search for: spillplacement

Displaying 14 results from an estimated 14 matches for "spillplacement".

2011 Jan 20
0
[LLVMdev] Minor warning reduction in lib/CodeGen/SpillPlacement.cpp
MSVC complains of truncating a double to a float. Let's just give it a float. -- GCS a+ e++ d- C++ ULS$ L+$ !E- W++ P+++$ w++$ tv+ b++ DI D++ 5++ Life is complex, with real and imaginary parts. "Ok, it boots. Which means it must be bug-free and perfect. " -- Linus Torvalds "People disagree with me. I just ignore them." -- Linus Torvalds -------------- next part
2013 Jun 12
3
[LLVMdev] RFC - Profile Guided Optimization in LLVM
...abilities? > Agreed. An important consideration here is WPO vs. LTO vs. TU-at-a-time call graphs. > FWIW, the greedy register allocator’s live range splitting algorithm is > designed to consume profile information so it can push spill code into cold > blocks. The primary interface is SpillPlacement::getBlockFrequency() which > currently returns an estimate based on loop depth only. > It doesn't use MachineBlockFrequency? If it does, it will get a lot more than loop depth: __builtin_expect, cold function attribute, and static branch heuristics. If it doesn't it should, and then...
2013 Jun 12
0
[LLVMdev] RFC - Profile Guided Optimization in LLVM
...seems like we might need a way of representing call graph profiling in addition to the existing branch probabilities? FWIW, the greedy register allocator’s live range splitting algorithm is designed to consume profile information so it can push spill code into cold blocks. The primary interface is SpillPlacement::getBlockFrequency() which currently returns an estimate based on loop depth only. /jakob
2013 Jun 17
0
[LLVMdev] RFC - Profile Guided Optimization in LLVM
...named, it returns a fixpoint number. I think you should scale it to be relative to the entry block frequency. +LiveIntervals::getSpillWeight(bool isDef, bool isUse, BlockFrequency freq) { + return (isDef + isUse) * freq.getFrequency(); } This computation can overflow. @@ -178,9 +180,10 @@ bool SpillPlacement::runOnMachineFunction(MachineFunction &mf) { // Compute total ingoing and outgoing block frequencies for all bundles. BlockFrequency.resize(mf.getNumBlockIDs()); + MachineBlockFrequencyInfo &MBFI = getAnalysis<MachineBlockFrequencyInfo>(); for (MachineFunction::iterator I...
2013 Jun 17
2
[LLVMdev] RFC - Profile Guided Optimization in LLVM
On 17.06.2013, at 15:56, Diego Novillo <dnovillo at google.com> wrote: > On 2013-06-15 16:39 , Benjamin Kramer wrote: >> Do you want to take over this effort or should I poke more at it? > > Since you've already started, it's easier if you poke more at it. Thanks. I've got a whole bunch of other things to go through. OK, will do. Jakob any comments on the
2011 Jan 20
0
[LLVMdev] Warnings when compiling llvm+clang (Mac OS X system gcc)
...ing RegAllocGreedy.cpp for Release+Asserts build /Users/filcab/dev/stuff/llvm/llvm/include/llvm/ADT/SmallVector.h: In member function ‘void<unnamed>::RAGreedy::calcLiveBlockInfo(llvm::LiveInterval&)’: /Users/filcab/dev/stuff/llvm/llvm/include/llvm/ADT/SmallVector.h:325: warning: ‘BC.llvm::SpillPlacement::BlockConstraint::Exit’ is used uninitialized in this function ---------------------------- ^^ The error repeats several times. It seems the error occurs when push_back copies the received object, which is a struct: struct BlockConstraint { unsigned Number; ///< Basic block number (from MBB::g...
2013 Jun 12
0
[LLVMdev] RFC - Profile Guided Optimization in LLVM
...abilities? > > Agreed. An important consideration here is WPO vs. LTO vs. TU-at-a-time call graphs. > > FWIW, the greedy register allocator’s live range splitting algorithm is designed to consume profile information so it can push spill code into cold blocks. The primary interface is SpillPlacement::getBlockFrequency() which currently returns an estimate based on loop depth only. > > It doesn't use MachineBlockFrequency? If it does, it will get a lot more than loop depth: __builtin_expect, cold function attribute, and static branch heuristics. If it doesn't it should, and then...
