Merge pull request #5591 from apple/eng/davide/slp-fix-october-branch

dcci · web-flow · commit efa83a2cdeaf · 2022-11-11T13:18:18.000-08:00
[SLP]Fix PR58863: Mask index beyond mask size for non-power-2 inserte…
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -6457,7 +6457,8 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
             if (Mask[I] != UndefMaskElem)
               Mask[I] = I + VecSz;
           for (unsigned I = OffsetEnd + 1 - Offset; I < VecSz; ++I)
-            Mask[I] = InMask.test(I) ? UndefMaskElem : I;
+            Mask[I] =
+                ((I >= InMask.size()) || InMask.test(I)) ? UndefMaskElem : I;
           Cost += TTI->getShuffleCost(TTI::SK_PermuteTwoSrc, InsertVecTy, Mask);
         }
       }

Original file line number	Diff line number	Diff line change
`@@ -6457,7 +6457,8 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,`
`6457`	`6457`	`if (Mask[I] != UndefMaskElem)`
`6458`	`6458`	`Mask[I] = I + VecSz;`
`6459`	`6459`	`for (unsigned I = OffsetEnd + 1 - Offset; I < VecSz; ++I)`
`6460`		`- Mask[I] = InMask.test(I) ? UndefMaskElem : I;`
	`6460`	`+ Mask[I] =`
	`6461`	`+ ((I >= InMask.size()) \|\| InMask.test(I)) ? UndefMaskElem : I;`
`6461`	`6462`	`Cost += TTI->getShuffleCost(TTI::SK_PermuteTwoSrc, InsertVecTy, Mask);`
`6462`	`6463`	`}`
`6463`	`6464`	`}`