Crossgen method bodies having ARM64 intrinsic #38060

kunalspathak · 2020-06-17T23:56:44Z

Today, method bodies containing ARM64 intrinsics are not crossgened. Currently there are not many methods that uses ARM64 intrinsics, but gradually once all the APIs #33308 are intrinsified , it will impact startup time of those methods because they would get JITted first before executing. This PR makes such methods bodies to get crossgened for faster startup time. Additionally, methods of Vector64 and Vector128 were being treated as method calls but since they are intrinsified already (again as part of #33308), it is safe to just crossgen their method bodies.

Below is the code size difference. There seems to be regression of 1% but it will still give us good startup time benefits.


Summary of Code Size diffs:
(Lower is better)

Total bytes of diff: 53216 (0.935% of base)
    diff is a regression.

Total byte diff includes 59916 bytes from reconciling methods
        Base had    0 unique methods,        0 unique bytes
        Diff had 1477 unique methods,    59916 unique bytes

Top file regressions (bytes):
       53216 : System.Private.CoreLib.dasm (0.935% of base)

1 total files with Code Size differences (0 improved, 1 regressed), 0 unchanged.

Top method regressions (bytes):
         500 (10.113% of base) : System.Private.CoreLib.dasm - Internal.Runtime.CompilerServices.Unsafe:As(byref):byref (292 base, 323 diff methods)
         220 (117.021% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:GetHashCode():int:this
         208 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:WithLower(System.Runtime.Intrinsics.Vector128`1[__Canon],System.Runtime.Intrinsics.Vector64`1[__Canon]):System.Runtime.Intrinsics.Vector128`1[__Canon] (0 base, 1 diff methods)
         208 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:WithUpper(System.Runtime.Intrinsics.Vector128`1[__Canon],System.Runtime.Intrinsics.Vector64`1[__Canon]):System.Runtime.Intrinsics.Vector128`1[__Canon] (0 base, 1 diff methods)
         180 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte,ubyte):System.Runtime.Intrinsics.Vector128`1[Byte] (0 base, 1 diff methods)
         180 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte,byte):System.Runtime.Intrinsics.Vector128`1[SByte] (0 base, 1 diff methods)
         136 (75.556% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:<Equals>g__SoftwareFallback|12_0(byref,System.Runtime.Intrinsics.Vector128`1[UInt32]):bool
         136 (75.556% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:<Equals>g__SoftwareFallback|12_0(byref,System.Runtime.Intrinsics.Vector128`1[Int32]):bool
         112 (70.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int16][System.Int16]:GetHashCode():int:this
         112 (70.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:GetHashCode():int:this
         112 (70.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:GetHashCode():int:this
         112 (70.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[UInt16][System.UInt16]:GetHashCode():int:this
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Byte],ubyte,System.Runtime.Intrinsics.Vector64`1[Byte],ubyte):System.Runtime.Intrinsics.Vector128`1[Byte] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Byte],ubyte,System.Runtime.Intrinsics.Vector128`1[Byte],ubyte):System.Runtime.Intrinsics.Vector128`1[Byte] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Double],ubyte,System.Runtime.Intrinsics.Vector128`1[Double],ubyte):System.Runtime.Intrinsics.Vector128`1[Double] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Int16],ubyte,System.Runtime.Intrinsics.Vector64`1[Int16],ubyte):System.Runtime.Intrinsics.Vector128`1[Int16] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Int16],ubyte,System.Runtime.Intrinsics.Vector128`1[Int16],ubyte):System.Runtime.Intrinsics.Vector128`1[Int16] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Int32],ubyte,System.Runtime.Intrinsics.Vector64`1[Int32],ubyte):System.Runtime.Intrinsics.Vector128`1[Int32] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Int32],ubyte,System.Runtime.Intrinsics.Vector128`1[Int32],ubyte):System.Runtime.Intrinsics.Vector128`1[Int32] (0 base, 1 diff methods)
         108 (     ∞ of base) : System.Private.CoreLib.dasm - Arm64:InsertSelectedScalar(System.Runtime.Intrinsics.Vector128`1[Int64],ubyte,System.Runtime.Intrinsics.Vector128`1[Int64],ubyte):System.Runtime.Intrinsics.Vector128`1[Int64] (0 base, 1 diff methods)

