Breakup python pytorch deps (#2582)

This lifts the core of the jit_ir_importer and ltc out of the pt1
project, making them peers to it. As a side-effect of this layering, now
the "MLIR bits" (dialects, etc) are not commingled with the various
parts of the pt1 project, allowing pt1 and ltc to overlay cleanly onto a
more fundamental "just MLIR" Python core. Prior to this, the Python
namespace was polluted to the point that this could not happen.

That "just MLIR" Python core will be introduced in a followup, which
will create the space to upstream the FX and ONNX pure Python importers.

This primary non-NFC change to the API is:

* `torch_mlir.dialects.torch.importer.jit_ir` ->
`torch_mlir.jit_ir_importer`.

The rest is source code layering so that we can make the pt1 project
optional without losing the other features.

Progress on #2546.

Author: stellaraccident

.github/workflows/buildAndTest.yml

@@ -27,15 +27,12 @@ jobs:
       matrix:
         os-arch: [ubuntu-x86_64, macos-arm64, windows-x86_64]
         llvm-build: [in-tree, out-of-tree]
-        torch-binary: [ON, OFF]
+        torch-binary: [ON]
         torch-version: [nightly, stable]
         exclude:
-          # Exclude llvm in-tree and pytorch source
-          - llvm-build: in-tree
-            torch-binary: OFF
-          # Exclude llvm out-of-tree and pytorch binary
+          # Exclude llvm out-of-tree and pytorch stable (to save resources)
           - llvm-build: out-of-tree
-            torch-binary: ON
+            torch-version: stable
           # Exclude macos-arm64 and llvm out-of-tree altogether
           - os-arch: macos-arm64
             llvm-build: out-of-tree
@@ -45,9 +42,6 @@ jobs:
             llvm-build: out-of-tree
           - os-arch: windows-x86_64
             torch-version: stable
-          # For PyTorch stable builds, we don't build PyTorch from source
-          - torch-version: stable
-            torch-binary: OFF
         include:
           # Specify OS versions
           - os-arch: ubuntu-x86_64

.gitignore

@@ -26,7 +26,7 @@ __pycache__
 bazel-*
 
 # Autogenerated files
-/projects/pt1/python/torch_mlir/csrc/base_lazy_backend/generated
+/projects/ltc/csrc/base_lazy_backend/generated
 
 #Docker builds
 build_oot/

CMakeLists.txt

@@ -149,10 +149,12 @@ endfunction()
 # Configure CMake.
 list(APPEND CMAKE_MODULE_PATH ${MLIR_MAIN_SRC_DIR}/cmake/modules)
 list(APPEND CMAKE_MODULE_PATH ${LLVM_MAIN_SRC_DIR}/cmake)
+list(APPEND CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/build_tools/cmake)
 
 include(TableGen)
 include(AddLLVM)
 include(AddMLIR)
+include(AddMLIRPython)
 
 ################################################################################
 # Setup python.
@@ -231,6 +233,4 @@ endif()
 # Sub-projects
 #-------------------------------------------------------------------------------
 
-if(TORCH_MLIR_ENABLE_PROJECT_PT1)
-  add_subdirectory(projects/pt1)
-endif()
+add_subdirectory(projects)

build_tools/autogen_ltc_backend.py

@@ -29,7 +29,6 @@
     TORCH_INCLUDE_DIR = TORCH_DIR
 TORCHGEN_DIR = Path(torchgen.__path__[0]).resolve()
 TORCH_MLIR_DIR = Path(__file__).resolve().parent.parent
-TORCH_MLIR_PT1_DIR = TORCH_MLIR_DIR / "projects" / "pt1"
 
 def reindent(text, prefix=""):
     return indent(dedent(text), prefix)
@@ -114,12 +113,12 @@ def __init__(self, binary_dir):
         self.binary_dir = Path(binary_dir)
         assert self.binary_dir.is_dir(), f"Binary directory not found: {self.binary_dir}"
         self.source_yaml = self.binary_dir.joinpath("generated_native_functions.yaml")
-        self.backend_path = TORCH_MLIR_PT1_DIR.joinpath(
-            "python", "torch_mlir", "csrc", "base_lazy_backend"
+        self.backend_path = TORCH_MLIR_DIR.joinpath(
+            "projects", "ltc", "csrc", "base_lazy_backend"
         )
         assert self.backend_path.is_dir(), f"Backend path not found: {self.backend_path}"
         self.generated_path = self.binary_dir.joinpath(
-            "projects", "pt1", "python", "torch_mlir", "csrc", "base_lazy_backend", "generated"
+            "projects", "ltc", "csrc", "base_lazy_backend", "generated"
         )
         self.generated_path.mkdir(parents=True, exist_ok=True)
 
@@ -415,7 +414,7 @@ def extract_signatures(text):
                     // for ops that dont have a corresponding structured kernel or shape definition
 
