lowering.cpp

#include <iterator>
#include <migraphx/gpu/lowering.hpp>
#include <migraphx/manage_ptr.hpp>
#include <migraphx/instruction.hpp>
#include <migraphx/make_op.hpp>

#include <migraphx/op/abs.hpp>
#include <migraphx/op/batch_norm_inference.hpp>
#include <migraphx/op/convolution.hpp>
#include <migraphx/op/deconvolution.hpp>
#include <migraphx/op/dot.hpp>
#include <migraphx/op/elu.hpp>
#include <migraphx/op/if_op.hpp>
#include <migraphx/op/leaky_relu.hpp>
#include <migraphx/op/lrn.hpp>
#include <migraphx/op/pooling.hpp>
#include <migraphx/op/reshape.hpp>
#include <migraphx/op/quant_convolution.hpp>
#include <migraphx/op/quant_dot.hpp>

#include <migraphx/gpu/abs.hpp>
#include <migraphx/gpu/batch_norm_inference.hpp>
#include <migraphx/gpu/context.hpp>
#include <migraphx/gpu/convolution.hpp>
#include <migraphx/gpu/deconvolution.hpp>
#include <migraphx/gpu/device_name.hpp>
#include <migraphx/gpu/elu.hpp>
#include <migraphx/gpu/equal.hpp>
#include <migraphx/gpu/gemm.hpp>
#include <migraphx/gpu/greater.hpp>
#include <migraphx/gpu/int8_conv_pack.hpp>
#include <migraphx/gpu/leaky_relu.hpp>
#include <migraphx/gpu/less.hpp>
#include <migraphx/gpu/logical_and.hpp>
#include <migraphx/gpu/logical_or.hpp>
#include <migraphx/gpu/logical_xor.hpp>
#include <migraphx/gpu/lrn.hpp>
#include <migraphx/gpu/miopen.hpp>
#include <migraphx/gpu/quant_convolution.hpp>
#include <migraphx/gpu/rocblas.hpp>
#include <migraphx/gpu/unary_not.hpp>
#include <migraphx/gpu/where.hpp>
#include <migraphx/gpu/compiler.hpp>
#include <migraphx/iterator_for.hpp>
#include <migraphx/program.hpp>
#include <utility>
#include <functional>
#include <algorithm>
#include <map>

namespace migraphx {
inline namespace MIGRAPHX_INLINE_NS {
namespace gpu {

struct miopen_apply
{
    module* mod          = nullptr;
    const lowering* pass = nullptr;
    std::unordered_map<std::string, std::function<instruction_ref(instruction_ref)>> apply_map{};
    instruction_ref last{};
    bool offload_copy   = false;
    bool int8_x4_format = true;
    bool compute_fp32   = false;

    context& get_context() const
    {
        assert(pass != nullptr);
        assert(pass->ctx != nullptr);
        return *pass->ctx;
    }

    void check_shape(shape x, instruction_ref i)
    {
        assert(x == i->get_shape());
        (void)x;
        (void)i;
    }

    const std::unordered_set<std::string>& get_rocblas_fp32_archs()
    {
        static std::unordered_set<std::string> supported_archs{"gfx908", "gfx90a"};
        return supported_archs;
    }

    void init()
    {
        assert(mod != nullptr);
        assert(pass != nullptr);

#if ROCBLAS_VERSION_MAJOR >= 2 && ROCBLAS_VERSION_MINOR >= 38
        auto& ctx              = get_context();
        const auto device_name = trim(split_string(get_device_name(), ':').front());
        if(contains(get_rocblas_fp32_archs(), device_name))
            compute_fp32 = true;
        rocblas_gemm_flags flag;
        rocblas_query_int8_layout_flag(ctx.get_stream().get_rocblas(), &flag);
        int8_x4_format = (flag == rocblas_gemm_flags_pack_int8x4);
#endif

        offload_copy = (mod->name() == "main") ? pass->offload_copy : false;

