Fix regression from RuntimeType.AllocateValueType rewrite

[jkoritzinsky]

Testing out fixes for #101134 and #104777

[jkoritzinsky]

/benchmark help

[jkoritzinsky]

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

[pr-benchmarks]

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

[pr-benchmarks]

micro - aspnet-perf-win

| benchmark                                                 | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio |
| --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- |
| StaticMethod4_arrayNotCached_int_string_struct_class      |          31.80 ns |        30.08 ns |  0.95 |                  104 B |                104 B |  1.00 |
| StaticMethod4_ByRefParams_int_string_struct_class         |          138.6 ns |        137.7 ns |  0.99 |                   48 B |                 48 B |  1.00 |
| StaticMethod4_int_string_struct_class                     |          19.52 ns |        19.80 ns |  1.01 |                    0 B |                  0 B |       |
| StaticMethod5_arrayNotCached_int_string_struct_class_bool |          45.05 ns |        43.55 ns |  0.97 |                  136 B |                136 B |  1.00 |
| StaticMethod5_ByRefParams_int_string_struct_class_bool    |          197.7 ns |        206.7 ns |  1.05 |                   72 B |                 72 B |  1.00 |

[jkoritzinsky]

Looks like this helps, but not enough.

[jkoritzinsky]

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

[pr-benchmarks]

Benchmark started for micro on aspnet-perf-win with runtime and arguments --variable filter=*System.Reflection.Invoke.StaticMethod*. Logs: link

[pr-benchmarks]

micro - aspnet-perf-win

| benchmark                                                 | mean (micro.base) | mean (micro.pr) | ratio | allocated (micro.base) | allocated (micro.pr) | ratio |
| --------------------------------------------------------- | ----------------- | --------------- | ----- | ---------------------- | -------------------- | ----- |
| StaticMethod4_arrayNotCached_int_string_struct_class      |          29.41 ns |        30.07 ns |  1.02 |                  104 B |                104 B |  1.00 |
| StaticMethod4_ByRefParams_int_string_struct_class         |         142.15 ns |        87.21 ns |  0.61 |                   48 B |                 48 B |  1.00 |
| StaticMethod4_int_string_struct_class                     |          20.62 ns |        19.83 ns |  0.96 |                    0 B |                  0 B |       |
| StaticMethod5_arrayNotCached_int_string_struct_class_bool |          42.85 ns |        43.83 ns |  1.02 |                  136 B |                136 B |  1.00 |
| StaticMethod5_ByRefParams_int_string_struct_class_bool    |          192.8 ns |        120.2 ns |  0.62 |                   72 B |                 72 B |  1.00 |

[dotnet-policy-service]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

[jkotas]

This block is Windows specific, so this change only helps on Windows. We need an equivalent fix for non-Windows platforms too.

[jkoritzinsky]

Yeah I'll probably revert this particular change and make an equivalent one that's more xplat-friendly.

[jkoritzinsky]

I've inlined the trail allocation logic into JIT_Box and factored out the rest of the logic to be in a separate framed helper instead of using the dynamic FCALL impl logic.

Let me know if this looks good to you.

[jkoritzinsky]

/benchmark micro aspnet-perf-win runtime --variable filter=System.Reflection.Invoke.StaticMethod

[jkotas]

Thanks!

[jkoritzinsky]

/ba-g timeouts unrelated