Runtime: Vector128<byte>:get_Zero() doesn't inline (or intrinsicify) at crossgen
Nor does Vector128<byte>.Count() or Vector128:AsByte(Vector128`1):Vector128`1
ASCIIUtility.WidenAsciiToUtf16_Sse2 calls Vector128<byte>.Zero
Which ends up reserving stack, making call and reading the stack back to zero a xmm register:
G_M55642_IG05:
lea rcx, [rsp+20H]
call [Vector128`1:get_Zero():Vector128`1]
movaps xmm0, xmmword ptr [rsp+20H]
movaps xmm1, xmm6
punpcklbw xmm1, xmm0
movdqu xmmword ptr [rsi], xmm1
mov rax, rsi
shr rax, 1
and rax, 7
mov edx, 8
sub rdx, rax
mov rax, rdx
sub rbx, 16
Which is quite inefficient
/cc @tannergooding @GrabYourPitchforks
area-CodeGen-coreclr
benaadams
All 32 comments
Switching it for default produces
G_M55642_IG05:
xorps xmm1, xmm1
movaps xmm2, xmm0
punpcklbw xmm2, xmm1
movdqu xmmword ptr [rdx], xmm2
mov r9, rdx
shr r9, 1
and r9, 7
mov eax, 8
sub rax, r9
sub r8, 16
Seems to be getting optimized correctly under Tier 1, perhaps a crossgen-specific issue?
GrabYourPitchforks
on 23 Feb 2020
Seems to be getting optimized correctly under Tier 1, perhaps a crossgen-specific issue?
Yes
benaadams
on 14 Mar 2020
Nor does Vector128<byte>.Count() or Vector128:AsByte(Vector128`1):Vector128`1
G_M37173_IG09:
call [Vector128`1:get_Count():int]
movsxd rax, eax
cmp rax, rbx
jg G_M37173_IG12
xor rbp, rbp
call [Vector128`1:get_Count():int]
movsxd rcx, eax
sub rbx, rcx
test rbx, rbx
je SHORT G_M37173_IG11
;; bbWeight=0.50 PerfScore 4.75
G_M37173_IG10:
movups xmm6, xmmword ptr [rsi+rbp]
movups xmm7, xmmword ptr [rdi+rbp]
lea rcx, [rsp+60H]
movaps xmmword ptr [rsp+20H], xmm6
lea rdx, bword ptr [rsp+20H]
call [Vector128:AsByte(Vector128`1):Vector128`1]
lea rcx, [rsp+50H]
movaps xmmword ptr [rsp+20H], xmm7
lea rdx, bword ptr [rsp+20H]
call [Vector128:AsByte(Vector128`1):Vector128`1]
movaps xmm0, xmmword ptr [rsp+60H]
You can mark these as [NonVersionable]. It will allow inlining in version-resilient R2R, but we have to guarantee that the methods are set in stone - that is safe guarantee to make for these specific ones.
jkotas
on 14 Mar 2020
Ignore my comment. [NonVersionable] is not going to help for these.
jkotas
on 14 Mar 2020
Its in System.Private.CoreLib so that shouldn't be the issue?
benaadams
on 14 Mar 2020
@tannergooding @CarolEidt What is the expected behavior here for intrinsics and crossgen?
BruceForstall
on 4 Apr 2020
This is currently "by design" as per https://github.com/dotnet/runtime/blob/master/src/coreclr/src/zap/zapinfo.cpp#L2143-L2153. which was originally added in https://github.com/dotnet/coreclr/pull/25365.
Basically, we don't currently treat any of the HWIntrinsic "helper methods" as crossgenable with the reasoning that the code often changes based on what ISA is supported.
That being said, all code paths are equivalent, so we should be able to target the "current baseline" and rely on rejit to produce the better code, like we are doing elsewhere.
@davidwrighton has also been working in the area so he may be able to comment if there would be any problem with doing so.
tannergooding
on 4 Apr 2020
Switching
Vector128<byte> zeroVector = Vector128<byte>.Zero;
to
Vector128<byte> zeroVector = default;
Crossgens it to xorps xmm1, xmm1 so should the C# change?
Also Vector128<byte>.Count() is always 16 regardless, should the C# change there (e.g. using in a loop conditional)
benaadams
on 4 Apr 2020
Also
Vector128:AsByte(Vector128`1):Vector128`1
could maybe switch to
Unsafe.As<Vector<ushort>, Vector<byte>>(ref vector)
to be more crossgen friendly?
benaadams
on 4 Apr 2020
crossgen1 is very conservative with handling anything related to intrinsics or Vector types. Tweaking the C# code won't help. crossgen2 is going to make it much better.
jkotas
on 5 Apr 2020
Yes, crossgen2 should handle this case much better. It doesn't quite make intrinsics useable without paying attention to the rules easy, but it broadens the set of workable cases. I hope to enhance the support to allow Avx2 and above in R2R'd code (with an opt-in) in .NET 5.0. There may be significant restrictions on the scenarios under which the Avx2 and above intrinsics are available, as the ABI can shift based on which instruction sets are available.
davidwrighton
on 7 Apr 2020
Tweaking the C# code won't help.
It does though which is what was bothering me; and they are Sse instructions not Avx.
Vector128<byte> zeroVector = Vector128<byte>.Zero;
utf16FirstHalfVector = Sse2.UnpackLow(asciiVector, zeroVector);
Generates at crossgen
G_M55642_IG05:
lea rcx, [rsp+20H]
call [Vector128`1:get_Zero():Vector128`1]
movaps xmm0, xmmword ptr [rsp+20H]
movaps xmm1, xmm6
punpcklbw xmm1, xmm0
movdqu xmmword ptr [rsi], xmm1
mov rax, rsi
shr rax, 1
and rax, 7
mov edx, 8
sub rdx, rax
mov rax, rdx
sub rbx, 16
Changing Zero to default instead
Vector128<byte> zeroVector = default;
utf16FirstHalfVector = Sse2.UnpackLow(asciiVector, zeroVector);
Generates the much better
G_M55642_IG05:
xorps xmm1, xmm1
movaps xmm2, xmm0
punpcklbw xmm2, xmm1
movdqu xmmword ptr [rdx], xmm2
mov r9, rdx
shr r9, 1
and r9, 7
mov eax, 8
sub rax, r9
sub r8, 16
For ARM64. we generate unnecessary memory load/store for Vector128<byte>.Count.
Vector128
550b0: a9be7bfd stp x29, x30, [sp, #-32]!
550b4: 910003fd mov x29, sp
550b8: 52800200 movz w0, #0x10
550bc: b9001fa0 str w0, [x29, #28]
550c0: b9401fa0 ldr w0, [x29, #28]
550c4: a8c27bfd ldp x29, x30, [sp], #32
550c8: d65f03c0 ret
Vector128.Size
550b0: a9bf7bfd stp x29, x30, [sp, #-16]!
550b4: 910003fd mov x29, sp
550b8: 52800200 movz w0, #0x10
550bc: a8c17bfd ldp x29, x30, [sp], #16
550c0: d65f03c0 ret
kunalspathak
on 12 Jun 2020
Right now all Vector128 and Vector256 are treated as regular calls on x86/x64 per this condition
@davidwrighton @jkotas
Do you think that we can relax this condition and allow some of Vector128 methods to be treated as intrinsics when compiling System.Private.CoreLib?
