… from the documentation online to check what library is present

…page, this commit didn't work. They also have something similar - adding Tensor core backend from Nvidia as an external codegen

…ddition

…ion of the pass

…t commit

… the next commit"

This reverts commit 429737d.

…f experimental changes which try to understand the SDFGIR and the changes to make IPUCodeGen registry into the frame_targets.

	The fix was to call 'self._frame.generate_state' recursively from ipu.generate_state. This goes into framecode.py and
	calls the recursive function which traverses substates and calls the codegen respectively.
	Based on this learning a point to note is to remember to call the recursive functions inside
	generate_*() functions.
	example is 'self._dispatcher.dispatch_subgraph' in 'generate_scope'.

2. Fix ipu/ipu 2 folders were created recursively.
	Fix - Remove 'target_type='ipu' from CodeObject.

…ok right now into GPU/MPI

…ers and the IPUDevice, this goes in __init__/exit part and not in the SDFG

…ested might be buggy

Code snippet below checks if the file is frame(.cpp vs .cu) if so
matches target and dumps the headers.

```
if backend == 'frame':  #on cpu.cpp file
+            for target in self.targets:
+                if target.target_name == 'ipu':
```

Some cosmetic changes removed the prints.

… add more handcrafted tests, remove the header generation from framecode.py and add into ipu.py

…f how cpu codegen works for a tasklet

This reverts commit 8416d82.

This reverts commit 6a254c3.

…r is from sve/codegen.py the later is much readable and simpler to understand

…atch_copy.

Allocate memory on device/host
Copy memory from one to another

Used Storage=IPU_Memory

1. register_array_dispatcher -> allocate_array/deallocate_array
2. register_copy_dispatcher -> copy_memory

Notes: using ipu_storage vs gpu_storage breaks.
links with last 2 commits

Intelligently maps the data on tiles based on scalar/array nodes and maps to
setTileMapping

1. Mapping a vertex will also be similar
2. This the not the most efficient mapping, currently arrays are just mapped equally on 10 tiles.
3. Add new test - simplified

2. Modify test to be "FLOAT" and not "DOUBLE"
3. Add new interface file which helps to add #include<> vs #include "".

…ret it use {{ instead

 during compilation - extra ../ present 3. Fix libraries issue partially,
 almost there after this fix

 Fix IPUConfig.cmake, ipu-config.cmake not found error, was
 wrongly using IPU but poplar libraries start with poplar

…puts to the library and the inputs to SDFG, seems like we need to dig into some other ways of allocation of variables and such details.

… one generates the golden code for poplar.

2. In the Host/ side the pipeline + functions are present.
3. The pipeline is not yet set, next patch might set it.
4. Lots of bugs still
5. Cerete Host/ Device/ files along with cpu/

…t the inputs to the library and the inputs to SDFG, seems like we need to dig into some other ways of allocation of variables and such details."

Reverting this as this is making it hard to understand a node as of now.

This reverts commit 63fad92.

…ave IPUCodegen + library codegen

together.

…m fpga and cuda. This creates a codegen which dumps both library + nodes

This reverts commit 2ffc0c9.

…lude, debugging issue on real IPU machine, pushing incomplete changes"

…ted on another testcase

… for transients and scope must be inside state

…snode is not IPU_Memory

…s no schedule= in code

…_memory() triggers