Retrieving LLVM IR from AOT tfcompile

[williamjeremy]

Is it possible to get LLVM intermediate representation (.ll) files from tfcompile instead of object code? If so, how can it be done? Alternatively, can one get source code instead of object code from tfcompile?

[tatatodd]

@williamjeremy there isn't a mechanism right now, but I'm adding simple support for this right now. Once it's in, you'll be able to set the --xla_dump_ir_to flag to specify a directory to dump out the LLVM IR, pre and post optimization.

There is no way to dump source code, if by "source code" you mean something like C or C++. tfcompile takes as input the TensorFlow graph, which is lowered to XLA HLO, which is lowered to LLVM IR, which is lowered to machine code.

I'll update this issue once the support has been merged.

[tatatodd]

201fa22 added the --xla_dump_ir_to support to the AOT codepath, so I'm closing this out.

[vsahil]

Is it possible to get LLVM intermediate representation (.ll) files from tfcompile instead of object code? If so, how can it be done? Alternatively, can one get source code instead of object code from tfcompile?

@williamjeremy were you able to get the LLVM files, if yes, can you please send a complete example with the commands. Thank you in advance!