As far as I understand, Sse and Sse2 are required ISAs on x86/x64 platforms, so it appears to be safe to do this, at least, for the following ones:
Vector128<T>.As*Vector128<T>.get_CountVector128<T>.get_Zero
echesakovMSFT
on 6 Aug 2020
👍1
Also Vector128<T>.Equals as that's Sse || Sse2?
benaadams
on 6 Aug 2020
I don't know enough about the rules of our code generator to say whether or not we can relax those restrictions. My sense is that the particular calls called out here would be acceptable to allow, but I'm no expert. (The major concern is that if the calls may cause usage of SSE3, or SSE4.2 instructions that code generation will be incorrect in System.Private.CoreLib as we compile CoreLib with the flags set that if any code in the SIMD intrinsics would use higher level SSE that they in fact WILL use them.) @tannergooding could you comment here?
davidwrighton
on 6 Aug 2020
If I understand this right, emission instructions belonging to an ISA outside of baseline (i.e. Sse Sse2) in codegen should be guarded with compOpportunisticallyDependsOn
For example,
https://github.com/dotnet/runtime/blob/2f7215e37a1060249ac12531ec2cbaf4db38fa17/src/coreclr/src/jit/hwintrinsiccodegenxarch.cpp#L1254-L1270
If crossgen returns false for such call compOpportunisticallyDependsOn(InstructionSet_AVX) JIT shouldn't be using these.
echesakovMSFT
on 6 Aug 2020
Zero, Size and As don't particularly use any processor intrinsics as they are (with some throws for incorrect T):
Vector128<T>.get_Zero => default;
Vector128<T>.get_Count => Vector128.Size / Unsafe.SizeOf<T>();
Vector128.AsByte<T>(this Vector128<T> vector) => vector.As<T, byte>();
Vector128.As<T, U>(this Vector128<T> vector) => Unsafe.As<Vector128<T>, Vector128<U>>(ref vector);
Vector128<T>.Equals does use processor intrinsics, but that's Sse || Sse2; which may be allowable?
benaadams
on 6 Aug 2020
@benaadams Correct, the ones you mentioned don't use any processor instructions when generated.
In fact, when treated as intrinsics get_Count and As* shouldn't produce any code - JIT will fold get_Count into a const node and when importing AsByte, AsDouble and alike JIT treats them as "casts" by changing the effectife type of a SIMD value.
get_Zero does produce code though
https://github.com/dotnet/runtime/blob/a69d919a62ee358b3d6786170577c92fc56ef988/src/coreclr/src/jit/hwintrinsiccodegenxarch.cpp#L1246-L1252
And from the codegen code it's not immediately clear what version of xorps will be used when generating Vector128<T>.Zero - it is VXORPS xmm1,xmm1, xmm1 (AVX) or JIT will fold this to XORPS xmm1, xmm1 (SSE)?
I totally see what David was concerned about here and I agree that we need to be very careful when allowing some of these to be pre-jitted.
echesakovMSFT
on 6 Aug 2020
And from the codegen code it's not immediately clear what version ...
Is that really relevant as that can be controlled from C#. If it is then there is a bigger issue at play?
For example as in the original summary ASCIIUtility.cs is using
Vector128<byte> zeroVector = Vector128<byte>.Zero;
And that generates at R2R
lea rcx, [rsp+20H]
call [Vector128`1:get_Zero():Vector128`1]
movaps xmm0, xmmword ptr [rsp+20H]
If the C# is instead changed to:
Vector128<byte> zeroVector = default;
At R2R it generates:
xorps xmm1, xmm1
So R2R is already handing xorps just not if you use Vector128<byte>.Zero
benaadams
on 6 Aug 2020
There was a specific issue in Crossgen 1 that prevented us from doing this. I'd have to find the last conversation around this to remember the exact reasoning.
I believe we determined it should be fine for Crossgen 2, however.
tannergooding
on 6 Aug 2020
Would aggressive inlining achieve the same effect?
benaadams
on 6 Aug 2020
@davidwrighton explained the reasoning here: https://github.com/dotnet/runtime/pull/38060#discussion_r443062650
We compile System.Private.CoreLib with the instruction set such that it thinks its compiling with support for SSE4.1 and such. However, we disabled the non-platform intrinsics as usages of those are not wrapped in IsSupported checks.
tannergooding
on 6 Aug 2020
The opportunistic checks are only such on Crossgen2 and represent an issue for Crossgen1.
It sounds like allowing certain methods which only use SSE/SSE2 (and therefore have no opportunistic checks, etc) might be fine, but the rules still aren't super clear to me as to whether that might cause some issue on Crossgen1.
tannergooding
on 6 Aug 2020
@tannergooding Thanks for finding this! If Crossgen1 makes JIT to think that its target platform supports SSE4.1 than definitely only no-codegen intrinsics can be allowed to be pre-jitted.
echesakovMSFT
on 6 Aug 2020
Would aggressive inlining achieve the same effect?
@benaadams I haven't though about this - but let me try this and some other ideas and I will be back with the results.
echesakovMSFT
on 6 Aug 2020
Here is what I found so far.
Allowing Vector128.As, Vector128.AsByte - Vector128.AsUInt64 to be unconditionally expanded when compiling S.P.C.dll leads to these results
Found 1 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -7601 (-0.199% of base)
diff is an improvement.
Top file improvements (bytes):
-7601 : System.Private.CoreLib.dasm (-0.199% of base)
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
Top method improvements (bytes):
-429 (-34.513% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
-429 (-34.347% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:NarrowUtf16ToLatin1_Sse2(long,long,long):long
-372 (-10.632% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf8(long,int,long,int,byref,byref):int
-283 (-7.453% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf16(long,int,long,int,byref,byref):int
-250 (-18.671% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:GetIndexOfFirstNonLatin1Char_Sse2(long,long):long
-244 (-18.141% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:GetIndexOfFirstNonAsciiChar_Sse2(long,long):long
-173 (-14.686% of base) : System.Private.CoreLib.dasm - System.Number:FormatDouble(byref,double,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-173 (-14.965% of base) : System.Private.CoreLib.dasm - System.Number:FormatSingle(byref,float,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-140 (-12.302% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunDouble(double,int,byref):bool
-140 (-12.891% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunSingle(float,int,byref):bool
-126 (-13.801% of base) : System.Private.CoreLib.dasm - System.Number:NumberToFloatingPointBits(byref,byref):long
-109 (-26.456% of base) : System.Private.CoreLib.dasm - System.Half:op_Explicit(System.Half):float
-108 (-2.032% of base) : System.Private.CoreLib.dasm - System.Diagnostics.Tracing.EventPipePayloadDecoder:DecodePayload(byref,System.ReadOnlySpan`1[Byte]):System.Object[]
-103 (-20.276% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:WidenLatin1ToUtf16_Sse2(long,long,long)
-102 (-43.966% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-102 (-47.442% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-94 (-7.198% of base) : System.Private.CoreLib.dasm - System.Variant:MarshalHelperConvertObjectToVariant(System.Object,byref)
-91 (-3.669% of base) : System.Private.CoreLib.dasm - System.Variant:MarshalHelperCastVariant(System.Object,int,byref)
-85 (-15.071% of base) : System.Private.CoreLib.dasm - System.Variant:ToObject():System.Object:this
-79 (-9.371% of base) : System.Private.CoreLib.dasm - System.Threading.ProcessorIdCache:ProcessorNumberSpeedCheck():bool
Top method improvements (percentages):
-58 (-71.605% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:Equals(System.Runtime.Intrinsics.Vector128`1[Single]):bool:this
-58 (-69.880% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:Equals(System.Runtime.Intrinsics.Vector128`1[Double]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:Equals(System.Runtime.Intrinsics.Vector128`1[Int64]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:Equals(System.Runtime.Intrinsics.Vector128`1[Int32]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int16][System.Int16]:Equals(System.Runtime.Intrinsics.Vector128`1[Int16]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Byte][System.Byte]:Equals(System.Runtime.Intrinsics.Vector128`1[Byte]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt16][System.UInt16]:Equals(System.Runtime.Intrinsics.Vector128`1[UInt16]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:Equals(System.Runtime.Intrinsics.Vector128`1[UInt64]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:Equals(System.Runtime.Intrinsics.Vector128`1[UInt32]):bool:this
-58 (-69.048% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[SByte][System.SByte]:Equals(System.Runtime.Intrinsics.Vector128`1[SByte]):bool:this
-29 (-67.442% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(ubyte):System.Runtime.Intrinsics.Vector128`1[Byte]
-29 (-67.442% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(ushort):System.Runtime.Intrinsics.Vector128`1[UInt16]
-29 (-61.702% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(short):System.Runtime.Intrinsics.Vector128`1[Int16]
-29 (-61.702% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(byte):System.Runtime.Intrinsics.Vector128`1[SByte]
-102 (-47.442% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-26 (-44.828% of base) : System.Private.CoreLib.dasm - System.BitConverter:SingleToInt32Bits(float):int
-26 (-44.068% of base) : System.Private.CoreLib.dasm - System.BitConverter:DoubleToInt64Bits(double):long
-102 (-43.966% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-67 (-41.875% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:Equals(System.Object):bool:this
-67 (-41.358% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:Equals(System.Object):bool:this
121 total methods with Code Size differences (121 improved, 0 regressed), 28561 unchanged.
