Merge branch 'branch-0.4.0' into 'branch-0.4.0'

MS-442 make it work

See merge request megasearch/milvus!458

Former-commit-id: c79aaefba6e0c78622b531a9d2ebe58af9126fed

cpp/CMakeLists.txt

@@ -34,7 +34,7 @@ else()
 endif()
 message(STATUS "Build type = ${BUILD_TYPE}")
 
-#add_definitions(-DNEW_SCHEDULER)
+add_definitions(-DNEW_SCHEDULER)
 
 project(milvus VERSION "${MILVUS_VERSION}")
 project(milvus_engine LANGUAGES CUDA CXX)

cpp/conf/server_config.template

@@ -66,24 +66,24 @@ resource_config:
       enable_loader: true
       enable_executor: true
 
-    gtx1060:
-      type: GPU
-      memory: 6
-      device_id: 0
-      enable_loader: true
-      enable_executor: true
+#    gtx1060:
+#      type: GPU
+#      memory: 6
+#      device_id: 0
+#      enable_loader: true
+#      enable_executor: true
 
-    gtx1660:
-      type: GPU
-      memory: 6
-      device_id: 1
-      enable_loader: true
-      enable_executor: true
+#    gtx1660:
+#      type: GPU
+#      memory: 6
+#      device_id: 1
+#      enable_loader: false
+#      enable_executor: false
 
   # connection list, length: 0~N
   # format: -${resource_name}===${resource_name}
   connections:
     - ssda===cpu
-    - cpu===gtx1060
-    - cpu===gtx1660
+#    - cpu===gtx1060
+#    - cpu===gtx1660
 

cpp/src/scheduler/SchedInst.cpp

@@ -7,6 +7,7 @@
 #include "SchedInst.h"
 #include "server/ServerConfig.h"
 #include "ResourceFactory.h"
+#include "knowhere/index/vector_index/gpu_ivf.h"
 
 namespace zilliz {
 namespace milvus {
@@ -36,8 +37,12 @@ SchedServInit() {
                                                                device_id,
                                                                enable_loader,
                                                                enable_executor));
+
+        knowhere::FaissGpuResourceMgr::GetInstance().InitDevice(device_id);
     }
 
+    knowhere::FaissGpuResourceMgr::GetInstance().InitResource();
+
     auto default_connection = Connection("default_connection", 500.0);
     auto connections = config.GetSequence(server::CONFIG_RESOURCE_CONNECTIONS);
     for (auto &conn : connections) {

cpp/src/wrapper/knowhere/vec_impl.cpp

@@ -241,8 +241,9 @@ server::KnowhereError IVFMixIndex::BuildAll(const long &nb,
         index_->Add(dataset, cfg);
 
         if (auto device_index = std::dynamic_pointer_cast<GPUIVF>(index_)) {
-            auto host_index = device_index->Copy_index_gpu_to_cpu();
+            auto host_index = device_index->CopyGpuToCpu(Config());
             index_ = host_index;
+            type = TransferToCpuIndexType(type);
         } else {
             WRAPPER_LOG_ERROR << "Build IVFMIXIndex Failed";
         }

