Closed sdjordjevicTT closed 3 weeks ago
You can add your sqrt.mlir to automatically generate ttnn files. I added docs on how to do that here: https://docs.tenstorrent.com/tt-mlir/ttrt.html#adding-llvm-lit-config-options-inside-a-mlir-file-to-create-flatbuffer-binaries.
--ttir-load-system-desc="path=%system_desc_path%"
)To test (see https://docs.tenstorrent.com/tt-mlir/ttrt.html#generate-flatbuffer-files-using-llvm-lit)
Inside build/test/ttmlir/Silicon, your sqrt.ttnn file should've been generated.
You can add your sqrt.mlir to automatically generate ttnn files. I added docs on how to do that here: https://docs.tenstorrent.com/tt-mlir/ttrt.html#adding-llvm-lit-config-options-inside-a-mlir-file-to-create-flatbuffer-binaries.
- Copy your .mlir file into /test/ttmlir/Silicon/TTNN
- Add relevant llvm-lit configs (for ex:
--ttir-load-system-desc="path=%system_desc_path%"
)To test (see https://docs.tenstorrent.com/tt-mlir/ttrt.html#generate-flatbuffer-files-using-llvm-lit)
- cmake --build build -- check-ttmlir
- ttrt run build/test/ttmlir/Silicon/TTNN
Inside build/test/ttmlir/Silicon, your sqrt.ttnn file should've been generated.
Great, thanks for this @tapspatel
We need to implement the square root operation end-to-end for the next milestone of bringing up LLama.
Solves: https://github.com/tenstorrent/tt-mlir/issues/418