If, in addition to As* methods, also treat the following as intrinsics - Vector128.Create, Vector128.CreateScalarUnsafe, Vector128.ToScalar:
Found 1 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -12573 (-0.329% of base)
diff is an improvement.
Top file improvements (bytes):
-12573 : System.Private.CoreLib.dasm (-0.329% of base)
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
Top method improvements (bytes):
-410 (-32.985% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
-410 (-32.826% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:NarrowUtf16ToLatin1_Sse2(long,long,long):long
-376 (-10.746% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf8(long,int,long,int,byref,byref):int
-352 (-29.881% of base) : System.Private.CoreLib.dasm - System.Number:FormatDouble(byref,double,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-349 (-30.190% of base) : System.Private.CoreLib.dasm - System.Number:FormatSingle(byref,float,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-344 (-6.472% of base) : System.Private.CoreLib.dasm - System.Diagnostics.Tracing.EventPipePayloadDecoder:DecodePayload(byref,System.ReadOnlySpan`1[Byte]):System.Object[]
-329 (-28.910% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunDouble(double,int,byref):bool
-295 (-27.164% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunSingle(float,int,byref):bool
-283 (-7.453% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf16(long,int,long,int,byref,byref):int
-279 (-46.656% of base) : System.Private.CoreLib.dasm - System.Math:Round(double):double
-262 (-51.779% of base) : System.Private.CoreLib.dasm - System.MathF:Round(float):float
-259 (-19.343% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:GetIndexOfFirstNonLatin1Char_Sse2(long,long):long
-253 (-18.810% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:GetIndexOfFirstNonAsciiChar_Sse2(long,long):long
-205 (-49.757% of base) : System.Private.CoreLib.dasm - System.Half:op_Explicit(System.Half):float
-189 (-87.907% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-187 (-80.603% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-177 (-19.387% of base) : System.Private.CoreLib.dasm - System.Number:NumberToFloatingPointBits(byref,byref):long
-152 (-65.801% of base) : System.Private.CoreLib.dasm - System.MathF:BitIncrement(float):float
-151 (-66.520% of base) : System.Private.CoreLib.dasm - System.MathF:BitDecrement(float):float
-149 (-60.081% of base) : System.Private.CoreLib.dasm - System.Math:BitDecrement(double):double
Top method improvements (percentages):
-65 (-92.857% of base) : System.Private.CoreLib.dasm - System.BitConverter:Int32BitsToSingle(int):float
-65 (-91.549% of base) : System.Private.CoreLib.dasm - System.BitConverter:Int64BitsToDouble(long):double
-53 (-91.379% of base) : System.Private.CoreLib.dasm - System.BitConverter:SingleToInt32Bits(float):int
-53 (-89.831% of base) : System.Private.CoreLib.dasm - System.BitConverter:DoubleToInt64Bits(double):long
-189 (-87.907% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-94 (-81.739% of base) : System.Private.CoreLib.dasm - System.MathF:CopySign(float,float):float
-187 (-80.603% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-53 (-80.303% of base) : System.Private.CoreLib.dasm - System.Single:IsNegative(float):bool
-53 (-77.941% of base) : System.Private.CoreLib.dasm - System.Double:IsNegative(double):bool
-115 (-75.163% of base) : System.Private.CoreLib.dasm - System.Math:Min(float,float):float
-57 (-75.000% of base) : System.Private.CoreLib.dasm - System.BitConverter:ToSingle(System.Byte[],int):float
-57 (-75.000% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:ReadSingle(long):float:this
-61 (-74.390% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:Write(long,float):this
-61 (-74.390% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:Write(long,double):this
-57 (-74.026% of base) : System.Private.CoreLib.dasm - System.BitConverter:ToDouble(System.Byte[],int):double
-57 (-74.026% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:ReadDouble(long):double:this
-115 (-72.785% of base) : System.Private.CoreLib.dasm - System.Math:Min(double,double):double
-53 (-71.622% of base) : System.Private.CoreLib.dasm - System.Single:IsFinite(float):bool
-53 (-71.622% of base) : System.Private.CoreLib.dasm - System.Single:IsInfinity(float):bool
-58 (-71.605% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:Equals(System.Runtime.Intrinsics.Vector128`1[Single]):bool:this
128 total methods with Code Size differences (128 improved, 0 regressed), 28554 unchanged.
Do this for Vector128<T>.Count and Vector256<T>.Count only
Found 1 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -658 (-0.017% of base)
diff is an improvement.
Top file improvements (bytes):
-658 : System.Private.CoreLib.dasm (-0.017% of base)
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
Top method regressions (bytes):
178 (136.923% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:GetHashCode():int:this
76 (46.914% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:GetHashCode():int:this
Top method improvements (bytes):
-57 (-61.957% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:UnalignedCountVector128(byref):long (2 methods)
-32 (-4.097% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:ToString():System.String:this
-31 (-3.861% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[SByte][System.SByte]:ToString():System.String:this
-31 (-3.914% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int32][System.Int32]:ToString():System.String:this
-31 (-3.890% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int64][System.Int64]:ToString():System.String:this
-31 (-3.861% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int16][System.Int16]:ToString():System.String:this
-31 (-3.995% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:ToString():System.String:this
-31 (-3.939% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int16][System.Int16]:ToString():System.String:this
-31 (-3.939% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[SByte][System.SByte]:ToString():System.String:this
-28 (-11.475% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[__Canon][System.__Canon]:<Equals>g__SoftwareFallback|12_0(byref,System.Runtime.Intrinsics.Vector128`1[__Canon]):bool
-27 (-71.053% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector128SpanLength(long,long):long
-27 (-71.053% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector256SpanLength(long,long):long
-25 (-9.542% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[__Canon][System.__Canon]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[__Canon]):bool
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector128SpanLength(long,int):long
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector256SpanLength(long,int):long
-24 (-3.230% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[__Canon][System.__Canon]:ToString():System.String:this
-23 (-3.030% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[__Canon][System.__Canon]:ToString():System.String:this
-22 (-12.500% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[__Canon][System.__Canon]:GetHashCode():int:this
-22 (-11.892% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[__Canon][System.__Canon]:GetHashCode():int:this
-22 (-3.121% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:ToString():System.String:this
Top method regressions (percentages):
178 (136.923% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:GetHashCode():int:this
76 (46.914% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:GetHashCode():int:this
Top method improvements (percentages):
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector128SpanLength(long,int):long
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector256SpanLength(long,int):long
-27 (-71.053% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector128SpanLength(long,long):long
-27 (-71.053% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetCharVector256SpanLength(long,long):long
-57 (-61.957% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:UnalignedCountVector128(byref):long (2 methods)
-15 (-15.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Byte][System.Byte]:GetHashCode():int:this
-15 (-15.000% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt16][System.UInt16]:GetHashCode():int:this
-15 (-14.851% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int16][System.Int16]:GetHashCode():int:this
-15 (-14.851% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[SByte][System.SByte]:GetHashCode():int:this
-15 (-13.889% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[UInt32][System.UInt32]:GetHashCode():int:this
-15 (-13.889% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int32][System.Int32]:GetHashCode():int:this
-15 (-13.761% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[UInt16][System.UInt16]:GetHashCode():int:this
-15 (-13.761% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Byte][System.Byte]:GetHashCode():int:this
-15 (-13.636% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[SByte][System.SByte]:GetHashCode():int:this
-15 (-13.636% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int16][System.Int16]:GetHashCode():int:this
-15 (-13.043% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[UInt64][System.UInt64]:GetHashCode():int:this
-15 (-13.043% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[Int64][System.Int64]:GetHashCode():int:this
-22 (-12.500% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[__Canon][System.__Canon]:GetHashCode():int:this
-22 (-11.892% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[__Canon][System.__Canon]:GetHashCode():int:this
-15 (-11.628% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector256`1[UInt32][System.UInt32]:<Equals>g__SoftwareFallback|14_0(byref,System.Runtime.Intrinsics.Vector256`1[UInt32]):bool
61 total methods with Code Size differences (55 improved, 6 regressed), 28621 unchanged.
