similar to: [RFC] Switching to MemorySSA-backed Dead Store Elimination (aka cross-bb DSE)

Hi Florian, Following up on D86967, I missed that all the timings were using the legacy pass manager. Did you do any testing on the compile and run time impact for the new pass manager? Thank you, Alina On Tue, Aug 25, 2020 at 12:51 PM Florian Hahn <florian_hahn at apple.com> wrote: > Hi, > > Thanks for all the responses! > > My understanding is that there were no

Hi Florian, First, thank you for working on this. I'm really glad to see this work so close to being enabled. I think the numbers look good for run time, and the benefits of switching for all configurations are clear. For compile time, the current regressions are noticeable, but not a deal breaker in my opinion. I'm very much in favor of switching in all configurations. To address some

> On Aug 18, 2020, at 16:59, Michael Kruse <llvmdev at meinersbur.de> wrote: > > Thanks for all the work. The reductions in stores look promising. Do you also have performance numbers how much this improves the execution time? Did you observe any regressions where MSSA resulted in fewer removed stores? I did not gather numbers for execution time yet, but I’ll try to share some

Can you share you compile-time and memory footprint measurements at least for CTMark? For a new pass/feature it would be great to share this with the community before you commit. Or did I miss them? Thanks Gerolf > On Jun 27, 2017, at 3:26 PM, Geoff Berry via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > EarlyCSE w/ MemorySSA has been enabled by default as of r306477 >

memoryIsNotModifiedBetween is precisely the sort of expensive walk we shouldn't be doing... I'm surprised it hasn't caused any serious issues yet.  Ideally, what we should be doing is using MemorySSA to find a dependency from the memset: if the closest dependency is the malloc, there aren't any stores between the memset and the malloc.  (But we aren't using MemorySSA in

Can you help a bit? I try to work with DSE but I got the following assert: opt: /home/xbolva00/LLVM/llvm/include/llvm/ADT/Optional.h:176: T* llvm::Optional<T>::getPointer() [with T = llvm::MemoryLocation]: Assertion `Storage.hasVal' failed. static bool eliminateStrlen(CallInst *CI, BasicBlock::iterator &BBI, AliasAnalysis *AA, MemoryDependenceResults

* if (isStringFromCalloc(Dst, TLI)) should be if (!isStringFromCalloc(Dst, TLI)) but still asserting... 2018-05-22 23:06 GMT+02:00 Dávid Bolvanský <david.bolvansky at gmail.com>: > Can you help a bit? > > I try to work with DSE but I got the following assert: > opt: /home/xbolva00/LLVM/llvm/include/llvm/ADT/Optional.h:176: T* > llvm::Optional<T>::getPointer() [with T

It looks like the memoryIsNotModifiedBetween assumes the second argument is a store, or some other instruction supported by MemoryLocation::get.  If you're passing in something else, you'll have to compute the MemoryLocation some other way. (Generally, if you're asking a question about an assertion, please include the whole stack trace; it's hard to guess what's happening

Full stack trace: opt: /home/xbolva00/LLVM/llvm/include/llvm/ADT/Optional.h:176: T* llvm::Optional<T>::getPointer() [with T = llvm::MemoryLocation]: Assertion `Storage.hasVal' failed. Stack dump: 0. Program arguments: opt aaa.ll -dse -S 1. Running pass 'Function Pass Manager' on module 'aaa.ll'. 2. Running pass 'Dead Store Elimination' on function

Looks like there are many overloads for "get". http://llvm.org/doxygen/MemoryLocation_8cpp_source.html But nothing for CallInst. Any suggestions how to do a proper one? I will look at it too. 2018-05-22 23:34 GMT+02:00 Dávid Bolvanský <david.bolvansky at gmail.com>: > Full stack trace: > > opt: /home/xbolva00/LLVM/llvm/include/llvm/ADT/Optional.h:176: T* >

> On May 10, 2019, at 8:59 PM, Vitaly Buka via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > Sorry for delay, I was busy with other stuff. > CTMark results. > > dse is the current DSE. > dsem is my experimental module level DSE. > dsem runs after dse, so it's additionally deleted stores. > > -O3 > dse - Number of stores deleted

"memory accesses between the malloc and the memset without an expensive linear scan of the block/function" (1) do you mean just use "memoryIsNotModifiedBetween" function in DSE to check it? x = maloc(..); memset(x, ...) (2) GetUnderlyingObject would give me Value * (from malloc) ? Also another case: memset(s, 0, len); // len > 1 return strlen(s); // optimize to 0 (3)

On 09/27/2018 12:24 AM, Chris Lattner via llvm-dev wrote: On Sep 26, 2018, at 11:55 AM, Reid Kleckner <rnk at google.com<mailto:rnk at google.com>> wrote: As suggested in the bug, if we were to rewrite these passes to use MemorySSA, this bottleneck would go away. I rebased a patch to do that for DSE, but finishing it off and enabling it by default is probably out of scope for me.

Can you post numbers for how many stores get eliminated from CTMark? > On Apr 16, 2019, at 11:45 AM, Vitaly Buka <vitalybuka at google.com> wrote: > > I tried -Os and effect of new approach significantly increases. > I run regular DSE and immediately myDSE. With -Os myDSE removes more than 50% of DSE number. > Which is expected as -Os inlines less and regular DSE can't

It works with MemoryLocation MemoryLocation::get(const CallInst *CI) { AAMDNodes AATags; CI->getAAMetadata(AATags); const auto &DL = CI->getModule()->getDataLayout(); return MemoryLocation(CI, DL.getTypeStoreSize(CI->getType()), AATags); } Is it fine? :) 2018-05-22 23:56 GMT+02:00 Dávid Bolvanský <david.bolvansky at gmail.com>: > Looks like there are many overloads

IR: define i32 @calloc_strlen_write_between() { %call = tail call noalias i8* @calloc(i32 10, i32 1) store i8 97, i8* %call, align 1 %call1 = tail call i32 @strlen(i8* %call) ret i32 %call1 } static bool eliminateStrlen(CallInst *CI, BasicBlock::iterator &BBI, AliasAnalysis *AA, MemoryDependenceResults *MD, const DataLayout &DL, const TargetLibraryInfo *TLI,

Yeah, sorry for that. Better "It compiles ok (but maybe incorrect code)", not "It works" as I wrote. 2018-05-23 1:08 GMT+02:00 Friedman, Eli <efriedma at codeaurora.org>: > You might want to look more carefully at how you're constructing the > MemoryLocation. The first argument is a pointer, and the second argument > is the number of bytes pointed to by

You might want to look more carefully at how you're constructing the MemoryLocation.   The first argument is a pointer, and the second argument is the number of bytes pointed to by that pointer (or MemoryLocation::UnknownSize if the number of bytes accessed isn't known). More generally, copy-pasting code you don't understand isn't a good idea. -Eli On 5/22/2018 4:02 PM, Dávid

On Tue, Sep 25, 2018 at 12:16 PM Sanjoy Das <sanjoy at playingwithpointers.com> wrote: > Let's not assume a dichotomy between storing a int64 in > llvm::Instruction and bitwise tricks -- they're both ways of caching > position within llvm::Instruction. I think we first need to establish > that we need such a cache in llvm::Instruction/llvm::BasicBlock at > all. >

On Mon, Apr 15, 2019 at 11:02 PM Amara Emerson via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > > > On Apr 15, 2019, at 1:51 PM, Vitaly Buka via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > > > Hi JF, > > > > I've heard that you are interested DSE improvements and maybe we need to be in sync. > > So far I experimented with