[Perf -1,517%] System.Numerics.Tests.Perf_Vector2.DistanceBenchmark

[DrewScoggins]

Run Information

Architecture	x64
OS	ubuntu 18.04
Changes	diff

Regressions in System.Numerics.Tests.Perf_Vector2

Benchmark	Baseline	Test	Test/Base	Modality	Baseline Outlier
DistanceBenchmark	0.61 ns	9.83 ns	16.17		True

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Numerics.Tests.Perf_Vector2*'

Histogram

System.Numerics.Tests.Perf_Vector2.DistanceBenchmark

[-0.526 ;  1.545) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[ 1.545 ;  3.549) | 
[ 3.549 ;  5.554) | 
[ 5.554 ;  7.559) | 
[ 7.559 ;  8.828) | 
[ 8.828 ; 10.852) | @@@@@@@@@@@@@@@@@@@@@@

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

Tagging subscribers to this area: @tannergooding, @pgovind
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Run Information

Architecture	x64
OS	ubuntu 18.04
Changes	diff

Regressions in System.Numerics.Tests.Perf_Vector2

Benchmark	Baseline	Test	Test/Base	Modality	Baseline Outlier
DistanceBenchmark	0.61 ns	9.83 ns	16.17		True

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Numerics.Tests.Perf_Vector2*'

Histogram

System.Numerics.Tests.Perf_Vector2.DistanceBenchmark

[-0.526 ;  1.545) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[ 1.545 ;  3.549) | 
[ 3.549 ;  5.554) | 
[ 5.554 ;  7.559) | 
[ 7.559 ;  8.828) | 
[ 8.828 ; 10.852) | @@@@@@@@@@@@@@@@@@@@@@

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

Author:	DrewScoggins
Assignees:	-
Labels:	arch-x64, area-System.Numerics, os-linux, os-windows, tenet-performance, tenet-performance-benchmarks, untriaged
Milestone:	-

[tannergooding]

@DrewScoggins, do we have disassembly easily accessible?

[DrewScoggins]

Thanks to @adamsitnik for noticing this. Look related to this PR #41898

[DrewScoggins]

No, because this is so old all the artifacts, if we had them, would have been purged. And this was before we added disassembly to the report.

[adamsitnik]

do we have disassembly easily accessible?

in this case we are lucky and the regression is reproducible on Windows. The disassembly:

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net5.0 net6.0 --filter System.Numerics.Tests.Perf_Vector2.DistanceBenchmark --bdn-arguments "--disasm true"

.NET 5.0.5 (5.0.521.16609), X64 RyuJIT

; System.Numerics.Tests.Perf_Vector2.DistanceBenchmark()
       vzeroupper
       mov       rax,16AD17D7C78
       mov       rax,[rax]
       vmovsd    xmm0,qword ptr [rax+8]
       mov       rax,16AD17D7C80
       mov       rax,[rax]
       vmovsd    xmm1,qword ptr [rax+8]
       vsubps    xmm0,xmm0,xmm1
       vdpps     xmm0,xmm0,xmm0,31
       vsqrtss   xmm0,xmm0,xmm0
       ret
; Total bytes of code 54

.NET 6.0.0 (6.0.21.20503), X64 RyuJIT

; System.Numerics.Tests.Perf_Vector2.DistanceBenchmark()
       sub       rsp,18
       vzeroupper
       mov       rax,13834F694A8
       mov       rax,[rax]
       add       rax,8
       vmovss    xmm0,dword ptr [rax]
       vmovss    dword ptr [rsp+10],xmm0
       vmovss    xmm0,dword ptr [rax+4]
       vmovss    dword ptr [rsp+14],xmm0
       mov       rax,13834F694B0
       mov       rax,[rax]
       add       rax,8
       vmovss    xmm0,dword ptr [rax]
       vmovss    dword ptr [rsp+8],xmm0
       vmovss    xmm0,dword ptr [rax+4]
       vmovss    dword ptr [rsp+0C],xmm0
       vmovsd    xmm0,qword ptr [rsp+10]
       vmovsd    xmm1,qword ptr [rsp+8]
       vsubps    xmm0,xmm0,xmm1
       vdpps     xmm0,xmm0,xmm0,31
       vsqrtss   xmm0,xmm0,xmm0
       add       rsp,18
       ret
; Total bytes of code 114

[DrewScoggins]

Run Information

Architecture	x64
OS	ubuntu 18.04
Changes	diff

Regressions in System.Numerics.Tests.Perf_Matrix3x2

Benchmark	Baseline	Test	Test/Base	Modality	Baseline Outlier
IsIdentityBenchmark	3.87 ns	8.17 ns	2.11	Bimodal	True
CreateScaleFromScalarXYWithCenterBenchmark	5.26 ns	6.73 ns	1.28	Bimodal	False

Related Issue on x64 Windows

[Perf 16%] System.Numerics.Tests.Perf_Matrix3x2.EqualsBenchmark

Related Issue on x86 Windows

[Perf 37%] System.Numerics.Tests.Perf_Matrix3x2 (2)

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'System.Numerics.Tests.Perf_Matrix3x2*'

Histogram

System.Numerics.Tests.Perf_Matrix3x2.IsIdentityBenchmark

[3.447 ; 4.154) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[4.154 ; 4.860) | 
[4.860 ; 5.566) | 
[5.566 ; 6.309) | 
[6.309 ; 7.015) | @@@@@@@@@@@@@@@
[7.015 ; 7.867) | @@@@@@
[7.867 ; 8.528) | @

System.Numerics.Tests.Perf_Matrix3x2.CreateScaleFromScalarXYWithCenterBenchmark

[5.141 ; 5.395) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[5.395 ; 5.559) | 
[5.559 ; 5.813) | @@@@@@@@@@@@@@@@@@@@@@@@@@@
[5.813 ; 6.170) | @@
[6.170 ; 6.321) | @
[6.321 ; 6.575) | @@@@@@@@@@@@@@@@@@@
[6.575 ; 6.856) | @@

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

[DrewScoggins]

These tests also regressed over the same commit range.

