TheRock/compiler/amd-llvm/mlir/docs/Dialects/emitc.md

The EmitC dialect allows to convert operations from other MLIR dialects to EmitC ops. Those can be translated to C/C++ via the Cpp emitter.

The following convention is followed:

• If template arguments are passed to an emitc.call_opaque operation, C++ is generated.
• If tensors are used, C++ is generated.
• If multiple return values are used within in a functions or an emitc.call_opaque operation, C++11 is required.
• If floating-point type template arguments are passed to an emitc.call_opaque operation, C++20 is required.
• If ssize_t is used, then the code requires the POSIX header sys/types.h or any of the C++ headers in which the type is defined.
• If _Float16 is used, the code requires the support of C additional floating types.
• If __bf16 is used, the code requires a compiler that supports it, such as GCC or Clang.
• If emitc.array with a dimension of size zero is used, then the code requires a GCC extension.
• If aligned_alloc is passed to an emitc.call_opaque operation, then C++17 or C11 is required.
• Else the generated code is compatible with C99.

These restrictions are neither inherent to the EmitC dialect itself nor to the Cpp emitter and therefore need to be considered while implementing conversions.

Type conversions are provided for the MLIR type index into the unsigned size_t type and its signed counterpart ptrdiff_t. Conversions between these two types are only valid if the index-typed values are within [PTRDIFF_MIN, PTRDIFF_MAX].

After the conversion, C/C++ code can be emitted with mlir-translate. The tool supports translating MLIR to C/C++ by passing -mlir-to-cpp. Furthermore, code with variables declared at top can be generated by passing the additional argument -declare-variables-at-top.

Besides operations part of the EmitC dialect, the Cpp targets supports translating the following operations:

• 'cf' Dialect
  • cf.br
  • cf.cond_br
• 'func' Dialect
  • func.call
  • func.func
  • func.return