Files
tbb/examples/graph/binpack/binpack.cpp
T
Hans Johnson e32d75f876 ENH: Update to TBB 2017 (2016-09-08 release)
The new release TBB is now under a new more
open license.
                                 Apache License
                           Version 2.0, January 2004
                        http://www.apache.org/licenses/

The list of most significant changes made over time in
Intel(R) Threading Building Blocks (Intel(R) TBB).

Intel TBB 2017
TBB_INTERFACE_VERSION == 9100

Changes (w.r.t. Intel TBB 4.4 Update 5):

- static_partitioner class is now a fully supported feature.
- async_node class is now a fully supported feature.
- Improved dynamic memory allocation replacement on Windows* OS to skip
    DLLs for which replacement cannot be done, instead of aborting.
- Intel TBB no longer performs dynamic memory allocation replacement
    for Microsoft* Visual Studio* 2008.
- For 64-bit platforms, quadrupled the worst-case limit on the amount
    of memory the Intel TBB allocator can handle.
- Added TBB_USE_GLIBCXX_VERSION macro to specify the version of GNU
    libstdc++ when it cannot be properly recognized, e.g. when used
    with Clang on Linux* OS. Inspired by a contribution from David A.
- Added graph/stereo example to demostrate tbb::flow::async_msg.
- Removed a few cases of excessive user data copying in the flow graph.
- Reworked split_node to eliminate unnecessary overheads.
- Added support for C++11 move semantics to the argument of
    tbb::parallel_do_feeder::add() method.
- Added C++11 move constructor and assignment operator to
    tbb::combinable template class.
- Added tbb::this_task_arena::max_concurrency() function and
    max_concurrency() method of class task_arena returning the maximal
    number of threads that can work inside an arena.
- Deprecated tbb::task_arena::current_thread_index() static method;
    use tbb::this_task_arena::current_thread_index() function instead.
- All examples for commercial version of library moved online:
    https://software.intel.com/en-us/product-code-samples. Examples are
    available as a standalone package or as a part of Intel(R) Parallel
    Studio XE or Intel(R) System Studio Online Samples packages.

Changes affecting backward compatibility:

- Renamed following methods and types in async_node class:
    Old                   New
    async_gateway_type => gateway_type
    async_gateway()    => gateway()
    async_try_put()    => try_put()
    async_reserve()    => reserve_wait()
    async_commit()     => release_wait()
- Internal layout of some flow graph nodes has changed; recompilation
    is recommended for all binaries that use the flow graph.

Preview Features:

- Added template class streaming_node to the flow graph API. It allows
    a flow graph to offload computations to other devices through
    streaming or offloading APIs.
- Template class opencl_node reimplemented as a specialization of
    streaming_node that works with OpenCL*.
- Added tbb::this_task_arena::isolate() function to isolate execution
    of a group of tasks or an algorithm from other tasks submitted
    to the scheduler.

Bugs fixed:

- Added a workaround for GCC bug #62258 in std::rethrow_exception()
    to prevent possible problems in case of exception propagation.
- Fixed parallel_scan to provide correct result if the initial value
    of an accumulator is not the operation identity value.
- Fixed a memory corruption in the memory allocator when it meets
    internal limits.
- Fixed the memory allocator on 64-bit platforms to align memory
    to 16 bytes by default for all allocations bigger than 8 bytes.
- As a workaround for crashes in the Intel TBB library compiled with
    GCC 6, added -flifetime-dse=1 to compilation options on Linux* OS.
- Fixed a race in the flow graph implementation.

Open-source contributions integrated:

- Enabling use of C++11 'override' keyword by Raf Schietekat.

------------------------------------------------------------------------
2016-09-28 21:12:48 +02:00

