bufferize.mlir

// -----// IR Dump After FuncBufferize //----- //
module {
  func @collapse_dynamic_shape_of_slice(%arg0: memref<?x?x?x?xf32>, %arg1: index, %arg2: index, %arg3: index) -> memref<2x?x?xf32> {
    %0 = bufferization.to_tensor %arg0 : memref<?x?x?x?xf32>
    %1 = tensor.extract_slice %0[0, 0, %arg1, %arg1] [%arg2, %arg2, %arg3, %arg3] [1, 1, 1, 1] : tensor<?x?x?x?xf32> to tensor<?x?x?x?xf32>
    %2 = tensor.cast %1 : tensor<?x?x?x?xf32> to tensor<2x?x?x?xf32>
    %3 = tensor.collapse_shape %2 [[0], [1, 2], [3]] : tensor<2x?x?x?xf32> into tensor<2x?x?xf32>
    %4 = bufferization.to_memref %3 : memref<2x?x?xf32>
    return %4 : memref<2x?x?xf32>
  }
}


ImplicitTypeIDRegistry::lookupOrInsert(mlir::OpTrait::ConstantLike<Empty>)
// -----// IR Dump After TensorBufferize //----- //
func @collapse_dynamic_shape_of_slice(%arg0: memref<?x?x?x?xf32>, %arg1: index, %arg2: index, %arg3: index) -> memref<2x?x?xf32> {
  %0 = bufferization.to_tensor %arg0 : memref<?x?x?x?xf32>
  %1 = memref.subview %arg0[0, 0, %arg1, %arg1] [%arg2, %arg2, %arg3, %arg3] [1, 1, 1, 1] : memref<?x?x?x?xf32> to memref<?x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %2 = bufferization.to_tensor %1 : memref<?x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %3 = memref.cast %1 : memref<?x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>> to memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %c1 = arith.constant 1 : index
  %4 = memref.dim %3, %c1 : memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %c2 = arith.constant 2 : index
  %5 = memref.dim %3, %c2 : memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %c3 = arith.constant 3 : index
  %6 = memref.dim %3, %c3 : memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  %7 = memref.alloc(%4, %5, %6) {alignment = 128 : i64} : memref<2x?x?x?xf32>
  %8 = bufferization.to_tensor %3 : memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>>
  memref.copy %3, %7 : memref<2x?x?x?xf32, affine_map<(d0, d1, d2, d3)[s0, s1, s2, s3] -> (d0 * s1 + s0 + d1 * s2 + d2 * s3 + d3)>> to memref<2x?x?x?xf32>
  %9 = memref.collapse_shape %7 [[0], [1, 2], [3]] : memref<2x?x?x?xf32> into memref<2x?x?xf32>
  %10 = bufferization.to_tensor %9 : memref<2x?x?xf32>
  return %9 : memref<2x?x?xf32>
}