[P0] Add attributes header file to manage attributes in neura dialects

[ShangkunLi]

We need to create a header file for many attributes (e.g., dataflow_mode, mapping_mode, mapping_strategy, etc.) in our dialect.

[YanzhouTang]

Hi Shangkun, is this new feature supports float type now? I tested it in test/c2llvm2mlir/simple_loop but failed. Here below is my test-mlir in test/c2llvm2mlir/simple_loop

// Compiles the original kernel.
// RUN: clang++ kernel.cpp -o %t-kernel.out

// Compiles the original kernel to mlir, then lower back to llvm, eventually binary.
// RUN: clang++ -S -emit-llvm -o %t-kernel.ll kernel.cpp
// RUN: mlir-translate --import-llvm %t-kernel.ll -o %t-kernel.mlir
// RUN: mlir-opt %t-kernel.mlir | mlir-translate -mlir-to-llvmir -o %t-kernel_back.ll
// RUN: llc %t-kernel_back.ll -relocation-model=pic -filetype=obj -o %t-kernel_back.o
// RUN: clang++ %t-kernel_back.o -o %t-kernel_back.out

// RUN: %t-kernel.out > %t-dumped_output.txt
// RUN: %t-kernel_back.out >> %t-dumped_output.txt
// RUN: FileCheck %s < %t-dumped_output.txt

// Verifies the output values are the same for the original and re-compiled kernel.
// CHECK: output: [[OUTPUT:[0-9]+\.[0-9]+]]
// CHECK: output: [[OUTPUT]]

// Test LLVM to NEURA lowering
// RUN: clang++ -S -emit-llvm -ffp-contract=off kernel.cpp -o %t-kernel.ll
// RUN: mlir-translate --import-llvm %t-kernel.ll -o %t-kernel.mlir

// RUN: mlir-neura-opt --assign-accelerator \
// RUN:   --lower-llvm-to-neura \
// RUN:   --canonicalize-live-in \
// RUN:   --leverage-predicated-value \
// RUN:   --transform-ctrl-to-data-flow \
// RUN:   --fold-constant \
// RUN:   --transform-to-steer-control \
// RUN:   --remove-predicated-type \
// RUN:   --insert-data-mov %t-kernel.mlir -o %t-kernel-neura.mlir
// RUN: FileCheck %s --check-prefix=CHECK-LLVM2NEURA < %t-kernel-neura.mlir

// RUN: mlir-neura-opt --assign-accelerator \
// RUN:   --lower-llvm-to-neura \
// RUN:   --canonicalize-live-in \
// RUN:   --leverage-predicated-value \
// RUN:   --transform-ctrl-to-data-flow \
// RUN:   --fold-constant \
// RUN:   --transform-to-steer-control \
// RUN:   --remove-predicated-type \
// RUN:   --insert-data-mov \
// RUN:   --map-to-accelerator="mapping-strategy=heuristic backtrack-config=simple" %t-kernel.mlir -o %t-kernel-mapped.mlir
// RUN: FileCheck %s --check-prefix=CHECK-LLVM2NEURA-MAP < %t-kernel-mapped.mlir

And the program encountered Segmentation fault when running with PASS transform-to-steer-control and remove-predicated-type

[ShangkunLi]

For transform-to-steer-control pass, we utilize the source predicate dataflow ir without any constant folding optimizations as input.

And remember to use spatial-only mapping mode for steering-based ir.

So please use:

// RUN: mlir-neura-opt --assign-accelerator \
// RUN:   --lower-llvm-to-neura \
// RUN:   --canonicalize-live-in \
// RUN:   --leverage-predicated-value \
// RUN:   --transform-ctrl-to-data-flow \
// RUN:   --transform-to-steer-control \
// RUN:   --remove-predicated-type \
// RUN:   --insert-data-mov \
// RUN:   --map-to-accelerator="mapping-strategy=heuristic mapping-mode=spatial-only backtrack-config=customized" %t-kernel.mlir -o %t-kernel-mapped.mlir

[tancheng]

Hi @ShangkunLi, can we have this cleanup be checked in ASAP? More attributes are created using string: https://github.com/coredac/dataflow/pull/157/files#r2434701297

[ShangkunLi]

Hi @ShangkunLi, can we have this cleanup be checked in ASAP? More attributes are created using string: https://github.com/coredac/dataflow/pull/157/files#r2434701297

I will do it this weekend, I am writing funding proposals today since the ddl is close...