// Map routines to iterate a function over an array, potentially splitting the array slice into
// chunks so that each chunk is processed concurrently. When using concurrent processing the
// Chunk size is either the nearest even division of the total array over the chosen concurrent
// processing goroutines or a specified maximum chunk size, whichever is smaller. Reducing
// chunk size can reduce the impact of divergence in time for processing chunks, but may add
// to overhead.
func Map(set Mapper, threads, maxChunkSize int) (results []interface{}, err error) {
queue := make(chan Operator, 1)
p := NewProcessor(queue, 0, threads)
defer p.Stop()
chunkSize := util.Min(int(math.Ceil(float64(set.Len())/float64(threads))), maxChunkSize)
quit := make(chan struct{})
go func() {
for s := 0; s*chunkSize < set.Len(); s++ {
select {
case <-quit:
break
default:
endChunk := util.Min(chunkSize*(s+1), set.Len())
queue <- set.Slice(chunkSize*s, endChunk)
}
}
}()
for r := 0; r*chunkSize < set.Len(); r++ {
result := <-p.out
if result.Err != nil {
err = fmt.Errorf("concurrent: map failed: %v", err)
close(quit)
break
}
results = append(results, result.Value)
}
return
}
// Pack a sequence into the Packed sequence. Returns a string giving diagnostic information.
func (pa *Packer) Pack(seq *linear.Seq) (string, error) {
if pa.packed.Alpha == nil {
pa.packed.Alpha = seq.Alpha
} else if pa.packed.Alpha != seq.Alpha {
return "", errors.New("pals: alphabet mismatch")
}
c := contig{Seq: seq}
padding := binSize - seq.Len()%binSize
if padding < minPadding {
padding += binSize
}
pa.length += pa.lastPad
c.from = pa.length
pa.length += seq.Len()
pa.lastPad = padding
m := &pa.packed.seqMap
bins := make([]int, (padding+seq.Len())/binSize)
for i := 0; i < len(bins); i++ {
bins[i] = len(m.contigs)
}
m.binMap = append(m.binMap, bins...)
m.contigs = append(m.contigs, c)
return fmt.Sprintf("%20s\t%10d\t%7d-%-d", seq.ID[:util.Min(20, len(seq.ID))], seq.Len(), len(m.binMap)-len(bins), len(m.binMap)-1), nil
}
