edopao
edopao
Four-lines bugfix and associated test case for map_fusion transformation. Without this change, the test would fail in SDFG validation with error: `dace.sdfg.validation.InvalidSDFGEdgeError: Memlet data does not match source or destination...
The Simplify pass introduces an error in the attached SDFG: [sdfg1_before_simplify.json](https://github.com/spcl/dace/files/14434365/sdfg1_before_simplify.json) The error is introduces by ConstantPropagagation subpass. See the SDFG before ([sdfg2_before_constant_propagation.json](https://github.com/spcl/dace/files/14434364/sdfg2_before_constant_propagation.json)) and after ([sdfg3_after_constant_propagation.json](https://github.com/spcl/dace/files/14434363/sdfg3_after_constant_propagation.json)) applying ConstantPropagagation.
With dynamic memlet, a scalar is not always read, which depends on the symbolic subset.
The below SDFG produces invalid CUDA code, after applying `apply_gpu_transformations`:  The problem is that the `tlet_1_scalar_expr` nodes in the second nested level results in a symbolic expression computed as...
The values retrieved from `Config` are strings. For block_size configuration, the values need to be converted to `int` type.