TensorRT-LLMs/cpp/tensorrt_llm/plugins/ncclPlugin/allreducePlugin.cpp

/*
 * SPDX-FileCopyrightText: Copyright (c) 1993-2022 NVIDIA CORPORATION &
 * AFFILIATES. All rights reserved. SPDX-License-Identifier: Apache-2.0
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
#include "allreducePlugin.h"
#include "mpi.h"
#include "plugin.h"
#include "tensorrt_llm/common/mpiUtils.h"
#include "tensorrt_llm/kernels/customAllReduceKernels.h"
#include <unistd.h>

using namespace nvinfer1;
using tensorrt_llm::plugins::AllreducePluginCreator;
using tensorrt_llm::plugins::AllreducePlugin;
using tensorrt_llm::kernels::AllReduceStrategyType;

static const char* ALLREDUCE_PLUGIN_VERSION{"1"};
static const char* ALLREDUCE_PLUGIN_NAME{"AllReduce"};
PluginFieldCollection AllreducePluginCreator::mFC{};
std::vector<nvinfer1::PluginField> AllreducePluginCreator::mPluginAttributes;

AllreducePlugin::AllreducePlugin(
    std::set<int> group, nvinfer1::DataType type, AllReduceStrategyType strategy, int32_t counter)
    : mGroup(group)
    , mType(type)
    , mStrategy(strategy)
    , mCounter(counter)
{
}

// Parameterized constructor
AllreducePlugin::AllreducePlugin(const void* data, size_t length)
{
    const char *d = reinterpret_cast<const char*>(data), *a = d;
    read(d, mType);
    read(d, mStrategy);
    read(d, mCounter);
    mGroup.clear();
    int groupItem = 0;
    while (d != a + length)
    {
        read(d, groupItem);
        mGroup.insert(groupItem);
    }
    TLLM_CHECK(d == a + length);
}

// IPluginV2DynamicExt Methods
nvinfer1::IPluginV2DynamicExt* AllreducePlugin::clone() const noexcept
{
    auto* plugin = new AllreducePlugin(*this);
    plugin->setPluginNamespace(mNamespace.c_str());
    return plugin;
}

nvinfer1::DimsExprs AllreducePlugin::getOutputDimensions(
    int outputIndex, const nvinfer1::DimsExprs* inputs, int nbInputs, nvinfer1::IExprBuilder& exprBuilder) noexcept
{
    return inputs[0];
}

bool AllreducePlugin::supportsFormatCombination(
    int pos, const nvinfer1::PluginTensorDesc* inOut, int nbInputs, int nbOutputs) noexcept
{
    if (mStrategy == AllReduceStrategyType::RING)
    {
        TLLM_CHECK_WITH_INFO(nbInputs == 1, "RING (aka. NCCL) strategy only accepts one input.");
    }
    else
    {
        TLLM_CHECK_WITH_INFO(nbInputs == 2, "Non-RING (aka. NCCL) strategies require a workspace tensor.");
    }

    if (nbInputs == 2 && pos == 1)
    {
        return (inOut[pos].type == nvinfer1::DataType::kINT64) && (inOut[pos].format == TensorFormat::kLINEAR);
    }
    else
    {
        return (inOut[pos].type == mType) && (inOut[pos].format == TensorFormat::kLINEAR);
    }
}

void AllreducePlugin::configurePlugin(const nvinfer1::DynamicPluginTensorDesc* in, int nbInputs,
    const nvinfer1::DynamicPluginTensorDesc* out, int nbOutputs) noexcept
{
}

size_t AllreducePlugin::getWorkspaceSize(const nvinfer1::PluginTensorDesc* inputs, int nbInputs,
    const nvinfer1::PluginTensorDesc* outputs, int nbOutputs) const noexcept
{
    return 0;
}

AllReduceStrategyType AllreducePlugin::selectImplementation(size_t messageSize, int worldSize) const noexcept
{
    if (worldSize <= 2)
    {
        if (messageSize < 16 * 1000 * 1000)
        {
            return AllReduceStrategyType::ONESHOT;
        }
    }

    if (worldSize > 2 && worldSize <= 4)
    {
        if (messageSize < 1 * 1000 * 1000)
        {
            return AllReduceStrategyType::ONESHOT;
        }
        if (messageSize < 8 * 1000 * 1000)
        {
            return AllReduceStrategyType::TWOSHOT;
        }
    }

    if (worldSize > 4)
    {
        if (messageSize < 500 * 1000)
        {
            return AllReduceStrategyType::ONESHOT;
        }
        if (messageSize < 8 * 1000 * 1000)
        {
            return AllReduceStrategyType::TWOSHOT;
        }
    }

