rocm-systems/projects/rocprofiler-sdk/tests/pc_sampling/address_translation.cpp

// MIT License
//
// Copyright (c) 2024-2025 ROCm Developer Tools
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in all
// copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.

// undefine NDEBUG so asserts are implemented
#ifdef NDEBUG
#    undef NDEBUG
#endif

#include "address_translation.hpp"
#include "pcs.hpp"
#include "utils.hpp"

#include <cassert>
#include <cstdio>
#include <iostream>
#include <memory>
#include <sstream>
#include <unordered_set>

namespace client
{
namespace address_translation
{
namespace
{
struct FlatProfiler
{
    FlatProfiler()  = default;
    ~FlatProfiler() = default;

    CodeobjAddressTranslate translator        = {};
    KernelObjectMap         kernel_object_map = {};
    FlatProfile             flat_profile      = {};
    std::mutex              global_mut        = {};
};
}  // namespace

// Raw pointer to prevent early destruction of static objects
FlatProfiler* flat_profiler = nullptr;

void
init()
{
    flat_profiler = new FlatProfiler();
}

void
fini()
{
    delete flat_profiler;
    flat_profiler = nullptr;
}

CodeobjAddressTranslate&
get_address_translator()
{
    return flat_profiler->translator;
}

KernelObjectMap&
get_kernel_object_map()
{
    return flat_profiler->kernel_object_map;
}

FlatProfile&
get_flat_profile()
{
    return flat_profiler->flat_profile;
}

std::mutex&
get_global_mutex()
{
    return flat_profiler->global_mut;
}

KernelObject::KernelObject(uint64_t    code_object_id,
                           std::string kernel_name,
                           uint64_t    begin_address,
                           uint64_t    end_address)
: code_object_id_(code_object_id)
, kernel_name_(kernel_name)
, begin_address_(begin_address)
, end_address_(end_address)
{
    auto&    translator = get_address_translator();
    uint64_t vaddr      = begin_address;
    while(vaddr < end_address)
    {
        auto inst = translator.get(code_object_id, vaddr);
        vaddr += inst->size;
        this->add_instruction(std::move(inst));
    }
}

void
dump_flat_profile()
{
    // It seems that an instruction can be part of multiple
    // instances of the same kernel loaded on two different devices.
    // We need to prevent counting the same instruction multiple times.
    std::unordered_set<Instruction*> visited_instructions;

    const auto& kernel_object_map = get_kernel_object_map();
    const auto& flat_profile      = get_flat_profile();

    std::stringstream ss;
    uint64_t          samples_num = 0;
    kernel_object_map.iterate_kernel_objects([&](const KernelObject* kernel_obj) {
        ss << "\n====================================";
        ss << "The kernel: " << kernel_obj->kernel_name()
           << " with the begin address: " << kernel_obj->begin_address()
           << " from code object with id: " << kernel_obj->code_object_id() << std::endl;
        kernel_obj->iterate_instrunctions([&](const Instruction& inst) {
            ss << "\t";
            ss << inst.inst << "\t";
            ss << inst.comment << "\t";
            ss << "samples: ";
            const auto* _sample_instruction = flat_profile.get_sample_instruction(inst);
            if(_sample_instruction == nullptr)
                ss << "0";
            else
            {
                _sample_instruction->process([&](const SampleInstruction& sample_instruction) {
                    ss << sample_instruction.sample_count();
                    // Each instruction should be visited exactly once.
                    // Otherwise, code object loading/unloading and relocations
                    // are not handled properly.
                    assert(visited_instructions.count(sample_instruction.inst()) == 0);
                    // Assure that each instruction is counted once.
                    if(visited_instructions.count(sample_instruction.inst()) == 0)
                    {
                        samples_num += sample_instruction.sample_count();
                        visited_instructions.insert(sample_instruction.inst());
                    }

                    if(sample_instruction.exec_mask_counts().size() <= 1)
                    {
                        ss << ", exec_mask: " << std::hex;
                        ss << sample_instruction.exec_mask_counts().begin()->first;
                        ss << std::dec;
                        assert(sample_instruction.sample_count() ==
                               sample_instruction.exec_mask_counts().begin()->second);
                    }
                    else
                    {
                        uint64_t num_samples_sum = 0;
                        // More than one exec_mask
                        for(auto& [exec_mask, samples_per_exec] :
                            sample_instruction.exec_mask_counts())
                        {
                            ss << std::endl;
                            ss << "\t\t"
                               << "exec_mask: " << std::hex << exec_mask;
                            ss << "\t"
                               << "samples: " << std::dec << samples_per_exec;
                            num_samples_sum += samples_per_exec;
                            ss << std::endl;
                        }
                        assert(sample_instruction.sample_count() == num_samples_sum);
                    }
                });
            }
            ss << std::endl;
        });
        ss << "====================================\n" << std::endl;
    });

    ss << "The total number of valid decoded samples: "
       << flat_profile.get_valid_decoded_samples_num() << std::endl;
    ss << "The total number of invalid samples      : " << flat_profile.get_invalid_samples_num()
       << std::endl;

    *utils::get_output_stream() << ss.str() << std::endl;

    utils::pcs_assert(
        samples_num == flat_profile.get_valid_decoded_samples_num(),
        "Number of collected valid samples different than the number of decoded samples.");
    utils::pcs_assert(samples_num > 0, "No valid samples collected/decoded.");
    // Temporarily disabling the check, until the trap handler latency is fixed.
    // utils::pcs_assert(flat_profile.more_valid_decoded_samples_expected(),
    //                   "More invalid samples observed.");
}

}  // namespace address_translation
}  // namespace client