300 lines
14 KiB
C++

/*
Copyright (c) 2005-2016 Intel Corporation
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/
/* Bin-packing algorithm that attempts to use minimal number of bins B of
size V to contain N items of varying sizes. */
#include <string>
#include <iostream>
#include <cmath>
#include "tbb/atomic.h"
#include "tbb/task_scheduler_init.h"
#include "tbb/tick_count.h"
#include "tbb/flow_graph.h"
#include "../../common/utility/utility.h"
using namespace std;
using namespace tbb;
using namespace tbb::flow;
typedef size_t size_type; // to represent non-zero indices, capacities, etc.
typedef size_t value_type; // the type of items we are attempting to pack into bins
typedef vector<value_type> bin; // we use a simple vector to represent a bin
// Our bin packers will be function nodes in the graph that take value_type items and
// return a dummy value. They will also implicitly send packed bins to the bin_buffer
// node, and unused items back to the value_pool node:
typedef multifunction_node<value_type, tuple<value_type, bin>, rejecting> bin_packer;
// Items are placed into a pool that all bin packers grab from, represent by a queue_node:
typedef queue_node<value_type> value_pool;
// Packed bins are placed in this buffer waiting to be serially printed and/or accounted for:
typedef buffer_node<bin> bin_buffer;
// Packed bins are taken from the_bin_buffer and processed by the_writer:
typedef function_node<bin, continue_msg, rejecting> bin_writer;
// Items are injected into the graph when this node sends them to the_value_pool:
typedef source_node<value_type> value_source;
// User-specified globals with default values
size_type V = 42; // desired capacity for each bin
size_type N = 1000; // number of elements to generate
bool verbose = false; // prints bin details and other diagnostics to screen
bool silent = false; // suppress all output except for time
int num_bin_packers=-1; // number of concurrent bin packers in operation; default is #threads;
// larger values can result in more bins at less than full capacity
size_type optimality=1; // 1 (default) is highest the algorithm can obtain; larger numbers run faster
// Calculated globals
size_type min_B; // lower bound on the optimal number of bins
size_type B; // the answer, i.e. number of bins used by the algorithm
size_type *input_array; // stores randomly generated input values
value_type item_sum; // sum of all randomly generated input values
tbb::atomic<value_type> packed_sum; // sum of all values currently packed into all bins
tbb::atomic<size_type> packed_items; // number of values currently packed into all bins
tbb::atomic<size_type> active_bins; // number of active bin_packers
bin_packer **bins; // the array of bin packers
// This class is the Body type for bin_packer
class bin_filler {
typedef bin_packer::output_ports_type ports_type;
bin my_bin; // the current bin that this bin_filler is packing
size_type my_used; // capacity of bin used by current contents (not to be confused with my_bin.size())
size_type relax, relax_val; // relaxation counter for determining when to settle for a non-full bin
bin_packer* my_bin_packer; // ptr to the bin packer that this body object is associated with
size_type bin_index; // index of the encapsulating bin packer in the global bins array
value_type looking_for; // the minimum size of item this bin_packer will accept
value_pool* the_value_pool; // the queue of incoming values
bool done; // flag to indicate that this binpacker has been deactivated
public:
bin_filler(size_t bidx, value_pool* _q) :
my_used(0), relax(0), relax_val(0), my_bin_packer(NULL), bin_index(bidx), looking_for(V), the_value_pool(_q), done(false) {}
void operator()(const value_type& item, ports_type& p) {
if (!my_bin_packer) my_bin_packer = bins[bin_index];
if (done) get<0>(p).try_put(item); // this bin_packer is done packing items; put item back to pool
else if (item > V) { // signal that packed_sum has reached item_sum at some point
size_type remaining = active_bins--;
if (remaining == 1 && packed_sum == item_sum) { // this is the last bin and it has seen everything
// this bin_packer may not have seen everything, so stay active
if (my_used>0) get<1>(p).try_put(my_bin);
my_bin.clear();
my_used = 0;
looking_for = V;
++active_bins;
}
else if (remaining == 1) { // this is the last bin, but there are remaining items
get<0>(p).try_put(V+1); // send out signal
++active_bins;
}
else if (remaining > 1) { // this is not the last bin; deactivate
if (my_used < V/(1+optimality*.1)) { // this bin is ill-utilized; throw back items and deactivate
packed_sum -= my_used;
packed_items -= my_bin.size();
for (size_type i=0; i<my_bin.size(); ++i)
get<0>(p).try_put(my_bin[i]);
the_value_pool->remove_successor(*my_bin_packer); // deactivate
done = true;
get<0>(p).try_put(V+1); // send out signal
}
else { // this bin is well-utilized; send out bin and deactivate
the_value_pool->remove_successor(*my_bin_packer); // build no more bins
done = true;
if (my_used>0) get<1>(p).try_put(my_bin);
get<0>(p).try_put(V+1); // send out signal
}
}
}
else if (item <= V-my_used && item >= looking_for) { // this item can be packed
my_bin.push_back(item);
my_used += item;
packed_sum += item;
++packed_items;
