Thanks, Craig. I will address this in the patch. It is unfortunate that MachineOutliner needs to keep all MachineFunction in memory and it is located just before emitting. So once it is used, all MachineFunctions are in memory almost for the whole codegen process. AFAIK, there is no good way to address this. Hopefully, we don't have many exceptions like this.
[RFC] Introducing classes for the codegen driven by new pass manager
Hi Matt, which analysis is this?
I checked again, and it currently is not a real analysis. It’s just a CallGraphResourceInfo map in AMDGPUAsmPrinter, which happens to rely on CodeGenSCCOrder. I think I intended to move this to a separate analysis at one point
Thanks. I'll take a look.
I want to know the current status of the New Pass Manager for Machine functions.
Is the basic infrastructure ready in the LLVM so that targets can start porting the codegen passes?
I believe this is the main page and there had been some discussions earlier, the last update was made nearly 3 years ago.
I hope there has been some progress in the meanwhile.