[PATCH] D25172: [MSSA] Initial version of avoiding unnecessary use walks when calling getClobberingMemoryAccess

[Daniel Berlin via llvm-commits]

On Mon, Oct 3, 2016 at 5:04 PM, George Burgess IV <
george.burgess.iv at gmail.com> wrote:

> > First, it looks like your attempt to add basicAA in setupAnalyses is
> not doing anything
>
> r283158 should fix that -- thanks.
>
> > We have no way to catch all the Use.set's that i can find
>
> Can we keep an `OldDef` member that tracks the ID of the Def/Phi we last
> optimized to?
>

So then isOptimized is "OldDef == Def"?


> It makes isOptimized() a bit more expensive, but I can't think of a valid
> transformation that swaps the DefiningAccess of Foo from Bar to Baz to Bar
> while making Bar not the clobber for Foo.
>



>
> On Sun, Oct 2, 2016 at 8:21 PM, Daniel Berlin <dberlin at dberlin.org> wrote:
>
>> Sigh.
>>
>> Two things need your thoughts on george.
>>
>> First, it looks like your attempt to add basicAA in setupAnalyses is not
>> doing anything:
>>  76   AliasResult AAResults::alias(const MemoryLocation &LocA,
>>    77                               const MemoryLocation &LocB) {
>> -> 78    for (const auto &AA : AAs) {
>>    79      auto Result = AA->alias(LocA, LocB);
>>    80      if (Result != MayAlias)
>>    81        return Result;
>> (lldb) p AAs
>> (std::__1::vector<std::__1::unique_ptr<llvm::AAResults::Concept,
>> std::__1::default_delete<llvm::AAResults::Concept> >,
>> std::__1::allocator<std::__1::unique_ptr<llvm::AAResults::Concept,
>> std::__1::default_delete<llvm::AAResults::Concept> > > >) $33 = size=0 {}
>>
>>
>>
>> Simple way to see this:
>>
>> In the unit test, the following:
>>   Type *Int8 = Type::getInt8Ty(C);
>>   Value *AllocaA = B.CreateAlloca(Int8, ConstantInt::get(Int8, 1), "A");
>>   Instruction *SIA = B.CreateStore(ConstantInt::get(Int8, 0), AllocaA);
>>   Value *AllocaB = B.CreateAlloca(Int8, ConstantInt::get(Int8, 1), "B");
>>   Instruction *SIB = B.CreateStore(ConstantInt::get(Int8, 0), AllocaB);
>>   Instruction *LIA = B.CreateLoad(AllocaA);
>>
>> results in:
>> define void @F() {
>>   %A = alloca i8, i8 1
>> ; 1 = MemoryDef(liveOnEntry)
>>   store i8 0, i8* %A
>>   %B = alloca i8, i8 1
>> ; 2 = MemoryDef(1)
>>   store i8 0, i8* %B
>> ; MemoryUse(2)
>>   %1 = load i8, i8* %A
>> }
>>
>> (The load is not being optimized because there is no AA).
>>
>> Second, we can only reset optimized properly if all operand setting is
>> done through setDefinedAccess.
>>
>> We have no way to catch all the Use.set's that i can find.
>>
>> If that's correct, i'll drop the optimized flag, and just keep the part
>> to reset setDefiningAccess.
>>
>> This means we will still pay a cost for calling getclobberingmemoryaccess
>> on loads.
>>
>>
>> On Sun, Oct 2, 2016 at 7:39 PM, Daniel Berlin <dberlin at dberlin.org>
>> wrote:
>>
>>> dberlin added a comment.
>>>
>>> (Unittest that resets the defining access, and calls
>>> getClobberingAccess, coming)
>>>
>>>
>>> https://reviews.llvm.org/D25172
>>>
>>>
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161003/46638802/attachment.html>