[llvm-commits] [llvm] r128332 - /llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp
Bill Wendling
isanbard at gmail.com
Sat Mar 26 02:31:18 PDT 2011
It was added earlier (see r128319). And there's all ready a test directory for DSE.
-bw
On Mar 26, 2011, at 1:12 AM, Cameron Zwarich wrote:
> Test?
>
> Cameron
>
> On 2011-03-26, at 1:02 AM, Bill Wendling wrote:
>
>> Author: void
>> Date: Sat Mar 26 03:02:59 2011
>> New Revision: 128332
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=128332&view=rev
>> Log:
>> Rework the logic that determines if a store completely overlaps an ealier store.
>>
>> There are two ways that a later store can comletely overlap a previous store:
>>
>> 1. They both start at the same offset, but the earlier store's size is <= the
>> later's size, or
>> 2. The earlier store's offset is > the later's offset, but it's offset + size
>> doesn't extend past the later's offset + size.
>>
>> Modified:
>> llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp
>>
>> Modified: llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp?rev=128332&r1=128331&r2=128332&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp (original)
>> +++ llvm/trunk/lib/Transforms/Scalar/DeadStoreElimination.cpp Sat Mar 26 03:02:59 2011
>> @@ -340,26 +340,34 @@
>> // Okay, we have stores to two completely different pointers. Try to
>> // decompose the pointer into a "base + constant_offset" form. If the base
>> // pointers are equal, then we can reason about the two stores.
>> - int64_t Off1 = 0, Off2 = 0;
>> - const Value *BP1 = GetPointerBaseWithConstantOffset(P1, Off1, TD);
>> - const Value *BP2 = GetPointerBaseWithConstantOffset(P2, Off2, TD);
>> + int64_t EarlierOff = 0, LaterOff = 0;
>> + const Value *BP1 = GetPointerBaseWithConstantOffset(P1, EarlierOff, TD);
>> + const Value *BP2 = GetPointerBaseWithConstantOffset(P2, LaterOff, TD);
>>
>> // If the base pointers still differ, we have two completely different stores.
>> if (BP1 != BP2)
>> return false;
>> -
>> - // Otherwise, we might have a situation like:
>> - // store i16 -> P + 1 Byte
>> - // store i32 -> P
>> - // In this case, we see if the later store completely overlaps all bytes
>> - // stored by the previous store.
>> - if (Off1 < Off2 || // Earlier starts before Later.
>> - Off2 < 0 || // Later is -.
>> - Off1+Earlier.Size > Off2+Later.Size) // Earlier goes beyond Later.
>> - return false;
>>
>> - // Otherwise, we have complete overlap.
>> - return true;
>> + // The later store completely overlaps the earlier store if:
>> + //
>> + // 1. Both start at the same offset and the later one's size is greater than
>> + // or equal to the earlier one's, or
>> + //
>> + // |--earlier--|
>> + // |-- later --|
>> + //
>> + // 2. The earlier store has an offset greater than the later offset, but which
>> + // still lies completely within the later store.
>> + //
>> + // |--earlier--|
>> + // |----- later ------|
>> + if ((EarlierOff == LaterOff && Earlier.Size <= Later.Size) ||
>> + (EarlierOff > LaterOff &&
>> + EarlierOff + Earlier.Size <= LaterOff + Later.Size))
>> + return true;
>> +
>> + // Otherwise, they don't completely overlap.
>> + return false;
>> }
>>
>> /// isPossibleSelfRead - If 'Inst' might be a self read (i.e. a noop copy of a
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
> _______________________________________________
> 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