Matchers

Applies to Linux

2024-05-08

3 min read time

Introduction

The matchers provide a way to compose several predicates together. A matcher such as m(m1, m2) first checks a match for m followed by a match for m1 and m2 subsequently.

The most commonly used matcher is the name matcher. It matches the instruction with the operator equal to the name specified:

auto match_sum = name("sum");

The above matcher finds sum operators. To find sum operators with the output standard_shape, use:

auto match_sum = name(“sum”)(standard_shape());

Arguments

To match arguments in the instructions, match each argument using the arg matcher:

auto match_sum = name("sum")(arg(0)(name("@literal"), arg(1)(name("@literal"))));

The above matcher matches a sum operator with two arguments that are literals. Note that the args matcher eliminates the need to write arg(0) and arg(1) everytime:

auto match_sum = name("sum")(args(name("@literal"), name("@literal")));

Binding

To reference other instructions encountered while traversing through the instructions, use .bind:

auto match_sum = name("sum")(args(
                                name("@literal").bind("one"),
                                name("@literal").bind("two")
                            )).bind("sum");

This associates the instruction to a name that can be read from the matcher_result when it matches.

Finding matches

To use the matchers to find instructions, write a callback object that contains the matcher and an apply function that takes the matcher_result when the match is found:

struct match_find_sum
{
    auto matcher() const { return name("sum"); }

    void apply(program& p, matcher_result r) const
    {
        // Do something with the result
    }
};

find_matches(prog, match_find_sum{});

Creating matchers

The macros MIGRAPH_BASIC_MATCHER and MIGRAPH_PRED_MATCHER help in the creation of the matchers. Here is how you can create a matcher for shapes that are broadcasted:

MIGRAPH_PRED_MATCHER(broadcasted_shape, instruction_ref ins)
{
    return ins->get_shape().broadcasted();
}

For parameters to the predicate, use make_basic_pred_matcher to create the matcher. Here is how you can create a matcher to check the number of dimensions of the shape:

inline auto number_of_dims(std::size_t n)
{
    return make_basic_pred_matcher([=](instruction_ref ins) {
        return ins->get_shape().lens().size() == n;
    });
}