2015 Oct 05
2
handling "overlapping" register classes
I'm working on generating code for a machine that has a register set kind of like the 68000. For those who don't recall, the 68K has 8 Data registers that can be used for ordinary integer instructions like add, subtract, multiply, shift, etc., and 8 Address registers that can be use for integer addition and a few other things, especially base registers for addressing modes. The Data
2013 Jun 17
2
[LLVMdev] BlockFrequency spill weights
...getSpillWeight(bool isDef, bool isUse, BlockFrequency freq) { > + return (isDef + isUse) * freq.getFrequency(); > } > > This computation can overflow. Yep, I went down the easy route and converted it to floating point arithmetic. Is that OK here? > > @@ -178,9 +180,10 @@ bool SpillPlacement::runOnMachineFunction(MachineFunction &mf) { > > // Compute total ingoing and outgoing block frequencies for all bundles. > BlockFrequency.resize(mf.getNumBlockIDs()); > + MachineBlockFrequencyInfo &MBFI = getAnalysis<MachineBlockFrequencyInfo>(); > for (Machine...
2013 Jun 12
3
[LLVMdev] RFC - Profile Guided Optimization in LLVM
On Wed, Jun 12, 2013 at 3:10 PM, Jakob Stoklund Olesen <stoklund at 2pi.dk>wrote: > It predates the block frequency interface. It just needs to be hooked up, > patches welcome. It would also be nice to remove the floating point > computations from the spill placement code. Cool, if Diego doesn't beat me to it, I may send you a patch as that seems easy and obviously
2013 Jun 12
6
[LLVMdev] RFC - Profile Guided Optimization in LLVM
I have started looking at the state of PGO (Profile Guided Optimization) in LLVM.**I want to discuss my high-level plan and make sure I'm not missing anything interesting out. I appreciate any feedback on this, pointers to existing work, patches and anything related to PGO in LLVM. I will be keeping changes to this plan in this web document
2013 Aug 07
5
[LLVMdev] Live range splitting with Ising models
...g model, and that means we can get away with using a very fast, very stupid algorithm (stolen from Hopfield networks) that simply converges on a nearby local minimum. We don’t actually need to use simulated annealing or other fancy optimization algorithms. Stupid Ising model optimizer: lib/CodeGen/SpillPlacement.cpp More on D-Wave: http://arstechnica.com/science/2013/08/d-waves-black-box-starts-to-open-up/ More on Ising models and Hopfield networks: http://www.inference.phy.cam.ac.uk/mackay/itila/ Thanks, /jakob
2018 Dec 05
3
Strange regalloc behaviour: one more available register causes much worse allocation
enableAdvancedRASplitCost() does the same thing as ConsiderLocalIntervalCost, but as a subtarget option instead of a command-line option, and as I’ve said it doesn’t help because it’s a non-local interval causing the eviction chain (RAGreedy::splitCanCauseEvictionChain only considers the local interval for a single block, and it’s unclear to me how to make it handle a non-local interval). John
2015 Jul 29
1
[LLVMdev] Error when i am using command make -j4 command in cygwin to compile safecode
...for Release+Asserts build llvm[2]: Compiling ShadowStackGCLowering.cpp for Release+Asserts build llvm[2]: Compiling ShrinkWrap.cpp for Release+Asserts build llvm[2]: Compiling SjLjEHPrepare.cpp for Release+Asserts build llvm[2]: Compiling SlotIndexes.cpp for Release+Asserts build llvm[2]: Compiling SpillPlacement.cpp for Release+Asserts build llvm[2]: Compiling SplitKit.cpp for Release+Asserts build llvm[2]: Compiling StackColoring.cpp for Release+Asserts build llvm[2]: Compiling StackMapLivenessAnalysis.cpp for Release+Asserts build llvm[2]: Compiling StackMaps.cpp for Release+Asserts build llvm[2]: Compil...