This commit rebases the compiler onto commit f69328049e9e from
llvm-project.
Changes:
* Use of the one-shot bufferizer for improved memory management
* A new pass `OneShotBufferizeDPSWrapper` that converts functions
returning tensors to destination-passing-style as required by the
one-shot bufferizer
* A new pass `LinalgGenericOpWithTensorsToLoopsPass` that converts
`linalg.generic` operations with value semantics to loop nests
* Rebase onto a fork of llvm-project at f69328049e9e with local
modifications to enable bufferization of `linalg.generic` operations
with value semantics
* Workaround for the absence of type propagation after type conversion
via extra patterns in all dialect conversion passes
* Printer, parser and verifier definitions moved from inline
declarations in ODS to the respective source files as required by
upstream changes
* New tests for functions with a large number of inputs
* Increase the number of allowed task inputs as required by new tests
* Use upstream function `mlir_configure_python_dev_packages()` to
locate Python development files for compatibility with various CMake
versions
Co-authored-by: Quentin Bourgerie <quentin.bourgerie@zama.ai>
Co-authored-by: Ayoub Benaissa <ayoub.benaissa@zama.ai>
Co-authored-by: Antoniu Pop <antoniu.pop@zama.ai>
This is currently lowering to our custom linalg conv operation, since
linalg ops doesn't support custom types. But as linalg will support
custom types in the future, we may want to lower to the native linalg op
instead
Quick fix due to ordering of includes, had to add #include
<mlir/Transforms/DialectConversion.h> to include/concretelang/Conversion/Utils/GenericOpTypeConversionPattern.h
This introduces a new header file `zamalang/Support/Constants.h` for
constants, currently only populated with a constant for the default
pattern rewriting benefit of 1.
Add a rewrite pattern that transforms an instance of
`HLFHELinalg.zero` into an instance of `linalg.generate` with an
appropriate region yielding a zero value.
Example:
%out = "HLFHELinalg.zero"() : () -> tensor<MxNx!HLFHE.eint<p>>
becomes:
%0 = tensor.generate {
^bb0(%arg2: index, %arg3: index):
%zero = "HLFHE.zero"() : () -> !HLFHE.eint<p>
tensor.yield %zero : !HLFHE.eint<p>
} : tensor<MxNx!HLFHE.eint<p>>
Operations with regions currently need explicit patterns for type
conversion throughout lowering. This change adds the required patterns
for `tensor.generate` to the lowering passes, such that the operation
can be used starting from the lowering from HLFHE to MidLFHE.
Replace `LinalgGenericTypeConverterPattern`, specialized for
`linalg.generic` with a generic type converter pattern
`RegionOpTypeConverterPattern` that can be instantiated for any
operation that has one or more regions.
Further enhancements:
- Supports multiple regions
- Uses more idiomatic instantiations of `llvm::for_each` instead of
manual iterations using for loops
