diff --git a/src/common/node_conf.c b/src/common/node_conf.c
index a3e2cdf86f28fd5a5ecf1fd07868346ea5b687d2..eea68a9e6362f0e93de738b650fb911ee22e8145 100644
--- a/src/common/node_conf.c
+++ b/src/common/node_conf.c
@@ -1094,3 +1094,18 @@ extern uint32_t cr_get_coremap_offset(uint32_t node_index)
xassert(cr_node_cores_offset);
return cr_node_cores_offset[node_index];
}
+
+/* The value of threads should be from config_ptr->threads if slurm.conf has fast schedule turned on (config_record);
+ * otherwise, it should be from node_ptr->threads (node_record).
+ */
+extern int _adjust_cpus_nppcu(uint16_t ntasks_per_core, uint16_t threads, int cpus ) {
+
+ if (ntasks_per_core == 0xffff)
+ ntasks_per_core = threads;
+
+ /* Adjust the number of CPUs according to the percentage of the hwthreads/core being used. */
+ cpus *= ntasks_per_core;
+ cpus /= threads;
+
+ return cpus;
+}
diff --git a/src/common/node_conf.h b/src/common/node_conf.h
index bc3afcf227db61cd3707e7df63aab5fa13a8b477..22b44c6ff2434165601b6d736c420c7a74a048a6 100644
--- a/src/common/node_conf.h
+++ b/src/common/node_conf.h
@@ -281,4 +281,9 @@ extern void cr_fini_global_core_data(void);
/*return the coremap index to the first core of the given node */
extern uint32_t cr_get_coremap_offset(uint32_t node_index);
+/* Given the number of tasks per core and the actual number of hw threads, compute
+ * how many CPUs are "visible" and, hence, usable on the node.
+ */
+extern int _adjust_cpus_nppcu(uint16_t ntasks_per_core, uint16_t threads, int cpus);
+
#endif /* !_HAVE_NODE_CONF_H */
diff --git a/src/common/slurm_resource_info.c b/src/common/slurm_resource_info.c
index 9b6c05332121c51e6314eeb2521b24c8cc732892..66d6d358d4ff013efba2a96d7622d5be451a5fb0 100644
--- a/src/common/slurm_resource_info.c
+++ b/src/common/slurm_resource_info.c
@@ -137,21 +137,9 @@ int slurm_get_avail_procs(const uint16_t socket_cnt,
uint32_t job_id,
char *name)
{
- uint16_t avail_cpus = 0, max_cpus = 0;
- uint16_t allocated_cpus = 0, allocated_cores = 0, allocated_sockets = 0;
- uint16_t max_avail_cpus = 0xffff; /* for alloc_* accounting */
- uint16_t min_sockets = 1, max_sockets = 0xffff;
- uint16_t min_cores = 1, max_cores = 0xffff;
- uint16_t max_threads = 0xffff;
- int i;
+ uint16_t avail_cpus = 0;
/* pick defaults for any unspecified items */
- if (socket_cnt != (uint16_t) NO_VAL)
- min_sockets = max_sockets = socket_cnt;
- if (core_cnt != (uint16_t) NO_VAL)
- min_cores = max_cores = core_cnt;
- if (thread_cnt != (uint16_t) NO_VAL)
- max_threads = thread_cnt;
if (cpus_per_task <= 0)
cpus_per_task = 1;
if (*threads <= 0)
@@ -160,11 +148,7 @@ int slurm_get_avail_procs(const uint16_t socket_cnt,
*cores = 1;
if (*sockets <= 0)
*sockets = *cpus / *cores / *threads;
- for (i = 0 ; alloc_cores && i < *sockets; i++) {
- allocated_cores += alloc_cores[i];
- if (alloc_cores[i])
- allocated_sockets++;
- }
+
#if (DEBUG)
info("get_avail_procs %u %s User_ sockets %u cores %u threads %u",
