summaryrefslogtreecommitdiffstats
path: root/decompress.cc
diff options
context:
space:
mode:
Diffstat (limited to 'decompress.cc')
-rw-r--r--decompress.cc453
1 files changed, 398 insertions, 55 deletions
diff --git a/decompress.cc b/decompress.cc
index 0a125d0..85baaf6 100644
--- a/decompress.cc
+++ b/decompress.cc
@@ -38,86 +38,429 @@
namespace {
-int do_decompress( LZ_Decoder * const decoder, const int infd, const int outfd,
- const Pretty_print & pp, const bool testing )
+long long in_size = 0;
+long long out_size = 0;
+
+
+struct Packet // data block
{
- const int in_buffer_size = 65536, out_buffer_size = 8 * in_buffer_size;
- uint8_t in_buffer[in_buffer_size], out_buffer[out_buffer_size];
+ uint8_t * data; // data == 0 means end of member
+ int size; // number of bytes in data (if any)
+ };
- while( true )
+
+class Packet_courier // moves packets around
+ {
+public:
+ unsigned long icheck_counter;
+ unsigned long iwait_counter;
+ unsigned long ocheck_counter;
+ unsigned long owait_counter;
+private:
+ int receive_worker_id; // worker queue currently receiving packets
+ int deliver_worker_id; // worker queue currently delivering packets
+ Slot_tally slot_tally;
+ std::vector< std::queue< Packet * > > ipacket_queues;
+ std::vector< std::queue< Packet * > > opacket_queues;
+ int num_working; // Number of workers still running
+ const int num_workers; // Number of workers
+ const int num_slots; // max packets in circulation
+ pthread_mutex_t imutex;
+ pthread_cond_t iav_or_eof; // input packet available or splitter done
+ pthread_mutex_t omutex;
+ pthread_cond_t oav_or_exit; // output packet available or all workers exited
+ bool eof; // splitter done
+
+public:
+ Packet_courier( const int workers, const int slots )
+ : icheck_counter( 0 ), iwait_counter( 0 ),
+ ocheck_counter( 0 ), owait_counter( 0 ),
+ receive_worker_id( 0 ), deliver_worker_id( 0 ),
+ slot_tally( slots ), ipacket_queues( workers ),
+ opacket_queues( workers ), num_working( workers ),
+ num_workers( workers ), num_slots( slots ), eof( false )
+ { xinit( &iav_or_eof, &imutex ); xinit( &oav_or_exit, &omutex ); }
+
+ ~Packet_courier()
+ { xdestroy( &iav_or_eof, &imutex ); xdestroy( &oav_or_exit, &omutex ); }
+
+ // make a packet with data received from splitter
+ // if data == 0, move to next queue
+ void receive_packet( uint8_t * const data, const int size )
+ {
+ Packet * ipacket = new Packet;
+ ipacket->data = data;
+ ipacket->size = size;
+ if( data != 0 )
+ { in_size += size; slot_tally.get_slot(); } // wait for a free slot
+ xlock( &imutex );
+ ipacket_queues[receive_worker_id].push( ipacket );
+ xbroadcast( &iav_or_eof );
+ xunlock( &imutex );
+ if( data == 0 && ++receive_worker_id >= num_workers )
+ receive_worker_id = 0;
+ }
+
+ // distribute a packet to a worker
+ Packet * distribute_packet( const int worker_id )
{
- int in_size = std::min( LZ_decompress_write_size( decoder ), in_buffer_size );
- if( in_size > 0 )
+ Packet * ipacket = 0;
+ xlock( &imutex );
+ ++icheck_counter;
+ while( ipacket_queues[worker_id].empty() && !eof )
{
- const int max_in_size = in_size;
- in_size = readblock( infd, in_buffer, max_in_size );
- if( in_size != max_in_size && errno )
- { pp(); show_error( "read error", errno ); return 1; }
- if( in_size == 0 ) LZ_decompress_finish( decoder );
- else if( in_size != LZ_decompress_write( decoder, in_buffer, in_size ) )
- internal_error( "library error (LZ_decompress_write)" );
+ ++iwait_counter;
+ xwait( &iav_or_eof, &imutex );
}
- int out_size = LZ_decompress_read( decoder, out_buffer, out_buffer_size );
-// std::fprintf( stderr, "%5d in_size, %6d out_size.\n", in_size, out_size );
- if( out_size < 0 )
+ if( !ipacket_queues[worker_id].empty() )
{
- const LZ_Errno lz_errno = LZ_decompress_errno( decoder );
- if( lz_errno == LZ_header_error )
+ ipacket = ipacket_queues[worker_id].front();
+ ipacket_queues[worker_id].pop();