Top method improvements (bytes):
        -236 (-19.601% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int64][System.Int64]:ToString():System.String:this
        -232 (-21.561% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Double][System.Double]:ToString():System.String:this
        -212 (-20.623% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[UInt64][System.UInt64]:ToString():System.String:this
        -176 (-3.392% of base) : System.Private.CoreLib.dasm - System.Numerics.Matrix4x4:<Invert>g__SseImpl|59_0(System.Numerics.Matrix4x4,byref):bool
        -156 (-9.466% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:GetIndexOfFirstNonAsciiChar_Sse2(long,long):long
        -148 (-8.959% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:GetIndexOfFirstNonLatin1Char_Sse2(long,long):long
        -104 (-54.167% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:UnalignedCountVector128(byref):long (2 methods)
        -104 (-57.778% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int64][System.Int64]:Equals(System.Runtime.Intrinsics.Vector64`1[Int64]):bool:this
        -104 (-57.778% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[UInt64][System.UInt64]:Equals(System.Runtime.Intrinsics.Vector64`1[UInt64]):bool:this
         -88 (-44.898% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Double][System.Double]:Equals(System.Runtime.Intrinsics.Vector64`1[Double]):bool:this
         -84 (-21.429% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
         -84 (-20.588% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:NarrowUtf16ToLatin1_Sse2(long,long,long):long
         -76 (-14.179% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:WidenLatin1ToUtf16_Sse2(long,long,long)
         -76 (-38.776% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Double][System.Double]:GetHashCode():int:this
         -76 (-46.341% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int64][System.Int64]:GetHashCode():int:this
         -76 (-46.341% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[UInt64][System.UInt64]:GetHashCode():int:this
         -68 (-30.909% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[SByte][System.SByte]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[SByte]):bool
         -68 (-30.909% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Byte][System.Byte]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[Byte]):bool
         -64 (-29.091% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[UInt16][System.UInt16]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[UInt16]):bool
         -64 (-29.091% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int16][System.Int16]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[Int16]):bool

Top method regressions (percentages):
          40 (     ∞ of base) : System.Private.CoreLib.dasm - System.Math:CopySign(double,double):double (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.MathF:CopySign(float,float):float (0 base, 1 diff methods)
          80 (     ∞ of base) : System.Private.CoreLib.dasm - System.SpanHelpers:TryFindFirstMatchedLane(System.Runtime.Intrinsics.Vector128`1[Byte],System.Runtime.Intrinsics.Vector128`1[Byte],byref):bool (0 base, 1 diff methods)
          72 (     ∞ of base) : System.Private.CoreLib.dasm - System.SpanHelpers:TryFindFirstMatchedLane(System.Runtime.Intrinsics.Vector128`1[UInt16],byref):bool (0 base, 1 diff methods)
          20 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:LeadingZeroCount(int):int (0 base, 1 diff methods)
          20 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:LeadingZeroCount(long):int (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:Log2(int):int (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:Log2(long):int (0 base, 1 diff methods)
          36 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:PopCount(int):int (0 base, 1 diff methods)
          32 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.BitOperations:PopCount(long):int (0 base, 1 diff methods)
          48 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.VectorMath:ConditionalSelectBitwise(System.Runtime.Intrinsics.Vector128`1[Single],System.Runtime.Intrinsics.Vector128`1[Single],System.Runtime.Intrinsics.Vector128`1[Single]):System.Runtime.Intrinsics.Vector128`1[Single] (0 base, 1 diff methods)
          48 (     ∞ of base) : System.Private.CoreLib.dasm - System.Numerics.VectorMath:ConditionalSelectBitwise(System.Runtime.Intrinsics.Vector128`1[Double],System.Runtime.Intrinsics.Vector128`1[Double],System.Runtime.Intrinsics.Vector128`1[Double]):System.Runtime.Intrinsics.Vector128`1[Double] (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(ubyte):System.Runtime.Intrinsics.Vector128`1[Byte] (0 base, 1 diff methods)
          24 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(double):System.Runtime.Intrinsics.Vector128`1[Double] (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(short):System.Runtime.Intrinsics.Vector128`1[Int16] (0 base, 1 diff methods)
          24 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(int):System.Runtime.Intrinsics.Vector128`1[Int32] (0 base, 1 diff methods)
          24 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(long):System.Runtime.Intrinsics.Vector128`1[Int64] (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(byte):System.Runtime.Intrinsics.Vector128`1[SByte] (0 base, 1 diff methods)
          24 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(float):System.Runtime.Intrinsics.Vector128`1[Single] (0 base, 1 diff methods)
          28 (     ∞ of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(ushort):System.Runtime.Intrinsics.Vector128`1[UInt16] (0 base, 1 diff methods)

