Preallocate jump stubs for dynamic methods #9883
Conversation
kouvel
commented
Mar 1, 2017
•
kouvel
commented
- This eliminates the possibility of running into an out-of-memory situation after compiling the method
- The temporary entry points block containing FixupPrecodes is extended for dynamic methods to include sufficient space for jump stubs
- When the target is too far for the FixupPrecode to encode a short relative jump, it instead does a short relative call or jump to the corresponding jump stub, which does an absolute jump to the target
- This eliminates the possibility of running into an out-of-memory situation after compiling the method - The temporary entry points block containing FixupPrecodes is extended for dynamic methods to include sufficient space for jump stubs - When the target is too far for the FixupPrecode to encode a short jump, it instead does a short call or jump to the corresponding jump stub, which does a far jump to the target
src/vm/amd64/cgenamd64.cpp
Outdated
| { | ||
| CONTRACTL | ||
| { | ||
| THROWS; // emptBackToBackJump may throw (see emitJump) |
There was a problem hiding this comment.
emptBackToBackJump [](start = 19, length = 18)
nit: typo: emptBackToBackJump
|
LGTM |
|
Did you compute/measure the additional size impact of this? Does NGEN store FixupPrecode? What is the size increase, say, of System.Private.CoreLib.ni.dll? Does nidump still work? Did you test on ngen'ed images? |
|
Since the change only applies to dynamically created methods, my understanding is that it should have no impact on ngen/crossgen. I'll check what you suggested. |
|
Regarding runtime memory impact, since we were allocating an additional 8 bytes per FixupPrecode, which wasn't being used, after this change the additional impact is 4 bytes per FixupPrecode for a dynamic method (12 bytes per jump stub). |
Looks like it, but it shouldn't be saving any FixupPrecodes for dynamic methods
No change
It should, but while I found the code for it I didn't find info on how to initiate a dump. Could you please point me to some info on this?
No, I can maybe diff the nidump before and after and see if there is anything to test |
|
Given that ngen images shouldn't include FixupPrecode, I wouldn't worry about it. I normally use "nidump /?" or similar to remind myself of the commands. Maybe it's not built in the open? I haven't looked. |
|
Fragile ngen images do include FixupPrecode - a lot of them. However, this change have not changed anything near NGen or nidump. |
Preallocate jump stubs for dynamic methods - This eliminates the possibility of running into an out-of-memory situation after compiling the method - The temporary entry points block containing FixupPrecodes is extended for dynamic methods to include sufficient space for jump stubs - When the target is too far for the FixupPrecode to encode a short relative jump, it instead does a short relative call or jump to the corresponding jump stub, which does an absolute jump to the target
Preallocate jump stubs for dynamic methods - This eliminates the possibility of running into an out-of-memory situation after compiling the method - The temporary entry points block containing FixupPrecodes is extended for dynamic methods to include sufficient space for jump stubs - When the target is too far for the FixupPrecode to encode a short relative jump, it instead does a short relative call or jump to the corresponding jump stub, which does an absolute jump to the target Commit migrated from dotnet/coreclr@4bafc10