+ }
+ xunlock( &imutex );
+ if( ipacket != 0 )
+ { if( ipacket->data != 0 ) slot_tally.leave_slot(); }
+ else
+ {
+ // Notify muxer when last worker exits
+ xlock( &omutex );
+ if( --num_working == 0 )
+ xsignal( &oav_or_exit );
+ xunlock( &omutex );
+ }
+ return ipacket;
+ }
+
+ // collect a packet from a worker
+ void collect_packet( Packet * const opacket, const int worker_id )
+ {
+ xlock( &omutex );
+ opacket_queues[worker_id].push( opacket );
+ if( worker_id == deliver_worker_id ) xsignal( &oav_or_exit );
+ xunlock( &omutex );
+ }
+
+ // deliver a packet to muxer
+ // if packet data == 0, move to next queue and wait again
+ Packet * deliver_packet()
+ {
+ Packet * opacket = 0;
+ xlock( &omutex );
+ ++ocheck_counter;
+ while( true )
+ {
+ while( opacket_queues[deliver_worker_id].empty() && num_working > 0 )
{
- if( LZ_decompress_total_out_size( decoder ) > 0 )
- break; // trailing garbage
- pp( "error reading member header" );
- return 1;
+ ++owait_counter;
+ xwait( &oav_or_exit, &omutex );
}
- if( lz_errno == LZ_mem_error )
+ if( opacket_queues[deliver_worker_id].empty() ) break;
+ opacket = opacket_queues[deliver_worker_id].front();
+ opacket_queues[deliver_worker_id].pop();
+ if( opacket->data != 0 ) break;
+ else
{
- pp( "not enough memory. Find a machine with more memory" );
- return 1;
+ if( ++deliver_worker_id >= num_workers ) deliver_worker_id = 0;
+ delete opacket; opacket = 0;
}
- pp();
- if( lz_errno == LZ_unexpected_eof )
+ }
+ xunlock( &omutex );
+ return opacket;
+ }
+
+ void finish() // splitter has no more packets to send
+ {
+ xlock( &imutex );
+ eof = true;
+ xbroadcast( &iav_or_eof );
+ xunlock( &imutex );
+ }
+
+ bool finished() // all packets delivered to muxer
+ {
+ if( !slot_tally.all_free() || !eof || num_working != 0 ) return false;
+ for( int i = 0; i < num_workers; ++i )
+ if( !ipacket_queues[i].empty() ) return false;
+ for( int i = 0; i < num_workers; ++i )
+ if( !opacket_queues[i].empty() ) return false;
+ return true;
+ }
+
+ const Slot_tally & tally() const { return slot_tally; }
+ };
+
+
+struct Splitter_arg
+ {
+ Packet_courier * courier;
+ const Pretty_print * pp;
+ int infd;
+ int packet_size;
+ };
+
+
+ // split data from input file into chunks and pass them to
+ // courier for packaging and distribution to workers.
+void * splitter( void * arg )
+ {
+ const Splitter_arg & tmp = *(Splitter_arg *)arg;
+ Packet_courier & courier = *tmp.courier;
+ const Pretty_print & pp = *tmp.pp;
+ const int infd = tmp.infd;
+ const int hsize = 6; // header size
+ const int tsize = 20; // trailer size
+ const int buffer_size = tmp.packet_size;
+ const int base_buffer_size = tsize + buffer_size + hsize;
+ uint8_t * const base_buffer = new( std::nothrow ) uint8_t[base_buffer_size];
+ if( base_buffer == 0 ) { pp( "not enough memory" ); fatal(); }
+ uint8_t * const buffer = base_buffer + tsize;
+
+ int size = readblock( infd, buffer, buffer_size + hsize ) - hsize;
+ bool at_stream_end = ( size < buffer_size );
+ if( size != buffer_size && errno )
+ { pp(); show_error( "read error", errno ); fatal(); }
+ if( size <= tsize || buffer[0] != 'L' || buffer[1] != 'Z' ||
+ buffer[2] != 'I' || buffer[3] != 'P' )
+ { pp( "bad magic number (file not in lzip format)" ); fatal(); }
+
+ long long partial_member_size = 0;
+ while( true )
+ {
+ int pos = 0;
+ for( int newpos = 1; newpos <= size; ++newpos )
+ if( buffer[newpos] == 'L' && buffer[newpos+1] == 'Z' &&
+ buffer[newpos+2] == 'I' && buffer[newpos+3] == 'P' )
{
- if( verbosity >= 0 )
- std::fprintf( stderr, "file ends unexpectedly at pos %lld\n",
- LZ_decompress_total_in_size( decoder ) );
- return 2;
+ long long member_size = 0;
+ for( int i = 1; i <= 8; ++i )
+ { member_size <<= 8; member_size += base_buffer[tsize+newpos-i]; }
+ if( partial_member_size + newpos - pos == member_size )
+ { // header found