Note that the code size increase in GetHashCode() methods is due to loop unrolling kicking in when compiling the methods since the loop boundary becomes constant.
Only Vector128<T>.Zero
Found 1 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -801 (-0.021% of base)
diff is an improvement.
Top file improvements (bytes):
-801 : System.Private.CoreLib.dasm (-0.021% of base)
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
Top method improvements (bytes):
-76 (-2.002% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf16(long,int,long,int,byref,byref):int
-37 (-27.007% of base) : System.Private.CoreLib.dasm - System.Numerics.Matrix4x4:op_UnaryNegation(System.Numerics.Matrix4x4):System.Numerics.Matrix4x4
-25 (-36.765% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(double):System.Runtime.Intrinsics.Vector128`1[Double]
-25 (-36.765% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(float):System.Runtime.Intrinsics.Vector128`1[Single]
-23 (-4.528% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:WidenLatin1ToUtf16_Sse2(long,long,long)
-23 (-42.593% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<CreateScalar>g__SoftwareFallback|51_0(double):System.Runtime.Intrinsics.Vector128`1[Double]
-23 (-42.593% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<CreateScalar>g__SoftwareFallback|56_0(float):System.Runtime.Intrinsics.Vector128`1[Single]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Byte],System.Runtime.Intrinsics.Vector64`1[Byte]):System.Runtime.Intrinsics.Vector128`1[Byte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Double],System.Runtime.Intrinsics.Vector64`1[Double]):System.Runtime.Intrinsics.Vector128`1[Double]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int16],System.Runtime.Intrinsics.Vector64`1[Int16]):System.Runtime.Intrinsics.Vector128`1[Int16]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int32],System.Runtime.Intrinsics.Vector64`1[Int32]):System.Runtime.Intrinsics.Vector128`1[Int32]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int64],System.Runtime.Intrinsics.Vector64`1[Int64]):System.Runtime.Intrinsics.Vector128`1[Int64]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[SByte],System.Runtime.Intrinsics.Vector64`1[SByte]):System.Runtime.Intrinsics.Vector128`1[SByte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Single],System.Runtime.Intrinsics.Vector64`1[Single]):System.Runtime.Intrinsics.Vector128`1[Single]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt16],System.Runtime.Intrinsics.Vector64`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[UInt16]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt32],System.Runtime.Intrinsics.Vector64`1[UInt32]):System.Runtime.Intrinsics.Vector128`1[UInt32]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt64],System.Runtime.Intrinsics.Vector64`1[UInt64]):System.Runtime.Intrinsics.Vector128`1[UInt64]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|40_0(System.Runtime.Intrinsics.Vector64`1[Byte],System.Runtime.Intrinsics.Vector64`1[Byte]):System.Runtime.Intrinsics.Vector128`1[Byte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|41_0(System.Runtime.Intrinsics.Vector64`1[Double],System.Runtime.Intrinsics.Vector64`1[Double]):System.Runtime.Intrinsics.Vector128`1[Double]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|42_0(System.Runtime.Intrinsics.Vector64`1[Int16],System.Runtime.Intrinsics.Vector64`1[Int16]):System.Runtime.Intrinsics.Vector128`1[Int16]
Top method improvements (percentages):
-23 (-42.593% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<CreateScalar>g__SoftwareFallback|51_0(double):System.Runtime.Intrinsics.Vector128`1[Double]
-23 (-42.593% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<CreateScalar>g__SoftwareFallback|56_0(float):System.Runtime.Intrinsics.Vector128`1[Single]
-25 (-36.765% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(double):System.Runtime.Intrinsics.Vector128`1[Double]
-25 (-36.765% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(float):System.Runtime.Intrinsics.Vector128`1[Single]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Byte],System.Runtime.Intrinsics.Vector64`1[Byte]):System.Runtime.Intrinsics.Vector128`1[Byte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Double],System.Runtime.Intrinsics.Vector64`1[Double]):System.Runtime.Intrinsics.Vector128`1[Double]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int16],System.Runtime.Intrinsics.Vector64`1[Int16]):System.Runtime.Intrinsics.Vector128`1[Int16]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int32],System.Runtime.Intrinsics.Vector64`1[Int32]):System.Runtime.Intrinsics.Vector128`1[Int32]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Int64],System.Runtime.Intrinsics.Vector64`1[Int64]):System.Runtime.Intrinsics.Vector128`1[Int64]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[SByte],System.Runtime.Intrinsics.Vector64`1[SByte]):System.Runtime.Intrinsics.Vector128`1[SByte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[Single],System.Runtime.Intrinsics.Vector64`1[Single]):System.Runtime.Intrinsics.Vector128`1[Single]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt16],System.Runtime.Intrinsics.Vector64`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[UInt16]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt32],System.Runtime.Intrinsics.Vector64`1[UInt32]):System.Runtime.Intrinsics.Vector128`1[UInt32]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:Create(System.Runtime.Intrinsics.Vector64`1[UInt64],System.Runtime.Intrinsics.Vector64`1[UInt64]):System.Runtime.Intrinsics.Vector128`1[UInt64]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|40_0(System.Runtime.Intrinsics.Vector64`1[Byte],System.Runtime.Intrinsics.Vector64`1[Byte]):System.Runtime.Intrinsics.Vector128`1[Byte]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|41_0(System.Runtime.Intrinsics.Vector64`1[Double],System.Runtime.Intrinsics.Vector64`1[Double]):System.Runtime.Intrinsics.Vector128`1[Double]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|42_0(System.Runtime.Intrinsics.Vector64`1[Int16],System.Runtime.Intrinsics.Vector64`1[Int16]):System.Runtime.Intrinsics.Vector128`1[Int16]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|43_0(System.Runtime.Intrinsics.Vector64`1[Int32],System.Runtime.Intrinsics.Vector64`1[Int32]):System.Runtime.Intrinsics.Vector128`1[Int32]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|44_0(System.Runtime.Intrinsics.Vector64`1[Int64],System.Runtime.Intrinsics.Vector64`1[Int64]):System.Runtime.Intrinsics.Vector128`1[Int64]
-21 (-36.207% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:<Create>g__SoftwareFallback|45_0(System.Runtime.Intrinsics.Vector64`1[SByte],System.Runtime.Intrinsics.Vector64`1[SByte]):System.Runtime.Intrinsics.Vector128`1[SByte]
37 total methods with Code Size differences (37 improved, 0 regressed), 28645 unchanged.
All of these together
Found 1 files with textual diffs.
Summary of Code Size diffs:
(Lower is better)
Total bytes of diff: -14074 (-0.368% of base)
diff is an improvement.
Top file improvements (bytes):
-14074 : System.Private.CoreLib.dasm (-0.368% of base)
1 total files with Code Size differences (1 improved, 0 regressed), 0 unchanged.
