[llvm-commits] [llvm] r135625 - in /llvm/trunk: lib/Transforms/Scalar/LICM.cpp test/Transforms/LICM/scalar-promote-memmodel.ll

Eli Friedman eli.friedman at gmail.com
Wed Jul 20 16:42:26 PDT 2011


On Wed, Jul 20, 2011 at 3:20 PM, Cameron Zwarich <zwarich at apple.com> wrote:
> Did you do any performance testing for this change?

I don't have full performance results, but I took a brief look, and
this transformation triggers pretty rarely.

-Eli

> Cameron
>
> On Jul 20, 2011, at 2:37 PM, Eli Friedman wrote:
>
>> Author: efriedma
>> Date: Wed Jul 20 16:37:47 2011
>> New Revision: 135625
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=135625&view=rev
>> Log:
>> Bring LICM into compliance with the new "Memory Model for Concurrent Operations" in LangRef.
>>
>>
>> Added:
>>    llvm/trunk/test/Transforms/LICM/scalar-promote-memmodel.ll
>> Modified:
>>    llvm/trunk/lib/Transforms/Scalar/LICM.cpp
>>
>> Modified: llvm/trunk/lib/Transforms/Scalar/LICM.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LICM.cpp?rev=135625&r1=135624&r2=135625&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Transforms/Scalar/LICM.cpp (original)
>> +++ llvm/trunk/lib/Transforms/Scalar/LICM.cpp Wed Jul 20 16:37:47 2011
>> @@ -151,6 +151,11 @@
>>     ///
>>     bool isSafeToExecuteUnconditionally(Instruction &I);
>>
>> +    /// isGuaranteedToExecute - Check that the instruction is guaranteed to
>> +    /// execute.
>> +    ///
>> +    bool isGuaranteedToExecute(Instruction &I);
>> +
>>     /// pointerInvalidatedByLoop - Return true if the body of this loop may
>>     /// store into the memory location pointed to by V.
>>     ///
>> @@ -577,6 +582,10 @@
>>   if (Inst.isSafeToSpeculativelyExecute())
>>     return true;
>>
>> +  return isGuaranteedToExecute(Inst);
>> +}
>> +
>> +bool LICM::isGuaranteedToExecute(Instruction &Inst) {
>>   // Otherwise we have to check to make sure that the instruction dominates all
>>   // of the exit blocks.  If it doesn't, then there is a path out of the loop
>>   // which does not execute this instruction, so we can't hoist it.
>> @@ -713,33 +722,36 @@
>>
>>       // If there is an non-load/store instruction in the loop, we can't promote
>>       // it.
>> -      unsigned InstAlignment;
>> -      if (LoadInst *load = dyn_cast<LoadInst>(Use)) {
>> +      if (isa<LoadInst>(Use)) {
>>         assert(!cast<LoadInst>(Use)->isVolatile() && "AST broken");
>> -        InstAlignment = load->getAlignment();
>>       } else if (StoreInst *store = dyn_cast<StoreInst>(Use)) {
>>         // Stores *of* the pointer are not interesting, only stores *to* the
>>         // pointer.
>>         if (Use->getOperand(1) != ASIV)
>>           continue;
>> -        InstAlignment = store->getAlignment();
>> +        unsigned InstAlignment = store->getAlignment();
>>         assert(!cast<StoreInst>(Use)->isVolatile() && "AST broken");
>> -      } else
>> -        return; // Not a load or store.
>>
>> -      // If the alignment of this instruction allows us to specify a more
>> -      // restrictive (and performant) alignment and if we are sure this
>> -      // instruction will be executed, update the alignment.
>> -      // Larger is better, with the exception of 0 being the best alignment.
>> -      if ((InstAlignment > Alignment || InstAlignment == 0)
>> -          && (Alignment != 0))
>> -        if (isSafeToExecuteUnconditionally(*Use)) {
>> -          GuaranteedToExecute = true;
>> -          Alignment = InstAlignment;
>> -        }
>> +        // Note that we only check GuaranteedToExecute inside the store case
>> +        // so that we do not introduce stores where they did not exist before
>> +        // (which would break the LLVM concurrency model).
>> +
>> +        // If the alignment of this instruction allows us to specify a more
>> +        // restrictive (and performant) alignment and if we are sure this
>> +        // instruction will be executed, update the alignment.
>> +        // Larger is better, with the exception of 0 being the best alignment.
>> +        if ((InstAlignment > Alignment || InstAlignment == 0)
>> +            && (Alignment != 0))
>> +          if (isGuaranteedToExecute(*Use)) {
>> +            GuaranteedToExecute = true;
>> +            Alignment = InstAlignment;
>> +          }
>> +
>> +        if (!GuaranteedToExecute)
>> +          GuaranteedToExecute = isGuaranteedToExecute(*Use);
>>
>> -      if (!GuaranteedToExecute)
>> -        GuaranteedToExecute = isSafeToExecuteUnconditionally(*Use);
>> +      } else
>> +        return; // Not a load or store.
>>
>>       LoopUses.push_back(Use);
>>     }
>>
>> Added: llvm/trunk/test/Transforms/LICM/scalar-promote-memmodel.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LICM/scalar-promote-memmodel.ll?rev=135625&view=auto
>> ==============================================================================
>> --- llvm/trunk/test/Transforms/LICM/scalar-promote-memmodel.ll (added)
>> +++ llvm/trunk/test/Transforms/LICM/scalar-promote-memmodel.ll Wed Jul 20 16:37:47 2011
>> @@ -0,0 +1,37 @@
>> +; RUN: opt < %s -basicaa -licm -S | FileCheck %s
>> +
>> +; Make sure we don't hoist a conditionally-executed store out of the loop;
>> +; it would violate the concurrency memory model
>> +
>> + at g = common global i32 0, align 4
>> +
>> +define void @bar(i32 %n, i32 %b) nounwind uwtable ssp {
>> +entry:
>> +  br label %for.cond
>> +
>> +for.cond:                                         ; preds = %for.inc, %entry
>> +  %i.0 = phi i32 [ 0, %entry ], [ %inc5, %for.inc ]
>> +  %cmp = icmp slt i32 %i.0, %n
>> +  br i1 %cmp, label %for.body, label %for.end
>> +
>> +for.body:                                         ; preds = %for.cond
>> +  %tobool = icmp eq i32 %b, 0
>> +  br i1 %tobool, label %for.inc, label %if.then
>> +
>> +if.then:                                          ; preds = %for.body
>> +  %tmp3 = load i32* @g, align 4
>> +  %inc = add nsw i32 %tmp3, 1
>> +  store i32 %inc, i32* @g, align 4
>> +  br label %for.inc
>> +
>> +; CHECK: load i32*
>> +; CHECK-NEXT: add
>> +; CHECK-NEXT: store i32
>> +
>> +for.inc:                                          ; preds = %for.body, %if.then
>> +  %inc5 = add nsw i32 %i.0, 1
>> +  br label %for.cond
>> +
>> +for.end:                                          ; preds = %for.cond
>> +  ret void
>> +}
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>




More information about the llvm-commits mailing list