test/Examples/transform/Ch2/sequence.mlir

// RUN: transform-opt-ch2 %s \
// RUN:   --pass-pipeline="builtin.module(transform-interpreter{ \
// RUN:        debug-bind-trailing-args=linalg.matmul,linalg.elementwise},\
// RUN:        canonicalize,cse,symbol-dce)" |\
// RUN: FileCheck %s

// ****************************** IMPORTANT NOTE ******************************
//
// If you are changing this file, you may also need to change
// mlir/docs/Tutorials/Transform accordingly.
//
// ****************************************************************************

// Original function to optimize.
func.func @fc_relu(%lhs: tensor<512x512xf32>, %rhs: tensor<512x512xf32>,
                   %bias: tensor<512x512xf32>, %output: tensor<512x512xf32>)
                   -> tensor<512x512xf32> {
  // Matrix-matrix multiplication.
  %matmul = linalg.matmul ins(%lhs, %rhs: tensor<512x512xf32>, tensor<512x512xf32>)
                          outs(%output: tensor<512x512xf32>) -> tensor<512x512xf32>

  // Elementwise addition.
  %biased = linalg.elementwise kind=#linalg.elementwise_kind<add>
    ins(%matmul, %bias : tensor<512x512xf32>, tensor<512x512xf32>)
    outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>

  // Elementwise max with 0 (ReLU).
  %c0f = arith.constant dense<0.0> : tensor<512x512xf32>
  %relued = linalg.elementwise kind=#linalg.elementwise_kind<max_signed>
    ins(%biased, %c0f : tensor<512x512xf32>, tensor<512x512xf32>)
    outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>
  func.return %relued : tensor<512x512xf32>
}

// CHECK-LABEL: func @fc_relu
// CHECK: scf.forall
// CHECK:   scf.forall
// CHECK:     %[[SLICE4:.+]] = tensor.extract_slice
// CHECK:     %[[SLICE5:.+]] = tensor.extract_slice
// CHECK:     %[[SLICE6:.+]] = tensor.extract_slice
// CHECK:     %[[SLICE7:.+]] = tensor.extract_slice
// CHECK:     %[[SLICE8:.+]] = tensor.extract_slice
// CHECK:     func.call @microkernel(%[[SLICE4]], %[[SLICE5]], %[[SLICE6]], %[[SLICE7]], %[[SLICE8]])
// CHECK-NOT: linalg.matmul
// CHECK-NOT: linalg.elementwise
// CHECK:     scf.forall.in_parallel
// CHECK:   linalg.elementwise kind=#linalg.elementwise_kind<max_signed>
// CHECK:   scf.forall.in_parallel

// Declaration of the "microkernel" function that we will be targeting.
func.func private @microkernel(
    %lhs: tensor<4x512xf32>,
    %rhs: tensor<512x4xf32>,
    %bias: tensor<4x4xf32>,
    %init: tensor<4x4xf32>,
    %output: tensor<4x4xf32>) -> tensor<4x4xf32>

module attributes {transform.with_named_sequence} {
  transform.named_sequence @__transform_main(
      %arg0: !transform.any_op,
      %arg1: !transform.op<"linalg.matmul">,
      %arg2: !transform.op<"linalg.elementwise">) {
    // Since the %arg2 handle is associated with both elementwise operations,
    // we need to split it into two handles so we can target only the second
    // elementwise operation.
    %add, %max = transform.split_handle %arg2 : (!transform.op<"linalg.elementwise">)
        -> (!transform.any_op, !transform.any_op)

    // The actual tiling transformation takes tile sizes as attributes. It produces a
    // handle to the loop generated during tiling.
    %tiled, %loop = transform.structured.tile_using_forall %max tile_sizes [8, 32]
        : (!transform.any_op) -> (!transform.any_op, !transform.any_op)

    // We can now fuse the other operations into the loop. Here, we fuse
    // operations one-by-one. This requires the operation that is being fused
    // to define the value used within the loop, so the order of such fusions
    // is important. We could also use "transform.merge_handles" to obtain
    // a single handle to all operations and give it to `fuse_into_containing_op`
    // that would take care of the ordering in this case.
    %add_fused, %loop2 = transform.structured.fuse_into_containing_op %add into %loop
        : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)
    %matmul_fused, %loop3 = transform.structured.fuse_into_containing_op %arg1 into %loop2
        : (!transform.op<"linalg.matmul">, !transform.any_op) -> (!transform.any_op, !transform.any_op)

    // Tile again to get the desired size. Note that this time this tiles the
    // "add" operation and fuses matmul into the loop, but doesn't affect the
    // "max" operation. This illustrates the precise targeting with the transform
    // dialect. Otherwise, it is difficult to differentiate "add" and "max", both
    // of which having the same kind.
    %tiled_second, %loop_second = transform.structured.tile_using_forall %add_fused tile_sizes [4, 4]
        : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
    %matmul_fused_2, %loop_second_2 =
        transform.structured.fuse_into_containing_op %matmul_fused into %loop_second
        : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)

    // Since outlining is currently only implemented for region-holding operations
    // such as loops, use tiling to size 1 to materialize the outer loop that is
    // going to be outlined.
    %_0, %loop_third = transform.structured.tile_using_forall %tiled_second tile_sizes [1]
        : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
    %_1, %outline_target = transform.structured.fuse_into_containing_op %matmul_fused_2 into %loop_third
        : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)
    %func, %call = transform.loop.outline %outline_target {func_name = "outlined"}
        : (!transform.any_op) -> (!transform.any_op, !transform.any_op)

    // Rewrite the call target.
    transform.my.change_call_target %call, "microkernel" : !transform.any_op

    transform.yield
  }
}