Top method regressions (bytes):
178 (136.923% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:GetHashCode():int:this
76 (46.914% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:GetHashCode():int:this
Top method improvements (bytes):
-410 (-32.985% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
-410 (-32.826% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:NarrowUtf16ToLatin1_Sse2(long,long,long):long
-376 (-10.746% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf8(long,int,long,int,byref,byref):int
-365 (-9.613% of base) : System.Private.CoreLib.dasm - System.Text.Unicode.Utf8Utility:TranscodeToUtf16(long,int,long,int,byref,byref):int
-352 (-29.881% of base) : System.Private.CoreLib.dasm - System.Number:FormatDouble(byref,double,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-349 (-30.190% of base) : System.Private.CoreLib.dasm - System.Number:FormatSingle(byref,float,System.ReadOnlySpan`1[Char],System.Globalization.NumberFormatInfo):System.String
-344 (-6.472% of base) : System.Private.CoreLib.dasm - System.Diagnostics.Tracing.EventPipePayloadDecoder:DecodePayload(byref,System.ReadOnlySpan`1[Byte]):System.Object[]
-329 (-28.910% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunDouble(double,int,byref):bool
-295 (-27.164% of base) : System.Private.CoreLib.dasm - Grisu3:TryRunSingle(float,int,byref):bool
-279 (-46.656% of base) : System.Private.CoreLib.dasm - System.Math:Round(double):double
-262 (-51.779% of base) : System.Private.CoreLib.dasm - System.MathF:Round(float):float
-259 (-19.343% of base) : System.Private.CoreLib.dasm - System.Text.Latin1Utility:GetIndexOfFirstNonLatin1Char_Sse2(long,long):long
-253 (-18.810% of base) : System.Private.CoreLib.dasm - System.Text.ASCIIUtility:GetIndexOfFirstNonAsciiChar_Sse2(long,long):long
-205 (-49.757% of base) : System.Private.CoreLib.dasm - System.Half:op_Explicit(System.Half):float
-189 (-87.907% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-187 (-80.603% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-177 (-19.387% of base) : System.Private.CoreLib.dasm - System.Number:NumberToFloatingPointBits(byref,byref):long
-152 (-65.801% of base) : System.Private.CoreLib.dasm - System.MathF:BitIncrement(float):float
-151 (-66.520% of base) : System.Private.CoreLib.dasm - System.MathF:BitDecrement(float):float
-149 (-60.081% of base) : System.Private.CoreLib.dasm - System.Math:BitDecrement(double):double
Top method regressions (percentages):
178 (136.923% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Single][System.Single]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int32][System.Int32]:GetHashCode():int:this
85 (85.859% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt32][System.UInt32]:GetHashCode():int:this
76 (46.914% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Double][System.Double]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[Int64][System.Int64]:GetHashCode():int:this
18 (16.981% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128`1[UInt64][System.UInt64]:GetHashCode():int:this
Top method improvements (percentages):
-65 (-92.857% of base) : System.Private.CoreLib.dasm - System.BitConverter:Int32BitsToSingle(int):float
-65 (-91.549% of base) : System.Private.CoreLib.dasm - System.BitConverter:Int64BitsToDouble(long):double
-53 (-91.379% of base) : System.Private.CoreLib.dasm - System.BitConverter:SingleToInt32Bits(float):int
-53 (-89.831% of base) : System.Private.CoreLib.dasm - System.BitConverter:DoubleToInt64Bits(double):long
-189 (-87.907% of base) : System.Private.CoreLib.dasm - System.MathF:<CopySign>g__SoftwareFallback|36_0(float,float):float
-94 (-81.739% of base) : System.Private.CoreLib.dasm - System.MathF:CopySign(float,float):float
-187 (-80.603% of base) : System.Private.CoreLib.dasm - System.Math:<CopySign>g__SoftwareFallback|46_0(double,double):double
-53 (-80.303% of base) : System.Private.CoreLib.dasm - System.Single:IsNegative(float):bool
-54 (-79.412% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(double):System.Runtime.Intrinsics.Vector128`1[Double]
-54 (-79.412% of base) : System.Private.CoreLib.dasm - System.Runtime.Intrinsics.Vector128:CreateScalar(float):System.Runtime.Intrinsics.Vector128`1[Single]
-53 (-77.941% of base) : System.Private.CoreLib.dasm - System.Double:IsNegative(double):bool
-115 (-75.163% of base) : System.Private.CoreLib.dasm - System.Math:Min(float,float):float
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector128SpanLength(long,int):long
-24 (-75.000% of base) : System.Private.CoreLib.dasm - System.SpanHelpers:GetByteVector256SpanLength(long,int):long
-57 (-75.000% of base) : System.Private.CoreLib.dasm - System.BitConverter:ToSingle(System.Byte[],int):float
-57 (-75.000% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:ReadSingle(long):float:this
-61 (-74.390% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:Write(long,float):this
-61 (-74.390% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:Write(long,double):this
-57 (-74.026% of base) : System.Private.CoreLib.dasm - System.BitConverter:ToDouble(System.Byte[],int):double
-57 (-74.026% of base) : System.Private.CoreLib.dasm - System.IO.UnmanagedMemoryAccessor:ReadDouble(long):double:this
221 total methods with Code Size differences (215 improved, 6 regressed), 28461 unchanged.
As an example, the following is what such change would do for pre-jitted code of System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
@@ -8,75 +8,50 @@
;
; V00 arg0 [V00,T02] ( 6, 11.50) long -> rsi
; V01 arg1 [V01,T04] ( 8, 8.50) long -> rdi
-; V02 arg2 [V02,T09] ( 6, 4.50) long -> rbx
-;* V03 loc0 [V03,T35] ( 0, 0 ) int -> zero-ref
-;* V04 loc1 [V04,T36] ( 0, 0 ) long -> zero-ref
-; V05 loc2 [V05 ] ( 5, 4.50) simd16 -> [rsp+0x110] do-not-enreg[XS] addr-exposed
-; V06 loc3 [V06 ] ( 5, 4.50) simd16 -> [rsp+0x100] do-not-enreg[XS] addr-exposed
-; V07 loc4 [V07,T37] ( 17, 19.50) simd16 -> mm6
-; V08 loc5 [V08,T39] ( 8, 11 ) simd16 -> mm7
+; V02 arg2 [V02,T07] ( 6, 4.50) long -> rbx
+;* V03 loc0 [V03,T24] ( 0, 0 ) int -> zero-ref
