search for: instsimplify

Today, InstSimplify calls computeKnownBits to try to fold to a constant ONLY if it can't find any other simplifications for an instruction. I think this means if we are able to find an earlier simplification we are LESS aggressive about creating a constant because we won't even look at known bits. This seems...

...king because building a tool to generate the instruction simplifier > automatically is relatively high on my TODO list and I don't want to duplicate > work :) The auto-simplifier is a very simple kind of super-optimizer. It tries to find optimizations suitable for InstructionSimplify [instsimplify]. The difference between instsimplify and instcombine is that instcombine is allowed to create new instructions while instsimplify needs to simplify expressions to already existing subexpressions present inside the existing expression. For example instsimplify can turn this (X + Y) - X into...

...rying to find a solution for this bug http://llvm.org/bugs/show_bug.cgi?id=11235 , I came to the conclusion that the following things can happen while optimizing: - after gvn, I get new constants which are inserted into br and add/sum/..., so there should be at least one more jump-threading and/or instsimplify - after instsimplify, I get new constants which are inserted into br, so there should be one more jump-threading pass - after jump-threading, new bigger blocks occur with redundant loads which need an other gvn pass At least for -O3 we will need those optimizations. Is there a kind of "I ne...

This can't be an instsimplify though? The values we want in these cases do not exist already: %res = or i8 %b, %a %res = or i1 %cmp, %c On Thu, Jul 13, 2017 at 5:10 PM, Daniel Berlin <dberlin at dberlin.org> wrote: > > > On Thu, Jul 13, 2017 at 2:12 PM, Sanjay Patel <spatel at rotateright.com> >...

...vm.org/D57504 The prototype demonstrates: -  Predicated vector intrinsics with an explicit mask and vector length parameter on IR level. -  First-class predicated SDNodes on ISel level. Mask and vector length are value operands. -  An incremental strategy to generalize PatternMatch/InstCombine/InstSimplify and DAGCombiner to work on both regular instructions and EVL intrinsics. -  DAGCombiner example: FMA fusion. -  InstCombine/InstSimplify example: FSub pattern re-writes. -  Early experiments on the LNT test suite (Clang static release, O3 -ffast-math) indicate that compile time on non-EVL IR is n...

...his bug > http://llvm.org/bugs/show_bug.cgi?id=11235 , > I came to the conclusion that the following things can happen while > optimizing: >  - after gvn, I get new constants which are inserted into br and > add/sum/..., so there should be at least one more jump-threading and/or > instsimplify >  - after instsimplify, I get new constants which are inserted into br, so > there should be one more jump-threading pass >  - after jump-threading, new bigger blocks occur with redundant loads which > need an other gvn pass > > At least for -O3 we will need those optimizations....

I'm looking for some opportunities to contribute to the LLVM core without too big of a learning curve. The open projects page<http://llvm.org/OpenProjects.html#misc_imp>suggests moving logic from the instruction combining transformation to the InstructionSimplify analysis. But it looks like some work has been done there since the projects page was last edited. Is there more benefit to

On Wed, Dec 12, 2012 at 01:59:55PM -0800, Dan Gohman wrote: > The bug here isn't in clang's use of noalias or in BasicAliasAnalysis' > implementation of noalias; it's in the code that's optimizing the > icmp. Let's come back to this. The attached patch decouples InstSimplify from the alias analysis and provides the conservative logic for when pointers are not equal. The only really problematic part would be InlineCost.cc, since that could try to replace NoAlias/ByVal arguments with allocas in the same function. For the moment, the inline cost estimation doesn't use...

...es. 3. I assumed that matching an (xor icmp, icmp) pattern can be done efficiently in InstCombine. In fact, I knew that we already did that match, so there is zero incremental cost to find this pattern in InstCombine. 4. I knew that we already handle related and-of-cmps and or-of-cmps patterns in InstSimplify/InstCombine. 5. Based on that, I have proposed a patch that mostly uses existing logic in those passes to solve this bug in the most general way I am aware of. It makes 2 calls to InstSimplify to find a potential fold before morphing/creating new instructions that may get folded by other existing...

...ind a solution for this bug http://llvm.org/bugs/show_bug.cgi?id=11235 , > I came to the conclusion that the following things can happen while optimizing: > - after gvn, I get new constants which are inserted into br and add/sum/..., so there should be at least one more jump-threading and/or instsimplify > - after instsimplify, I get new constants which are inserted into br, so there should be one more jump-threading pass > - after jump-threading, new bigger blocks occur with redundant loads which need an other gvn pass > > At least for -O3 we will need those optimizations. > Is t...

