Passes
- General StreamTensor Passes
- 'tensor_ext' Dialect Passes
- 'dataflow' Dialect Passes
- -streamtensor-convert-itensor-empty-to-instance
- -streamtensor-convert-tensor-to-kernel
- -streamtensor-ensure-itensor-single-use
- -streamtensor-fold-itensor
- -streamtensor-lower-itensor-to-stream
- -streamtensor-materialize-kernel
- -streamtensor-merge-task-chain
- -streamtensor-pack-kernel-interface
- -streamtensor-simplify-task-structure
- -streamtensor-vectorize-itensor
- -streamtensor-widen-kernel-interface
- 'runtime' Dialect Passes
- 'hls' Dialect Passes
General StreamTensor Passes¶
-streamtensor-comprehensive-bufferize¶
Comprehensively bufferize the program
-streamtensor-fuse-linalg-fill¶
Fuse linalg fill op into generic op
-streamtensor-linalg-fake-quantize¶
Convert to quantized model (only for testing use)
Options¶
-quan-bits : Number of bits for quantization of non-constant values
-const-quan-bits : Number of bits for quantization of constant values
-streamtensor-raise-scf-to-affine¶
Raise SCF to affine
-streamtensor-strip-annotations¶
Strip all annotations with the given name
Options¶
-annotation-name :
-streamtensor-transform-interpreter¶
Interprete transform sequence with the given entry point
Options¶
-debug-payload-root-tag : Select the operation with 'transform.target_tag' attribute having the given value as payload IR root. If empty select the pass anchor operation as the payload IR root.
-disable-expensive-checks : Disable expensive checks in the interpreter for a faster run.
-entry-point : Entry point of the pass pipeline.
-delete-entry-point : Delete the entry point after transformation.
'tensor_ext' Dialect Passes¶
-streamtensor-convert-tensor-empty-to-instance¶
Convert tensor empty op to instance op
-streamtensor-convert-tensor-primitive-to-linalg¶
Convert tensor primitive op to linalg op
-streamtensor-decompose-tensor-ops¶
Decompose tensor ops
-streamtensor-lower-pack-unpack¶
Lower tensor pack/unpack ops
Options¶
-constant-folding : Whether to apply constant folding
-streamtensor-lower-quantize-ops¶
Lower quantize ops to normal arithmetic ops
-streamtensor-raise-extract-slice-to-chunk¶
Raise extract_slice to chunk if possible
'dataflow' Dialect Passes¶
-streamtensor-convert-itensor-empty-to-instance¶
Convert itensor empty op to instance op
-streamtensor-convert-tensor-to-kernel¶
Convert tensor ops to kernel ops
-streamtensor-ensure-itensor-single-use¶
Ensure each itensor has a single use
-streamtensor-fold-itensor¶
Fold trivial itensor communications
-streamtensor-lower-itensor-to-stream¶
Lower itensor operations to stream operations
-streamtensor-materialize-kernel¶
Materialize kernel ops to task ops
-streamtensor-merge-task-chain¶
Merge chain of tasks into a single task
-streamtensor-pack-kernel-interface¶
Pack/unpack the interface of kernel ops
-streamtensor-simplify-task-structure¶
Simplify task structure
-streamtensor-vectorize-itensor¶
Vectorize itensor elements
Options¶
-max-vectorize-bitwidth : Maximum bitwidth of vectorization
-streamtensor-widen-kernel-interface¶
Widen/unwiden the interface of kernel ops
Options¶
-max-widen-bitwidth : Maximum bitwidth of widening
'runtime' Dialect Passes¶
-streamtensor-convert-memref-to-pointer¶
Convert memref views to pointer passing for C++ emission
-streamtensor-generate-runtime-host-func¶
Generate runtime host function
Options¶
-host-func-name : Name of the host function
-timeout-in-ms : Timeout in ms when waiting the kernel run to complete
'hls' Dialect Passes¶
-streamtensor-convert-dataflow-to-func¶
Convert structural dataflow to function for C++ emission
Options¶
-create-kernel-wrapper : Whether to create kernel wrapper function
-streamtensor-generate-connectivity¶
Generate connectivity ops from kernel tasks
Options¶
-num-hbm-port : Number of system port for HBM access
-streamtensor-generate-directive¶
Generate HLS directives
Options¶
-num-uram : Number of URAMs available on chip
-num-bram : Number of BRAMs available on chip
-size-lutram-in-bit : The size of LUTRAMs in bits
-streamtensor-materialize-directive¶
Materialize HLS directives