Perf improvements in CCW creation.

[manodasanW]

• Caching the vtable entries per type and as part of that also moving vtable cleanup to per type rather than per object.
• Also moving to store IInspectableInfo per type rather than per object.

Based on a perf and memory analysis, it can be seen the memory consumption for storing the vtable entries is down as expected and in the repro scenario where CCW creation was found to be expensive, it is no longer near the top of the CPU usage and the execution time of the repro seems to have went down by half.

Fixes #438
Fixes #554

[AaronRobinsonMSFT]

FYI @jkotas

[AaronRobinsonMSFT]

Based on a perf and memory analysis, it can be seen the memory consumption for storing the vtable entries is down as expected and in the repro scenario where CCW creation was found to be expensive, it is no longer near the top of the CPU usage and the execution time of the repro seems to have went down by half.

@manodasanW I appreciate the following request may be annoying, but would it be possible to get a sense of the raw numbers we are looking at here? Like Scenario A vs Scenario B and then % improvement? Memory used and CPU usage are all that is interesting at the moment. It helps the .NET team talk about the investments we make in APIs and scenarios they unlock.

[manodasanW]

Based on a perf and memory analysis, it can be seen the memory consumption for storing the vtable entries is down as expected and in the repro scenario where CCW creation was found to be expensive, it is no longer near the top of the CPU usage and the execution time of the repro seems to have went down by half.

@manodasanW I appreciate the following request may be annoying, but would it be possible to get a sense of the raw numbers we are looking at here? Like Scenario A vs Scenario B and then % improvement? Memory used and CPU usage are all that is interesting at the moment. It helps the .NET team talk about the investments we make in APIs and scenarios they unlock.

I plan on asking Bartek if he can do an updated run of this scenario which he has been tracking for various CsWinRT versions and the previous built-in version. I will also look at getting some metrics for memory usage.

[AaronRobinsonMSFT]

I plan on asking Bartek if he can do an updated run of this scenario which he has been tracking for various CsWinRT versions and the previous built-in version. I will also look at getting some metrics for memory usage.

Perfect. That would be most insightful. Thanks!