+ uint8_t * data = new( std::nothrow ) uint8_t[newpos - pos];
+ if( data == 0 ) { pp( "not enough memory" ); fatal(); }
+ std::memcpy( data, buffer + pos, newpos - pos );
+ courier.receive_packet( data, newpos - pos );
+ courier.receive_packet( 0, 0 ); // end of member token
+ partial_member_size = 0;
+ pos = newpos;
+ }
}
- if( verbosity >= 0 )
- std::fprintf( stderr, "LZ_decompress_read error: %s.\n",
- LZ_strerror( LZ_decompress_errno( decoder ) ) );
- return 1;
+
+ if( at_stream_end )
+ {
+ uint8_t * data = new( std::nothrow ) uint8_t[size + hsize - pos];
+ if( data == 0 ) { pp( "not enough memory" ); fatal(); }
+ std::memcpy( data, buffer + pos, size + hsize - pos );
+ courier.receive_packet( data, size + hsize - pos );
+ courier.receive_packet( 0, 0 ); // end of member token
+ break;
}
- else if( out_size > 0 && outfd >= 0 )
+ if( pos < buffer_size )
{
- const int wr = writeblock( outfd, out_buffer, out_size );
- if( wr != out_size )
- { pp(); show_error( "write error", errno ); return 1; }
+ partial_member_size += buffer_size - pos;
+ uint8_t * data = new( std::nothrow ) uint8_t[buffer_size - pos];
+ if( data == 0 ) { pp( "not enough memory" ); fatal(); }
+ std::memcpy( data, buffer + pos, buffer_size - pos );
+ courier.receive_packet( data, buffer_size - pos );
}
- if( LZ_decompress_finished( decoder ) == 1 ) break;
- if( in_size == 0 && out_size == 0 )
- internal_error( "library error (LZ_decompress_read)" );
+ std::memcpy( base_buffer, base_buffer + buffer_size, tsize + hsize );
+ size = readblock( infd, buffer + hsize, buffer_size );
+ at_stream_end = ( size < buffer_size );
+ if( size != buffer_size && errno )
+ { pp(); show_error( "read error", errno ); fatal(); }
}
- if( verbosity >= 1 )
- { if( testing ) std::fprintf( stderr, "ok\n" );
- else std::fprintf( stderr, "done\n" ); }
+ delete[] base_buffer;
+ courier.finish(); // no more packets to send
return 0;
}
-} // end namespace
+
+struct Worker_arg
+ {
+ Packet_courier * courier;
+ const Pretty_print * pp;
+ int worker_id;
+ int packet_size;
+ };
-int decompress( const int infd, const int outfd, const Pretty_print & pp,
- const bool testing )
+ // consume packets from courier, decompress their contents, and
+ // give the produced packets to courier.
+void * worker( void * arg )
{
+ const Worker_arg & tmp = *(Worker_arg *)arg;
+ Packet_courier & courier = *tmp.courier;
+ const Pretty_print & pp = *tmp.pp;
+ const int worker_id = tmp.worker_id;
+ const int new_data_size = tmp.packet_size;
+
+ uint8_t * new_data = new( std::nothrow ) uint8_t[new_data_size];
LZ_Decoder * const decoder = LZ_decompress_open();
- int retval;
+ if( !new_data || !decoder || LZ_decompress_errno( decoder ) != LZ_ok )
+ { pp( "not enough memory" ); fatal(); }
+ int new_pos = 0;
- if( !decoder || LZ_decompress_errno( decoder ) != LZ_ok )
+ while( true )
{
- pp( "not enough memory" );
- retval = 1;
+ Packet * ipacket = courier.distribute_packet( worker_id );
+ if( ipacket == 0 ) break; // no more packets to process
+ if( ipacket->data == 0 ) LZ_decompress_finish( decoder );
+
+ int written = 0;
+ while( true )
+ {
+ if( LZ_decompress_write_size( decoder ) > 0 && written < ipacket->size )
+ {
+ const int wr = LZ_decompress_write( decoder, ipacket->data + written,
+ ipacket->size - written );
+ if( wr < 0 ) internal_error( "library error (LZ_decompress_write)" );
+ written += wr;
+ if( written > ipacket->size )
+ internal_error( "ipacket size exceeded in worker" );
+ }
+ while( true ) // read and pack decompressed data
+ {
+ const int rd = LZ_decompress_read( decoder, new_data + new_pos,
+ new_data_size - new_pos );
+ if( rd < 0 )
+ {
+ pp();
+ if( verbosity >= 0 )
+ std::fprintf( stderr, "LZ_decompress_read error in worker %d: %s.\n",
+ worker_id, LZ_strerror( LZ_decompress_errno( decoder ) ) );
+ fatal();
+ }
+ new_pos += rd;
+ if( new_pos > new_data_size )
+ internal_error( "opacket size exceeded in worker" );
