Extracted more helper functions into test_utils and separated test.cpp and test.py versions

[jackl-xilinx]

No description provided.

[jackl-xilinx]

So I reviewed the helper/ convenience functions in xrtUtils and the additional ones that would also be helpful and consolidated them into test_utils.h/.cpp instead of xrtUtils.h/.cpp because theses functions are not strictly xrt related. I've updated vision kernels to match this (there's more changes needed to support it).

Another choice I made was that in test_utils, I've also put it in its own namespace to avoid possible future conflicts.

Finally, section-3 test.cpp/test.py have been simplified with run loop/timer and trace stuff moved to section-4.

There may be some debate about how "simple" the new test.cpp/.py is. In particular, some feel that we should have more details about the xrt calls in the top level host code. But that seems like a lot of code that we can wrap up and prevent type-o's from creeping in. We can still refer to the xrt calls in test_utils and explain it but no need to have it in every test.cpp. I've also moved camelCase function names to underscore to make it more consistent with other helper functions.

[makslevental]

Is everyone aware that we have extremely simple in-tree Python bindings to XRT https://github.com/Xilinx/mlir-aie/blob/main/python/XRTModule.cpp#L146-L174 since ~December?

from aie.xrt import XCLBin
xclbin = XCLBin(xclbin_path, "MLIR_AIE")

xclbin.load_ipu_instructions(ipu_insts)
[c] = xclbin.mmap_buffers([(K,)], np.int32)
wrap_C = np.asarray(c)
C = np.zeros((K,), dtype=np.int32)
np.copyto(wrap_C, C, casting="no")

xclbin.sync_buffers_to_device()
xclbin.run()
print("Running kernel")
xclbin.wait(30)
xclbin.sync_buffers_from_device()

[jackl-xilinx]

Is everyone aware that we have extremely simple in-tree Python bindings to XRT https://github.com/Xilinx/mlir-aie/blob/main/python/XRTModule.cpp#L146-L174 since ~December?

from aie.xrt import XCLBin
xclbin = XCLBin(xclbin_path, "MLIR_AIE")

xclbin.load_ipu_instructions(ipu_insts)
[c] = xclbin.mmap_buffers([(K,)], np.int32)
wrap_C = np.asarray(c)
C = np.zeros((K,), dtype=np.int32)
np.copyto(wrap_C, C, casting="no")

xclbin.sync_buffers_to_device()
xclbin.run()
print("Running kernel")
xclbin.wait(30)
xclbin.sync_buffers_from_device()

Sorry, Max. I was not aware. This looks really good for the python host code. I'll bring it up as something we can use. Thanks for pointing it out.