job_id, name, socket_cnt, core_cnt, thread_cnt);
@@ -173,139 +157,31 @@ int slurm_get_avail_procs(const uint16_t socket_cnt,
info("get_avail_procs %u %s Ntask node %u sockets %u core %u",
job_id, name, ntaskspernode, ntaskspersocket,
ntaskspercore);
- info("get_avail_procs %u %s cr_type %d cpus %u alloc_ c %u s %u",
- job_id, name, cr_type, *cpus, allocated_cores,
- allocated_sockets);
+ info("get_avail_procs %u %s cr_type %d cpus %u",
+ job_id, name, cr_type, *cpus);
for (i = 0; alloc_cores && i < *sockets; i++)
info("get_avail_procs %u %s alloc_cores[%d] = %u",
job_id, name, i, alloc_cores[i]);
#endif
- allocated_cpus = allocated_cores * (*threads);
-
- /* For the following CR types, nodes have no notion of socket, core,
- and thread. Only one level of logical processors */
- if (cr_type & CR_CORE) {
- if (*cpus >= allocated_cpus)
- *cpus -= allocated_cpus;
- else {
- *cpus = 0;
- error("cons_res: *cpus underflow");
- }
- if (allocated_cores > 0) {
- max_avail_cpus = 0;
- int tmp_diff = 0;
- for (i=0; i<*sockets; i++) {
- tmp_diff = *cores - alloc_cores[i];
- if (min_cores <= tmp_diff) {
- tmp_diff *= (*threads);
- max_avail_cpus += tmp_diff;
- }
- }
- }
-
- /*** honor socket/core/thread maximums ***/
- *sockets = MIN(*sockets, max_sockets);
- *cores = MIN(*cores, max_cores);
- *threads = MIN(*threads, max_threads);
-
- if (min_sockets > *sockets) {
- *cpus = 0;
- } else {
- int max_cpus_socket = 0;
- max_cpus = 0;
- for (i=0; i<*sockets; i++) {
- max_cpus_socket = 0;
- if (min_cores <= *cores) {
- int num_threads = *threads;
- if (ntaskspercore > 0) {
- num_threads = MIN(num_threads,
- ntaskspercore);
- }
- max_cpus_socket = *cores * num_threads;
- }
- if (ntaskspersocket > 0) {
- max_cpus_socket = MIN(max_cpus_socket,
- ntaskspersocket);
- }
- max_cpus += max_cpus_socket;
- }
- max_cpus = MIN(max_cpus, max_avail_cpus);
- }
-
- /*** honor any availability maximum ***/
- max_cpus = MIN(max_cpus, max_avail_cpus);
-
- if (ntaskspernode > 0) {
- max_cpus = MIN(max_cpus, ntaskspernode);
- }
- } else if (cr_type & CR_SOCKET) {
- if (*sockets >= allocated_sockets)
- *sockets -= allocated_sockets; /* sockets count */
- else {
- *sockets = 0;
- error("cons_res: *sockets underflow");
- }
- if (*cpus >= allocated_cpus)
- *cpus -= allocated_cpus;
- else {
- *cpus = 0;
- error("cons_res: *cpus underflow");
- }
-
- if (min_sockets > *sockets)
- *cpus = 0;
-
- /*** honor socket/core/thread maximums ***/
- *sockets = MIN(*sockets, max_sockets);
- *cores = MIN(*cores, max_cores);
- *threads = MIN(*threads, max_threads);
+ uint32_t nppcu = ntaskspercore;
- /*** compute an overall maximum cpu count honoring ntasks* ***/
- max_cpus = *threads;
- if (ntaskspercore > 0) {
- max_cpus = MIN(max_cpus, ntaskspercore);
- }
- max_cpus *= *cores;
- if (ntaskspersocket > 0) {
- max_cpus = MIN(max_cpus, ntaskspersocket);
- }
- max_cpus *= *sockets;
- if (ntaskspernode > 0) {
- max_cpus = MIN(max_cpus, ntaskspernode);
- }
-
