Files
clang-p2996/mlir/test/Dialect/LLVM/transform-e2e.mlir
Alex Zinenko 2fe4d90cac [mlir] make structured transform ops use types
Types have been introduced a while ago and provide for better
readability and transform-time verification. Use them in the ops from
the structured transform dialect extension.

In most cases, the types are appended as trailing functional types or a
derived format of the functional type that allows for an empty right
hand size without the annoying `-> ()` syntax (similarly to `func.func`
declaration that may omit the arrow). When handles are used inside mixed
static/dynamic lists, such as tile sizes, types of those handles follow
them immediately as in `sizes [%0 : !transform.any_value, 42]`. This
allows for better readability than matching the trailing type.

Update code to remove hardcoded PDL dependencies and expunge PDL from
structured transform op code.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D144515
2023-05-16 08:16:56 +00:00

60 lines
2.5 KiB
MLIR

// RUN: mlir-opt %s --test-transform-dialect-interpreter -test-transform-dialect-erase-schedule --test-lower-to-llvm --split-input-file | FileCheck %s
// CHECK-LABEL: llvm.func @matmul_tensors
func.func @matmul_tensors(
%arg0: tensor<2x4xf32>, %arg1: tensor<4x6xf32>, %arg2: tensor<2x6xf32>)
-> tensor<2x6xf32> {
// CHECK-NOT: linalg
// CHECK: llvm.intr.fmuladd{{.*}}
%0 = linalg.matmul ins(%arg0, %arg1: tensor<2x4xf32>, tensor<4x6xf32>)
outs(%arg2: tensor<2x6xf32>)
-> tensor<2x6xf32>
return %0 : tensor<2x6xf32>
}
transform.sequence failures(propagate) {
^bb1(%module_op: !transform.any_op):
%0 = transform.structured.match ops{["linalg.matmul"]} in %module_op : (!transform.any_op) -> !transform.any_op
%1, %loops:3 = transform.structured.tile %0 [2, 2, 2] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)
%2 = get_closest_isolated_parent %1 : (!transform.any_op) -> !transform.any_op
transform.structured.vectorize %2 : (!transform.any_op) -> !transform.any_op
%b = transform.bufferization.one_shot_bufferize layout{IdentityLayoutMap}
%module_op {bufferize_function_boundaries = true}
: (!transform.any_op) -> !transform.any_op
%f = transform.structured.match ops{["func.func"]} in %b
: (!transform.any_op) -> !transform.any_op
// TODO: group these lower-level controls into various properly named vector
// lowering TD macros.
%func = transform.vector.lower_contraction %f
lowering_strategy = "outerproduct"
: (!transform.any_op) -> !transform.any_op
%func_2 = transform.vector.apply_transfer_permutation_patterns %func
: (!transform.any_op) -> !transform.any_op
%func_3 = transform.vector.lower_multi_reduction %func_2
lowering_strategy = "innerparallel"
: (!transform.any_op) -> !transform.any_op
%func_4 = transform.vector.split_transfer_full_partial %func_3
split_transfer_strategy = "linalg-copy"
: (!transform.any_op) -> !transform.any_op
%func_5 = transform.vector.transfer_to_scf %func_4
max_transfer_rank = 1 full_unroll = true
: (!transform.any_op) -> !transform.any_op
%func_6 = transform.vector.lower_transfer %func_5
max_transfer_rank = 1
: (!transform.any_op) -> !transform.any_op
%func_7 = transform.vector.lower_shape_cast %func_6
: (!transform.any_op) -> !transform.any_op
%func_8 = transform.vector.lower_transpose %func_7
lowering_strategy = "shuffle_1d"
: (!transform.any_op) -> !transform.any_op
}