Thanks for your reply David. Would you please clarify what do you mean by saying: David Chisnall-5 wrote> Increasingly, it seems that nodes have several metadata nodes and so they > are all paying the time penalty for the lookup.Arsen -- View this message in context: http://llvm.1065342.n5.nabble.com/LLVM-Value-relationship-with-metadata-tp50737p50793.html Sent from the LLVM - Dev mailing list archive at Nabble.com.
On 6 Nov 2012, at 04:54, Arsen wrote:> Thanks for your reply David. Would you please clarify what do you mean by > saying: > > David Chisnall-5 wrote >> Increasingly, it seems that nodes have several metadata nodes and so they >> are all paying the time penalty for the lookup.I mean that metadata has gone from being something that occasionally appeared, and then mainly in debug mode, to something that appears on a large number of instructions. The current design was created when metadata was both rarely created and rarely queried. It is now far more frequently created, although it may still be relatively infrequently queried (mainly TBAA, debug info writing). It is therefore probably worth profiling whether the original design decisions are still correct. David
Thanks for clarifications Arsen -- View this message in context: http://llvm.1065342.n5.nabble.com/LLVM-Value-relationship-with-metadata-tp50737p50907.html Sent from the LLVM - Dev mailing list archive at Nabble.com.
Possibly Parallel Threads
- [LLVMdev] LLVM Value relationship with metadata
- [LLVMdev] LLVM Value relationship with metadata
- [LLVMdev] LLVM Value relationship with metadata
- [LLVMdev] Passing specific register for an Instruction in target description files.
- [LLVMdev] Passing specific register for an Instruction in target description files.