                     #include "shape_inference.h"
-                    #include "torch_mlir/csrc/base_lazy_backend/utils/exception.h"
+                    #include "base_lazy_backend/utils/exception.h"
                     namespace torch {{
                     namespace lazy {{
                     {}
@@ -467,7 +466,7 @@ def gen_fallback_code(*args, **kwargs):
             node_base="torch::lazy::TorchMlirNode",
             node_base_hdr=str(self.backend_path.joinpath("mlir_node.h")),
             tensor_class=self.tensor_class,
-            tensor_class_hdr="torch_mlir/csrc/base_lazy_backend/tensor.h",
+            tensor_class_hdr="base_lazy_backend/tensor.h",
             create_aten_from_ltc_tensor="CreateFunctionalizedAtenFromLtcTensor",
             shape_inference_hdr=str(self.generated_path.joinpath("shape_inference.h")),
             lazy_ir_generator=GenMlirLazyIr,

projects/pt1/python/torch_mlir/cmake/modules/TorchMLIRPyTorch.cmake renamed to build_tools/cmake/TorchMLIRPyTorch.cmake

@@ -364,9 +364,9 @@ function setup_venv() {
 function build_out_of_tree() {
   local torch_from_bin="$1"
   local python_version="$2"
-  echo ":::: Build out-of-tree Torch from binary: $torch_from_bin with Python: $python_version"
-
   local torch_version="$3"
+  echo ":::: Build out-of-tree Torch from binary: $torch_from_bin with Python: $python_version ($torch_version)"
+
   local enable_ltc="ON"
   if [[ "${torch_version}" == "stable" ]]
   then

build_tools/python_deploy/build_linux_packages.sh

@@ -42,6 +42,6 @@ if [ ! -z ${TORCH_MLIR_EXT_MODULES} ]; then
 fi
 
 PYTHONPATH="${pypath}" python \
-  -m torch_mlir.dialects.torch.importer.jit_ir.build_tools.abstract_interp_lib_gen \
+  -m torch_mlir.jit_ir_importer.build_tools.abstract_interp_lib_gen \
   --pytorch_op_extensions=${ext_module:-""} \
   --torch_transforms_cpp_dir="${torch_transforms_cpp_dir}"

build_tools/update_abstract_interp_lib.sh

@@ -43,7 +43,7 @@ fi
 
 set +u
 PYTHONPATH="${PYTHONPATH}:${pypath}" python \
-  -m torch_mlir.dialects.torch.importer.jit_ir.build_tools.torch_ods_gen \
+  -m torch_mlir.jit_ir_importer.build_tools.torch_ods_gen \
   --torch_ir_include_dir="${torch_ir_include_dir}" \
   --pytorch_op_extensions="${ext_module}" \
   --debug_registry_dump="${torch_ir_include_dir}/JITOperatorRegistryDump.txt"

build_tools/update_torch_ods.sh

@@ -17,7 +17,7 @@ The end-to-end test is important to check the correctness of the other steps.
 
 ### Step 2. Update ods
 
-Update [torch_ods_gen.py](https://github.com/llvm/torch-mlir/blob/main/projects/pt1/python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/torch_ods_gen.py) with the new op and run [update_torch_ods.sh](https://github.com/llvm/torch-mlir/blob/main/build_tools/update_torch_ods.sh) to generate the ods. Running `update_torch_ods.sh` would dump all the operators with schema into `JITOperatorRegistryDump.txt`. It’s convenient to look for ops signatures and operands names in this file.
+Update [torch_ods_gen.py](https://github.com/llvm/torch-mlir/blob/main/projects/pt1/python/torch_mlir/jit_ir_importer/build_tools/torch_ods_gen.py) with the new op and run [update_torch_ods.sh](https://github.com/llvm/torch-mlir/blob/main/build_tools/update_torch_ods.sh) to generate the ods. Running `update_torch_ods.sh` would dump all the operators with schema into `JITOperatorRegistryDump.txt`. It’s convenient to look for ops signatures and operands names in this file.
 
 ### Step 3. Propagate types
 It’s essential to make sure the new op implements shape and dtype inference. See [abstract_interp_lib](https://github.com/llvm/torch-mlir/blob/main/docs/abstract_interp_lib.md) for information on adding shape and dtype inference.

docs/Torch-ops-E2E-implementation.md

@@ -26,7 +26,7 @@ The two main use cases are:
 ## Architecture
 
 Functions are defined as TorchScript-able Python functions in
-`python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/abstract_interp_lib_gen.py`.
+`python/torch_mlir/jit_ir_importer/build_tools/abstract_interp_lib_gen.py`.
 The signatures of the functions are systematically derived from Torch JIT
 operator registry. Most shape functions are expected to reuse the upstream
 helper functions