt32D(ReadOnlySpan`1,byref,byref):bool
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 2, 2020
 Closed

@AndyAyersMS
AndyAyersMS
commented
Mar 9, 2020

Copy link


Copy Markdown

MemberAuthor
@dotnet/jit-contrib any other comments? I figured some of this might need re-expressing...
@erozenfeld
erozenfeld
commented
Mar 10, 2020

Copy link


Copy Markdown

Contributor
I'll review this tomorrow.
erozenfeld
erozenfeld
suggested changes
Mar 10, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I didn't find any holes in the algorithm but left some questions and suggestions.
Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/lclmorph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
 dspTreeID(arg->GetNode()), lcl->GetLclNum());

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would move this JITDUMP to under the next else.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

 return result;
 }

 // Will return different answers if x and y refer to same struct
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

y is not a struct here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

'y' is a ref so it parsed ok to me, but will revise to make it clearer.
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCallsAliasing.cs
Outdated

Comment threadsrc/coreclr/src/jit/gentree.cpp
 lcl = addr->AsLclVarCommon();
 }
 else if (addr->OperIs(GT_ADDR))
 {
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a semantic difference between OBJ(lcl) and OBJ(ADDR(lcl)) where lcl is an implicit byref parameter? Do they both represent a copy of the struct? When do we end up with one vs. the other?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What this means changes over time, and we're straddling that change.
See above?
Comment threadsrc/coreclr/tests/src/JIT/opt/Tailcall/ImplicitByrefTailCalls.cs
 // Here the call to ZZZ would need to be a slow tail call,
 // so we won't tail call at all. But the only
 // reference to x is at the call, and it's not in 
 // a loop, so we can still can avoid making a copy.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we also have a test case when the call is in a loop and a test case for NoReturnLastUse?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sure, will add these.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // make a local copy of the struct for the callee.
 hasByrefParameter = true;
 break;

@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would rename this var to something like copyByRefStructParameter.
sandreenko
sandreenko
reviewed
Mar 10, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

a few questions.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated


 if (addr->OperIs(GT_LCL_VAR))
 {
 lcl = addr->AsLclVarCommon();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't understand this case, LCL_VAR is used as an address here (in other cases it is used as a value), so it has to be 'byref' or 'ref' type. Then how can it be lvaIsImplicitByRefLocal if it always should return false for TYP_IMPL size lclVars?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, this is potentially confusing.
The short answer is that before morph, arguments passed implicitly by-reference are represented by-value in the IR.

This gets changed in fgMorphImplicitByRefArgs where we make the by-reference aspect explicit. But the above helper gets called both before and after, and so needs to recognize cases where the local either represents a struct or a pointer to a struct.
Note the first two "match" cases here are pre-existing logic; all I did was move this to a helper method and add the last case.
Comment threadsrc/coreclr/src/jit/gentree.cpp
Outdated

// Return Value:
// GenTreeLclVarCommon node for the local, or nullptr.

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You can change GenTreeLclVarCommon to GenTreeLclVar, it is more precise for this case.
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 if (!call->IsTailCallViaHelper() && (varDsc->lvRefCnt(RCS_EARLY) == 1) && !fgMightHaveLoop())
 {
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No need for AsLclVarCommon.
Comment threadsrc/coreclr/src/jit/morph.cpp
 assert(!call->IsTailCall());
 unsigned varNum = lcl->AsLclVarCommon()->GetLclNum();
 LclVarDsc* varDsc = lvaGetDesc(varNum);
 const unsigned short totalAppearances = varDsc->lvRefCnt(RCS_EARLY);
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

question: how do you decide what is marked as const and what is not? Like why varNum is not in this case.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I try to mark all locals that won't be redefined as const, but am more likely to do this for lines I explicitly edit.
Comment threadsrc/coreclr/src/jit/morph.cpp
 {
 // First, see if this arg is a byref param.
 GenTreeLclVarCommon* const lcl = arg->GetNode()->IsImplicitByrefParameterValue(this);