cpp/src/wrapper/knowhere/vec_index.cpp

@@ -106,6 +106,10 @@ VecIndexPtr GetVecIndexFactory(const IndexType &type) {
             index = std::make_shared<zilliz::knowhere::GPUIVFSQ>(0);
             return std::make_shared<IVFMixIndex>(index, IndexType::FAISS_IVFSQ8_MIX);
         }
+        case IndexType::FAISS_IVFSQ8: {
+            index = std::make_shared<zilliz::knowhere::IVFSQ>();
+            break;
+        }
         case IndexType::NSG_MIX: { // TODO(linxj): bug.
             index = std::make_shared<zilliz::knowhere::NSG>(0);
             break;
@@ -194,10 +198,10 @@ server::KnowhereError write_index(VecIndexPtr index, const std::string &location
 // TODO(linxj): redo here.
 void AutoGenParams(const IndexType &type, const long &size, zilliz::knowhere::Config &cfg) {
     auto nlist = cfg.get_with_default("nlist", 0);
-    if (size <= TYPICAL_COUNT/16384 + 1) {
+    if (size <= TYPICAL_COUNT / 16384 + 1) {
         //handle less row count, avoid nlist set to 0
         cfg["nlist"] = 1;
-    } else if (int(size/TYPICAL_COUNT) * nlist == 0) {
+    } else if (int(size / TYPICAL_COUNT) * nlist == 0) {
         //calculate a proper nlist if nlist not specified or size less than TYPICAL_COUNT
         cfg["nlist"] = int(size / TYPICAL_COUNT * 16384);
     }
@@ -225,6 +229,20 @@ void AutoGenParams(const IndexType &type, const long &size, zilliz::knowhere::Co
     }
 }
 
+IndexType TransferToCpuIndexType(const IndexType &type) {
+    switch (type) {
+        case IndexType::FAISS_IVFFLAT_MIX: {
+            return IndexType::FAISS_IVFFLAT_CPU;
+        }
+        case IndexType::FAISS_IVFSQ8_MIX: {
+            return IndexType::FAISS_IVFSQ8;
+        }
+        default: {
+            return IndexType::INVALID;
+        }
+    }
+}
+
 }
 }
 }

cpp/src/wrapper/knowhere/vec_index.h

@@ -32,6 +32,7 @@ enum class IndexType {
     FAISS_IVFPQ_GPU,
     SPTAG_KDT_RNT_CPU,
     FAISS_IVFSQ8_MIX,
+    FAISS_IVFSQ8,
     NSG_MIX,
 };
 
@@ -88,6 +89,8 @@ extern VecIndexPtr LoadVecIndex(const IndexType &index_type, const zilliz::knowh
 
 extern void AutoGenParams(const IndexType& type, const long& size, Config& cfg);
 
+extern IndexType TransferToCpuIndexType(const IndexType& type);
+
 }
 }
 }

cpp/unittest/knowhere/knowhere_test.cpp

@@ -8,13 +8,14 @@
 #include <easylogging++.h>
 
 #include <wrapper/knowhere/vec_index.h>
+#include "knowhere/index/vector_index/gpu_ivf.h"
 
 #include "utils.h"
 
 INITIALIZE_EASYLOGGINGPP
 
 using namespace zilliz::milvus::engine;
-using namespace zilliz::knowhere;
+//using namespace zilliz::knowhere;
 
 using ::testing::TestWithParam;
 using ::testing::Values;
@@ -66,8 +67,8 @@ class KnowhereWrapperTest
     Config train_cfg;
     Config search_cfg;
 
