Drop support for memrefs from JitRunner
authorAlex Zinenko <zinenko@google.com>
Thu, 26 Sep 2019 12:41:26 +0000 (05:41 -0700)
committerA. Unique TensorFlower <gardener@tensorflow.org>
Thu, 26 Sep 2019 12:42:01 +0000 (05:42 -0700)
commit99be3351b874444498c03a87e2aeec6f2f8c208d
tree67e537ff443b8535a422942a24bee5674b082ea7
parent116dac00baa6870aec2a2b469b2d6f95c2fbb316
Drop support for memrefs from JitRunner

The support for functions taking and returning memrefs of floats was introduced
in the first version of the runner, created before MLIR had reliable lowering
of allocation/deallocation to library calls.  It forcibly runs MLIR
transformation convering affine, loop and standard dialects into the LLVM
dialect, unlike the other runner flows that accept the LLVM dialect directly.
Memref support leads to more complex layering and is generally fragile.  Drop
it in favor of functions returning a scalar, or library-based function calls to
print memrefs and other data structures.

PiperOrigin-RevId: 271330839
mlir/lib/Support/JitRunner.cpp
mlir/test/mlir-cpu-runner/simple.mlir
mlir/tools/mlir-cpu-runner/CMakeLists.txt