+;* V04 loc1 [V04,T25] ( 0, 0 ) long -> zero-ref
+; V05 loc2 [V05,T30] ( 5, 4.50) simd16 -> mm6
+; V06 loc3 [V06,T31] ( 5, 4.50) simd16 -> mm7
+; V07 loc4 [V07,T26] ( 17, 19.50) simd16 -> mm8
+; V08 loc5 [V08,T28] ( 8, 11 ) simd16 -> mm9
; V09 loc6 [V09,T00] ( 16, 29 ) long -> r14
-; V10 loc7 [V10,T11] ( 2, 4.50) long -> rbx
-; V11 loc8 [V11,T38] ( 3, 12 ) simd16 -> mm7
-; V12 loc9 [V12,T40] ( 3, 8 ) simd16 -> mm8
+; V10 loc7 [V10,T09] ( 2, 4.50) long -> rbx
+; V11 loc8 [V11,T27] ( 3, 12 ) simd16 -> mm9
+; V12 loc9 [V12,T29] ( 3, 8 ) simd16 -> mm10
; V13 OutArgs [V13 ] ( 1, 1 ) lclBlk (32) [rsp+0x00] "OutgoingArgSpace"
-; V14 tmp1 [V14 ] ( 2, 2 ) simd16 -> [rsp+0xF0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V15 tmp2 [V15 ] ( 2, 2 ) simd16 -> [rsp+0xE0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V16 tmp3 [V16 ] ( 2, 2 ) simd16 -> [rsp+0xD0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V17 tmp4 [V17,T30] ( 3, 1.50) int -> rax
-; V18 tmp5 [V18 ] ( 2, 8 ) simd16 -> [rsp+0xC0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V19 tmp6 [V19 ] ( 2, 8 ) simd16 -> [rsp+0xB0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V20 tmp7 [V20 ] ( 2, 2 ) simd16 -> [rsp+0xA0] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V21 tmp8 [V21 ] ( 2, 2 ) simd16 -> [rsp+0x90] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V22 tmp9 [V22 ] ( 2, 2 ) simd16 -> [rsp+0x80] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V23 tmp10 [V23 ] ( 2, 2 ) simd16 -> [rsp+0x70] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V24 tmp11 [V24 ] ( 2, 2 ) simd16 -> [rsp+0x60] do-not-enreg[XS] addr-exposed "struct address for call/obj"
-; V25 tmp12 [V25 ] ( 2, 2 ) simd16 -> [rsp+0x50] do-not-enreg[XS] addr-exposed "struct address for call/obj"
+; V14 tmp1 [V14,T19] ( 3, 1.50) int -> rax
+;* V15 tmp2 [V15 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+;* V16 tmp3 [V16 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+; V17 tmp4 [V17,T11] ( 2, 2.50) ref -> r15 class-hnd "Inlining Arg"
+; V18 tmp5 [V18,T12] ( 2, 2.50) ref -> r14 class-hnd "Inlining Arg"
+;* V19 tmp6 [V19 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
+;* V20 tmp7 [V20 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+;* V21 tmp8 [V21 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+; V22 tmp9 [V22,T20] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
+;* V23 tmp10 [V23 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+;* V24 tmp11 [V24 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
+; V25 tmp12 [V25,T21] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
;* V26 tmp13 [V26 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
;* V27 tmp14 [V27 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-; V28 tmp15 [V28,T13] ( 2, 2.50) ref -> r15 class-hnd "Inlining Arg"
-; V29 tmp16 [V29,T14] ( 2, 2.50) ref -> r14 class-hnd "Inlining Arg"
-;* V30 tmp17 [V30 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
-;* V31 tmp18 [V31 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-;* V32 tmp19 [V32 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-; V33 tmp20 [V33,T31] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
-;* V34 tmp21 [V34 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-;* V35 tmp22 [V35 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-; V36 tmp23 [V36,T32] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
-;* V37 tmp24 [V37 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-;* V38 tmp25 [V38 ] ( 0, 0 ) bool -> zero-ref "Inlining Arg"
-; V39 tmp26 [V39,T33] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
-;* V40 tmp27 [V40 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
-; V41 tmp28 [V41,T01] ( 2, 16 ) bool -> rax "Inlining Arg"
-; V42 tmp29 [V42,T05] ( 2, 10 ) ref -> r12 class-hnd "Inlining Arg"
-;* V43 tmp30 [V43 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
-; V44 tmp31 [V44,T15] ( 2, 2 ) bool -> rax "Inlining Arg"
-; V45 tmp32 [V45,T34] ( 2, 1.50) ref -> rsi class-hnd "Inlining Arg"
-; V46 tmp33 [V46,T25] ( 3, 1.50) ref -> rcx "argument with side effect"
-; V47 tmp34 [V47 ] ( 8, 14 ) simd16 -> [rsp+0x40] do-not-enreg[XS] addr-exposed "by-value struct argument"
-; V48 tmp35 [V48,T16] ( 2, 2 ) long -> rcx "argument with side effect"
-; V49 tmp36 [V49 ] ( 8, 14 ) simd16 -> [rsp+0x30] do-not-enreg[XS] addr-exposed "by-value struct argument"
-; V50 tmp37 [V50,T17] ( 2, 2 ) long -> rcx "argument with side effect"
-; V51 tmp38 [V51 ] ( 6, 6 ) simd16 -> [rsp+0x20] do-not-enreg[XS] addr-exposed "by-value struct argument"
-; V52 tmp39 [V52,T18] ( 2, 2 ) long -> rcx "argument with side effect"
-; V53 tmp40 [V53,T19] ( 2, 2 ) long -> rcx "argument with side effect"
-; V54 tmp41 [V54,T20] ( 2, 2 ) long -> rcx "argument with side effect"
-; V55 tmp42 [V55,T21] ( 2, 2 ) long -> rcx "argument with side effect"
-; V56 tmp43 [V56,T26] ( 3, 1.50) ref -> rcx "argument with side effect"
-; V57 tmp44 [V57,T27] ( 3, 1.50) ref -> rcx "argument with side effect"
-; V58 tmp45 [V58,T28] ( 3, 1.50) ref -> rcx "argument with side effect"
-; V59 tmp46 [V59,T07] ( 2, 8 ) long -> rcx "argument with side effect"
-; V60 tmp47 [V60,T08] ( 2, 8 ) long -> rcx "argument with side effect"
-; V61 tmp48 [V61,T10] ( 3, 6 ) ref -> rcx "argument with side effect"
-; V62 tmp49 [V62,T22] ( 2, 2 ) long -> rcx "argument with side effect"
-; V63 tmp50 [V63,T23] ( 2, 2 ) long -> rcx "argument with side effect"
-; V64 tmp51 [V64,T29] ( 3, 1.50) ref -> rcx "argument with side effect"
-; V65 tmp52 [V65,T24] ( 2, 2 ) long -> rcx "argument with side effect"
-; V66 cse0 [V66,T12] ( 3, 3 ) ref -> r14 "CSE - conservative"
-; V67 cse1 [V67,T06] ( 7, 8 ) long -> rbp "CSE - aggressive"
-; V68 cse2 [V68,T03] ( 6, 13.50) long -> r15 "CSE - aggressive"
+; V28 tmp15 [V28,T22] ( 2, 1.50) ref -> r15 class-hnd "Inlining Arg"
+;* V29 tmp16 [V29 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
+; V30 tmp17 [V30,T01] ( 2, 16 ) bool -> rax "Inlining Arg"
+; V31 tmp18 [V31,T05] ( 2, 10 ) ref -> r12 class-hnd "Inlining Arg"
+;* V32 tmp19 [V32 ] ( 0, 0 ) long -> zero-ref "NewObj constructor temp"