[DrewScoggins]

.NET 5.0.5 (5.0.521.16609), X64 RyuJIT

; System.Numerics.Tests.Perf_Matrix3x2.IsIdentityBenchmark()
       sub       rsp,38
       vxorps    xmm4,xmm4,xmm4
       vmovdqa   xmmword ptr [rsp+20],xmm4
       xor       eax,eax
       mov       [rsp+30],rax
       lea       rcx,[rsp+20]
       call      System.Numerics.Matrix3x2.get_Identity()
       lea       rcx,[rsp+20]
       call      System.Numerics.Matrix3x2.get_IsIdentity()
       nop
       add       rsp,38
       ret
; Total bytes of code 47

; System.Numerics.Matrix3x2.get_Identity()
       vzeroupper
       mov       rax,1ECEDC01420
       mov       rax,[rax]
       vmovdqu   xmm0,xmmword ptr [rax+8]
       vmovdqu   xmmword ptr [rcx],xmm0
       mov       rdx,[rax+18]
       mov       [rcx+10],rdx
       mov       rax,rcx
       ret
; Total bytes of code 37

; System.Numerics.Matrix3x2.get_IsIdentity()
       vzeroupper
       vmovss    xmm0,dword ptr [rcx]
       vucomiss  xmm0,dword ptr [7FFEC58B2528]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+0C]
       vucomiss  xmm0,dword ptr [7FFEC58B252C]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+4]
       vxorps    xmm1,xmm1,xmm1
       vucomiss  xmm0,xmm1
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+8]
       vxorps    xmm1,xmm1,xmm1
       vucomiss  xmm0,xmm1
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+10]
       vxorps    xmm1,xmm1,xmm1
       vucomiss  xmm0,xmm1
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+14]
       vxorps    xmm1,xmm1,xmm1
       vucomiss  xmm0,xmm1
       setnp     al
       jp        short M02_L00
       sete      al
M02_L00:
       movzx     eax,al
       ret
M02_L01:
       xor       eax,eax
       ret
; Total bytes of code 115

.NET 6.0.0 (6.0.21.20503), X64 RyuJIT

; System.Numerics.Tests.Perf_Matrix3x2.IsIdentityBenchmark()
       sub       rsp,98
       vzeroupper
       vxorps    xmm4,xmm4,xmm4
       vmovdqa   xmmword ptr [rsp+80],xmm4
       xor       eax,eax
       mov       [rsp+90],rax
       lea       rcx,[rsp+80]
       call      System.Numerics.Matrix3x2.get_Identity()
       vmovdqu   xmm0,xmmword ptr [rsp+80]
       vmovdqu   xmmword ptr [rsp+68],xmm0
       mov       rcx,[rsp+90]
       mov       [rsp+78],rcx
       lea       rcx,[rsp+50]
       call      System.Numerics.Matrix3x2.get_Identity()
       vmovdqu   xmm0,xmmword ptr [rsp+68]
       vmovdqu   xmmword ptr [rsp+38],xmm0
       mov       rcx,[rsp+78]
       mov       [rsp+48],rcx
       vmovdqu   xmm0,xmmword ptr [rsp+50]
       vmovdqu   xmmword ptr [rsp+20],xmm0
       mov       rcx,[rsp+60]
       mov       [rsp+30],rcx
       lea       rcx,[rsp+38]
       lea       rdx,[rsp+20]
       call      System.Numerics.Matrix3x2.op_Equality(System.Numerics.Matrix3x2, System.Numerics.Matrix3x2)
       movzx     eax,al
       add       rsp,98
       ret
; Total bytes of code 154

; System.Numerics.Matrix3x2.get_Identity()
       vzeroupper
       mov       rax,1DE240E1428
       mov       rax,[rax]
       vmovdqu   xmm0,xmmword ptr [rax+8]
       vmovdqu   xmmword ptr [rcx],xmm0
       mov       rdx,[rax+18]
       mov       [rcx+10],rdx
       mov       rax,rcx
       ret
; Total bytes of code 37

; System.Numerics.Matrix3x2.op_Equality(System.Numerics.Matrix3x2, System.Numerics.Matrix3x2)
       vzeroupper
       vmovss    xmm0,dword ptr [rcx]
       vucomiss  xmm0,dword ptr [rdx]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+0C]
       vucomiss  xmm0,dword ptr [rdx+0C]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+4]
       vucomiss  xmm0,dword ptr [rdx+4]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+8]
       vucomiss  xmm0,dword ptr [rdx+8]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+10]
       vucomiss  xmm0,dword ptr [rdx+10]
       jp        short M02_L01
       jne       short M02_L01
       vmovss    xmm0,dword ptr [rcx+14]
       vucomiss  xmm0,dword ptr [rdx+14]
       setnp     al
       jp        short M02_L00
       sete      al
M02_L00:
       movzx     eax,al
       ret
M02_L01:
       xor       eax,eax
       ret
; Total bytes of code 96