Fix h_vmem: requested total instead of per cpu

meaning we requested $cpus times as much as necessary.

nextflow.config

@@ -73,11 +73,15 @@ process.time = '5m'
 // Default cluster options (change this if you use an executor other than sge
 // or local)
 // Note: requesting h_stack avoids http://www.parallelpython.com/component/option,com_smf/Itemid,1/topic,395.0/wap2,wap2
-String h_vmem(def memory) {
-    memory = memory?.toMega()
-    memory ?  "-l h_vmem=${memory}M" : ""
+String h_vmem(def task) {
+    def memory = task.memory?.toMega()
+    if (!memory) {
+        return ""
+    }
+    memory /= task.cpus ?: 1
+    return "-l h_vmem=${memory}M"
 }
-process.clusterOptions = { "-S /bin/bash ${h_vmem(task.memory)} -l h_stack=8M" }
+process.clusterOptions = { "-S /bin/bash ${h_vmem(task)} -l h_stack=8M" }
 process.penv = 'serial'
 
 // Per process config