MLIR: dynamic shared memory

MLIR
1

I am trying to replicate a portion of the research paper that says, "The support to allocate shared memory dynamically is currently absent from MLIR. " titled “HIGH PERFORMANCE GPU CODE GENERATION FOR
MATRIX -MATRIX MULTIPLICATION USING MLIR: SOME
EARLY RESULTS
1
202guA32]CD.sc[1v19131.8012:viXraNavdeep Katel1,2
Vivek Khandelwal1
Uday Bondhugula1,2”

Is it still true?

2

Maybe take a look at gpu.dynamic_shared_memory, this might be what you are looking for. :slight_smile:

1 Like