// calcInChunck does calculation in a chunck of SNPs.
func (cmd *cmdCt) calcInChunck(snpChan chan *pileup.SNP) *calc.Calculator {
// Create job channel.
// Each job is a array of SNPs, which we will calculate
// correlations of the first SNP with the rest ones.
jobChan := make(chan []*pileup.SNP)
go func() {
defer close(jobChan)
arr := []*pileup.SNP{}
for s := range snpChan {
arr = append(arr, s)
lag := s.Pos - arr[0].Pos
if lag < 0 {
cmd.panic("SNPs are not in order.")
}
if lag >= cmd.maxl {
jobChan <- arr
arr = arr[1:]
}
}
jobChan <- arr
}()
// Make ncpu workers.
// For each worker, do the calculation,
// and push the result into a channel.
ncpu := runtime.GOMAXPROCS(0)
c := make(chan *calc.Calculator)
for i := 0; i < ncpu; i++ {
go func() {
covs := calc.New(cmd.maxl)
maxN := 10000
xArr := make([]float64, maxN)
yArr := make([]float64, maxN)
for arr := range jobChan {
cmd.calcSNPArr(arr, covs, xArr, yArr)
}
c <- covs
}()
}
// Wait for all the worker,
// and collect their results.
var covs *calc.Calculator
for i := 0; i < ncpu; i++ {
cc := <-c
if i == 0 {
covs = cc
} else {
covs.Append(cc)
}
}
return covs
}
// calcSNPArr calculate correlation of the first SNP with the rest.
// It finds pairs of bases that are from the same read,
// compare every two pairs of bases,
// and compute several correlations, which is contained in a calculator.
// A calculator here is a black box.
// calcSNPArr only push inputs into the calculator.
func (cmd *cmdCt) calcSNPArr(snpArr []*pileup.SNP, calculator *calc.Calculator, xArr, yArr []float64) {
s1 := snpArr[0]
m := make(map[string]pileup.Allele)
for _, a := range s1.Alleles {
if isATGC(a.Base) {
m[a.QName] = a
}
}
pairs := make([]AllelePair, len(m))
for k := 0; k < len(snpArr); k++ {
s2 := snpArr[k]
// double check the lag.
// it expects an order array of SNPs.
l := s2.Pos - s1.Pos
if l >= cmd.maxl {
break
} else if l < 0 {
cmd.panic("SNPs is not in order.")
}
numPair := cmd.findPairs(m, s2.Alleles, pairs)
// check the coverage.
// if less than min coverage, skip.
if numPair < cmd.minCoverage {
continue
}
k := 0
for i := 0; i < numPair && k < len(xArr); i++ {
p1 := pairs[i]
for j := i + 1; j < numPair && k < len(xArr); j++ {
p2 := pairs[j]
x := cmd.diffBases(p1.A.Base, p2.A.Base)
y := cmd.diffBases(p1.B.Base, p2.B.Base)
xArr[k] = x
yArr[k] = y
k++
}
}
calculator.Increment(xArr[:k], yArr[:k], l)
}
}