Hi, this is a very minor thing, I just wanted to signal the presence of a duplicated parameter in the user guide mapping section:
https://kuleuven-micas.github.io/zigzag/mapping.html
In "User-defined mapping constraints", many parameters are repeated.
Specifically, I'm referring to the text:
-
core_allocation: The accelerator core id onto which this ONNX node is mapped (the id provided when creating the core in the hardware description file). Since ZigZag only supports single-core architectures, the core allocation must be set to 1.
-
spatial_mapping: The spatial parallelization strategy to execute the node with (this can be automated through the SpatialMappingGeneratorStage).
-
memory_operand_links: The memory operand links, which link the memory operands (defined in the memory hierarchy of the core) to the layer operands (which are generated in the ONNXModelParserStage and are typically ‘O’, ‘I’, ‘W’ for a convolutional layer). This extra memory mapping is added to allow flexible memory allocation schemes. If left empty, a default value will be used instead.
so the points 4,5, and 6 are the same as the text above
Hi, this is a very minor thing, I just wanted to signal the presence of a duplicated parameter in the user guide mapping section:
https://kuleuven-micas.github.io/zigzag/mapping.html
In "User-defined mapping constraints", many parameters are repeated.
Specifically, I'm referring to the text:
core_allocation: The accelerator core id onto which this ONNX node is mapped (the id provided when creating the core in the hardware description file). Since ZigZag only supports single-core architectures, the core allocation must be set to 1.
spatial_mapping: The spatial parallelization strategy to execute the node with (this can be automated through the SpatialMappingGeneratorStage).
memory_operand_links: The memory operand links, which link the memory operands (defined in the memory hierarchy of the core) to the layer operands (which are generated in the ONNXModelParserStage and are typically ‘O’, ‘I’, ‘W’ for a convolutional layer). This extra memory mapping is added to allow flexible memory allocation schemes. If left empty, a default value will be used instead.
so the points 4,5, and 6 are the same as the text above