- /*** honor any availability maximum ***/
- max_cpus = MIN(max_cpus, max_avail_cpus);
- } else { /* CR_CPU (default) */
- if ((cr_type & CR_CPU) ||
- (!(cr_type & CR_MEMORY))) {
- if (*cpus >= allocated_cpus)
- *cpus -= allocated_cpus;
- else {
- *cpus = 0;
- error("cons_res: *cpus underflow");
- }
- }
-
- /*** compute an overall maximum cpu count honoring ntasks* ***/
- max_cpus = *cpus;
- if (ntaskspernode > 0) {
- max_cpus = MIN(max_cpus, ntaskspernode);
- }
+ if (nppcu == 0xffff) { /* nppcu was not explicitly set, use all threads */
+ info( "***** nppcu was not explicitly set, use all threads *****");
+ nppcu = *threads;
}
- /*** factor cpus_per_task into max_cpus ***/
- max_cpus *= cpus_per_task;
- /*** round down available based on cpus_per_task ***/
- avail_cpus = (*cpus / cpus_per_task) * cpus_per_task;
- avail_cpus = MIN(avail_cpus, max_cpus);
+ avail_cpus = *sockets * *cores * *threads;
+ avail_cpus = avail_cpus * nppcu / *threads;
+
+ if(ntaskspernode>0)
+ avail_cpus = MIN(avail_cpus, ntaskspernode * cpus_per_task);
#if (DEBUG)
info("get_avail_procs %u %s return cpus %u sockets %u cores %u threads %u",
job_id, name, *cpus, *sockets, *cores, *threads);
info("get_avail_procs %d %s avail_cpus %u", job_id, name, avail_cpus);
#endif
- return(avail_cpus);
+ return avail_cpus;
}
/*
diff --git a/src/plugins/select/cray/basil_alps.h b/src/plugins/select/cray/basil_alps.h
index 6660fbc8cefaf5c5281cda3813a7fd7a265b1a6a..470fa3323d0ae9e1e3fdf1c6b5e86f74c76fa46e 100644
--- a/src/plugins/select/cray/basil_alps.h
+++ b/src/plugins/select/cray/basil_alps.h
@@ -448,7 +448,8 @@ struct basil_rsvn_param {
depth, /* depth > 0, -d */
nppn, /* nppn > 0, -N */
npps, /* PEs per segment, -S */
- nspn; /* segments per node, -sn */
+ nspn, /* segments per node, -sn */
+ nppcu; /* Processors Per Compute Unit. BASIL 1.3 */
char *nodes; /* NodeParamArray */
struct basil_label *labels; /* LabelParamArray */
@@ -622,7 +623,7 @@ extern void free_inv(struct basil_inventory *inv);
extern long basil_reserve(const char *user, const char *batch_id,
uint32_t width, uint32_t depth, uint32_t nppn,
- uint32_t mem_mb, struct nodespec *ns_head,
+ uint32_t mem_mb, uint32_t nppcu, struct nodespec *ns_head,
struct basil_accel_param *accel_head);
extern int basil_confirm(uint32_t rsvn_id, int job_id, uint64_t pagg_id);
extern const struct basil_rsvn *basil_rsvn_by_id(const struct basil_inventory *inv,
diff --git a/src/plugins/select/cray/basil_interface.c b/src/plugins/select/cray/basil_interface.c
index a91376d9b63d0383e4fa0e102b969b3e2428cdb0..4f48e9ba2f73c38c11302a56b4c43091052bf963 100644
--- a/src/plugins/select/cray/basil_interface.c
+++ b/src/plugins/select/cray/basil_interface.c
@@ -741,6 +741,8 @@ extern int do_basil_reserve(struct job_record *job_ptr)
long rc;
char *user, batch_id[16];
struct basil_accel_param* bap;
+ uint16_t nppcu = 0;
+ uint16_t hwthreads_per_core = 1;
if (!job_ptr->job_resrcs || job_ptr->job_resrcs->nhosts == 0)
return SLURM_SUCCESS;
