-aie-vectorize

Vectorize the output of affine supervectorizer to AIE vector abstraction

Options

-shift                 : Shift parameter for rounding and saturation
-zero-offset           : Zero offset for indicating the location of zeroes in convolution filter (useful for 16x16 scheme)
-dup-factor            : Duplication factor for each value in convolution filter (useful for 8x8 scheme)
-unaligned-loads-check : Enable the unaligned loads check.
-aieml                 : 

-aievec-split-load-ups-chains

Split vector.load + aievec.ups chains to reduce shuffle operations

This pass optimizes chains of vector.load followed by aievec.ups operations for AIE2p targets. Instead of loading a 1024-bit vector and then shuffling it into two halves for separate UPS operations (3 shuffles total), it splits both the load and UPS into two 512-bit halves, requiring only 1 shuffle for concatenation.