+ if( new_pos == new_data_size || LZ_decompress_finished( decoder ) == 1 )
+ {
+ if( new_pos > 0 ) // make data packet
+ {
+ Packet * opacket = new Packet;
+ opacket->data = new_data;
+ opacket->size = new_pos;
+ courier.collect_packet( opacket, worker_id );
+ new_pos = 0;
+ new_data = new( std::nothrow ) uint8_t[new_data_size];
+ if( new_data == 0 ) { pp( "not enough memory" ); fatal(); }
+ }
+ if( LZ_decompress_finished( decoder ) == 1 )
+ {
+ LZ_decompress_reset( decoder );
+ Packet * opacket = new Packet; // end of member token
+ opacket->data = 0;
+ opacket->size = 0;
+ courier.collect_packet( opacket, worker_id );
+ break;
+ }
+ }
+ if( rd == 0 ) break;
+ }
+ if( ipacket->data == 0 ) { delete ipacket; break; }
+ if( written == ipacket->size )
+ { delete[] ipacket->data; delete ipacket; break; }
+ }
}
- else retval = do_decompress( decoder, infd, outfd, pp, testing );
+
+ delete[] new_data;
+ if( LZ_decompress_total_in_size( decoder ) != 0 )
+ { pp( "error, remaining data in decoder" ); fatal(); }
LZ_decompress_close( decoder );
- return retval;
+ return 0;
+ }
+
+
+ // get from courier the processed and sorted packets, and write
+ // their contents to the output file.
+void muxer( Packet_courier & courier, const Pretty_print & pp, const int outfd )
+ {
+ while( true )
+ {
+ Packet * opacket = courier.deliver_packet();
+ if( opacket == 0 ) break; // queue is empty. all workers exited
+
+ out_size += opacket->size;
+
+ if( outfd >= 0 )
+ {
+ const int wr = writeblock( outfd, opacket->data, opacket->size );
+ if( wr != opacket->size )
+ { pp(); show_error( "write error", errno ); fatal(); }
+ }
+ delete[] opacket->data;
+ delete opacket;
+ }
+ }
+
+} // end namespace
+
+
+ // init the courier, then start the splitter and the workers and
+ // call the muxer.
+int decompress( const int num_workers, const int num_slots,
+ const int infd, const int outfd, const Pretty_print & pp,
+ const int debug_level, const bool testing )
+ {
+ in_size = 0;
+ out_size = 0;
+ const int packet_size = 1 << 20;
+ Packet_courier courier( num_workers, num_slots );
+
+ Splitter_arg splitter_arg;
+ splitter_arg.courier = &courier;
+ splitter_arg.pp = &pp;
+ splitter_arg.infd = infd;
+ splitter_arg.packet_size = packet_size;
+
+ pthread_t splitter_thread;
+ xcreate( &splitter_thread, splitter, &splitter_arg );
+
+ Worker_arg * worker_args = new( std::nothrow ) Worker_arg[num_workers];
+ pthread_t * worker_threads = new( std::nothrow ) pthread_t[num_workers];
+ if( worker_args == 0 || worker_threads == 0 )
+ { pp( "not enough memory" ); fatal(); }
+ for( int i = 0; i < num_workers; ++i )
+ {
+ worker_args[i].courier = &courier;
+ worker_args[i].pp = &pp;
+ worker_args[i].worker_id = i;
+ worker_args[i].packet_size = packet_size;
+ xcreate( &worker_threads[i], worker, &worker_args[i] );
+ }
+
+ muxer( courier, pp, outfd );
+
+ for( int i = num_workers - 1; i >= 0; --i )
+ xjoin( worker_threads[i] );
+ delete[] worker_threads; worker_threads = 0;
+ delete[] worker_args; worker_args = 0;
+
+ xjoin( splitter_thread );
+
+ if( verbosity >= 2 )
+ std::fprintf( stderr, "decompressed size %9lld, size %9lld. ",
+ out_size, in_size );
+
+ if( verbosity >= 1 )
+ { if( testing ) std::fprintf( stderr, "ok\n" );
+ else std::fprintf( stderr, "done\n" ); }
+
+ if( debug_level & 1 )
+ std::fprintf( stderr,
+ "splitter tried to send a packet %8lu times\n"
+ "splitter had to wait %8lu times\n"
+ "any worker tried to consume from splitter %8lu times\n"
+ "any worker had to wait %8lu times\n"
+ "muxer tried to consume from workers %8lu times\n"
+ "muxer had to wait %8lu times\n",
+ courier.tally().check_counter,
+ courier.tally().wait_counter,
+ courier.icheck_counter,
+ courier.iwait_counter,
+ courier.ocheck_counter,
+ courier.owait_counter );
+
+ if( !courier.finished() ) internal_error( "courier not finished" );
+ return 0;
}