@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: create a temp for lcl->GetLclNum()?
Comment threadsrc/coreclr/src/jit/morph.cpp
Outdated

 // a local struct assembled from the promoted fields.
 if (lcl != nullptr)
 {
 JITDUMP("Arg [%06u] is unpromoted implicit byref V%02u, seeing if we can still tail call\n",
@sandreenkosandreenkoMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sentence is confusing: "Arg 000001 is unpromoted implicit byref V002, seeing if we can still tail call ... no, V002 was promoted. "

Can we check varDsc->lvPromoted before we print that it was not promoted?
Comment threadsrc/coreclr/src/jit/morph.cpp
 JITDUMP("... yes, arg is the only appearance of V%02u\n", lcl->GetLclNum());
 hasByrefParameter = false;
 }
 // If no other call arg refers to this byref, and no other arg is
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I can't find it in https://github.com/dotnet/runtime/blob/master/docs/coding-guidelines/clr-jit-coding-conventions.md#73-commenting-code-blocks, but my personal opinion is that we should not allow comments after } and before 'else {`.
Comment threadsrc/coreclr/src/jit/morph.cpp
 // Do we pass 'lcl' more than once to the callee?
 if (arg2->isStruct && arg2->passedByRef)
 {
 GenTreeLclVarCommon* const lcl2 =
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm, in the previous use of IsImplicitByrefParameterValue was for optimizations, so if it was not able to find the lcl it was correct, it was just rejecting the optimization. Here it looks different: if it doesn't find the local, then it is optimizing. There are cases where locals are buried under IND(ADDR(IND(ADDR())) sequences, I think IsImplicitByrefParameterValue won't give us the right result for such cases and it could lead to an incorrect optimization.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
@erozenfelderozenfeldMar 10, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it looks different: if it doesn't find the local, then it is optimizing.
No, if we don't find the local, we are being conservative and checking arg for exposure if arg2 is a TYP_BYREF or TYP_I_IMPL.
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note that if IND long(ADDR(IND long(ADDR(LCL_VAR long))) doesn't widen, then lclmorph doesn't mark LCL_VAR as lvaSetVarAddrExposed, so the next check won't catch it also.
I think that's fine, there is no real address exposure here.
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I mean a case like call(LCL_VAR v01, IND(ADDR(IND(ADDR(LCL_VAR v01))) ), when we are checking arg2 for interference with arg1, how will we catch it?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. Yes, that may be a problem. Also, can we have call(LCL_VAR V01, ADD(LCL_VAR V01, INT x)) ?
@erozenfelderozenfeldMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Actually, won't we have lvHasLdAddrOp set in these cases?
@sandreenkosandreenkoMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't set lvHasLdAddrOp if the next instruction consumes it:
runtime/src/coreclr/src/jit/flowgraph.cpp

Lines 4797 to 4809
in
73268c7
if (notStruct && notLastInstr && notDebugCode && 
impILConsumesAddr(codeAddr + sz, impTokenLookupContextHandle, info.compScopeHnd)) 
 { 
// We can skip the addrtaken, as next IL instruction consumes
// the address.
 } 
else
 { 
 lvaTable[varNum].lvHasLdAddrOp = 1; 
if (!info.compIsStatic && (varNum == 0)) 
 { 
// Addr taken on "this" pointer is significant,
// go ahead to mark it as permanently addr-exposed here.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we don't recognize an implicit byref arg via IsImplicitByrefParameterValue we won't tail call. So for instance, passing a sub-struct of an implicit byref param currently blocks tail calling. I thought about unblocking this but left it as a future enhancement.
If there's an explicit byref arg1 that could alias some other implicit byref arg2 (which is also an implicit byref param), then the only way for arg1 to alias arg2 is for the param local to be address exposed or address taken. This must be an "unconsumed" case per above.
AndyAyersMSand others
added 4 commits
March 10, 2020 18:52
@AndyAyersMS@erozenfeld
Apply suggestions from code review
cf98b3e
Co-Authored-By: Eugene Rozenfeld <erozen@microsoft.com>
@AndyAyersMS
update test cases
50da720
@AndyAyersMS
Tweak the tests a bit.
7084ebf
Use no-opt to ensure tail called methods don't promote their struct
args and so undo the aliasing checks.
Note that one of the tests that could tail call doesn't because of
the way morph reuses struct temps.
@AndyAyersMS
code changes per review feedback
a33348c
@AndyAyersMS
AndyAyersMS
commented
Mar 11, 2020

