[wasm] Review compiler options (again)

[lewing]

We're currently generating and processing full dwarf debug information for native and aot'd methods, then in a separate pass stripping that information in release mode (depending on several options) this can add substantial time and memory overhead. We need to review the emscripten compiler options and optimize the default case better. As part of the process we should review some of the properties we use and see if there is an easy way to surface some of them to the customer.

• debug info
• initial memory
• TrimMode -> remove partial setting [mono] Try enabling full trimming by default when not using blazor #89110
• Memory size

see #79874 #80103

[dotnet-issue-labeler]

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

[ghost]

Tagging subscribers to 'arch-wasm': @lewing
See info in area-owners.md if you want to be subscribed.

Issue Details

---

We're currently generating and processing full dwarf debug information for native and aot'd methods, then in a separate pass stripping that information in release mode (depending on several options) this can add substantial time and memory overhead. We need to review the emscripten compiler options and optimize the default case better. As part of the process we should review some of the properties we use and see if there is an easy way to surface some of them to the customer.

see #79874

Author:	lewing
Assignees:	radical
Labels:	arch-wasm, untriaged
Milestone:	-

[pavelsavara]

• explore -Oz vs -O2 in dotnet.wasm of the runtime pack.
• figure out if -fPIC is a good thing for us
• figure out what makes the data section break into thousands of sections vs just two in AOT. Maybe wasm-ld ?

[lewing]

to expand a bit:

1. the tests are exceptional in that they by design use most of the api surface and as a result are hard to AOT so we use special options to the compiler to stay inside container limits
2. our default settings generate and process dwarf info in the first linker pass and that info causes some really degenerate behavior in wasm-opt when it has to process all the relocations for all the symbols. we really only use the name and possibly line information in the dwarf symbols, and we strip that information on publish so we should probably change our options to stop doing the expensive in time and space relocation processing
   that patch fixes 1. for some of the tests (-01 skips wasm-opt in finalize)
   there might be other cases we should tweak
   fixing 2. would fix the default case

def should_run_binaryen_optimizer():
  # run the binaryen optimizer in -O2+. in -O0 we don't need it obviously, while
  # in -O1 we don't run it as the LLVM optimizer has been run, and it does the
  # great majority of the work; not running the binaryen optimizer in that case
  # keeps -O1 mostly-optimized while compiling quickly and without rewriting
  # DWARF etc.

see #80143 for initial attempt

[pavelsavara]

emscripten-core/emscripten#15269 (comment)