@@ -773,6 +775,17 @@ extern int do_basil_reserve(struct job_record *job_ptr)
node_min_mem = job_ptr->details->pn_min_memory;
}
+ if (slurmctld_conf.select_type_param & CR_ONE_TASK_PER_CORE) {
+ if ( job_ptr->details && job_ptr->details->mc_ptr && job_ptr->details->mc_ptr->ntasks_per_core == 0xffff ) {
+ nppcu = 1;
+ debug("No explicit ntasks-per-core has been set, using nppcu=1.");
+ }
+ }
+
+ if ( job_ptr->details && job_ptr->details->mc_ptr && job_ptr->details->mc_ptr->ntasks_per_core != 0xffff ) {
+ nppcu = job_ptr->details->mc_ptr->ntasks_per_core;
+ }
+
for (i = first_bit; i <= last_bit; i++) {
struct node_record *node_ptr = node_record_table_ptr + i;
uint32_t basil_node_id;
@@ -836,6 +849,12 @@ extern int do_basil_reserve(struct job_record *job_ptr)
for (i = 0; i < job_ptr->job_resrcs->nhosts; i++) {
uint32_t node_tasks = job_ptr->job_resrcs->cpus[i] / mppdepth;
+ if ( job_ptr->job_resrcs->sockets_per_node[i] > 0 && job_ptr->job_resrcs->cores_per_socket[i] > 0 )
+ hwthreads_per_core = job_ptr->job_resrcs->cpus[i] / job_ptr->job_resrcs->sockets_per_node[i] / job_ptr->job_resrcs->cores_per_socket[i];
+
+ if (nppcu)
+ node_tasks = node_tasks * nppcu / hwthreads_per_core;
+
if (mppnppn && mppnppn < node_tasks)
node_tasks = mppnppn;
mppwidth += node_tasks;
@@ -850,7 +869,7 @@ extern int do_basil_reserve(struct job_record *job_ptr)
bap = NULL;
rc = basil_reserve(user, batch_id, mppwidth, mppdepth, mppnppn,
- mppmem, ns_head, bap);
+ mppmem, (uint32_t)nppcu, ns_head, bap);
xfree(user);
if (rc <= 0) {
/* errno value will be resolved by select_g_job_begin() */
diff --git a/src/plugins/select/cray/libalps/basil_request.c b/src/plugins/select/cray/libalps/basil_request.c
index d704ee837e82083be9866447ad4f38f27b691fba..15f66f0baefbbf803191036420a06e5870c75790 100644
--- a/src/plugins/select/cray/libalps/basil_request.c
+++ b/src/plugins/select/cray/libalps/basil_request.c
@@ -88,9 +88,9 @@ static void _rsvn_write_reserve_xml(FILE *fp, struct basil_reservation *r,
if (version >= BV_5_1)
_write_xml(fp, " <ReserveParam architecture=\"%s\" "
"width=\"%ld\" depth=\"%ld\" nppn=\"%ld\""
- " nppcu=\"0\"",
+ " nppcu=\"%d\"",
nam_arch[param->arch],
- param->width, param->depth, param->nppn);
+ param->width, param->depth, param->nppn, param->nppcu);
else
_write_xml(fp, " <ReserveParam architecture=\"%s\" "
"width=\"%ld\" depth=\"%ld\" nppn=\"%ld\"",
diff --git a/src/plugins/select/cray/libalps/do_reserve.c b/src/plugins/select/cray/libalps/do_reserve.c
index 337f092a8267cd6901b86580aaee288154f26616..fc896ab38b008a5a1a1f4b597eb13a8388fd28ec 100644
--- a/src/plugins/select/cray/libalps/do_reserve.c
+++ b/src/plugins/select/cray/libalps/do_reserve.c
@@ -44,7 +44,7 @@ static int _rsvn_add_mem_param(struct basil_rsvn_param *rp, uint32_t mem_mb)
*/
static int _rsvn_add_params(struct basil_reservation *resv,
uint32_t width, uint32_t depth, uint32_t nppn,
- uint32_t mem_mb, char *mppnodes,
+ uint32_t mem_mb, char *mppnodes, uint32_t nppcu,