+; V33 tmp20 [V33,T13] ( 2, 2 ) bool -> rax "Inlining Arg"
+; V34 tmp21 [V34,T23] ( 2, 1.50) ref -> rsi class-hnd "Inlining Arg"
+; V35 tmp22 [V35,T14] ( 3, 1.50) ref -> rcx "argument with side effect"
+; V36 tmp23 [V36,T15] ( 3, 1.50) ref -> rcx "argument with side effect"
+; V37 tmp24 [V37,T16] ( 3, 1.50) ref -> rcx "argument with side effect"
+; V38 tmp25 [V38,T17] ( 3, 1.50) ref -> rcx "argument with side effect"
+; V39 tmp26 [V39,T08] ( 3, 6 ) ref -> rcx "argument with side effect"
+; V40 tmp27 [V40,T18] ( 3, 1.50) ref -> rcx "argument with side effect"
+; V41 cse0 [V41,T10] ( 3, 3 ) ref -> r14 "CSE - aggressive"
+; V42 cse1 [V42,T06] ( 7, 8 ) long -> rbp "CSE - aggressive"
+; V43 cse2 [V43,T03] ( 6, 13.50) long -> r15 "CSE - aggressive"
;
-; Lcl frame size = 336
+; Lcl frame size = 112
G_M2855_IG01:
push r15
@@ -86,14 +61,16 @@ G_M2855_IG01:
push rsi
push rbp
push rbx
- sub rsp, 336
- movaps qword ptr [rsp+140H], xmm6
- movaps qword ptr [rsp+130H], xmm7
- movaps qword ptr [rsp+120H], xmm8
+ sub rsp, 112
+ movaps qword ptr [rsp+60H], xmm6
+ movaps qword ptr [rsp+50H], xmm7
+ movaps qword ptr [rsp+40H], xmm8
+ movaps qword ptr [rsp+30H], xmm9
+ movaps qword ptr [rsp+20H], xmm10
mov rsi, rcx
mov rdi, rdx
mov rbx, r8
- ;; bbWeight=1 PerfScore 17.00
+ ;; bbWeight=1 PerfScore 23.00
G_M2855_IG02:
mov rbp, qword ptr [(reloc)]
mov r14, gword ptr [rbp]
@@ -111,28 +88,25 @@ G_M2855_IG03:
call qword ptr [rax+32]System.Diagnostics.DebugProvider:Fail(System.String,System.String):this
;; bbWeight=0.25 PerfScore 3.13
G_M2855_IG04:
- lea rcx, [rsp+110H]
- mov edx, -128
- call [System.Runtime.Intrinsics.Vector128:Create(short):System.Runtime.Intrinsics.Vector128`1[Int16]]
- lea rcx, [rsp+100H]
- mov edx, 0x7F80
- call [System.Runtime.Intrinsics.Vector128:Create(ushort):System.Runtime.Intrinsics.Vector128`1[UInt16]]
- movdqu xmm6, xmmword ptr [rsi]
+ movups xmm6, xmmword ptr [reloc @RWD16]
+ movups xmm7, xmmword ptr [reloc @RWD48]
+ movdqu xmm8, xmmword ptr [rsi]
call [System.Runtime.Intrinsics.X86.Sse41:get_IsSupported():bool]
test al, al
je SHORT G_M2855_IG07
- ;; bbWeight=1 PerfScore 13.75
+ ;; bbWeight=1 PerfScore 12.25
G_M2855_IG05:
- movaps xmm0, xmmword ptr [rsp+110H]
- ptest xmm6, xmm0
- je G_M2855_IG09
+ ptest xmm8, xmm6
+ je SHORT G_M2855_IG09
xor rax, rax
- ;; bbWeight=0.50 PerfScore 3.63
+ ;; bbWeight=0.50 PerfScore 2.13
G_M2855_IG06:
- movaps xmm6, qword ptr [rsp+140H]
- movaps xmm7, qword ptr [rsp+130H]
- movaps xmm8, qword ptr [rsp+120H]
- add rsp, 336
+ movaps xmm6, qword ptr [rsp+60H]
+ movaps xmm7, qword ptr [rsp+50H]
+ movaps xmm8, qword ptr [rsp+40H]
+ movaps xmm9, qword ptr [rsp+30H]
+ movaps xmm10, qword ptr [rsp+20H]
+ add rsp, 112
pop rbx
pop rbp
pop rsi
@@ -141,30 +115,22 @@ G_M2855_IG06:
pop r14
pop r15
ret
- ;; bbWeight=0.50 PerfScore 8.38
+ ;; bbWeight=0.50 PerfScore 12.38
G_M2855_IG07:
- lea rcx, [rsp+F0H]
- movaps xmmword ptr [rsp+40H], xmm6
- lea rdx, bword ptr [rsp+40H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt16(System.Runtime.Intrinsics.Vector128`1[Int16]):System.Runtime.Intrinsics.Vector128`1[UInt16]]
- lea rcx, [rsp+E0H]
- movaps xmm0, xmmword ptr [rsp+F0H]
- movaps xmm1, xmmword ptr [rsp+100H]
- paddusw xmm0, xmm1
- movaps xmmword ptr [rsp+30H], xmm0
- lea rdx, bword ptr [rsp+30H]
- call [System.Runtime.Intrinsics.Vector128:AsByte(System.Runtime.Intrinsics.Vector128`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[Byte]]
- movaps xmm0, xmmword ptr [rsp+E0H]
+ movaps xmm0, xmm8
+ paddusw xmm0, xmm7
pmovmskb eax, xmm0
test eax, 0xAAAA
je SHORT G_M2855_IG09
xor rax, rax
- ;; bbWeight=0.50 PerfScore 11.92
+ ;; bbWeight=0.50 PerfScore 1.54
G_M2855_IG08:
- movaps xmm6, qword ptr [rsp+140H]
- movaps xmm7, qword ptr [rsp+130H]
- movaps xmm8, qword ptr [rsp+120H]
- add rsp, 336
+ movaps xmm6, qword ptr [rsp+60H]
+ movaps xmm7, qword ptr [rsp+50H]
+ movaps xmm8, qword ptr [rsp+40H]
+ movaps xmm9, qword ptr [rsp+30H]
+ movaps xmm10, qword ptr [rsp+20H]
+ add rsp, 112
pop rbx
pop rbp
pop rsi
@@ -173,55 +139,34 @@ G_M2855_IG08:
pop r14
pop r15
ret
- ;; bbWeight=0.50 PerfScore 8.38
+ ;; bbWeight=0.50 PerfScore 12.38
G_M2855_IG09:
- movaps xmm7, xmm6
- packuswb xmm7, xmm6
- lea rcx, [rsp+D0H]
- movaps xmmword ptr [rsp+20H], xmm7
- lea rdx, bword ptr [rsp+20H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt64(System.Runtime.Intrinsics.Vector128`1[Byte]):System.Runtime.Intrinsics.Vector128`1[UInt64]]
- movaps xmm0, xmmword ptr [rsp+D0H]
- movq xmmword ptr [rdi], xmm0
+ movaps xmm9, xmm8
+ packuswb xmm9, xmm8
+ movq xmmword ptr [rdi], xmm9
mov r14d, 8
test dil, 8
- jne G_M2855_IG12
- movdqu xmm6, xmmword ptr [rsi+16]
+ jne SHORT G_M2855_IG12
+ movdqu xmm8, xmmword ptr [rsi+16]
call [System.Runtime.Intrinsics.X86.Sse41:get_IsSupported():bool]
test al, al
je SHORT G_M2855_IG10
- movaps xmm0, xmmword ptr [rsp+110H]
- ptest xmm6, xmm0
+ ptest xmm8, xmm6
je SHORT G_M2855_IG11
jmp G_M2855_IG28
- ;; bbWeight=0.50 PerfScore 14.00
+ ;; bbWeight=0.50 PerfScore 8.00
G_M2855_IG10:
- lea rcx, [rsp+70H]
- movaps xmmword ptr [rsp+40H], xmm6
- lea rdx, bword ptr [rsp+40H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt16(System.Runtime.Intrinsics.Vector128`1[Int16]):System.Runtime.Intrinsics.Vector128`1[UInt16]]
- lea rcx, [rsp+60H]
- movaps xmm0, xmmword ptr [rsp+70H]
- movaps xmm1, xmmword ptr [rsp+100H]
- paddusw xmm0, xmm1
- movaps xmmword ptr [rsp+30H], xmm0
- lea rdx, bword ptr [rsp+30H]
- call [System.Runtime.Intrinsics.Vector128:AsByte(System.Runtime.Intrinsics.Vector128`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[Byte]]
- movaps xmm0, xmmword ptr [rsp+60H]
- pmovmskb ecx, xmm0
- test ecx, 0xAAAA
+ movaps xmm0, xmm8
+ paddusw xmm0, xmm7
+ pmovmskb eax, xmm0
+ test eax, 0xAAAA
jne G_M2855_IG28
- ;; bbWeight=0.50 PerfScore 11.79
+ ;; bbWeight=0.50 PerfScore 1.42
G_M2855_IG11:
- movaps xmm7, xmm6
- packuswb xmm7, xmm6
- lea rcx, [rsp+50H]
- movaps xmmword ptr [rsp+20H], xmm7
- lea rdx, bword ptr [rsp+20H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt64(System.Runtime.Intrinsics.Vector128`1[Byte]):System.Runtime.Intrinsics.Vector128`1[UInt64]]