On Wed, Jan 16, 2013 at 12:53:55AM +0100, Joerg Sonnenberger wrote: > Let's come back to this. The attached patch decouples InstSimplify from > the alias analysis and provides the conservative logic for when pointers > are not equal. Let's take the version with the cleanup from IRC. *sigh* Dealing with too many copies. Joerg -------------- next part -------------- Index: lib/Analysis/InlineCost.cpp ======================...

Hi Benjamin, > LLVM misses this optimization because ScalarEvolution's ComputeExitLimitFromICmp doesn't handle signed<= (SLE) and thus can't compute the number of times the loop is executed. I wonder if there's a reason for this, it seems like something simple to add. instsimplify could also be enhanced to clean it up in this particular case, but it would be better to make scev smarter. Ciao, Duncan.

...go all the way to fold if > both arguments are constants (I guess). > > So, is the idea to transform ConstantFolding into a new function pass, > or to recode another pass (like InstCombine) to deal with such things? No new passes; the idea is that all constant folding should be done by InstSimplify. -Eli

Hello, My name is Siddharth. I’m a student studying math and computer science at the University of Toronto. I’m interested in learning more about the LLVM compiler through GSoC. I have no prior experience with LLVM, but I have a strong interest in compilers, and I look forward to learning more. I would like to pursue one of two potential projects: rewriting bugpoint, and moving optimizations

...if some or all uses of the select would be simplified by InstructionSimplify, > and not perform the transform if so. Instead of turning this into a phase ordering issue, I'd rather increase the power of the folding logic to catch the general form. In this case, instead of handling this in instsimplify, why not handle this in reassociate? This is conceptually no different than (x & y & x) -> (x & y), it's just that in this case you have: (x & y & (x or -1)) which should fold to (x & y) just the same. -Chris

...the result of a value number > analysis (NewGVN) to assign numbers. > > While the problem NewGVN solves is that of finding all the Herbrand > equivalences (structural equivalence, same operation and operand > structurally congruent), it also does a canonicalization step calling > InstSimplify to canonicalize (so, in this sense it catches more > stuffs, e.g.: > > %x = add %a, 0 > %y = sub %b, 0 > not equivalent > > instSimplify(%y) -> add %a, 0 > now they're equivalent. > > Note: The current IR outlliner doesn't really use the approach I just &...

On 07/13/2017 06:40 PM, Daniel Berlin via llvm-dev wrote: > Ah yes, it can only return one instruction and you need two. > > NewGVN could still handle it if instsimplify was changed to return the > right binary operators because it has infrastructure that can handle > insertion. > (it would need a little work). > > > (At this point, InstCombine seems to have become worse to me than > GCC's combining infrastructure, so i'd really like...

...ussed with Dan) is leveraging the result of a value number analysis (NewGVN) to assign numbers. While the problem NewGVN solves is that of finding all the Herbrand equivalences (structural equivalence, same operation and operand structurally congruent), it also does a canonicalization step calling InstSimplify to canonicalize (so, in this sense it catches more stuffs, e.g.: %x = add %a, 0 %y = sub %b, 0 not equivalent instSimplify(%y) -> add %a, 0 now they're equivalent. Note: The current IR outlliner doesn't really use the approach I just described, yet, but it's not hard to imagine ex...

...t type is one of i8, i16, etc. So vectors of funky types, or not so funky types like i1, are no longer matched. I noticed this while working on PR11948. The good thing is that the testcase there no longer crashes the compiler because earlier patterns now fail to match :) The bad thing is that a -instsimplify testcase like the following no longer passes: define <2 x i1> @vectorselect1(<2 x i1> %cond) { ; CHECK: @vectorselect1 %invert = xor <2 x i1> %cond, <i1 1, i1 1> %s = select <2 x i1> %invert, <2 x i32> <i32 0, i32 0>, <2 x i32> <i32...

On Wed, Dec 12, 2012 at 1:26 PM, Joerg Sonnenberger <joerg at britannica.bec.de> wrote: > On Wed, Dec 12, 2012 at 11:01:01AM -0800, Dan Gohman wrote: >> > Is that >> > assumption violated if I explicitly cast away const and pass the result >> > to a function with NoAlias argument? >> >> Not immediately, no. It means that you can't access the