[FEA] Support for half-float mixed precise in brute-force (#225)

- distance supports half-float mixed precision
- prefiltered_brute_force supports half
- migrate the ann brute force test cases and support half

Authors:
  - rhdong (https://github.com/rhdong)
  - Ben Frederickson (https://github.com/benfred)

Approvers:
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #225

cpp/CMakeLists.txt

@@ -216,30 +216,43 @@ add_library(
   src/cluster/kmeans_transform_float.cu
   src/cluster/single_linkage_float.cu
   src/distance/detail/pairwise_matrix/dispatch_canberra_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_canberra_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_canberra_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_correlation_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_correlation_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_correlation_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_cosine_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_cosine_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_cosine_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_hamming_unexpanded_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_hamming_unexpanded_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_hamming_unexpanded_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_hellinger_expanded_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_hellinger_expanded_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_hellinger_expanded_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_jensen_shannon_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_jensen_shannon_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_jensen_shannon_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_kl_divergence_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_kl_divergence_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_kl_divergence_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l1_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_l1_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l1_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l2_expanded_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_l2_expanded_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l2_expanded_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l2_unexpanded_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_l2_unexpanded_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l2_unexpanded_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l_inf_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_l_inf_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_l_inf_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_lp_unexpanded_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_lp_unexpanded_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_lp_unexpanded_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_russel_rao_float_float_float_int.cu
+  src/distance/detail/pairwise_matrix/dispatch_russel_rao_half_float_float_int.cu
   src/distance/detail/pairwise_matrix/dispatch_russel_rao_double_double_double_int.cu
   src/distance/detail/pairwise_matrix/dispatch_rbf.cu
   src/distance/detail/fused_distance_nn.cu
@@ -413,6 +426,8 @@ add_library(
   src/selection/select_k_half_uint32_t.cu
 )
 
+target_compile_definitions(cuvs PRIVATE "CUVS_EXPLICIT_INSTANTIATE_ONLY")
+
 target_compile_options(
   cuvs INTERFACE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-extended-lambda
                  --expt-relaxed-constexpr>

cpp/bench/ann/CMakeLists.txt

@@ -35,6 +35,7 @@ option(CUVS_ANN_BENCH_USE_GGNN "Include ggnn algorithm in benchmark" OFF)
 option(CUVS_ANN_BENCH_SINGLE_EXE
        "Make a single executable with benchmark as shared library modules" OFF
 )
+option(CUVS_KNN_BENCH_USE_CUVS_BRUTE_FORCE "Include cuVS brute force knn in benchmark" ON)
 
 # ##################################################################################################
 # * Process options ----------------------------------------------------------
@@ -53,6 +54,9 @@ if(BUILD_CPU_ONLY)
   set(CUVS_ANN_BENCH_USE_CUVS_BRUTE_FORCE OFF)
   set(CUVS_ANN_BENCH_USE_CUVS_CAGRA_HNSWLIB OFF)
   set(CUVS_ANN_BENCH_USE_GGNN OFF)
+  set(CUVS_KNN_BENCH_USE_CUVS_BRUTE_FORCE OFF)
+else()
+  set(CUVS_FAISS_ENABLE_GPU ON)
 endif()
 
 set(CUVS_ANN_BENCH_USE_CUVS OFF)
@@ -61,6 +65,7 @@ if(CUVS_ANN_BENCH_USE_CUVS_IVF_PQ
    OR CUVS_ANN_BENCH_USE_CUVS_IVF_FLAT
    OR CUVS_ANN_BENCH_USE_CUVS_CAGRA
    OR CUVS_ANN_BENCH_USE_CUVS_CAGRA_HNSWLIB
+   OR CUVS_KNN_BENCH_USE_CUVS_BRUTE_FORCE
 )
   set(CUVS_ANN_BENCH_USE_CUVS ON)
 endif()
@@ -169,6 +174,8 @@ function(ConfigureAnnBench)
     )
   endif()
 
+  target_compile_definitions(${BENCH_NAME} PRIVATE "CUVS_EXPLICIT_INSTANTIATE_ONLY")
+
   target_include_directories(
     ${BENCH_NAME}
     PUBLIC "$<BUILD_INTERFACE:${CUVS_SOURCE_DIR}/include>"
@@ -223,6 +230,17 @@ if(CUVS_ANN_BENCH_USE_CUVS_BRUTE_FORCE)
   ConfigureAnnBench(NAME CUVS_BRUTE_FORCE PATH src/cuvs/cuvs_benchmark.cu LINKS cuvs)
 endif()
 
+if(CUVS_KNN_BENCH_USE_CUVS_BRUTE_FORCE)
+  ConfigureAnnBench(
+    NAME
+    CUVS_KNN_BRUTE_FORCE
+    PATH
+    $<$<BOOL:${CUVS_KNN_BENCH_USE_CUVS_BRUTE_FORCE}>:src/cuvs/cuvs_brute_force_knn.cu>
+    LINKS
+    cuvs
+  )
+endif()
+
 if(CUVS_ANN_BENCH_USE_CUVS_CAGRA)
   ConfigureAnnBench(
     NAME