[wasm] Add the illink substitutions for SIMD #79672
Conversation
This should fix size regression introduce with dotnet#78068 The untrimmed S.N.Vector class added few kilobytes to the assemblies of the bench sample. That wasn't that bad, OTOH the code produced by the aot compiler was much larger, 5MB in this case. So this will case default non-SIMD case. In the case of SIMD build we will need more investigation.
ghost
assigned 
|
/azp run runtime-wasm |
|
Azure Pipelines successfully started running 1 pipeline(s). |
| <_ExtraTrimmerArgs Condition="'$(WasmEnableSIMD)' == 'true'">$(_ExtraTrimmerArgs) --substitutions $(MSBuildThisFileDirectory)/ILLink.Substitutions.WasmIntrinsics.xml</_ExtraTrimmerArgs> | ||
| <_ExtraTrimmerArgs Condition="'$(WasmEnableSIMD)' != 'true'">$(_ExtraTrimmerArgs) --substitutions $(MSBuildThisFileDirectory)/ILLink.Substitutions.NoWasmIntrinsics.xml</_ExtraTrimmerArgs> |
There was a problem hiding this comment.
This is using private properties from the ILLink sdk. Since this file is shipped, we need to use an alternate way to pass these arguments.
There was a problem hiding this comment.
This property is suggested in the repo docs: https://github.com/dotnet/linker/blob/2db841fc5f6bc7592b66ffe3bd0e1c888d022f93/docs/illink-options.md?plain=1#L3-L7
The illink is IL linker version shipping with .NET Core or .NET 5 platforms. It's bundled with the .NET SDK and most of the options are accessible using msbuild properties but any option can also be passed using _ExtraTrimmerArgs property.
And it is being used by multiple other projects like xamarin-macios.
@marek-safar Is there an alternate property that should be used for passing arbitrary arguments to the linker?
There was a problem hiding this comment.
I asked @vitek-karas about it yesterday and he thinks there's currently no other way to set that.
There was a problem hiding this comment.
Adding @sbomer who knows best.
If this is an important scenario we could add more official support for it in the linker SDK bits.
|
Tagging subscribers to 'arch-wasm': @lewing Issue DetailsThis should fix size regression introduced with #78068 The untrimmed S.N.Vector class added few kilobytes to the assemblies of the bench sample. That wasn't that bad, OTOH the code produced by the aot compiler was much larger, 5MB in this case. So this will case default non-SIMD case. In the case of SIMD build we will need more investigation.
|
|
The AOT test failures are #79569, and unrelated. |
This should fix size regression introduced with #78068
The untrimmed S.N.Vector class added few kilobytes to the assemblies of the bench sample. That wasn't that bad, OTOH the code produced by the aot compiler was much larger, 5MB in this case.
So this will case default non-SIMD case. In the case of SIMD build we will need more investigation.