FuseMulAddFInBF16Pattern Struct Reference

Pattern to fold an f32 multiply into the add that is its only user, as a vector.fma with bf16 inputs. More...

Inheritance diagram for FuseMulAddFInBF16Pattern:

Collaboration diagram for FuseMulAddFInBF16Pattern:

Public Member Functions
LogicalResult	matchAndRewrite (arith::AddFOp op, PatternRewriter &rewriter) const override

Detailed Description

Pattern to fold an f32 multiply into the add that is its only user, as a vector.fma with bf16 inputs.

Only when both ops allow contraction: the fused result can differ where the product underflows or overflows.

Definition at line 893 of file VectorToVectorConversions.cpp.

Member Function Documentation

matchAndRewrite()

LogicalResult FuseMulAddFInBF16Pattern::matchAndRewrite ( arith::AddFOp op, PatternRewriter & rewriter ) const

inlineoverride

Definition at line 896 of file VectorToVectorConversions.cpp.

---

The documentation for this struct was generated from the following file:

• lib/Dialect/AIEVec/Transforms/VectorToVectorConversions.cpp