Provide the LLVM as a compiler backend (#73)

* Also rename ArrayAllocate.type to element_type

fuzz_tests/test_generate.py

@@ -1,7 +1,7 @@
 from hypothesis import given
 
+from tensora.compile import BroadcastTargetIndexError, TensorMethod
 from tensora.desugar import DiagonalAccessError, NoKernelFoundError
-from tensora.function import BroadcastTargetIndexError, TensorMethod
 
 from .strategies import problem_and_tensors
 

noxfile.py

@@ -1,7 +1,7 @@
 from nox import options, parametrize
 from nox_poetry import Session, session
 
-options.sessions = ["test", "test_taco", "test_numpy", "coverage", "lint"]
+options.sessions = ["test", "test_taco", "test_cffi", "test_numpy", "coverage", "lint"]
 
 
 @session(python=["3.10", "3.11", "3.12"])
@@ -18,6 +18,13 @@ def test_taco(s: Session):
     s.run("python", "-m", "pytest", "--cov", "tensora", "tests/taco")
 
 
+@session(python=["3.10", "3.11", "3.12"])
+def test_cffi(s: Session):
+    s.install(".[cffi]", "pytest", "pytest-cov")
+    s.env["COVERAGE_FILE"] = f".coverage.cffi.{s.python}"
+    s.run("python", "-m", "pytest", "--cov", "tensora", "tests_cffi")
+
+
 @session(python=["3.10", "3.11", "3.12"])
 def test_numpy(s: Session):
     s.install(".[numpy]", "pytest", "pytest-cov")

pyproject.toml

@@ -21,9 +21,10 @@ classifiers = [
 python = "^3.10"
 returns = "^0.20|^0.21|^0.22"
 cffi = "^1.16"
-# setuptools is required by cffi on Python 3.12 and up
+# setuptools is required for cffi compilation on Python 3.12 and up
 # https://cffi.readthedocs.io/en/latest/whatsnew.html#v1-16-0rc1
-setuptools = { version = "^69", python = ">=3.12" }
+setuptools = { version = "^69", python = ">=3.12", optional = true }
+llvmlite = "^0.42"
 parsita = "^2.0"
 typer = "^0.9"
 tensora-taco = { version = "^0.1.2", optional = true }
@@ -51,7 +52,8 @@ mkdocs-material = "^9"
 tensora = "tensora.cli:app"
 
 [tool.poetry.extras]
-taco = ["tensora-taco"]
+taco = ["tensora-taco", "setuptools"]
+cffi = ["setuptools"]
 numpy = ["numpy"]
 scipy = ["scipy"]
 
@@ -98,7 +100,7 @@ extend-immutable-calls = ["hypothesis.strategies.integers"]
 
 [tool.ruff.lint.per-file-ignores]
 # Allow glob imports and uppercase identifiers in tests
-"tests/*" = ["F403", "F405", "N802", "N806"]
+"tests*/*" = ["F403", "F405", "N802", "N806"]
 
 # F401: Allow unused imports in __init__.py
 "__init__.py" = ["F401"]

src/tensora/__init__.py

@@ -1,4 +1,4 @@
+from .compile import evaluate, evaluate_taco, evaluate_tensora, tensor_method
 from .format import Format, Mode
-from .function import evaluate, evaluate_taco, evaluate_tensora, tensor_method
 from .generate import TensorCompiler
 from .tensor import Tensor

src/tensora/cli.py

@@ -9,7 +9,7 @@
 
 from .expression import parse_assignment
 from .format import parse_named_format
-from .generate import TensorCompiler, generate_c_code
+from .generate import Language, TensorCompiler, generate_code
 from .kernel_type import KernelType
 from .problem import make_problem
 
@@ -52,6 +52,14 @@ def tensora(
             help="The tensor algebra compiler to use to generate the kernel.",
         ),
     ] = TensorCompiler.tensora,
+    language: Annotated[
+        Language,
+        typer.Option(
+            "--language",
+            "-l",
+            help="The language in which to generate the kernel.",
+        ),
+    ] = Language.c,
     output_path: Annotated[
         Optional[Path],
         typer.Option(
@@ -107,7 +115,7 @@ def tensora(
             raise NotImplementedError()
 
     # Generate code
-    match generate_c_code(problem, kernel_types, tensor_compiler):
+    match generate_code(problem, kernel_types, tensor_compiler, language):
         case Failure(error):
             typer.echo(str(error), err=True)
             raise typer.Exit(1)

src/tensora/codegen/__init__.py

@@ -1 +1,2 @@
 from ._ir_to_c import ir_to_c, ir_to_c_function_definition, ir_to_c_statement
+from ._ir_to_llvm import ir_to_llvm

src/tensora/codegen/_ir_to_c.py

@@ -160,13 +160,13 @@ def ir_to_c_boolean_to_integer(self: BooleanToInteger) -> str:
 
 @ir_to_c_expression.register(ArrayAllocate)
 def ir_to_c_array_allocate(self: ArrayAllocate) -> str:
-    return f"malloc(sizeof({type_to_c(self.type)}) * {parens(self.n_elements, (Add, Subtract))})"
+    return f"malloc(sizeof({type_to_c(self.element_type)}) * {parens(self.n_elements, (Add, Subtract))})"
 
 
 @ir_to_c_expression.register(ArrayReallocate)
 def ir_to_c_array_reallocate(self: ArrayReallocate) -> str:
     old = ir_to_c_expression(self.old)
-    return f"realloc({old}, sizeof({type_to_c(self.type)}) * {parens(self.n_elements, (Add, Subtract))})"
+    return f"realloc({old}, sizeof({type_to_c(self.element_type)}) * {parens(self.n_elements, (Add, Subtract))})"
 
 
 def ir_to_c_declaration(self: Declaration) -> str: