search for: ullman

I'm looking over the RR list scheduler and am wondering about how the Sethi-Ullman numbers are used. The comments and the code indicate that the node with the lowest S-I number is scheduled first. Traditionally S-I scheduling schedules the node with the highest S-I number first. What am I missing? -Dave

David Greene <dag at cray.com> writes: > What am I missing? Perhaps that this is bottom-up scheduling, so picking the lower-numbered node is the same as picking the higher-numbered node in a top-down scheduler? Just looking for confirmation. Thanks! -Dave

Dear LLVM'ers, I am browsing the instruction schedulers available in llc, and there are many: -pre-RA-sched = {default, none, simple, simple-noitin, list-burr, list-tdrr, list-td} I looked into the sources in lib/CodeGen/SelectionDAG, and I could find implementation of Sethi-Ullman numbering, list scheduling, etc. Now, I wish I could find some comparison between the schedulers. I guess you guys probably have some tests somewhere. Basically, I wish I could know when it is better to use a scheduler, and when it is better to use another. Even if someone could point me a...

...r as compiler texts, there are many newer texts to recommend as just about all the major optimization passes are done differently after SSA-form appeared in about 1991. However, for adding Cortex-A8 support, I don't think you'll need to know anything more than the old Dragon book (Aho/Sethi/Ullman) has in it. Instruction selection is done in LLVM using SelectionDAGs and the TableGen tool and it's domain-specific language. Start by looking at llvm/lib/Target/ARM/ARMInstrInfo.td and seeing how the basic instructions are matched. For armv7/armv7a, a new file will want to be created similar...

...register pressure--heuristics can only hurt--making it difficult to distinguish between a lucky scheduler and a good scheduler. It's not surprising that SelectionDAG scheduling with BURR reduces spill code on average. It is fully register pressure aggressive. It gives highest priority to Sethi-Ullman number, which is typically nonsense, but does prevent some of the worst register pressure situations. It then does an expensive check to determine the shortest live range. This is also inaccurate, but on average reduces pressure. The reason we switched from BURR to ILP a couple years ago was that...

...there are many newer texts to recommend as > just about all the major optimization passes are done differently > after SSA-form appeared in about 1991. However, for adding Cortex-A8 > support, I don't think you'll need to know anything more than the old > Dragon book (Aho/Sethi/Ullman) has in it. Instruction selection is > done in LLVM using SelectionDAGs and the TableGen tool and it's > domain-specific language. > > Start by looking at llvm/lib/Target/ARM/ARMInstrInfo.td and seeing how > the basic instructions are matched. For armv7/armv7a, a new file will &...

Hello, I'm looking to learn more about what is available for memory versioning or inducing some kind of SSA for memory objects. I found this page: http://llvm.org/docs/tutorial/OCamlLangImpl7.html#memory-in-llvm Which says: "[LLVM] does not require (or permit) memory objects to be in SSA form.... In LLVM, instead of encoding dataflow analysis of memory into the LLVM IR, it

...done at LLVM store instructions (e.g., two pointers could be stored into the same memory location). Back-tracking def-use chains isn't sufficient for finding this kind of merging, and since the points-to analysis would probably use classical data-flow analysis (as described in the Kam and Ullman paper), finding the exact merge point after the analysis has been done may not be possible. 2) Inter-procedural data-flow through function arguments and return values. You can generally track these through SSA values, but indirect function calls and vararg functions can make this difficult. -...

...OSes fairly well, or at least I've written 3, but it's true that I've been out of the software business for the last 4 years as I recover from severe medical problems. I never concentrated on compilers in school, but just after graduating, I had a chance to buy a new copy of Aho/Setho/Ullman's compiler book, and I'm now reading it. I did take the compiler course, but it seemed rather basic to me. The OS thing they let me do as indiependent study, with a prof watching me like I was doing a doctoral thing, and I got a whole lot out of that. So, I was wondering if someone has a...

