[llvm] [VectorCombine] New folding pattern for extract/binop/shuffle chains (PR #145232)
Rajveer Singh Bharadwaj via llvm-commits
llvm-commits at lists.llvm.org
Mon Jun 30 03:57:22 PDT 2025
================
@@ -2988,6 +2989,240 @@ bool VectorCombine::foldShuffleFromReductions(Instruction &I) {
return foldSelectShuffle(*Shuffle, true);
}
+bool VectorCombine::foldShuffleChainsToReduce(Instruction &I) {
+ auto *EEI = dyn_cast<ExtractElementInst>(&I);
+ if (!EEI)
+ return false;
+
+ std::queue<Value *> InstWorklist;
+ Value *InitEEV = nullptr;
+
+ unsigned int CommonCallOp = 0, CommonBinOp = 0;
+
+ bool IsFirstCallOrBinInst = true;
+ bool ShouldBeCallOrBinInst = true;
+
+ SmallVector<Value *, 3> PrevVecV(3, nullptr);
+ int64_t ShuffleMaskHalf = -1, ExpectedShuffleMaskHalf = 1;
+ int64_t VecSize = -1;
+
+ Value *VecOp;
+ if (!match(&I, m_ExtractElt(m_Value(VecOp), m_Zero())))
+ return false;
+
+ auto *FVT = dyn_cast<FixedVectorType>(VecOp->getType());
+ if (!FVT)
+ return false;
+
+ VecSize = FVT->getNumElements();
+ if (VecSize < 2 || (VecSize % 2) != 0)
----------------
Rajveer100 wrote:
At the moment it works for even integers (not limited to powers of 2). For odd, I will need to check again for a suitable pattern and resolve the parity.
https://github.com/llvm/llvm-project/pull/145232
More information about the llvm-commits
mailing list