looking_for = V-my_used;
relax = 0;
if (packed_sum == item_sum) {
get<0>(p).try_put(V+1); // send out signal
}
if (my_used == V) {
get<1>(p).try_put(my_bin);
my_bin.clear();
my_used = 0;
looking_for = V;
}
}
else { // this item can't be packed; relax constraints
++relax;
if (relax >= (N-packed_items)/optimality) { // this bin_packer has looked through enough items
relax = 0;
--looking_for; // accept a wider range of items
if (looking_for == 0 && my_used < V/(1+optimality*.1) && my_used > 0 && active_bins > 1) {
// this bin_packer is ill-utilized and can't find items; deactivate and throw back items
size_type remaining = active_bins--;
if (remaining > 1) { // not the last bin_packer
the_value_pool->remove_successor(*my_bin_packer); // deactivate
done = true;
}
else active_bins++; // can't deactivate last bin_packer
packed_sum -= my_used;
packed_items -= my_bin.size();
for (size_type i=0; i<my_bin.size(); ++i)
get<0>(p).try_put(my_bin[i]);
my_bin.clear();
my_used = 0;
}
else if (looking_for == 0 && (my_used >= V/(1+optimality*.1) || active_bins == 1)) {
// this bin_packer can't find items but is well-utilized, so send it out and reset
get<1>(p).try_put(my_bin);
my_bin.clear();
my_used = 0;
looking_for = V;
}
}
get<0>(p).try_put(item); // put unused item back to pool
}
}
};
// source node uses this to send the values to the value_pool
class item_generator {
size_type counter;
public:
item_generator() : counter(0) {}
bool operator()(value_type& m) {
if (counter<N) {
m = input_array[counter];
++counter;
return true;
}
return false;
}
};
// the terminal function_node uses this to gather stats and print bin information
class bin_printer {
value_type running_count;
size_type item_count;
value_type my_min, my_max;
double avg;
public:
bin_printer() : running_count(0), item_count(0), my_min(V), my_max(0), avg(0) {}
continue_msg operator()(bin b) {
value_type sum=0;
++B;
if (verbose)
cout << "[ ";
for (size_type i=0; i<b.size(); ++i) {
if (verbose)
cout << b[i] << " ";
sum+=b[i];
++item_count;
}
if (sum < my_min) my_min = sum;
if (sum > my_max) my_max = sum;
avg += sum;
running_count += sum;
if (verbose)
cout << "]=" << sum << "; Done/Packed/Total cap: " << running_count << "/" << packed_sum << "/" << item_sum
<< " items:" << item_count << "/" << packed_items << "/" << N << " B=" << B << endl;
if (item_count == N) { // should be the last; print stats
avg = avg/(double)B;
if (!silent)
cout << "SUMMARY: #Bins used: " << B << "; Avg size: " << avg << "; Max size: " << my_max
<< "; Min size: " << my_min << "\n Lower bound on optimal #bins: " << min_B
<< "; Start #bins: " << num_bin_packers << endl;
}
return continue_msg(); // need to return something
}
};
int get_default_num_threads() {
static int threads = 0;
if (threads == 0)
threads = task_scheduler_init::default_num_threads();
return threads;
}
int main(int argc, char *argv[]) {
try {
utility::thread_number_range threads(get_default_num_threads);
utility::parse_cli_arguments(argc, argv,
utility::cli_argument_pack()
//"-h" option for displaying help is present implicitly
.positional_arg(threads,"#threads",utility::thread_number_range_desc)
.arg(verbose,"verbose"," print diagnostic output to screen")
.arg(silent,"silent"," limits output to timing info; overrides verbose")
.arg(N,"N"," number of values to pack")
.arg(V,"V"," capacity of each bin")
.arg(num_bin_packers,"#packers"," number of concurrent bin packers to use "
"(default=#threads)")
.arg(optimality,"optimality","controls optimality of solution; 1 is highest, use\n"
" larger numbers for less optimal but faster solution")
);
if (silent) verbose = false; // make silent override verbose
// Generate random input data
srand(42);
input_array = new value_type[N];
item_sum = 0;
for (size_type i=0; i<N; ++i) {
input_array[i] = rand() % V + 1; // generate items that fit in a bin
item_sum += input_array[i];
}
min_B = (item_sum % V) ? item_sum/V + 1 : item_sum/V;
tick_count start = tick_count::now();
for(int p = threads.first; p <= threads.last; p = threads.step(p)) {
task_scheduler_init init(p);
packed_sum = 0;
packed_items = 0;
B = 0;
if (num_bin_packers == -1) num_bin_packers = p;
active_bins = num_bin_packers;
if (!silent)
cout << "binpack running with " << item_sum << " capacity over " << N << " items, optimality="
<< optimality << ", " << num_bin_packers << " bins of capacity=" << V << " on " << p
<< " threads.\n";
graph g;
value_source the_source(g, item_generator(), false);
value_pool the_value_pool(g);
make_edge(the_source, the_value_pool);
bin_buffer the_bin_buffer(g);
bins = new bin_packer*[num_bin_packers];
for (int i=0; i<num_bin_packers; ++i) {
bins[i] = new bin_packer(g, 1, bin_filler(i, &the_value_pool));
make_edge(the_value_pool, *(bins[i]));
make_edge(output_port<0>(*(bins[i])), the_value_pool);
make_edge(output_port<1>(*(bins[i])), the_bin_buffer);
}
bin_writer the_writer(g, 1, bin_printer());
make_edge(the_bin_buffer, the_writer);
the_source.activate();
g.wait_for_all();
for (int i=0; i<num_bin_packers; ++i) {
delete bins[i];
}
delete[] bins;
}
utility::report_elapsed_time((tick_count::now() - start).seconds());
delete[] input_array;
return 0;
} catch(std::exception& e) {
cerr<<"error occurred. error text is :\"" <<e.what()<<"\"\n";
return 1;
}
}