Function TEST_CASE

test/gpu/fuse_ops.cpp:40–108

}

TEST_CASE(layernorm_pointwise)
{
    migraphx::shape s{migraphx::shape::float_type, {2, 3, 4}};
    auto create_program = [=](bool first_arg_layernorm) {
        migraphx::program p;
        auto* mm       = p.get_main_module();
        auto x         = mm->add_parameter("x", s);
        auto y         = mm->add_parameter("y", s);
        auto z         = mm->add_parameter("z", s);
        auto alloc     = migraphx::make_op("allocate", {{"shape", to_value(s)}});
        auto alloc_ins = mm->add_instruction(alloc);
        auto* pw_add1 =
            create_pointwise_module(p, "main:pointwise0", {x, y}, single_pointwise("add"));
        auto add1 =
            mm->add_instruction(make_precompile_op("pointwise"), {x, y, alloc_ins}, {pw_add1});
        auto alloc_ins2 = mm->add_instruction(alloc);
        auto layernorm_ins =
            mm->add_instruction(make_precompile_op("gpu::prelayernorm"), add1, alloc_ins2);
        std::vector<migraphx::instruction_ref> pw_inputs = {layernorm_ins, z};
        if(not first_arg_layernorm)
        {
            pw_inputs = {z, layernorm_ins};
        }
        auto* pw_add2 =
            create_pointwise_module(p, "main:pointwise1", pw_inputs, single_pointwise("add"));
        auto alloc_ins3 = mm->add_instruction(alloc);
        pw_inputs.push_back(alloc_ins3);
        auto add2 = mm->add_instruction(make_precompile_op("pointwise"), pw_inputs, {pw_add2});
        mm->add_return({add2});
        return p;
    };

    auto create_fused_program = [=]() {
        migraphx::program p;
        auto* mm       = p.get_main_module();
        auto x         = mm->add_parameter("x", s);
        auto y         = mm->add_parameter("y", s);
        auto z         = mm->add_parameter("z", s);
        auto alloc     = migraphx::make_op("allocate", {{"shape", to_value(s)}});
        auto alloc_ins = mm->add_instruction(alloc);
        auto* pw_add1 =
            create_pointwise_module(p, "main:pointwise0", {x, y}, single_pointwise("add"));
        auto add1 =
            mm->add_instruction(make_precompile_op("pointwise"), {x, y, alloc_ins}, {pw_add1});
        auto alloc_ins2 = mm->add_instruction(alloc);
        auto* pw_add2 =
            create_pointwise_module(p, "main:pointwise1", {x, z}, single_pointwise("add"));
        auto layernorm_op = migraphx::make_op("gpu::prelayernorm");
        auto pre_comp_op  = migraphx::make_op(
            "gpu::precompile_op",
            {{"op", migraphx::to_value(layernorm_op)}, {"output_shape", migraphx::to_value(s)}});

        auto layernorm_ins = mm->add_instruction(pre_comp_op, {add1, z, alloc_ins2}, {pw_add2});
        mm->add_return({layernorm_ins});
        return p;
    };

    {