Use helpers in Guid's ROS<byte> constructor

[DaZombieKiller]

Closes #78440. (SharpLab demo)
Moves the big endian path and exception throwing to a helper, resulting in improved codegen when the constructor is inlined.

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this PR. If you have write-permissions please help me learn by adding exactly one area label.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Closes #78440. (SharpLab demo)
Moves the big endian path and exception throwing to a helper, resulting in improved codegen when the constructor is inlined.

Author:	DaZombieKiller
Assignees:	-
Labels:	area-System.Runtime, community-contribution
Milestone:	-

[EgorBo]

Technically, the BigEndian path doesn't need a helper, it's trimmed out anyway. It's the throw helper that messes inlining up:

*************** In fgFindBasicBlocks() for System.Guid:.ctor(System.ReadOnlySpan`1[ubyte]):this
weight= 77 : state  16 [ ldarga.s ]
weight= 79 : state  40 [ call ]
weight= 41 : state  32 [ ldc.i4.s ]
weight=  6 : state  46 [ beq.s ]
weight= 79 : state  40 [ call ]
weight= 66 : state 102 [ ldstr ]
weight= 79 : state  40 [ call ]
weight= 66 : state 102 [ ldstr ]
weight=227 : state 103 [ newobj ]
weight=210 : state 108 [ throw ]
weight=159 : state 112 [ ldsfld ]
weight=-24 : state  39 [ pop ]
weight= 10 : state   3 [ ldarg.0 ]
weight= 16 : state   4 [ ldarg.1 ]
weight= 79 : state  40 [ call ]
weight= 36 : state 115 [ stobj ]
weight= 19 : state  42 [ ret ]

multiplier in instance constructors increased to 1.5.
multiplier in methods of struct increased to 4.5.
1 arguments are structs passed by value.  Multiplier increased to 6.5.
Inline candidate has an arg that feeds a constant test.  Multiplier increased to 7.5.
Inline candidate callsite is boring.  Multiplier increased to 8.8.
calleeNativeSizeEstimate=1225
callsiteNativeSizeEstimate=135
benefit multiplier=8.8
threshold=1188
Native estimate for function size exceeds threshold for inlining 122.5 > 118.8 (multiplier = 8.8)

[tannergooding]

Technically, the BigEndian path doesn't need a helper, it's trimmed out anyway.

Is it properly trimmed out? The linker has a bad habit of leaving comparisons and branches that do nothing and that likely negatively impacts some analysis and opts we can do

[EgorBo]

Technically, the BigEndian path doesn't need a helper, it's trimmed out anyway.

Is it properly trimmed out? The linker has a bad habit of leaving comparisons and branches that do nothing and that likely negatively impacts some analysis and opts we can do

I've attached the IL ^, looks pretty normal to me 🤷
In fact, it's almost inlined as is

[stephentoub]

Thanks. I just read @EgorBo's comment. Given that, I think we should just outline the throw and keep the rest the same. Otherwise, we're possibly regressing Guid's ctor on big endian for little reason.

[DaZombieKiller]

That makes sense. When I added the ReadGuidLittleEndian helper I was in the "portable assembly" mindset and forgot that the runtime is always trimmed, rendering the helper redundant. I've removed it in 65763ac.

[stephentoub]

Thanks.