Make tutorial examples use dynamic linkage. (#343)

* Add shared lib target for GPU version.

* Dynamically link tutorial examples.

* Add shared lib to clean target in GPU Makefile.

* Add clean target to tutorial examples Makefile.

* Minor modifications to tutorial GPU examples.

gpu/Makefile

@@ -85,6 +85,10 @@ CUOBJ=      impl/BroadcastSum.o \
 $(LIBNAME).a: $(CPPOBJ) $(CUOBJ)
 	ar r $@ $^
 
+$(LIBNAME).$(SHAREDEXT): $(CPPOBJ) $(CUOBJ)
+	$(CC) -L$(CUDAROOT)/lib64 -lcublas -lcudart $(LDFLAGS) \
+	$(FAISSSHAREDFLAGS) -o $(LIBNAME).$(SHAREDEXT) $^
+
 
 test/demo_ivfpq_indexing_gpu: test/demo_ivfpq_indexing_gpu.cpp \
 	$(LIBNAME).a ../libfaiss.a
@@ -115,6 +119,7 @@ test/demo_ivfpq_indexing_gpu: test/demo_ivfpq_indexing_gpu.cpp \
 
 clean:
 	rm -rf *.o impl/*.o utils/*.o test/*.o $(LIBNAME).a \
+		 $(LIBNAME).$(SHAREDEXT) \
                  ../python/*swigfaiss_gpu* ../*swigfaiss_gpu*
 
 

tutorial/cpp/4-GPU.cpp

@@ -36,8 +36,6 @@ int main() {
         xq[d * i] += i / 1000.;
     }
 
-    int k = 4;
-
     faiss::gpu::StandardGpuResources res;
 
     // Using a flat index
@@ -48,6 +46,8 @@ int main() {
     index_flat.add(nb, xb);  // add vectors to the index
     printf("ntotal = %ld\n", index_flat.ntotal);
 
+    int k = 4;
+
     {       // search xq
         long *I = new long[k * nq];
         float *D = new float[k * nq];

tutorial/cpp/5-Multiple-GPUs.cpp

@@ -38,6 +38,8 @@ int main() {
 
     int ngpus = faiss::gpu::getNumDevices();
 
+    printf("Number of GPUs: %d\n", ngpus);
+
     std::vector<faiss::gpu::GpuResources*> res;
     std::vector<int> devs;
     for(int i = 0; i < ngpus; i++) {
@@ -55,7 +57,7 @@ int main() {
         );
 
     printf("is_trained = %s\n", gpu_index->is_trained ? "true" : "false");
-    gpu_index->add(nb, xb);  // vectors to the index
+    gpu_index->add(nb, xb);  // add vectors to the index
     printf("ntotal = %ld\n", gpu_index->ntotal);
 
     int k = 4;

tutorial/cpp/Makefile

@@ -2,25 +2,38 @@ MAKEFILE_INC=../../makefile.inc
 
 -include $(MAKEFILE_INC)
 
+NVCCLDFLAGS =	-Xcompiler \"-Wl,-rpath=../../:../../gpu/\" \
+		-L../.. -L../../gpu -lfaiss -lgpufaiss
+
+LDFLAGS = -L../.. -Wl,-rpath=../.. -lfaiss
+
 all: cpu gpu
 
 cpu: 1-Flat 2-IVFFlat 3-IVFPQ
 
 gpu: 4-GPU 5-Multiple-GPUs
 
-1-Flat: 1-Flat.cpp ../../libfaiss.a
-	$(CC) -o $@ $(CFLAGS) $^ -I../../../ $(LDFLAGS) $(BLASLDFLAGS)
+1-Flat: 1-Flat.cpp ../../libfaiss.$(SHAREDEXT)
+	$(CC) -o $@ $(CFLAGS) $< -I../../../ $(LDFLAGS)
+
+2-IVFFlat: 2-IVFFlat.cpp ../../libfaiss.$(SHAREDEXT)
+	$(CC) -o $@ $(CFLAGS) $< -I../../../ $(LDFLAGS)
+
+3-IVFPQ: 3-IVFPQ.cpp ../../libfaiss.$(SHAREDEXT)
+	$(CC) -o $@ $(CFLAGS) $< -I../../../ $(LDFLAGS)
+
+4-GPU: 4-GPU.cpp ../../libfaiss.$(SHAREDEXT) ../../gpu/libgpufaiss.$(SHAREDEXT)
+	$(NVCC) $(NVCCFLAGS) -o $@ $< $(NVCCLDFLAGS) -I../../../
 
-2-IVFFlat: 2-IVFFlat.cpp ../../libfaiss.a
-	$(CC) -o $@ $(CFLAGS) $^ -I../../../ $(LDFLAGS) $(BLASLDFLAGS)
+5-Multiple-GPUs: 5-Multiple-GPUs.cpp ../../libfaiss.$(SHAREDEXT) \
+	../../gpu/libgpufaiss.$(SHAREDEXT)
+	$(NVCC) $(NVCCFLAGS) -o $@ $< $(NVCCLDFLAGS) -I../../../
 
-3-IVFPQ: 3-IVFPQ.cpp ../../libfaiss.a
-	$(CC) -o $@ $(CFLAGS) $^ -I../../../ $(LDFLAGS) $(BLASLDFLAGS)
+../../libfaiss.$(SHAREDEXT):
+	cd ../../ && make libfaiss.$(SHAREDEXT)
 
-4-GPU: 4-GPU.cpp ../../libfaiss.a ../../gpu/libgpufaiss.a
-	$(NVCC) $(NVCCFLAGS) -o $@ $^ -I../../../ -Xcompiler -fopenmp -lcublas \
-	$(BLASLDFLAGSNVCC)
+../../gpu/libgpufaiss.$(SHAREDEXT):
+	cd ../../gpu/ && make libgpufaiss.$(SHAREDEXT)
 
-5-Multiple-GPUs: 5-Multiple-GPUs.cpp ../../libfaiss.a ../../gpu/libgpufaiss.a
-	$(NVCC) $(NVCCFLAGS) -o $@ $^ -I../../../ -Xcompiler -fopenmp -lcublas \
-	$(BLASLDFLAGSNVCC)
+clean:
+	rm -f 1-Flat 2-IVFFlat 3-IVFPQ 4-GPU 5-Multiple-GPUs

tutorial/python/4-GPU.py

@@ -17,8 +17,6 @@ xq[:, 0] += np.arange(nq) / 1000.
 
 import faiss                     # make faiss available
 
-print("number of GPUs:", faiss.get_num_gpus())
-
 res = faiss.StandardGpuResources()  # use a single GPU
 
 ## Using a flat index

tutorial/python/5-Multiple-GPUs.py

@@ -21,14 +21,16 @@ ngpus = faiss.get_num_gpus()
 
 print("number of GPUs:", ngpus)
 
-index = faiss.index_cpu_to_all_gpus( # build the index
-    faiss.IndexFlatL2(d)
+cpu_index = faiss.IndexFlatL2(d)
+
+gpu_index = faiss.index_cpu_to_all_gpus(  # build the index
+    cpu_index
 )
 
-index.add(xb)                  # add vectors to the index
-print(index.ntotal)
+gpu_index.add(xb)              # add vectors to the index
+print(gpu_index.ntotal)
 
 k = 4                          # we want to see 4 nearest neighbors
-D, I = index.search(xq, k)     # actual search
+D, I = gpu_index.search(xq, k) # actual search
 print(I[:5])                   # neighbors of the 5 first queries
 print(I[-5:])                  # neighbors of the 5 last queries