...there are many newer texts to recommend as > just about all the major optimization passes are done differently > after SSA-form appeared in about 1991. However, for adding Cortex-A8 > support, I don't think you'll need to know anything more than the old > Dragon book (Aho/Sethi/Ullman) has in it. Instruction selection is > done in LLVM using SelectionDAGs and the TableGen tool and it's > domain-specific language. > > Start by looking at llvm/lib/Target/ARM/ARMInstrInfo.td and seeing how > the basic instructions are matched. For armv7/armv7a, a new file will &...

...instructions (e.g., two pointers > could be stored into the same memory location). Back-tracking def-use > chains isn't sufficient for finding this kind of merging, and since > the points-to analysis would probably use classical data-flow analysis > (as described in the Kam and Ullman paper), finding the exact merge > point after the analysis has been done may not be possible. > > 2) Inter-procedural data-flow through function arguments and return > values. You can generally track these through SSA values, but > indirect function calls and vararg functions can...

Actually I have collected some pointer information in the form [ p -> a,c ]. Now suppose at some node I have information as [p->a,c]. Now i want to find a merge node above this node where this information is actually geting merged. So if I get a merge node above this, I can check in its predecessors if their out has only [p->a] or [p->c] and if not so then I will look for the next

Hi, - Cortex-A8 needs a specific instruction scheduler as dual issue forces you to interleave some instructions to allow to run two instructions in the same cycle for the best performance (Cortex-A9 is out-of-order so dual issue is not an issue (!) for performance). - Cortex-A8/A9 have several useful new instructions : for instance, bit operations like bitfield insertion/extraction or having

...9;ll feel better about the whole thing..OpenBSD is really nice with the pro-police stack and would like to see an alternative to the GCC only compiler chain of tools especially as it is based on a somewhat archaic optiminzation backend and procedural stuff is pretty dated in the RTL layer.. Aho and Ullman can only take you so far.. Then you need to use the FORCE.. LUKE use the force.. of llvm that is. :) cheers all. Jeff Cohen <jeffc at jolt-lang.org> wrote: I just added __OpenBSD__ everywhere __FreeBSD__ was being tested (there were about a dozen places). I suspect we'll have to add...

Hello all, I am wondering the use case of undef. Is there any C code that I could get undef if it is compiled by clang. Thanks -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20161204/7ab81194/attachment.html>

I am reading the Mining of Massive Datasets Book by Rajaraman and Ullman. It has a good explanation of Recommendation System at Chapter 9. But what are the relationship between 1) SVD (Singular Decomposition) 2) UV-Decomposition 3) NMF (Non-negative Matrix Factorization) In particular, it seems 2) and 3) can be very similar. Is it right? Thanks. -- View this...

hello, what about these functions lex/yacc which can parse and recognize a syntax? thanks _____________________________________________________________________________ [[alternative HTML version deleted]]

...ng for a fast network. We changed the value to 5 ms and recompiled. -> Success! Was this the right way to solve the problem? Or does anyone know a better solution? Ciao Marcus * We build our computers the way we build our cities -- over * time, without a plan, on top of ruins. * (Ellen Ullman, "The dumbing-down of programming")

...ignoring it - by the way, it's impossible to create an unix user with a whitespace in its name, and it's even difficult to create a username with more than 8 characters. We solved that problem by adding user map = /etc/samba/smbusers user level = 7 while our user map file looks like ullman = "maenz ullrich" So far, our solution works pretty good. However, if we want to add all users defined in that Windows domain, we need to maintain a smbusers file with approximately 500 entries. In addition, there are some usernames similar (like: schmid, schmidt, schmitt...) - this...

Thank you so much but could you tell me a little bit more in detail about that you suggested? Sorry, I'm just a greenhorn. Thanks, Seung J. Lee ---- Original message ---- >Date: Sat, 14 Jul 2007 21:26:14 -0500 >From: "David A. Greene" <greened at obbligato.org> >Subject: Re: [LLVMdev] not to break 'for' statement into basic blocks >To: llvmdev at