similar to: [LLVMdev] Number of spills/stores

Hi, guys, I am comparing the performance of the default scheduler (seems to be the one that minimizes register pressure) with no scheduler (-pre-RA-sched=none), and I got these numbers. The ratio is low_reg_pressure/none, that is, the lower the number, the better the performance with low register pressure: CFP2000/177.mesa/177.mesa 1.00 CFP2000/179.art/179.art

Fernando Magno Quintao Pereira wrote: >> I'm not sure about one thing: you assign stack slot to each new register you >> replace the spilled one with. And then you need to allocate physical >> registers to them. Is it possible to assign physical register to the virtual >> one which has a stack slot already? >> > > Yes. The stack slot is the place where the

Dear LLVM guys, I am coding a liveness analysis algorithm, and I found this comment on LiveVariables.cpp: Line 00195 - http://llvm.org/doxygen/LiveVariables_8cpp-source.html : // PhysRegInfo - Keep track of which instruction was the last use of a // physical register. This is a purely local property, because all physical // register references as presumed dead across basic blocks.

The short answer is this is not currently done. LLVM register allocator does not currently use a move to a different register class in lieu of spills. This is definitely something we should do in the future. Can you file a bugzilla enhancement request? Thanks, Evan On Jun 13, 2007, at 11:09 AM, Chris Lattner wrote: > On Tue, 12 Jun 2007, Fernando Magno Quintao Pereira wrote: >>

> I'm not sure about one thing: you assign stack slot to each new register you > replace the spilled one with. And then you need to allocate physical > registers to them. Is it possible to assign physical register to the virtual > one which has a stack slot already? > Yes. The stack slot is the place where the value will be stored in memory, but, when that value is effectively

On Thu, 12 Apr 2007, Fernando Magno Quintao Pereira wrote: >> I'm definitely interested in improving coalescing and it sounds like >> this would fall under that work. Do you have references to papers >> that talk about the various algorithms? > > Some suggestions: > > @InProceedings{Budimlic02, > AUTHOR = {Zoran Budimlic and Keith D. Cooper and Timothy

Dear LLVMers, I wish someone could help me to clarify a register allocation question. I will use X86 in this example, but it really does not matter too much. Imagine that I have a virtual v stored in register AL, and I am about to cross a function call, that effectively overwrites AL. Well, I have, say, register ESI free, but as it happens, ESI is bigger than AL. It should be

On 8/7/06, Chris Lattner <sabre at nondot.org> wrote: > > On Sun, 6 Aug 2006, Anton Vayvod wrote: > > I'm developing a register allocator that works iteratively. It spills > some > > virtual registers on each iteration until all the rest have physical > ones > > assigned. > > Take a look at the linear scan allocator. It is also iterative: it uses >

I'm not sure about one thing: you assign stack slot to each new register you replace the spilled one with. And then you need to allocate physical registers to them. Is it possible to assign physical register to the virtual one which has a stack slot already? On 8/21/06, Fernando Magno Quintao Pereira <fernando at cs.ucla.edu> wrote: > > > > So what addIntervalsToSpills

Dear guys, I want to coalesce some copies, and I would like to know if there is any method that I can call, like JoinCopy from the old (LLVM 1.9) LiveIntervals class. I found it in SimpleRegisterCoalescing (LLVM 2.1), but I do not want to call this analysis, as I have my own. basically, I can determine that two virtuals do not overlap, and I know that it is safe to join them. In

On 11/23/07, Fernando Magno Quintao Pereira <fernando at cs.ucla.edu> wrote: > > > Hi, Sanjiv, > > those passes operate on the whole machine function. Each machine > function contains many basic blocks. If a program has many functions, the > register allocator will be called as many times, i.e it does not do > interprocedural allocation. > > best, > >

Hi Florian, we though about using UIUC, like in LLVM. Do you guys know if that could be a problem, given that we are mining the functions from github? > Have you thought about integrating the benchmarks as external tests into LLVM’s test-suite? That would make it very easy to play around with. We did not think about it actually. But we would be happy to do it, if the community accepts

Does this depend on vreg subreg support? It sounds as if it could be vaguely related, but I'm not sure it's necessary. -- Chris On Jun 14, 2007, at 12:27 PM, Evan Cheng wrote: > The short answer is this is not currently done. LLVM register > allocator does not currently use a move to a different register class > in lieu of spills. This is definitely something we should do in

On Tue, 12 Jun 2007, Fernando Magno Quintao Pereira wrote: > Imagine that I have a virtual v stored in register AL, and I am about > to cross a function call, that effectively overwrites AL. Well, I have, > say, register ESI free, but as it happens, ESI is bigger than AL. It > should be technically possible to move the contents of AL into ESI, to > avoid spilling 'v',

Chris Lattner wrote: > On Thu, 12 Apr 2007, Fernando Magno Quintao Pereira wrote: >>> I'm definitely interested in improving coalescing and it sounds like >>> this would fall under that work. Do you have references to papers >>> that talk about the various algorithms? >> Some suggestions: >> >> @InProceedings{Budimlic02, >> AUTHOR =

Dear LLVMers, I need help with one error in my register allocator. I am getting errors when folding instructions for the x86 (it does not happen when I use the PowerPC as target). The problem is that, when I execute something like: mi = mbb->insert(mbb->erase(mi), fmi), where mbb is a MachineBasicBlock, valgrind gives this error message: Address 0x42C62B0 is 24 bytes inside a block

Hey, guys, I am creating some move instructions with MRegisterInfo::copyRegToReg. How do I get a pointer to the instruction that I just created? Is there a way to do something like: // mbb is MachineBasicBlock, reg_info is MRegisterInfo MachineBasicBlock::iterator iter = mbb.getFirstTerminator(); reg_info->copyRegToReg(mbb, iter, dst, src, rc); iter--; (???) MachineInstr *

>> And I have a quite fast algo that I believe is simpler than [Budimlic02] >> and I can share it with you :) > > Do you have a paper on this? I'd be interested in seeing it. > Yes, I have a tech report on this page: http://compilers/fernando/projects/soc/ and I have submitted a paper to SAS, and now I am waiting for the review. The coalescing algorithm is described in

Dear guys, what is the best way to implement a swap of floating point registers in X86? For the integer registers, I am using xchg. Is there a similar instruction for floating point? My function to insert swaps is like: void X86RegisterInfo::swapRegs( MachineBasicBlock & mbb, MachineBasicBlock::iterator mi, unsigned r1, unsigned r2, const TargetRegisterClass

Guys, I am in need of a no-op instruction: an instruction that does not do anything, and has no operands. Does LLVM predefine such an instruction? I want to transform the program so that there is no empty basic block. Fernando