- movaps xmm0, xmmword ptr [rsp+50H]
- movq xmmword ptr [rdi+8], xmm0
- ;; bbWeight=0.50 PerfScore 5.63
+ movaps xmm9, xmm8
+ packuswb xmm9, xmm8
+ movq xmmword ptr [rdi+8], xmm9
+ ;; bbWeight=0.50 PerfScore 1.13
G_M2855_IG12:
mov rax, rdi
and rax, 15
@@ -288,42 +233,30 @@ G_M2855_IG20:
sub rbx, 16
;; bbWeight=0.50 PerfScore 0.13
G_M2855_IG21:
- movdqu xmm6, xmmword ptr [rsi+2*r14]
- movdqu xmm7, xmmword ptr [rsi+2*r14+16]
- movaps xmm8, xmm6
- por xmm8, xmm7
+ movdqu xmm8, xmmword ptr [rsi+2*r14]
+ movdqu xmm9, xmmword ptr [rsi+2*r14+16]
+ movaps xmm10, xmm8
+ por xmm10, xmm9
call [System.Runtime.Intrinsics.X86.Sse41:get_IsSupported():bool]
test al, al
je SHORT G_M2855_IG24
;; bbWeight=4 PerfScore 35.33
G_M2855_IG22:
- movaps xmm0, xmmword ptr [rsp+110H]
- ptest xmm8, xmm0
+ ptest xmm10, xmm6
je SHORT G_M2855_IG25
- ;; bbWeight=2 PerfScore 14.00
+ ;; bbWeight=2 PerfScore 8.00
G_M2855_IG23:
jmp G_M2855_IG30
;; bbWeight=0.50 PerfScore 1.00
G_M2855_IG24:
- lea rcx, [rsp+C0H]
- movaps xmmword ptr [rsp+40H], xmm8
- lea rdx, bword ptr [rsp+40H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt16(System.Runtime.Intrinsics.Vector128`1[Int16]):System.Runtime.Intrinsics.Vector128`1[UInt16]]
- lea rcx, [rsp+B0H]
- movaps xmm0, xmmword ptr [rsp+C0H]
- movaps xmm1, xmmword ptr [rsp+100H]
- paddusw xmm0, xmm1
- movaps xmmword ptr [rsp+30H], xmm0
- lea rdx, bword ptr [rsp+30H]
- call [System.Runtime.Intrinsics.Vector128:AsByte(System.Runtime.Intrinsics.Vector128`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[Byte]]
- movaps xmm0, xmmword ptr [rsp+B0H]
- pmovmskb eax, xmm0
+ paddusw xmm10, xmm7
+ pmovmskb eax, xmm10
test eax, 0xAAAA
jne G_M2855_IG30
- ;; bbWeight=2 PerfScore 47.17
+ ;; bbWeight=2 PerfScore 5.17
G_M2855_IG25:
- packuswb xmm6, xmm7
- movaps xmm7, xmm6
+ packuswb xmm8, xmm9
+ movaps xmm9, xmm8
lea r15, [rdi+r14]
test r15b, 15
sete al
@@ -343,7 +276,7 @@ G_M2855_IG26:
call qword ptr [rax+32]System.Diagnostics.DebugProvider:Fail(System.String,System.String):this
;; bbWeight=1 PerfScore 16.25
G_M2855_IG27:
- movdqa xmmword ptr [r15], xmm7
+ movdqa xmmword ptr [r15], xmm9
add r14, 16
cmp r14, rbx
jbe G_M2855_IG21
@@ -352,10 +285,12 @@ G_M2855_IG28:
mov rax, r14
;; bbWeight=0.50 PerfScore 0.13
G_M2855_IG29:
- movaps xmm6, qword ptr [rsp+140H]
- movaps xmm7, qword ptr [rsp+130H]
- movaps xmm8, qword ptr [rsp+120H]
- add rsp, 336
+ movaps xmm6, qword ptr [rsp+60H]
+ movaps xmm7, qword ptr [rsp+50H]
+ movaps xmm8, qword ptr [rsp+40H]
+ movaps xmm9, qword ptr [rsp+30H]
+ movaps xmm10, qword ptr [rsp+20H]
+ add rsp, 112
pop rbx
pop rbp
pop rsi
@@ -364,36 +299,24 @@ G_M2855_IG29:
pop r14
pop r15
ret
- ;; bbWeight=0.50 PerfScore 8.38
+ ;; bbWeight=0.50 PerfScore 12.38
G_M2855_IG30:
call [System.Runtime.Intrinsics.X86.Sse41:get_IsSupported():bool]
test al, al
je SHORT G_M2855_IG31
- movaps xmm0, xmmword ptr [rsp+110H]
- ptest xmm6, xmm0
+ ptest xmm8, xmm6
je SHORT G_M2855_IG32
jmp SHORT G_M2855_IG28
- ;; bbWeight=0.50 PerfScore 6.63
+ ;; bbWeight=0.50 PerfScore 5.13
G_M2855_IG31:
- lea rcx, [rsp+A0H]
- movaps xmmword ptr [rsp+40H], xmm6
- lea rdx, bword ptr [rsp+40H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt16(System.Runtime.Intrinsics.Vector128`1[Int16]):System.Runtime.Intrinsics.Vector128`1[UInt16]]
- lea rcx, [rsp+90H]
- movaps xmm0, xmmword ptr [rsp+A0H]
- movaps xmm1, xmmword ptr [rsp+100H]
- paddusw xmm0, xmm1
- movaps xmmword ptr [rsp+30H], xmm0
- lea rdx, bword ptr [rsp+30H]
- call [System.Runtime.Intrinsics.Vector128:AsByte(System.Runtime.Intrinsics.Vector128`1[UInt16]):System.Runtime.Intrinsics.Vector128`1[Byte]]
- movaps xmm0, xmmword ptr [rsp+90H]
- pmovmskb eax, xmm0
+ paddusw xmm7, xmm8
+ pmovmskb eax, xmm7
test eax, 0xAAAA
- jne G_M2855_IG28
- ;; bbWeight=0.50 PerfScore 11.79
+ jne SHORT G_M2855_IG28
+ ;; bbWeight=0.50 PerfScore 1.29
G_M2855_IG32:
- movaps xmm7, xmm6
- packuswb xmm7, xmm6
+ movaps xmm9, xmm8
+ packuswb xmm9, xmm8
lea r15, [rdi+r14]
test r15b, 7
sete al
@@ -413,15 +336,15 @@ G_M2855_IG33:
call qword ptr [rax+32]System.Diagnostics.DebugProvider:Fail(System.String,System.String):this
;; bbWeight=0.25 PerfScore 4.06
G_M2855_IG34:
- lea rcx, [rsp+80H]
- movaps xmmword ptr [rsp+20H], xmm7
- lea rdx, bword ptr [rsp+20H]
- call [System.Runtime.Intrinsics.Vector128:AsUInt64(System.Runtime.Intrinsics.Vector128`1[Byte]):System.Runtime.Intrinsics.Vector128`1[UInt64]]
- movaps xmm0, xmmword ptr [rsp+80H]
- movq xmmword ptr [r15], xmm0
+ movq xmmword ptr [r15], xmm9
add r14, 8
jmp G_M2855_IG28
- ;; bbWeight=0.50 PerfScore 6.13
+ ;; bbWeight=0.50 PerfScore 1.63
+RWD00 db 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h
+RWD16 db 080h, 0FFh, 080h, 0FFh, 080h, 0FFh, 080h, 0FFh, 080h, 0FFh, 080h, 0FFh, 080h, 0FFh, 080h, 0FFh
+RWD32 db 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h, 000h
+RWD48 db 080h, 07Fh, 080h, 07Fh, 080h, 07Fh, 080h, 07Fh, 080h, 07Fh, 080h, 07Fh, 080h, 07Fh, 080h, 07Fh
+
-; Total bytes of code 1243, prolog size 42, PerfScore 438.33, (MethodHash=d994f4d8) for method System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
+; Total bytes of code 833, prolog size 42, PerfScore 316.58, (MethodHash=d994f4d8) for method System.Text.ASCIIUtility:NarrowUtf16ToAscii_Sse2(long,long,long):long
; ============================================================
👍1
Doesn't look to be introducing any instructions that aren't already there, just removing some cruft?
benaadams
on 11 Aug 2020
Doesn't look to be introducing any instructions that aren't already there, just removing some cruft?
Yes, they shouldn't be introducing any instructions outside of baseline ISAs - this is how I hand-picked the methods by looking at the corresponding codegen code.
echesakovMSFT
on 11 Aug 2020
That seems reasonable to do, although we need to take care to put documentation in the code about why this is ok, and notes in the codegen code about this.
davidwrighton
on 11 Aug 2020
👍1
Was this page helpful?
0 / 5 - 0 ratings
Related issues
nalywa
路
3Comments
btecu
路
3Comments
jzabroski
路
3Comments
Timovzl
路
3Comments
yahorsi
路
3Comments