Top method improvements (percentages):
         -44 (-64.706% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector128SpanLength(long,long):long
         -44 (-64.706% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector256SpanLength(long,long):long
         -40 (-58.824% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector128SpanLength(long,int):long
         -40 (-58.824% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector256SpanLength(long,int):long
        -104 (-57.778% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int64][System.Int64]:Equals(System.Runtime.Intrinsics.Vector64`1[Int64]):bool:this
        -104 (-57.778% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[UInt64][System.UInt64]:Equals(System.Runtime.Intrinsics.Vector64`1[UInt64]):bool:this
        -104 (-54.167% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:UnalignedCountVector128(byref):long (2 methods)
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Byte][System.Byte]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[Byte]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Double][System.Double]:get_AllBitsSet():System.Runtime.Intrinsics.Vector64`1[Double]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int16][System.Int16]:get_AllBitsSet():System.Runtime.Intrinsics.Vector64`1[Int16]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int64][System.Int64]:get_AllBitsSet():System.Runtime.Intrinsics.Vector64`1[Int64]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector64`1[Int32][System.Int32]:get_AllBitsSet():System.Runtime.Intrinsics.Vector64`1[Int32]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt16][System.UInt16]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[UInt16]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[SByte][System.SByte]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[SByte]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[Single]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[UInt64]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[UInt32]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[Double]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int16][System.Int16]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[Int16]
         -28 (-53.846% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:get_AllBitsSet():System.Runtime.Intrinsics.Vector128`1[Int64]

1825 total methods with Code Size differences (314 improved, 1511 regressed), 26424 unchanged.

src/coreclr/src/pal/src/misc/jitsupport.cpp

src/coreclr/src/zap/zapinfo.cpp

davidwrighton

Please fix my comment about wrapping the usage of the InstructionSet flags inside of a TARGET_ARM64 block.

kunalspathak added 2 commits June 17, 2020 16:47

enable methods containing arm64 intrinsic to crossgen

f9559cf

Unconditionally set ArmBase/AdvSimd

eea1983

Dotnet-GitSync-Bot added the area-crossgen2-coreclr label Jun 17, 2020

added a comment

0001fcf

kunalspathak changed the title ~~Arm64 crossgen~~ Crossgen method bodies having ARM64 intrinsic Jun 18, 2020

kunalspathak added 2 commits June 17, 2020 17:33

added the missing #endif

5675f00

Move ArmBase/AdvSimd setting inside host_arm64

26be29e

tannergooding reviewed Jun 18, 2020

View reviewed changes

src/coreclr/src/pal/src/misc/jitsupport.cpp Show resolved Hide resolved

tannergooding reviewed Jun 18, 2020

View reviewed changes

src/coreclr/src/zap/zapinfo.cpp Show resolved Hide resolved

kunalspathak mentioned this pull request Jun 18, 2020

Optimize WithUpper/WithLower with InsertSelectedScalar, SpanHelpers.Sequence APIs #38075

Merged

kunalspathak marked this pull request as ready for review June 19, 2020 00:16

davidwrighton requested changes Jun 19, 2020

View reviewed changes

review comments

c687034

davidwrighton approved these changes Jun 19, 2020

View reviewed changes

kunalspathak merged commit 8439668 into dotnet:master Jun 20, 2020

kunalspathak mentioned this pull request Jun 26, 2020

Move AdvSimd check from PAL to VM #38425

Merged

This was referenced Jul 21, 2020

Test failure: JIT/HardwareIntrinsics/Arm/ArmBase.Arm64/ArmBase.Arm64_ro/ArmBase.Arm64_ro.sh #39618

Closed

[Arm64] Don't generate hardware intrinsic method bodies #39753

Merged

tannergooding mentioned this pull request Aug 6, 2020

Vector128<byte>:get_Zero() doesn't inline (or intrinsicify) at crossgen #32714

Closed

echesakov mentioned this pull request Aug 6, 2020

[Arm64] Treat methods of non-generic Vector64 and Vector128 classes as intrinsics #40441

Merged

AntonLapounov mentioned this pull request Aug 10, 2020

Fix runtime crash on Linux ARM64 running in WSL #40597

Merged

ghost locked as resolved and limited conversation to collaborators Dec 8, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Crossgen method bodies having ARM64 intrinsic #38060

Crossgen method bodies having ARM64 intrinsic #38060

kunalspathak commented Jun 17, 2020 •

edited

Loading

davidwrighton left a comment

Crossgen method bodies having ARM64 intrinsic #38060

Crossgen method bodies having ARM64 intrinsic #38060

Conversation

kunalspathak commented Jun 17, 2020 • edited Loading

davidwrighton left a comment

Choose a reason for hiding this comment

kunalspathak commented Jun 17, 2020 •

edited

Loading