similar to: [LLVMdev] Updated Pluggable Coalescers Patch

Here's a proposed patch for reworking register coalescing to allow pluggable coalescers. I think I've got the interfaces where I want them and am reasonably sure I've squashed most of the bugs. I'm still doing some testing and want to get through a whole regimen before committing. As a reminder, this patch has several goals: - Allow user-specified register coalescers, similar

Hi David, Thanks for this patch! Some comments: 1. typedef std::set<const LiveInterval *> IntervalSet; Please use SmallPtrSet instead. 2. + virtual void mergeIntervals(const LiveInterval &a, + const LiveInterval &b, + const MachineInstr &copy) {}; I find the name misleading. It's not actually

On Monday 27 August 2007 19:02, Evan Cheng wrote: > > interfere() isn't sufficient because the allocation algorithm may > > place other > > constraints on coalescing. George and Appel's iterated register > > coalescing > > is a prime example. It wants to "freeze" copies that should not be > > coalesced > > because doing so might cause

On Monday 27 August 2007 15:13, Evan Cheng wrote: > 1. typedef std::set<const LiveInterval *> IntervalSet; > Please use SmallPtrSet instead. Ok. > 2. > + virtual void mergeIntervals(const LiveInterval &a, > + const LiveInterval &b, > + const MachineInstr &copy) {}; > > I find the name

On Aug 27, 2007, at 2:04 PM, David Greene wrote: > >> 3. >> >> + /// Allow the register allocator to communicate when it doesn't >> + /// want a copy coalesced. This may be due to assumptions >> made by >> + /// the allocator about various invariants and so this >> question is >> + /// a matter of legality, not performance.

On Aug 28, 2007, at 8:59 AM, David Greene wrote: > On Monday 27 August 2007 19:02, Evan Cheng wrote: > >>> interfere() isn't sufficient because the allocation algorithm may >>> place other >>> constraints on coalescing. George and Appel's iterated register >>> coalescing >>> is a prime example. It wants to "freeze" copies that

On Monday 20 August 2007 15:58, David A. Greene wrote: > Here's a proposed patch for reworking register coalescing to allow > pluggable coalescers. No comments? Or is everyone taking vacation before the fall? :) -Dave

On Wednesday 11 July 2007 12:41, Tanya M. Lattner wrote: > I think the coalescer should be flexible enough to be run independent of > the register allocator. For example, you may want to expose the copies > induced by transforming out of SSA to the scheduler. If the scheduler is > run before register allocation, then you would want the coalescing to > happen before as well. I

Hi David, On Mon, 2007-07-09 at 16:32 -0500, David Greene wrote: > Ok, I'm at a point now where I can implement plyggable register coalescers as > we originally wanted when I started the refactoring work. > > I'm in the midst of finishing up an implementation following the model of how > register allocators are selected. Thgere may be some more refactoring/code >

On Jul 17, 2007, at 12:21 PM, David Greene wrote: > >> Are you designing the coalescer so it always run before >> allocation? If so, >> please keep it simple. :-) > > Believe me, I'm trying to keep it as simple as possible. The > coalescer I'm > working on doesn't run "before" allocation. It runs in conjunction > with it. >

On Jul 16, 2007, at 9:12 AM, David Greene wrote: > On Friday 13 July 2007 13:32, David A. Greene wrote: >> On Wednesday 11 July 2007 15:07, Christopher Lamb wrote: >>> Could it be possible for there to be a harness type interface that >>> would allow coalescers that support both modes to be hooked into the >>> pass registration, and those that depend on the

On Jul 17, 2007, at 9:28 AM, David Greene wrote: > > [...] > 00369 // If the intervals were swapped by Join, swap them back so > that the > register > 00370 // mapping (in the r2i map) is correct. > 00371 if (Swapped) SrcInt.swap(DstInt); > > Whoops! At this point repSrcReg is not consistent with SrcInt and the > same goes for repDstReg! > > 00372

Ok, I'm at a point now where I can implement plyggable register coalescers as we originally wanted when I started the refactoring work. I'm in the midst of finishing up an implementation following the model of how register allocators are selected. Thgere may be some more refactoring/code sharing work to be done with this, but I want to get it working first. createRegisterAllocator() is

On Monday 09 July 2007 17:07, David Greene wrote: > On Monday 09 July 2007 16:49, Reid Spencer wrote: > > The only thing that comes to mind is that creating and running the > > coalescer are separate operations so you might want to do the creation > > of it in alias analysis style. Then, the allocator can a) determine if a > > coalescer was created, b) obtain the

On Mon, 2007-07-16 at 16:19 -0700, Evan Cheng wrote: > Property changes on: llvm/include/llvm/CodeGen/RegisterAllocator.h > ___________________________________________________________________ > Name: svn:eol-style > + LF > > What's this? Its a property change. In Subversion every file and directory can have properties associated with them. For example the svn:ignore

On Friday 13 July 2007 13:32, David A. Greene wrote: > On Wednesday 11 July 2007 15:07, Christopher Lamb wrote: > > Could it be possible for there to be a harness type interface that > > would allow coalescers that support both modes to be hooked into the > > pass registration, and those that depend on the allocator not be > > registered as passes? > > I have a

On Monday 09 July 2007 16:49, Reid Spencer wrote: > The only thing that comes to mind is that creating and running the > coalescer are separate operations so you might want to do the creation > of it in alias analysis style. Then, the allocator can a) determine if a > coalescer was created, b) obtain the coalescer that was created, if any, > and c) run it at the right time for the

On Jul 17, 2007, at 8:49 PM, David A. Greene wrote: > On Tuesday 17 July 2007 14:21, David Greene wrote: > >>> I don't care for a MachineFunctionPass that can be directly >>> called. I >>> think it's a very good idea to keep the coalescers independent from >>> the allocators. If that's desired, we should enhance passmanager so >>>

Hi David, Just as an idea, you could try to use a coloring register allocator submitted by Bill Wendling some time ago on the mailing list? See this thread for more information http://lists.cs.uiuc.edu/pipermail/llvmdev/2007-April/008489.html This allocator is already implemented using LLVM and could be probably useful for testing your new infrastructure and refactored interfaces. At least

On Monday 16 July 2007 18:20, Evan Cheng wrote: > > Does the current implementation follow a published algorithm? If > > so, do we > > have a pointer to it? The comments are rather sparse and it's > > difficult to > > figure out what's going on at times. > > I don't know. I suspect not. I agree the current implementation is a > bit of a