Modify logic to compute the count of CPUs allocated to a job step as

used to set SLURM_CPUS_ON_NODE env var and compute memory limit.
We compare the allocated CPUs information passed in the job step 
credential (typically cores) and compare with the CPUs configured
on the node and scale the value as needed.

src/slurmd/slurmd/req.c

@@ -735,7 +735,16 @@ _check_job_credential(launch_tasks_request_msg_t *req, uid_t uid,
 			      tasks_to_launch, alloc_lps, arg.jobid,
 			      arg.stepid, (long) arg.uid, arg.hostlist);
 		}
-        } else
+		/* NOTE: alloc_lps is the count of allocated resources
+		 * (typically cores). Convert to CPU count as needed */
+		if (i_last_bit <= i_first_bit)
+			error("step credential has no CPUs selected");
+		else {
+			i = conf->conf_cpus / (i_last_bit - i_first_bit);
+			if (i > 1)
+				alloc_lps *= i;
+		}
+	} else
 		alloc_lps = 1;
 
 	/* Overwrite any memory limits in the RPC with contents of the 
@@ -968,7 +977,7 @@ _set_batch_job_limits(slurm_msg_t *msg)
 		
 	if (arg.job_mem & MEM_PER_CPU) {
 		int i;
-		uint32_t alloc_lps = 0, last_bit;
+		uint32_t alloc_lps = 0, last_bit = 0;   
 		if (arg.job_nhosts > 0) {
 			last_bit = arg.sockets_per_node[0] * 
 				   arg.cores_per_socket[0];
@@ -981,6 +990,17 @@ _set_batch_job_limits(slurm_msg_t *msg)
 			error("_set_batch_job_limit: alloc_lps is zero");
 			alloc_lps = 1;
 		}
+
+		/* NOTE: alloc_lps is the count of allocated resources
+		 * (typically cores). Convert to CPU count as needed */
+		if (last_bit < 1)
+			error("Batch job credential allocates no CPUs");
+		else {
+			i = conf->conf_cpus / last_bit;
+			if (i > 1)
+				alloc_lps *= i;
+		}
+
 		req->job_mem = arg.job_mem & (~MEM_PER_CPU);
 		req->job_mem *= alloc_lps;
 	} else