    return AllReduceStrategyType::RING;
}

int AllreducePlugin::enqueue(const nvinfer1::PluginTensorDesc* inputDesc, const nvinfer1::PluginTensorDesc* outputDesc,
    const void* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept
{
    if (isBuilding())
    {
        return 0;
    }
    int size = 1;
    for (int i = 0; i < inputDesc[0].dims.nbDims; ++i)
    {
        size *= inputDesc[0].dims.d[i];
    }
    size_t sizePerElem = 0;
    using tensorrt_llm::common::datatype_enum;
    datatype_enum type;
    switch (mType)
    {
    case DataType::kFLOAT:
        sizePerElem = sizeof(float);
        type = datatype_enum::TYPE_FP32;
        break;
    case DataType::kHALF:
        sizePerElem = sizeof(half);
        type = datatype_enum::TYPE_FP16;
        break;
#ifdef ENABLE_BF16
    case DataType::kBF16:
        sizePerElem = sizeof(__nv_bfloat16);
        type = datatype_enum::TYPE_BF16;
        break;
#endif
    default: break;
    }

    auto runtimeStrategy = mStrategy;
    if (runtimeStrategy == AllReduceStrategyType::AUTO)
    {
        runtimeStrategy = selectImplementation(size * sizePerElem, mGroup.size());
    }

    if (runtimeStrategy == AllReduceStrategyType::RING)
    {
        NCCLCHECK(ncclAllReduce(inputs[0], outputs[0], size, (*getDtypeMap())[inputDesc[0].type], ncclSum,
            (*getCommMap())[mGroup], stream));
    }
    else
    {
        int myRank;
        int nRanks = inputDesc[1].dims.d[0] / 3;
        MPICHECK(MPI_Comm_rank(MPI_COMM_WORLD, &myRank));
        // FIXME: pass world config here
        myRank = myRank % nRanks;

        auto params = tensorrt_llm::kernels::AllReduceParams::deserialize(
            reinterpret_cast<const int32_t*>(inputs[1]), nRanks, myRank, mCounter);

        // Make sure all GPUs have finished using their peer_comm_buffer_ptrs in previous invocations
        tensorrt_llm::kernels::invokeMultiGpuBarrier(params, stream);
        cudaMemcpyAsync(
            params.peer_comm_buffer_ptrs[myRank], inputs[0], size * sizePerElem, cudaMemcpyDeviceToDevice, stream);

        tensorrt_llm::kernels::customAllReduce(params, outputs[0], size, sizePerElem, type, runtimeStrategy, stream);
    }

    return 0;
}

// IPluginV2Ext Methods
nvinfer1::DataType AllreducePlugin::getOutputDataType(
    int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept
{
    assert(index == 0);
    return inputTypes[0];
}

// IPluginV2 Methods

const char* AllreducePlugin::getPluginType() const noexcept
{
    return ALLREDUCE_PLUGIN_NAME;
}

const char* AllreducePlugin::getPluginVersion() const noexcept
{
    return ALLREDUCE_PLUGIN_VERSION;
}

int AllreducePlugin::getNbOutputs() const noexcept
{
    return 1;
}

bool AllreducePlugin::isCustomAllReduceSuported(int ranks_per_node) const noexcept
{
    constexpr bool isCudaVersionSupported =
#if defined(CUDART_VERSION) && CUDART_VERSION >= 11020
        true;
#else
        false;
#endif

    return isCudaVersionSupported && (ranks_per_node % 2 == 0) && (ranks_per_node <= kernels::MAX_RANKS_PER_NODE)
        && (ranks_per_node > 0);
}

int AllreducePlugin::initialize() noexcept
{
    if (isBuilding() || mStrategy == AllReduceStrategyType::ONESHOT || mStrategy == AllReduceStrategyType::TWOSHOT)
    {
        return 0;
    }

    int myRank, nRanks;
    MPICHECK(MPI_Comm_rank(MPI_COMM_WORLD, &myRank));
    MPICHECK(MPI_Comm_size(MPI_COMM_WORLD, &nRanks));

    int deviceId;
    cudaGetDevice(&deviceId);

    auto* commMap = getCommMap();
    // [] operator inserts T() if it does not exist
    if ((*commMap)[mGroup] != nullptr)
    {
        return 0;
    }

    int groupRank = 0;
    for (auto it = mGroup.begin(); it != mGroup.end(); ++it)
    {
        if (*it == myRank)
        {
            break;
        }
        ++groupRank;
    }

    ncclUniqueId id;
    if (myRank == *mGroup.begin())
    {
        ncclGetUniqueId(&id);
        for (auto it = std::next(std::begin(mGroup), 1); it != mGroup.end(); ++it)
        {
            MPICHECK(MPI_Send(&id, sizeof(id), MPI_BYTE, *it, 0, MPI_COMM_WORLD));
        }
    }
    else
    {
        MPI_Status status;
        MPICHECK(MPI_Recv(&id, sizeof(id), MPI_BYTE, *mGroup.begin(), 0, MPI_COMM_WORLD, &status));
    }
    (*commMap)[mGroup] = nullptr;
    NCCLCHECK(ncclCommInitRank(&((*commMap)[mGroup]), mGroup.size(), id, groupRank));

    return 0;
}

void AllreducePlugin::terminate() noexcept
{
    if (mStrategy == AllReduceStrategyType::RING || mStrategy == AllReduceStrategyType::AUTO)
    {
        auto* commMap = getCommMap();
        // [] operator inserts T() if it does not exist
        if (isBuilding() || (*commMap)[mGroup] == nullptr)
        {
            return;
        }
        NCCLCHECK(ncclCommDestroy((*commMap)[mGroup]));
        (*commMap)[mGroup] = nullptr;
    }
}

size_t AllreducePlugin::getSerializationSize() const noexcept
{
    return sizeof(int) * mGroup.size() + sizeof(mType) + sizeof(mStrategy) + sizeof(mCounter);
}

void AllreducePlugin::serialize(void* buffer) const noexcept
{
    char *d = static_cast<char*>(buffer), *a = d;
    write(d, mType);
    write(d, mStrategy);
    write(d, mCounter);
    for (auto it = mGroup.begin(); it != mGroup.end(); ++it)
    {
        write(d, *it);
    }
    assert(d == a + getSerializationSize());
}

void AllreducePlugin::destroy() noexcept
{
    // This gets called when the network containing plugin is destroyed
    delete this;
}

///////////////

AllreducePluginCreator::AllreducePluginCreator()
{
    // Fill PluginFieldCollection with PluginField arguments metadata
    mPluginAttributes.clear();
    mPluginAttributes.emplace_back(PluginField("group", nullptr, PluginFieldType::kINT32, 1));
    mPluginAttributes.emplace_back(PluginField("type_id", nullptr, PluginFieldType::kINT32, 1));
    mPluginAttributes.emplace_back(PluginField("strategy", nullptr, PluginFieldType::kINT8, 1));
    mPluginAttributes.emplace_back(PluginField("counter", nullptr, PluginFieldType::kINT32, 1));
    mFC.nbFields = mPluginAttributes.size();
    mFC.fields = mPluginAttributes.data();
}

const char* AllreducePluginCreator::getPluginName() const noexcept
{
    return ALLREDUCE_PLUGIN_NAME;
}

const char* AllreducePluginCreator::getPluginVersion() const noexcept
{
    return ALLREDUCE_PLUGIN_VERSION;
}

const PluginFieldCollection* AllreducePluginCreator::getFieldNames() noexcept
{
    return &mFC;
}

IPluginV2* AllreducePluginCreator::createPlugin(const char* name, const PluginFieldCollection* fc) noexcept
{
    const PluginField* fields = fc->fields;
    std::set<int> group;
    nvinfer1::DataType type;
    AllReduceStrategyType strategy;
    int32_t counter;
    // Read configurations from each fields
    for (int i = 0; i < fc->nbFields; ++i)
    {
        const char* attrName = fields[i].name;
        if (!strcmp(attrName, "group"))
        {
            TLLM_CHECK(fields[i].type == PluginFieldType::kINT32);
            const auto* r = static_cast<const int*>(fields[i].data);
            for (int j = 0; j < fields[i].length; ++j)
            {
                group.insert(*r);
                ++r;
            }
        }
        else if (!strcmp(attrName, "type_id"))
        {
            TLLM_CHECK(fields[i].type == PluginFieldType::kINT32);
            type = static_cast<nvinfer1::DataType>(*(static_cast<const nvinfer1::DataType*>(fields[i].data)));
        }
        else if (!strcmp(attrName, "strategy"))
        {
            TLLM_CHECK(fields[i].type == PluginFieldType::kINT8);
            strategy = static_cast<AllReduceStrategyType>(*static_cast<const int8_t*>(fields[i].data));
        }
        else if (!strcmp(attrName, "counter"))
        {
            TLLM_CHECK(fields[i].type == PluginFieldType::kINT32);
            counter = *static_cast<const int32_t*>(fields[i].data);
        }
    }

    try
    {
        auto* obj = new AllreducePlugin(group, type, strategy, counter);
        obj->setPluginNamespace(mNamespace.c_str());
        return obj;
    }
    catch (const std::exception& e)
    {
        caughtError(e);
    }
    return nullptr;
}

IPluginV2* AllreducePluginCreator::deserializePlugin(
    const char* name, const void* serialData, size_t serialLength) noexcept
{
    // This object will be deleted when the network is destroyed, which will
    // call AllreducePlugin::destroy()
    try
    {
        auto* obj = new AllreducePlugin(serialData, serialLength);
        obj->setPluginNamespace(mNamespace.c_str());
        return obj;
    }
    catch (const std::exception& e)
    {
        caughtError(e);
    }
    return nullptr;
}