struct basil_accel_param *accel)
{
struct basil_rsvn_param *rp = xmalloc(sizeof(*rp));
@@ -57,6 +57,7 @@ static int _rsvn_add_params(struct basil_reservation *resv,
rp->depth = depth;
rp->nppn = nppn;
rp->nodes = mppnodes;
+ rp->nppcu = nppcu;
rp->accel = accel;
if (mem_mb && _rsvn_add_mem_param(rp, mem_mb) < 0) {
@@ -89,7 +90,7 @@ static struct basil_reservation *_rsvn_new(const char *user,
const char *batch_id,
uint32_t width, uint32_t depth,
uint32_t nppn, uint32_t mem_mb,
- char *mppnodes,
+ uint32_t nppcu, char *mppnodes,
struct basil_accel_param *accel)
{
struct basil_reservation *res;
@@ -105,7 +106,7 @@ static struct basil_reservation *_rsvn_new(const char *user,
strncpy(res->batch_id, batch_id, sizeof(res->batch_id));
if (_rsvn_add_params(res, width, depth, nppn,
- mem_mb, mppnodes, accel) < 0) {
+ mem_mb, mppnodes, nppcu, accel) < 0) {
free_rsvn(res);
return NULL;
}
@@ -127,7 +128,7 @@ static struct basil_reservation *_rsvn_new(const char *user,
*/
long basil_reserve(const char *user, const char *batch_id,
uint32_t width, uint32_t depth, uint32_t nppn,
- uint32_t mem_mb, struct nodespec *ns_head,
+ uint32_t mem_mb, uint32_t nppcu, struct nodespec *ns_head,
struct basil_accel_param *accel_head)
{
struct basil_reservation *rsvn;
@@ -138,7 +139,7 @@ long basil_reserve(const char *user, const char *batch_id,
free_nodespec(ns_head);
rsvn = _rsvn_new(user, batch_id, width, depth, nppn, mem_mb,
- mppnodes, accel_head);
+ nppcu, mppnodes, accel_head);
if (rsvn == NULL)
return -BE_INTERNAL;
diff --git a/src/plugins/select/cray/libemulate/alps_emulate.c b/src/plugins/select/cray/libemulate/alps_emulate.c
index 9707e53d449d043c84de0f17f3fcb7e6646bbdf0..09d14ec65f5f32675eef8e56fbacdd25e1e41ca7 100644
--- a/src/plugins/select/cray/libemulate/alps_emulate.c
+++ b/src/plugins/select/cray/libemulate/alps_emulate.c
@@ -507,7 +507,7 @@ extern void free_inv(struct basil_inventory *inv)
extern long basil_reserve(const char *user, const char *batch_id,
uint32_t width, uint32_t depth, uint32_t nppn,
- uint32_t mem_mb, struct nodespec *ns_head,
+ uint32_t mem_mb, uint32_t nppcu, struct nodespec *ns_head,
struct basil_accel_param *accel_head)
{
int i;
@@ -516,8 +516,8 @@ extern long basil_reserve(const char *user, const char *batch_id,
#if _DEBUG
struct nodespec *my_node_spec;
info("basil_reserve user:%s batch_id:%s width:%u depth:%u nppn:%u "
- "mem_mb:%u",
- user, batch_id, width, depth, nppn, mem_mb);
+ "mem_mb:%u nppcu:%u",
+ user, batch_id, width, depth, nppn, mem_mb, nppcu);
my_node_spec = ns_head;
while (my_node_spec) {
info("basil_reserve node_spec:start:%u,end:%u",
diff --git a/src/plugins/select/linear/select_linear.c b/src/plugins/select/linear/select_linear.c
index b228974373871c5d126c45ab45b1d0a38cc8f81b..be2967feb658c42fbf9198847bb4e61d7c5dd72d 100644
--- a/src/plugins/select/linear/select_linear.c
+++ b/src/plugins/select/linear/select_linear.c
@@ -503,12 +503,11 @@ static uint16_t _get_avail_cpus(struct job_record *job_ptr, int index)
