Clr core (#376)

* add .net core support

* add linux nuget

* fix linux nuspec

* fix CsharpClient.vcxproj

* fix Linux and Windows conflict

* fix windows nuget package

* add dump and loadfromdump

* fix setup.py

* fix cuda LOG

* fix GPU log

* fix Dockerfile for ubuntu20.04

* fix Dockerfile

---------

Co-authored-by: cheqi <cheqi@SRGSSD-07>

Author: MaggieQi

AnnService/inc/Core/Common/cuda/GPUQuantizer.hxx

@@ -89,7 +89,7 @@ class GPU_Quantizer {
 
       // Make sure L2 is used, since other 
       if(metric != DistMetric::L2) {
-        LOG(Helper::LogLevel::LL_Error, "Only L2 distance currently supported for PQ or OPQ\n");
+        SPTAGLIB_LOG(Helper::LogLevel::LL_Error, "Only L2 distance currently supported for PQ or OPQ\n");
         exit(1);
       }
 
@@ -103,7 +103,7 @@ class GPU_Quantizer {
         m_BlockSize = pq_quantizer->GetBlockSize();
         m_DimPerSubvector = pq_quantizer->GetDimPerSubvector();
 
-        LOG(Helper::LogLevel::LL_Debug, "Using PQ - numSubVectors:%d, KsPerSub:%ld, BlockSize:%ld, DimPerSub:%d, total size of tables:%ld\n", m_NumSubvectors, m_KsPerSubvector, m_BlockSize, m_DimPerSubvector, m_BlockSize*m_NumSubvectors*sizeof(float));
+        SPTAGLIB_LOG(Helper::LogLevel::LL_Debug, "Using PQ - numSubVectors:%d, KsPerSub:%ld, BlockSize:%ld, DimPerSub:%d, total size of tables:%ld\n", m_NumSubvectors, m_KsPerSubvector, m_BlockSize, m_DimPerSubvector, m_BlockSize*m_NumSubvectors*sizeof(float));
 
         rType = pq_quantizer->GetReconstructType();
         CUDA_CHECK(cudaMalloc(&m_DistanceTables, m_BlockSize * m_NumSubvectors * sizeof(float)));
@@ -117,14 +117,14 @@ class GPU_Quantizer {
         m_BlockSize = opq_quantizer->GetBlockSize();
         m_DimPerSubvector = opq_quantizer->GetDimPerSubvector();
 
-        LOG(Helper::LogLevel::LL_Debug, "Using OPQ - numSubVectors:%d, KsPerSub:%ld, BlockSize:%ld, DimPerSub:%d, total size of tables:%ld\n", m_NumSubvectors, m_KsPerSubvector, m_BlockSize, m_DimPerSubvector, m_BlockSize*m_NumSubvectors*sizeof(float));
+        SPTAGLIB_LOG(Helper::LogLevel::LL_Debug, "Using OPQ - numSubVectors:%d, KsPerSub:%ld, BlockSize:%ld, DimPerSub:%d, total size of tables:%ld\n", m_NumSubvectors, m_KsPerSubvector, m_BlockSize, m_DimPerSubvector, m_BlockSize*m_NumSubvectors*sizeof(float));
 
         rType = opq_quantizer->GetReconstructType();
         CUDA_CHECK(cudaMalloc(&m_DistanceTables, m_BlockSize * m_NumSubvectors * sizeof(float)));
         CUDA_CHECK(cudaMemcpy(m_DistanceTables, opq_quantizer->GetL2DistanceTables(), m_BlockSize*m_NumSubvectors*sizeof(float), cudaMemcpyHostToDevice));
       }
       else {
-        LOG(Helper::LogLevel::LL_Error, "Only PQ and OPQ quantizers are supported for GPU build\n");
+        SPTAGLIB_LOG(Helper::LogLevel::LL_Error, "Only PQ and OPQ quantizers are supported for GPU build\n");
         exit(1);
       }
 

AnnService/inc/Core/Common/cuda/KNN.hxx

@@ -396,7 +396,7 @@ void refineGraphGPU(SPTAG::VectorIndex* index, Point<T,SUMTYPE,MAX_DIM>* d_point
   cudaMemcpy(d_candidates, candidates, dataSize*candidatesPerVector*sizeof(int), cudaMemcpyHostToDevice);
 
   auto t2 = std::chrono::high_resolution_clock::now();
-  LOG(SPTAG::Helper::LogLevel::LL_Info, "find candidates time (ms): %lld\n", std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count());
+  SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Info, "find candidates time (ms): %lld\n", std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count());
 
   // Use a number of batches of refinement to overlap CPU and GPU work (TODO)
   int NUM_BATCHES = 1;
@@ -414,13 +414,13 @@ void refineGraphGPU(SPTAG::VectorIndex* index, Point<T,SUMTYPE,MAX_DIM>* d_point
       refineBatch_kernel<T,SUMTYPE,MAX_DIM, REFINE_THREADS><<<REFINE_BLOCKS,REFINE_THREADS>>>(d_points, batch_size, i*batch_size, d_graph, d_candidates, listMem, candidatesPerVector, KVAL, refineDepth, metric);
       cudaError_t status = cudaDeviceSynchronize();
       if(status != cudaSuccess) {
-          LOG(SPTAG::Helper::LogLevel::LL_Error, "Refine error code:%d\n", status);
+          SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Error, "Refine error code:%d\n", status);
       }
     }
 
   }
   t2 = std::chrono::high_resolution_clock::now();
-  LOG(SPTAG::Helper::LogLevel::LL_Info, "GPU refine time (ms): %lld\n", std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count());
+  SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Info, "GPU refine time (ms): %lld\n", std::chrono::duration_cast<std::chrono::milliseconds>(t2 - t1).count());
 
   cudaFree(listMem);
   cudaFree(d_candidates);

AnnService/inc/Core/Common/cuda/Refine.hxx

@@ -379,9 +379,9 @@ __host__ void construct_trees_PQ(TPtree** d_trees, PointSet<T>** ps, int N, int
         size_t freeMem, totalMem;
         CUDA_CHECK(cudaMemGetInfo(&freeMem, &totalMem));
         size_t neededMem = N*reconDim*sizeof(R);
-        LOG(SPTAG::Helper::LogLevel::LL_Debug, "Memory needed for reconstructed vectors to build TPT: %ld, memory availalbe: %ld\n", neededMem, totalMem);
+        SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Debug, "Memory needed for reconstructed vectors to build TPT: %ld, memory availalbe: %ld\n", neededMem, totalMem);
         if(freeMem*0.9 < neededMem) {
-          LOG(SPTAG::Helper::LogLevel::LL_Error, "Insufficient memory for reconstructed vectors to build TPTree.\n");
+          SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Error, "Insufficient memory for reconstructed vectors to build TPTree.\n");
           exit(1);
         }
 
@@ -432,7 +432,7 @@ printf("Num leaves:%d, min leaf:%d, max leaf:%d\n", d_trees[0]->num_leaves, min_
 
         cudaDeviceProp prop;
         CUDA_CHECK(cudaGetDeviceProperties(&prop, gpuNum)); // Get avil. memory
-        LOG(SPTAG::Helper::LogLevel::LL_Info, "GPU %d - %s\n", gpuNum, prop.name);
+        SPTAGLIB_LOG(SPTAG::Helper::LogLevel::LL_Info, "GPU %d - %s\n", gpuNum, prop.name);
 
         size_t freeMem, totalMem;
         CUDA_CHECK(cudaMemGetInfo(&freeMem, &totalMem));