-    int dim = 64;
-    int nb = 10000;
+    int dim = 512;
+    int nb = 1000000;
     int nq = 10;
     int k = 10;
     std::vector<float> xb;
@@ -94,27 +95,27 @@ INSTANTIATE_TEST_CASE_P(WrapperParam, KnowhereWrapperTest,
                             //                Config::object{{"nlist", 100}, {"dim", 64}},
                             //                Config::object{{"dim", 64}, {"k", 10}, {"nprobe", 40}}
                             //),
-                            std::make_tuple(IndexType::FAISS_IVFFLAT_MIX, "Default",
-                                            64, 100000, 10, 10,
-                                            Config::object{{"nlist", 1000}, {"dim", 64}, {"metric_type", "L2"}},
-                                            Config::object{{"dim", 64}, {"k", 10}, {"nprobe", 5}}
-                            ),
-                            std::make_tuple(IndexType::FAISS_IDMAP, "Default",
-                                            64, 100000, 10, 10,
-                                            Config::object{{"dim", 64}, {"metric_type", "L2"}},
-                                            Config::object{{"dim", 64}, {"k", 10}}
-                            ),
+//                            std::make_tuple(IndexType::FAISS_IVFFLAT_MIX, "Default",
+//                                            64, 100000, 10, 10,
+//                                            Config::object{{"nlist", 1000}, {"dim", 64}, {"metric_type", "L2"}},
+//                                            Config::object{{"dim", 64}, {"k", 10}, {"nprobe", 5}}
+//                            ),
+//                            std::make_tuple(IndexType::FAISS_IDMAP, "Default",
+//                                            64, 100000, 10, 10,
+//                                            Config::object{{"dim", 64}, {"metric_type", "L2"}},
+//                                            Config::object{{"dim", 64}, {"k", 10}}
+//                            ),
                             std::make_tuple(IndexType::FAISS_IVFSQ8_MIX, "Default",
-                                            64, 100000, 10, 10,
-                                            Config::object{{"dim", 64}, {"nlist", 1000}, {"nbits", 8}, {"metric_type", "L2"}},
-                                            Config::object{{"dim", 64}, {"k", 10}, {"nprobe", 5}}
-                            ),
-                            std::make_tuple(IndexType::NSG_MIX, "Default",
-                                            128, 250000, 10, 10,
-                                            Config::object{{"dim", 128}, {"nlist", 8192}, {"nprobe", 16}, {"metric_type", "L2"},
-                                                           {"knng", 200}, {"search_length", 40}, {"out_degree", 60}, {"candidate_pool_size", 200}},
-                                            Config::object{{"k", 10}, {"search_length", 20}}
+                                            512, 1000000, 10, 10,
+                                            Config::object{{"dim", 512}, {"nlist", 1000}, {"nbits", 8}, {"metric_type", "L2"}},
+                                            Config::object{{"dim", 512}, {"k", 10}, {"nprobe", 5}}
                             )
+//                            std::make_tuple(IndexType::NSG_MIX, "Default",
+//                                            128, 250000, 10, 10,
+//                                            Config::object{{"dim", 128}, {"nlist", 8192}, {"nprobe", 16}, {"metric_type", "L2"},
+//                                                           {"knng", 200}, {"search_length", 40}, {"out_degree", 60}, {"candidate_pool_size", 200}},
+//                                            Config::object{{"k", 10}, {"search_length", 20}}
+//                            )
                             //std::make_tuple(IndexType::SPTAG_KDT_RNT_CPU, "Default",
                             //                64, 10000, 10, 10,
                             //                Config::object{{"TPTNumber", 1}, {"dim", 64}},
@@ -135,6 +136,34 @@ TEST_P(KnowhereWrapperTest, base_test) {
     AssertResult(res_ids, res_dis);
 }
 
+TEST_P(KnowhereWrapperTest, to_gpu_test) {
+    EXPECT_EQ(index_->GetType(), index_type);
+
+    zilliz::knowhere::FaissGpuResourceMgr::GetInstance().InitDevice(0);
+    zilliz::knowhere::FaissGpuResourceMgr::GetInstance().InitDevice(1);
+
+    auto elems = nq * k;
+    std::vector<int64_t> res_ids(elems);
+    std::vector<float> res_dis(elems);
+
+    index_->BuildAll(nb, xb.data(), ids.data(), train_cfg);
+    index_->Search(nq, xq.data(), res_dis.data(), res_ids.data(), search_cfg);
+    AssertResult(res_ids, res_dis);
+    {
+        index_->CopyToGpu(1);
+    }
+
+    std::string file_location = "/tmp/whatever";
+    write_index(index_, file_location);
+    auto new_index = read_index(file_location);
+
+    auto dev_idx = new_index->CopyToGpu(1);
+    for (int i = 0; i < 10000; ++i) {
+        dev_idx->Search(nq, xq.data(), res_dis.data(), res_ids.data(), search_cfg);
+    }
+    AssertResult(res_ids, res_dis);
+}
+
 TEST_P(KnowhereWrapperTest, serialize) {
     EXPECT_EQ(index_->GetType(), index_type);