[llvm-commits] [llvm] r165126 - /llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp

[Hal Finkel]

On Wed, 03 Oct 2012 16:11:44 -0000
Preston Gurd <preston.gurd at intel.com> wrote:

> Author: pgurd
> Date: Wed Oct  3 11:11:44 2012
> New Revision: 165126
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=165126&view=rev
> Log:
> This Patch corrects a problem whereby the optimization to use a
> faster divide instruction (for Intel Atom) was not being done by
> Clang, because the type context used by Clang is not the default
> context.

This is not a specific comment on this patch, but can you (or someone
else) please explain why Clang would be using a different default
context from LLVM?

Thanks again,
Hal

> 
> It fixes the problem by getting the global context types for each
> div/rem instruction in order to compare them against the types in the
> BypassTypeMap.
> 
> Tests for this will be done as a separate patch to Clang.
> 
> Patch by Tyler Nowicki.
> 
> 
> Modified:
>     llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp
> 
> Modified: llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp?rev=165126&r1=165125&r2=165126&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp (original)
> +++ llvm/trunk/lib/Transforms/Utils/BypassSlowDivision.cpp Wed Oct  3
> 11:11:44 2012 @@ -238,14 +238,24 @@ if (!UseDivOp && !UseRemOp)
>        continue;
>  
> +    // Skip division on vector types, only optimize integer
> instructions
> +    if (!J->getType()->isIntegerTy())
> +      continue;
> +
> +    // Get same type in global context
> +    IntegerType *T = cast<IntegerType>(J->getType());
> +    IntegerType *GT = IntegerType::get(getGlobalContext(),
> T->getBitWidth()); +
>      // Continue if div/rem type is not bypassed
> -    DenseMap<Type *, Type *>::const_iterator BT =
> -      BypassTypeMap.find(J->getType());
> -    if (BT == BypassTypeMap.end())
> +    DenseMap<Type *, Type *>::const_iterator BI =
> BypassTypeMap.find(GT);
> +    if (BI == BypassTypeMap.end())
>        continue;
>  
> -    IntegerType *BypassType = cast<IntegerType>(BT->second);
> -    MadeChange |= reuseOrInsertFastDiv(F, I, J, BypassType, UseDivOp,
> +    // Get the bypass type in the original context
> +    IntegerType *GBT = cast<IntegerType>(BI->second);
> +    IntegerType *BT = IntegerType::get(J->getContext(),
> GBT->getBitWidth()); +
> +    MadeChange |= reuseOrInsertFastDiv(F, I, J, BT, UseDivOp,
>                                         UseSignedOp, DivCache);
>    }
>  
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



-- 
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory