Commit 7f5c2fef authored by 余昆's avatar 余昆
Browse files

Remove OnlyGpuPass and add MILVUS_GPU_VERSION define

parents 0a858939 0d1923c6

core/conf/server_cpu_config.template

+8 −7
Original line number Diff line number Diff line
@@ -27,8 +27,7 @@ metric_config: 
    port: 8080                      # port prometheus uses to fetch metrics, must in range [1025, 65534]



cache_config:

  cpu_cache_capacity: 16            # GB, CPU memory used for cache, must be a positive integer

  cpu_cache_threshold: 0.85         # percentage of data that will be kept when cache cleanup is triggered, must be in range (0.0, 1.0]

  cpu_cache_capacity: 16            # GB, size of CPU memory used for cache, must be a positive integer

  cache_insert_data: false          # whether to load inserted data into cache, must be a boolean



engine_config:
@@ -36,8 +35,10 @@ engine_config: 
                                    # if nq >= use_blas_threshold, use OpenBlas, slower with stable response times

  gpu_search_threshold: 1000        # threshold beyond which the search computation is executed on GPUs only



resource_config:

  search_resources:                 # define the devices used for search computation, must be in format: cpu or gpux

    - cpu

  index_build_resources:            # define the devices used for index building, must be in format: cpu or gpux

    - cpu
 
gpu_resource_config:

  enable_gpu: false                 # whether to enable GPU resources

  cache_capacity: 4                 # GB, size of GPU memory per card used for cache, must be a positive integer

  search_resources:                 # define the GPU devices used for search computation, must be in format gpux

    - gpu0

  build_index_resources:            # define the GPU devices used for index building, must be in format gpux

    - gpu0

core/conf/server_gpu_config.template

+6 −8
Original line number Diff line number Diff line
@@ -27,10 +27,7 @@ metric_config: 
    port: 8080                      # port prometheus uses to fetch metrics, must in range [1025, 65534]



cache_config:

  cpu_cache_capacity: 16            # GB, CPU memory used for cache, must be a positive integer

  cpu_cache_threshold: 0.85         # percentage of data that will be kept when cache cleanup is triggered, must be in range (0.0, 1.0]

  gpu_cache_capacity: 4             # GB, GPU memory used for cache, must be a positive integer

  gpu_cache_threshold: 0.85         # percentage of data that will be kept when cache cleanup is triggered, must be in range (0.0, 1.0]

  cpu_cache_capacity: 16            # GB, size of CPU memory used for cache, must be a positive integer

  cache_insert_data: false          # whether to load inserted data into cache, must be a boolean



engine_config:
@@ -38,9 +35,10 @@ engine_config: 
                                    # if nq >= use_blas_threshold, use OpenBlas, slower with stable response times

  gpu_search_threshold: 1000        # threshold beyond which the search computation is executed on GPUs only



resource_config:

  search_resources:                 # define the devices used for search computation, must be in format: cpu or gpux

    - cpu

gpu_resource_config:

  enable_gpu: true                  # whether to enable GPU resources

  cache_capacity: 4                 # GB, size of GPU memory per card used for cache, must be a positive integer

  search_resources:                 # define the GPU devices used for search computation, must be in format gpux

    - gpu0

  index_build_resources:            # define the devices used for index building, must be in format: cpu or gpux

  build_index_resources:            # define the GPU devices used for index building, must be in format gpux

    - gpu0

core/src/cache/GpuCacheMgr.cpp

+2 −2
Original line number Diff line number Diff line
@@ -37,7 +37,7 @@ GpuCacheMgr::GpuCacheMgr() { 
    Status s;



    int64_t gpu_cache_cap;

    s = config.GetCacheConfigGpuCacheCapacity(gpu_cache_cap);

    s = config.GetGpuResourceConfigCacheCapacity(gpu_cache_cap);

    if (!s.ok()) {

        SERVER_LOG_ERROR << s.message();

    }
@@ -45,7 +45,7 @@ GpuCacheMgr::GpuCacheMgr() { 
    cache_ = std::make_shared<Cache<DataObjPtr>>(cap, 1UL << 32);



    float gpu_mem_threshold;

    s = config.GetCacheConfigGpuCacheThreshold(gpu_mem_threshold);

    s = config.GetGpuResourceConfigCacheThreshold(gpu_mem_threshold);

    if (!s.ok()) {

        SERVER_LOG_ERROR << s.message();

    }

core/src/db/engine/ExecutionEngineImpl.cpp

+11 −2
Original line number Diff line number Diff line
@@ -144,7 +144,14 @@ ExecutionEngineImpl::HybridLoad() const { 
    }



    const std::string key = location_ + ".quantizer";

    std::vector<uint64_t> gpus = scheduler::get_gpu_pool();



    server::Config& config = server::Config::GetInstance();

    std::vector<int32_t> gpus;

    Status s = config.GetGpuResourceConfigSearchResources(gpus);

    if (!s.ok()) {

        ENGINE_LOG_ERROR << s.message();

        return;

    }



    // cache hit

    {
@@ -578,7 +585,9 @@ ExecutionEngineImpl::GpuCache(uint64_t gpu_id) { 
// TODO(linxj): remove.

Status

ExecutionEngineImpl::Init() {

    auto gpu_ids = scheduler::get_build_resources();

    server::Config& config = server::Config::GetInstance();

    std::vector<int32_t> gpu_ids;

    Status s = config.GetGpuResourceConfigBuildIndexResources(gpu_ids);

    for (auto id : gpu_ids) {

        if (gpu_num_ == id) {

            return Status::OK();

core/src/scheduler/SchedInst.cpp

+8 −11
Original line number Diff line number Diff line
@@ -45,17 +45,6 @@ std::mutex BuildMgrInst::mutex_; 


void

load_simple_config() {

    server::Config& config = server::Config::GetInstance();

    std::string mode;

    config.GetResourceConfigMode(mode);

    std::vector<std::string> pool;

    config.GetResourceConfigSearchResources(pool);



    // get resources

    auto gpu_ids = get_gpu_pool();



    auto build_gpu_ids = get_build_resources();



    // create and connect

    ResMgrInst::GetInstance()->Add(ResourceFactory::Create("disk", "DISK", 0, true, false));
@@ -63,6 +52,13 @@ load_simple_config() { 
    ResMgrInst::GetInstance()->Add(ResourceFactory::Create("cpu", "CPU", 0, true, true));

    ResMgrInst::GetInstance()->Connect("disk", "cpu", io);



    // get resources

#ifdef MILVUS_GPU_VERSION

    server::Config& config = server::Config::GetInstance();

    std::vector<int32_t> gpu_ids;

    config.GetGpuResourceConfigSearchResources(gpu_ids);

    std::vector<int32_t> build_gpu_ids;

    config.GetGpuResourceConfigBuildIndexResources(build_gpu_ids);

    auto pcie = Connection("pcie", 12000);



    std::vector<int64_t> not_find_build_ids;
@@ -89,6 +85,7 @@ load_simple_config() { 
            ResourceFactory::Create(std::to_string(not_find_id), "GPU", not_find_id, true, true));

        ResMgrInst::GetInstance()->Connect("cpu", std::to_string(not_find_id), pcie);

    }

#endif

}



void