        add_generic_op("acos");
        add_generic_op("acosh");
        add_generic_op("add");
        add_generic_op("asin");
        add_generic_op("asinh");
        add_generic_op("atan");
        add_generic_op("atanh");
        add_generic_op("ceil");
        add_generic_op("contiguous");
        add_generic_op("cos");
        add_generic_op("cosh");
        add_generic_op("div");
        add_generic_op("equal");
        add_generic_op("erf");
        add_generic_op("exp");
        add_generic_op("floor");
        add_generic_op("greater");
        add_generic_op("less");
        add_generic_op("log");
        add_generic_op("logical_and");
        add_generic_op("logical_or");
        add_generic_op("logical_xor");
        add_generic_op("max");
        add_generic_op("min");
        add_generic_op("mul");
        add_generic_op("not");
        add_generic_op("pow");
        add_generic_op("prelu");
        add_generic_op("recip");
        add_generic_op("relu");
        add_generic_op("round");
        add_generic_op("rsqrt");
        add_generic_op("sigmoid");
        add_generic_op("sign");
        add_generic_op("sin");
        add_generic_op("sinh");
        add_generic_op("sqdiff");
        add_generic_op("sqrt");
        add_generic_op("sub");
        add_generic_op("tan");
        add_generic_op("tanh");
        add_generic_op("where");

        add_extend_op("abs");
        add_extend_op("argmax");
        add_extend_op("argmin");
        add_extend_op("clip");
        add_extend_op("concat");
        add_extend_op("convert");
        add_extend_op("elu");
        add_extend_op("gather");
        add_extend_op("leaky_relu");
        add_extend_op("logsoftmax");
        add_extend_op("lrn");
        add_extend_op("multinomial");
        add_extend_op("nonzero");
        add_extend_op("pad");
        add_extend_op("pooling");
        add_extend_op("prefix_scan_sum");
        add_extend_op("reverse");
        add_extend_op("rnn_var_sl_last_output");
        add_extend_op("rnn_var_sl_shift_output");
        add_extend_op("rnn_var_sl_shift_sequence");
        add_extend_op("scatter_none");
        add_extend_op("topk");

        add_batch_norm_inference_op();
        add_convolution_op();
        add_deconvolution_op();
        add_gemm_op<op::dot>("dot");
        add_gemm_op<op::quant_dot>("quant_dot");
        add_if_op();
        add_loop_op();
        add_neg_op();
        add_nms_op();
        add_quant_convolution_op();
    }

    void copy_params() const
    {
        if(not offload_copy)
            return;

        for(auto ins : iterator_for(*mod))
        {
            if(ins->name() != "@param")
                continue;

            // parameter no outputs, no need to insert copy to gpu
            if(ins->outputs().empty())
                continue;

            auto pos = std::next(ins);
            auto a   = insert_allocation(pos, ins->get_shape());
            auto c   = mod->insert_instruction(pos, make_op("hip::copy_to_gpu"), ins, a);
            mod->replace_instruction(ins, c);
        }

        // return instruction
        auto ret = std::prev(mod->end());
        if(ret->name() == "@return")
        {
            const auto& inputs = ret->inputs();

            // each input of ret need to be copied from gpu to host, and replace
            // output with copy output
            for(const auto& in : inputs)
            {
                auto p_output = mod->insert_instruction(ret, make_op("hip::copy_from_gpu"), in);
                instruction::replace_argument(ret, in, p_output);
            }
        }
        // else branch to handle legacy program without the return instruction
        else
        {
            mod->add_instruction(make_op("hip::copy_from_gpu"), ret);
        }
    }

    void apply()
    {
        init();
        for(auto it = mod->begin(); it != mod->end(); it++)
        {
            auto s = it->get_shape();
            if(apply_map.count(it->name()) > 0)
            {
                check_shape(s, apply_map.at(it->name())(it));
            }
            else if(has_compiler_for(it->name()))
            {
                check_shape(s, insert_precompile_op(it));
            }
        }

        copy_params();
    }

    instruction_ref insert_precompile_op(instruction_ref ins) const
    {
        auto output                       = insert_allocation(ins, ins->get_shape());
        std::vector<instruction_ref> refs = ins->inputs();
        refs.push_back(output);

        return mod->replace_instruction(
            ins,
            make_op("gpu::precompile_op", {{"op", to_value(ins->get_operator())}}),
            refs,
            ins->module_inputs());
    }

    instruction_ref insert_allocation(instruction_ref ins, const shape& s) const
    {
        return mod->insert_instruction(ins, make_op("allocate", {{"shape", to_value(s)}}));
    }

    void add_convolution_op()
    {
        apply_map.emplace("convolution", [=](instruction_ref ins) {
            auto&& op = any_cast<op::convolution>(ins->get_operator());

            auto conv = miopen_convolution{op, make_conv(op)};
            auto ws   = conv.find(get_context(), ins->get_shape(), to_shapes(ins->inputs()));

            auto workspace = insert_allocation(ins, ws);
            auto output    = insert_allocation(ins, ins->get_shape());

            return mod->replace_instruction(
                ins, conv, ins->inputs().at(0), ins->inputs().at(1), workspace, output);
        });
    }

    void add_deconvolution_op()
    {
        apply_map.emplace("deconvolution", [=](instruction_ref ins) {
            auto&& op = any_cast<op::deconvolution>(ins->get_operator());

            auto conv = miopen_deconvolution{op, make_deconv(op)};
            auto ws   = conv.compile(get_context(), ins->get_shape(), to_shapes(ins->inputs()));

            auto workspace = insert_allocation(ins, ws);
            auto output    = insert_allocation(ins, ins->get_shape());

            return mod->replace_instruction(
                ins, conv, ins->inputs().at(0), ins->inputs().at(1), workspace, output);
        });
    }

    template <typename Op>
    void add_gemm_op(const std::string& name)
    {
        apply_map.emplace(name, [=](instruction_ref ins) {
            std::vector<instruction_ref> refs = ins->inputs();
            assert(refs.size() == 2);
            auto output = insert_allocation(ins, ins->get_shape());
            refs.push_back(output);
            return mod->replace_instruction(
                ins, rocblas_gemm<Op>{Op{}, 1, 0, int8_x4_format, compute_fp32}, refs);
        });
    }

    void add_quant_convolution_op()
    {
        apply_map.emplace("quant_convolution", [=](instruction_ref ins) {
            auto&& op = any_cast<op::quant_convolution>(ins->get_operator());
            shape ws;
            miopen_quant_convolution conv;
            auto compile_quant_conv_with_format = [&](bool format) {
                conv = miopen_quant_convolution{op, format, make_conv(op)};
                ws   = conv.compile(get_context(), ins->get_shape(), to_shapes(ins->inputs()));
            };

            try
            {
                compile_quant_conv_with_format(int8_x4_format);
            }
            catch(migraphx::exception&)
            {
                // In case no solver supports the default format, retry using the other format.
                compile_quant_conv_with_format(!int8_x4_format);
            }

            auto args      = ins->inputs();
            auto workspace = insert_allocation(ins, ws);
            auto output    = insert_allocation(ins, ins->get_shape());

            return mod->replace_instruction(ins, conv, args[0], args[1], workspace, output);
        });
    }

    // add_generic_op just constructs the operator with no fields whereas add_extend_op copies over
    // the fields Since it doesn't have fields its default constructed

    void add_generic_op(const std::string& name) { add_generic_op(name, "gpu::" + name); }

    void add_generic_op(const std::string& op_name, const std::string& gpu_name)
    {
        apply_map.emplace(op_name, [=](instruction_ref ins) {
            auto output                       = insert_allocation(ins, ins->get_shape());
            std::vector<instruction_ref> refs = ins->inputs();
            refs.push_back(output);

            return mod->replace_instruction(ins, make_op(gpu_name), refs);
        });
    }

    void add_extend_op(const std::string& name) { add_extend_op(name, "gpu::" + name); }

    void add_extend_op(const std::string& op_name, const std::string& gpu_name)
    {
        apply_map.emplace(op_name, [=](instruction_ref ins) {
            auto&& op                         = ins->get_operator();
            auto output                       = insert_allocation(ins, ins->get_shape());
            std::vector<instruction_ref> refs = ins->inputs();
            refs.push_back(output);

            return mod->replace_instruction(ins, make_op(gpu_name, op.to_value()), refs);
        });
    }

    void add_batch_norm_inference_op()
    {
        apply_map.emplace("batch_norm_inference", [=](instruction_ref ins) {
            auto&& op       = any_cast<op::batch_norm_inference>(ins->get_operator());
            auto output     = insert_allocation(ins, ins->get_shape());
            shape old_shape = ins->inputs().at(1)->get_shape();
            auto input      = ins->inputs()[0];
            auto input_lens = input->get_shape().lens();
            std::vector<int64_t> rsp_lens(input_lens.size(), 1);
            // for per_activation case, also need to reshape input
            if(op.bn_mode == op::batch_norm_inference::per_activation)
            {
                std::copy(input_lens.begin() + 1, input_lens.end(), rsp_lens.begin() + 1);
            }
            else
            {
                rsp_lens[1] = static_cast<int64_t>(old_shape.elements());
            }