debug2("host %s HW_ cpus %u boards %u sockets %u cores %u threads %u ",
node_ptr->name, cpus, boards, sockets, cores, threads);
#endif
-
avail_cpus = slurm_get_avail_procs(
min_sockets, min_cores, min_threads, cpus_per_task,
ntasks_per_node, ntasks_per_socket, ntasks_per_core,
&cpus, &sockets, &cores, &threads, NULL,
- CR_CPU, job_ptr->job_id, node_ptr->name);
+ CR_CORE, job_ptr->job_id, node_ptr->name);
#if SELECT_DEBUG
debug("avail_cpus index %d = %d (out of %d %d %d %d)",
@@ -532,6 +531,32 @@ static uint16_t _get_total_cpus(int index)
return node_ptr->cpus;
}
+static uint16_t _get_total_threads(int index)
+{
+ struct node_record *node_ptr = &(select_node_ptr[index]);
+ if (select_fast_schedule)
+ return node_ptr->config_ptr->threads;
+ else
+ return node_ptr->threads;
+}
+
+/*
+ * _get_ntasks_per_core - Retrieve the value of ntasks_per_core from
+ * the given job_details record. If it wasn't set, return 0xffff.
+ * Intended for use with the _adjust_cpus_nppcu function.
+ */
+
+static uint16_t _get_ntasks_per_core(struct job_details *details) {
+ uint16_t ntasks_per_core;
+
+ if ((details->mc_ptr))
+ ntasks_per_core = details->mc_ptr->ntasks_per_core;
+ else
+ ntasks_per_core = 0xffff;
+
+ return ntasks_per_core;
+}
+
static job_resources_t *_create_job_resources(int node_cnt)
{
job_resources_t *job_resrcs_ptr;
@@ -553,7 +578,7 @@ static void _build_select_struct(struct job_record *job_ptr, bitstr_t *bitmap)
{
int i, j, k;
int first_bit, last_bit;
- uint32_t node_cpus, total_cpus = 0, node_cnt;
+ uint32_t node_cpus, total_cpus = 0, node_cnt, node_threads;
struct node_record *node_ptr;
uint32_t job_memory_cpu = 0, job_memory_node = 0;
job_resources_t *job_resrcs_ptr;
@@ -586,10 +611,15 @@ static void _build_select_struct(struct job_record *job_ptr, bitstr_t *bitmap)
if (!bit_test(bitmap, i))
continue;
node_ptr = &(select_node_ptr[i]);
- if (select_fast_schedule)
- node_cpus = node_ptr->config_ptr->cpus;
- else
- node_cpus = node_ptr->cpus;
+ if (select_fast_schedule) {
+ node_cpus = node_ptr->config_ptr->cpus;
+ node_threads = node_ptr->config_ptr->threads;
+ } else {
+ node_cpus = node_ptr->cpus;
+ node_threads = node_ptr->threads;
+ }
+
+ node_cpus = _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), node_threads, node_cpus);
job_resrcs_ptr->cpus[j] = node_cpus;
if ((k == -1) ||
(job_resrcs_ptr->cpu_array_value[k] != node_cpus)) {
@@ -861,7 +891,7 @@ static int _job_test(struct job_record *job_ptr, bitstr_t *bitmap,
rem_nodes--;
max_nodes--;
rem_cpus -= avail_cpus;
- total_cpus += _get_total_cpus(index);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(index), _get_total_cpus(index));
} else { /* node not required (yet) */
bit_clear(bitmap, index);
consec_cpus[consec_index] += avail_cpus;
@@ -968,7 +998,7 @@ static int _job_test(struct job_record *job_ptr, bitstr_t *bitmap,
if (best_fit_nodes == 0)
break;
if (job_ptr->details->contiguous &&
