Jan Sjodin
2011-Feb-24 12:05 UTC
[LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
----- Original Message ----> From: Chris Lattner <clattner at apple.com> > To: Yuri <yuri at rawbw.com> > Cc: llvmdev at cs.uiuc.edu > Sent: Sun, February 20, 2011 3:26:35 AM > Subject: Re: [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule > > > On Feb 19, 2011, at 8:05 PM, Yuri wrote: > > > On 02/19/2011 14:52, Yuri wrote: > >> Will MC path for JNI be included in 2.9? > >> > > > > Sorry. I meant: Will MC path for JIT be included in 2.9? > > While it would be nice, it doesn't seem like anyone is working on it at the >moment. > > -Chris >I have been working on my own stuff this month. This is still not the "proper" solution that has been proposed for MCJIT. I posted some patches a few weeks ago, but since there was no response I continued to work anyway. I decided to try and reuse the ELF code emission as much as possible to be able to run gdb on the generated code. By refactoring the code for MCELFStreamer and ELFObjectWriter I am now able to generate code in memory and the code can be executed and debugged with dgb. With a bit more cleanup I believe it is possible to get a "JIT" (it generates code for a whole module) with just a few hundred lines of code. I'm not sure if this is what people want in general but it works for my application. - Jan
Bill Wendling
2011-Feb-25 00:27 UTC
[LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
On Feb 24, 2011, at 4:05 AM, Jan Sjodin wrote:> On Feb 19, 2011, at 8:05 PM, Yuri wrote: >> >>> On 02/19/2011 14:52, Yuri wrote: >>>> Will MC path for JNI be included in 2.9? >>>> >>> >>> Sorry. I meant: Will MC path for JIT be included in 2.9? >> >> While it would be nice, it doesn't seem like anyone is working on it at the >> moment. >> >> -Chris >> > > I have been working on my own stuff this month. This is still not the "proper" > solution that has been proposed for MCJIT. I posted some patches a few weeks > ago, but since there was no response I continued to work anyway. I decided to > try and reuse the ELF code emission as much as possible to be able to run gdb > on the generated code. By refactoring the code for MCELFStreamer and > ELFObjectWriter I am now able to generate code in memory and the code can be > executed and debugged with dgb. With a bit more cleanup I believe it is > possible to get a "JIT" (it generates code for a whole module) with just a few > hundred lines of code. I'm not sure if this is what people want in general but > it works for my application. >Just a reminder: The deadline for "new features" is coming up (i.e., the date of the 2.9 branch). After branching, there will still be time to finish up features that are near complete. That said, a longer "bake-time" for new features is preferred. And at this point, it is probably up to the various maintainers to approve or reject them. If it's possible to turn a new feature off, that's a definite bonus. :-) Because we can then determine if we want it on or off by default without affecting the 2.9 release timeframe. -bw
Jan Sjodin
2011-Feb-25 15:16 UTC
[LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
----- Original Message ----> From: Bill Wendling <bwendling at apple.com> > To: Jan Sjodin <jan_sjodin at yahoo.com> > Cc: Chris Lattner <clattner at apple.com>; Yuri <yuri at rawbw.com>; >llvmdev at cs.uiuc.edu > Sent: Thu, February 24, 2011 7:27:43 PM > Subject: Re: [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule > > On Feb 24, 2011, at 4:05 AM, Jan Sjodin wrote: > > > On Feb 19, 2011, at 8:05 PM, Yuri wrote: > >> > >>> On 02/19/2011 14:52, Yuri wrote: > >>>> Will MC path for JNI be included in 2.9? > >>>> > >>> > >>> Sorry. I meant: Will MC path for JIT be included in 2.9? > >> > >> While it would be nice, it doesn't seem like anyone is working on it at >the > > >> moment. > >> > >> -Chris > >> > > > > I have been working on my own stuff this month. This is still not the >"proper" > > > solution that has been proposed for MCJIT. I posted some patches a few >weeks > > > ago, but since there was no response I continued to work anyway. I decided >to > > > try and reuse the ELF code emission as much as possible to be able to run >gdb > > > on the generated code. By refactoring the code for MCELFStreamer and > > ELFObjectWriter I am now able to generate code in memory and the code can >be > > > executed and debugged with dgb. With a bit more cleanup I believe it is > > possible to get a "JIT" (it generates code for a whole module) with just a >few > > > hundred lines of code. I'm not sure if this is what people want in general >but > > > it works for my application. > > > Just a reminder: > > The deadline for "new features" is coming up (i.e., the date of the 2.9 >branch). After branching, there will still be time to finish up features that >are near complete. That said, a longer "bake-time" for new features is >preferred. And at this point, it is probably up to the various maintainers to >approve or reject them. > > If it's possible to turn a new feature off, that's a definite bonus. :-) >Because we can then determine if we want it on or off by default without >affecting the 2.9 release timeframe. >I can certainly start subitting patches to refactor some of the code. For example splitting .cpp files into .h and .cpp for classes that are needed. I don't know if it will go into 2.9 or not. - Jan
Possibly Parallel Threads
- [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
- [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
- [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
- [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule
- [LLVMdev] Announcing: LLVM 2.9 Tentative Release Schedule