            auto reshape_op = op::reshape{rsp_lens};
            std::vector<instruction_ref> reshapes;
            std::transform(ins->inputs().begin() + 1,
                           ins->inputs().end(),
                           std::back_inserter(reshapes),
                           [&](auto i) { return mod->insert_instruction(ins, reshape_op, i); });

            return mod->replace_instruction(ins,
                                            miopen_batch_norm_inference{op},
                                            input,
                                            reshapes[0],
                                            reshapes[1],
                                            reshapes[2],
                                            reshapes[3],
                                            output);
        });
    }

    // use 0 - input to represent neg
    void add_neg_op()
    {
        apply_map.emplace("neg", [=](instruction_ref ins) {
            auto s = ins->get_shape();
            std::vector<float> zeros(s.elements(), 0.0f);
            auto l0     = mod->add_literal(literal(s, zeros));
            auto output = insert_allocation(ins, s);
            return mod->replace_instruction(
                ins, make_op("gpu::sub"), l0, ins->inputs().front(), output);
        });
    }

    // add input and output argument for the if operator
    void add_if_op()
    {
        apply_map.emplace("if", [=](instruction_ref ins) {
            std::vector<instruction_ref> inputs = ins->inputs();
            auto cpu_cond =
                mod->insert_instruction(ins, make_op("hip::copy_from_gpu"), inputs.front());
            auto sync_cond = mod->insert_instruction(ins, make_op("hip::sync_stream"), cpu_cond);
            inputs.front() = sync_cond;

            return mod->replace_instruction(ins, ins->get_operator(), inputs, ins->module_inputs());
        });
    }

    // replace the loop operator with gpu_loop operator
    void add_loop_op()
    {
        apply_map.emplace("loop", [=](instruction_ref ins) {
            std::vector<instruction_ref> inputs = ins->inputs();
            // copy max_iter from gpu to cpu
            auto cpu_max_iter =
                mod->insert_instruction(ins, make_op("hip::copy_from_gpu"), inputs.at(0));
            auto cpu_cond =
                mod->insert_instruction(ins, make_op("hip::copy_from_gpu"), inputs.at(1));
            auto synced_max_iter =
                mod->insert_instruction(ins, make_op("hip::sync_stream"), cpu_max_iter, cpu_cond);
            inputs.at(0)     = synced_max_iter;
            inputs.at(1)     = cpu_cond;
            auto copy_inputs = inputs;
            std::transform(copy_inputs.begin(),
                           copy_inputs.end(),
                           std::back_inserter(inputs),
                           [&](auto in) { return insert_allocation(ins, in->get_shape()); });

            auto mod_args = ins->module_inputs();
            auto output   = insert_allocation(ins, ins->get_shape());

            const auto* sub_mod = mod_args.front();
            auto cond_out       = insert_allocation(ins, sub_mod->get_output_shapes().front());

            // add cond and mod outputs to the argument list
            inputs.push_back(cond_out);
            inputs.push_back(output);

            return mod->replace_instruction(
                ins, make_op("gpu::loop", ins->get_operator().to_value()), inputs, mod_args);
        });
    }

    void add_nms_op()
    {
        apply_map.emplace("nonmaxsuppression", [=](instruction_ref ins) {
            auto s      = ins->get_shape();
            auto output = insert_allocation(ins, s);
            std::vector<instruction_ref> cpu_inputs;
            auto inputs = ins->inputs();
            std::transform(
                inputs.begin(), inputs.end(), std::back_inserter(cpu_inputs), [&](auto in) {
                    return mod->insert_instruction(ins, make_op("hip::copy_from_gpu"), in);
                });
            cpu_inputs.front() =
                mod->insert_instruction(ins, make_op("hip::sync_stream"), cpu_inputs);
            auto cpu_out = mod->insert_instruction(ins, ins->get_operator(), cpu_inputs);
            auto gpu_out =
                mod->insert_instruction(ins, make_op("hip::copy_to_gpu"), cpu_out, output);
            return mod->replace_instruction(ins, gpu_out);
        });
    }
};

void lowering::apply(module& m) const { miopen_apply{&m, this}.apply(); }

} // namespace gpu
} // namespace MIGRAPHX_INLINE_NS
} // namespace migraphx