Copy link


Copy Markdown

MemberAuthor
OSX builds failing with
Error: python 3.7.6_1 is already installed
To upgrade to 3.7.7, run `brew upgrade python`.

@jashook
jashook
commented
Mar 11, 2020

Copy link


Copy Markdown

Contributor
Fixed this morning with #33481
jashook
jashook
reviewed
Mar 11, 2020
View reviewed changes
Comment threadsrc/coreclr/src/jit/gentree.cpp

GenTreeLclVarCommon* GenTree::IsImplicitByrefParameterValue(Compiler* compiler)
{
#if defined(TARGET_AMD64) || defined(TARGET_ARM64)
@jashookjashookMar 10, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't this be limited to windows amd64?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Windows, yes.
We seem to think arm64 can have these too. I don't know the ABI well enough to say one way or another.
runtime/src/coreclr/src/jit/compiler.h

Lines 444 to 446
in
5ea5b57
 #if defined(TARGET_AMD64) || defined(TARGET_ARM64) 
unsignedchar lvIsImplicitByRef : 1; // Set if the argument is an implicit byref.
 #endif// defined(TARGET_AMD64) || defined(TARGET_ARM64)
Perhaps we need a define just for this and a cleanup of all references?
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like windows arm64 uses the standard ARM64 EABI, so there are no implicit byrefs there. I'll put up a separate PR to clean this up here and throughout the rest of the jit codebase.
@jashookjashookMar 11, 2020

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I was under the impression that arm64 has implicit byrefs, but defining it for unix amd64 seems incorrect for sure.
@jashookjashookMar 11, 2020•

edited

Loading

Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

> 16 byte structs on arm64 windows or linux I believe are passed by reference.
@AndyAyersMSAndyAyersMSMar 11, 2020

Copy link


Copy Markdown

MemberAuthor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, I see it now.. (from the EABI)
If the argument type is a Composite Type larger than 16 bytes, then the argument is copied to memory allocated by the caller, and the argument is replaced by a pointer to the copy.
So the right condition is windows x64, and all arm64.
sandreenko
sandreenko
approved these changes
Mar 11, 2020
View reviewed changes
@sandreenkosandreenko
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM
erozenfeld
erozenfeld
approved these changes
Mar 11, 2020
View reviewed changes
@erozenfelderozenfeld
left a comment


Copy link


Copy Markdown

Contributor
Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks for making the changes.
@AndyAyersMS
AndyAyersMS
merged commit 514db28
into
dotnet:masterMar 11, 2020
@AndyAyersMS
AndyAyersMS
deleted the

ImplicitByRefTailCallOpt

branch
March 11, 2020 20:16
@GrabYourPitchforksGrabYourPitchforks
mentioned this pull request
Mar 12, 2020
 Merged

@AndyAyersMSAndyAyersMS
mentioned this pull request
Jul 22, 2020

Closed

@AndyAyersMSAndyAyersMS
mentioned this pull request
Aug 9, 2020

Closed

@ghostghost
locked as resolved and limited conversation to collaborators
Dec 10, 2020
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.


Reviewers


3 more reviewers

@jashookjashookjashook approved these changes
@erozenfelderozenfelderozenfeld approved these changes
@sandreenkosandreenkosandreenko approved these changes
Reviewers whose approvals may not affect merge requirements

Assignees


No one assigned


Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI

Projects


None yet


Milestone


No milestone


Development

Successfully merging this pull request may close these issues.

7 participants

@AndyAyersMS@benaadams@jaykrell@GrabYourPitchforks@erozenfeld@jashook@sandreenko