112 lines · plain
1// RUN: transform-opt-ch2 %s \2// RUN: --pass-pipeline="builtin.module(transform-interpreter{ \3// RUN: debug-bind-trailing-args=linalg.matmul,linalg.elementwise},\4// RUN: canonicalize,cse,symbol-dce)" |\5// RUN: FileCheck %s6 7// ****************************** IMPORTANT NOTE ******************************8//9// If you are changing this file, you may also need to change10// mlir/docs/Tutorials/Transform accordingly.11//12// ****************************************************************************13 14// Original function to optimize.15func.func @fc_relu(%lhs: tensor<512x512xf32>, %rhs: tensor<512x512xf32>,16 %bias: tensor<512x512xf32>, %output: tensor<512x512xf32>)17 -> tensor<512x512xf32> {18 // Matrix-matrix multiplication.19 %matmul = linalg.matmul ins(%lhs, %rhs: tensor<512x512xf32>, tensor<512x512xf32>)20 outs(%output: tensor<512x512xf32>) -> tensor<512x512xf32>21 22 // Elementwise addition.23 %biased = linalg.elementwise kind=#linalg.elementwise_kind<add>24 ins(%matmul, %bias : tensor<512x512xf32>, tensor<512x512xf32>)25 outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>26 27 // Elementwise max with 0 (ReLU).28 %c0f = arith.constant dense<0.0> : tensor<512x512xf32>29 %relued = linalg.elementwise kind=#linalg.elementwise_kind<max_signed>30 ins(%biased, %c0f : tensor<512x512xf32>, tensor<512x512xf32>)31 outs(%output : tensor<512x512xf32>) -> tensor<512x512xf32>32 func.return %relued : tensor<512x512xf32>33}34 35// CHECK-LABEL: func @fc_relu36// CHECK: scf.forall37// CHECK: scf.forall38// CHECK: %[[SLICE4:.+]] = tensor.extract_slice39// CHECK: %[[SLICE5:.+]] = tensor.extract_slice40// CHECK: %[[SLICE6:.+]] = tensor.extract_slice41// CHECK: %[[SLICE7:.+]] = tensor.extract_slice42// CHECK: %[[SLICE8:.+]] = tensor.extract_slice43// CHECK: func.call @microkernel(%[[SLICE4]], %[[SLICE5]], %[[SLICE6]], %[[SLICE7]], %[[SLICE8]])44// CHECK-NOT: linalg.matmul45// CHECK-NOT: linalg.elementwise46// CHECK: scf.forall.in_parallel47// CHECK: linalg.elementwise kind=#linalg.elementwise_kind<max_signed>48// CHECK: scf.forall.in_parallel49 50// Declaration of the "microkernel" function that we will be targeting.51func.func private @microkernel(52 %lhs: tensor<4x512xf32>,53 %rhs: tensor<512x4xf32>,54 %bias: tensor<4x4xf32>,55 %init: tensor<4x4xf32>,56 %output: tensor<4x4xf32>) -> tensor<4x4xf32>57 58module attributes {transform.with_named_sequence} {59 transform.named_sequence @__transform_main(60 %arg0: !transform.any_op,61 %arg1: !transform.op<"linalg.matmul">,62 %arg2: !transform.op<"linalg.elementwise">) {63 // Since the %arg2 handle is associated with both elementwise operations,64 // we need to split it into two handles so we can target only the second65 // elementwise operation.66 %add, %max = transform.split_handle %arg2 : (!transform.op<"linalg.elementwise">)67 -> (!transform.any_op, !transform.any_op)68 69 // The actual tiling transformation takes tile sizes as attributes. It produces a70 // handle to the loop generated during tiling.71 %tiled, %loop = transform.structured.tile_using_forall %max tile_sizes [8, 32]72 : (!transform.any_op) -> (!transform.any_op, !transform.any_op)73 74 // We can now fuse the other operations into the loop. Here, we fuse75 // operations one-by-one. This requires the operation that is being fused76 // to define the value used within the loop, so the order of such fusions77 // is important. We could also use "transform.merge_handles" to obtain78 // a single handle to all operations and give it to `fuse_into_containing_op`79 // that would take care of the ordering in this case.80 %add_fused, %loop2 = transform.structured.fuse_into_containing_op %add into %loop81 : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)82 %matmul_fused, %loop3 = transform.structured.fuse_into_containing_op %arg1 into %loop283 : (!transform.op<"linalg.matmul">, !transform.any_op) -> (!transform.any_op, !transform.any_op)84 85 // Tile again to get the desired size. Note that this time this tiles the86 // "add" operation and fuses matmul into the loop, but doesn't affect the87 // "max" operation. This illustrates the precise targeting with the transform88 // dialect. Otherwise, it is difficult to differentiate "add" and "max", both89 // of which having the same kind.90 %tiled_second, %loop_second = transform.structured.tile_using_forall %add_fused tile_sizes [4, 4]91 : (!transform.any_op) -> (!transform.any_op, !transform.any_op)92 %matmul_fused_2, %loop_second_2 =93 transform.structured.fuse_into_containing_op %matmul_fused into %loop_second94 : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)95 96 // Since outlining is currently only implemented for region-holding operations97 // such as loops, use tiling to size 1 to materialize the outer loop that is98 // going to be outlined.99 %_0, %loop_third = transform.structured.tile_using_forall %tiled_second tile_sizes [1]100 : (!transform.any_op) -> (!transform.any_op, !transform.any_op)101 %_1, %outline_target = transform.structured.fuse_into_containing_op %matmul_fused_2 into %loop_third102 : (!transform.any_op, !transform.any_op) -> (!transform.any_op, !transform.any_op)103 %func, %call = transform.loop.outline %outline_target {func_name = "outlined"}104 : (!transform.any_op) -> (!transform.any_op, !transform.any_op)105 106 // Rewrite the call target.107 transform.my.change_call_target %call, "microkernel" : !transform.any_op108 109 transform.yield110 }111}112