Benchmarks for find() algorithm

Boost.Compute, STL and Thrust benchmarks for find() algorithm.

Author: jszuppe

perf/CMakeLists.txt

@@ -25,6 +25,7 @@ set(BENCHMARKS
   erase_remove
   exclusive_scan
   fill
+  find
   find_end
   includes
   inner_product
@@ -70,6 +71,7 @@ endforeach()
 set(STL_BENCHMARKS
   stl_accumulate
   stl_count
+  stl_find
   stl_find_end
   stl_includes
   stl_inner_product
@@ -118,6 +120,7 @@ if(${BOOST_COMPUTE_HAVE_CUDA})
     thrust_accumulate
     thrust_count
     thrust_exclusive_scan
+    thrust_find
     thrust_inner_product
     thrust_merge
     thrust_partial_sum

perf/perf.py

@@ -118,6 +118,7 @@ def run_benchmark(name, sizes, vs=[]):
             "accumulate",
             "count",
             "exclusive_scan",
+            "find",
             "inner_product",
             "merge",
             "partial_sum",
@@ -136,6 +137,7 @@ def run_benchmark(name, sizes, vs=[]):
         "stl": [
             "accumulate",
             "count",
+            "find",
             "find_end",
             "includes",
             "inner_product",

perf/perf_find.cpp

@@ -0,0 +1,87 @@
+//---------------------------------------------------------------------------//
+// Copyright (c) 2015 Jakub Szuppe <j.szuppe@gmail.com>
+//
+// Distributed under the Boost Software License, Version 1.0
+// See accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt
+//
+// See http://kylelutz.github.com/compute for more information.
+//---------------------------------------------------------------------------//
+
+#include <algorithm>
+#include <iostream>
+#include <vector>
+
+#include <boost/compute/system.hpp>
+#include <boost/compute/algorithm/find.hpp>
+#include <boost/compute/container/vector.hpp>
+
+#include "perf.hpp"
+
+// Max integer that can be generated by rand_int() function.
+int rand_int_max = 25;
+
+int rand_int()
+{
+    return static_cast<int>((rand() / double(RAND_MAX)) * rand_int_max);
+}
+
+int main(int argc, char *argv[])
+{
+    perf_parse_args(argc, argv);
+    std::cout << "size: " << PERF_N << std::endl;
+
+    // setup context and queue for the default device
+    boost::compute::device device = boost::compute::system::default_device();
+    boost::compute::context context(device);
+    boost::compute::command_queue queue(context, device);
+    std::cout << "device: " << device.name() << std::endl;
+
+    // create vector of random numbers on the host
+    std::vector<int> host_vector(PERF_N);
+    std::generate(host_vector.begin(), host_vector.end(), rand_int);
+
+    // create vector on the device and copy the data
+    boost::compute::vector<int> device_vector(PERF_N, context);
+    boost::compute::copy(
+        host_vector.begin(),
+        host_vector.end(),
+        device_vector.begin(),
+        queue
+    );
+
+    // trying to find element that isn't in vector (worst-case scenario)
+    int wanted = rand_int_max + 1;
+
+    // device iterator
+    boost::compute::vector<int>::iterator device_it;
+
+    perf_timer t;
+    for(size_t trial = 0; trial < PERF_TRIALS; trial++){
+        t.start();
+        device_it = boost::compute::find(
+            device_vector.begin(), device_vector.end(), wanted, queue
+        );
+        queue.finish();
+        t.stop();
+    }
+    std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
+
+    // verify if found index is correct by comparing it with std::find() result
+    size_t host_index = std::distance(host_vector.begin(),
+                                      std::find(host_vector.begin(),
+                                                host_vector.end(),
+                                                wanted));
+    size_t device_index = device_it.get_index();
+
+    if(device_index != host_index){
+        std::cout << "ERROR: "
+                  << "device_index (" << device_index << ") "
+                  << "!= "
+                  << "host_index (" << host_index << ")"
+                  << std::endl;
+        return -1;
+    }
+
+    return 0;
+}

perf/perf_stl_find.cpp

@@ -0,0 +1,45 @@
+//---------------------------------------------------------------------------//
+// Copyright (c) 2015 Jakub Szuppe <j.szuppe@gmail.com>
+//
+// Distributed under the Boost Software License, Version 1.0
+// See accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt
+//
+// See http://kylelutz.github.com/compute for more information.
+//---------------------------------------------------------------------------//
+
+#include <algorithm>
+#include <iostream>
+#include <vector>
+
+#include "perf.hpp"
+
+// Max integer that can be generated by rand_int() function.
+int rand_int_max = 25;
+
+int rand_int()
+{
+    return static_cast<int>((rand() / double(RAND_MAX)) * rand_int_max);
+}
+
+int main(int argc, char *argv[])
+{
+    perf_parse_args(argc, argv);
+    std::cout << "size: " << PERF_N << std::endl;
+
+    // create vector of random numbers on the host
+    std::vector<int> host_vector(PERF_N);
+    std::generate(host_vector.begin(), host_vector.end(), rand_int);
+
+    // trying to find element that isn't in vector (worst-case scenario)
+    int wanted = rand_int_max + 1;
+    perf_timer t;
+    for(size_t trial = 0; trial < PERF_TRIALS; trial++){
+        t.start();
+        std::find(host_vector.begin(), host_vector.end(), wanted);
+        t.stop();
+    }
+    std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
+
+    return 0;
+}

perf/perf_thrust_find.cu

@@ -0,0 +1,53 @@
+//---------------------------------------------------------------------------//
+// Copyright (c) 2015 Jakub Szuppe <j.szuppe@gmail.com>
+//
+// Distributed under the Boost Software License, Version 1.0
+// See accompanying file LICENSE_1_0.txt or copy at
+// http://www.boost.org/LICENSE_1_0.txt
+//
+// See http://kylelutz.github.com/compute for more information.
+//---------------------------------------------------------------------------//
+
+#include <algorithm>
+#include <iostream>
+#include <vector>
+
+#include <thrust/find.h>
+#include <thrust/host_vector.h>
+#include <thrust/device_vector.h>
+
+#include "perf.hpp"
+
+// Max integer that can be generated by rand_int() function.
+int rand_int_max = 25;
+
+int rand_int()
+{
+    return static_cast<int>((rand() / double(RAND_MAX)) * rand_int_max);
+}
+
+int main(int argc, char *argv[])
+{
+    perf_parse_args(argc, argv);
+    std::cout << "size: " << PERF_N << std::endl;
+
+    // create vector of random numbers on the host
+    thrust::host_vector<int> host_vector(PERF_N);
+    thrust::generate(host_vector.begin(), host_vector.end(), rand_int);
+
+    thrust::device_vector<int> v = host_vector;
+
+    // trying to find element that isn't in vector (worst-case scenario)
+    int wanted = rand_int_max + 1;
+    
+    perf_timer t;
+    for(size_t trial = 0; trial < PERF_TRIALS; trial++){
+        t.start();
+        thrust::find(v.begin(), v.end(), wanted);
+        cudaDeviceSynchronize();
+        t.stop();
+    }
+    std::cout << "time: " << t.min_time() / 1e6 << " ms" << std::endl;
+
+    return 0;
+}