- ((best_fit_cpus < rem_cpus) ||
+ ((best_fit_cpus < rem_cpus) ||
(!_enough_nodes(best_fit_nodes, rem_nodes,
min_nodes, req_nodes))))
break; /* no hole large enough */
@@ -988,7 +1018,7 @@ static int _job_test(struct job_record *job_ptr, bitstr_t *bitmap,
max_nodes--;
avail_cpus = _get_avail_cpus(job_ptr, i);
rem_cpus -= avail_cpus;
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
}
for (i = (best_fit_req - 1);
i >= consec_start[best_fit_location]; i--) {
@@ -1002,7 +1032,7 @@ static int _job_test(struct job_record *job_ptr, bitstr_t *bitmap,
max_nodes--;
avail_cpus = _get_avail_cpus(job_ptr, i);
rem_cpus -= avail_cpus;
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
}
} else {
for (i = consec_start[best_fit_location];
@@ -1017,7 +1047,7 @@ static int _job_test(struct job_record *job_ptr, bitstr_t *bitmap,
max_nodes--;
avail_cpus = _get_avail_cpus(job_ptr, i);
rem_cpus -= avail_cpus;
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
}
}
if (job_ptr->details->contiguous ||
@@ -1243,7 +1273,7 @@ static int _job_test_topo(struct job_record *job_ptr, bitstr_t *bitmap,
bit_set(bitmap, i);
alloc_nodes++;
rem_cpus -= avail_cpus;
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
}
}
/* Accumulate additional resources from leafs that
@@ -1278,7 +1308,7 @@ static int _job_test_topo(struct job_record *job_ptr, bitstr_t *bitmap,
bit_set(bitmap, i);
alloc_nodes++;
rem_cpus -= _get_avail_cpus(job_ptr, i);
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
if ((alloc_nodes > max_nodes) ||
((alloc_nodes >= want_nodes) &&
(rem_cpus <= 0)))
@@ -1348,7 +1378,7 @@ static int _job_test_topo(struct job_record *job_ptr, bitstr_t *bitmap,
bit_set(bitmap, i);
alloc_nodes++;
rem_cpus -= _get_avail_cpus(job_ptr, i);
- total_cpus += _get_total_cpus(i);
+ total_cpus += _adjust_cpus_nppcu(_get_ntasks_per_core(job_ptr->details), _get_total_threads(i), _get_total_cpus(i));
if ((alloc_nodes > max_nodes) ||
((alloc_nodes >= want_nodes) && (rem_cpus <= 0)))
break;
diff --git a/src/slurmctld/node_scheduler.c b/src/slurmctld/node_scheduler.c
index 93fb420bec3d44846be1511eabf45653dab7f1c4..120b4235e8232ee9cdd0a77d6fd5f1ed3b8ee5b7 100644
--- a/src/slurmctld/node_scheduler.c
+++ b/src/slurmctld/node_scheduler.c
@@ -130,6 +130,22 @@ static bitstr_t *_valid_features(struct job_details *detail_ptr,
static int _fill_in_gres_fields(struct job_record *job_ptr);
+/*
+ * _get_ntasks_per_core - Retrieve the value of ntasks_per_core from
+ * the given job_details record. If it wasn't set, return 0xffff.
+ * Intended for use with the _adjust_cpus_nppcu function.
+ */
+static uint16_t _get_ntasks_per_core(struct job_details *details) {
+ uint16_t ntasks_per_core;
+
+ if ((details->mc_ptr))
+ ntasks_per_core = details->mc_ptr->ntasks_per_core;
+ else
+ ntasks_per_core = 0xffff;
+
+ return ntasks_per_core;
+}
+
/*
* _build_gres_alloc_string - Fill in the gres_alloc string field for a
* given job_record
@@ -2073,7 +2089,7 @@ static int _build_node_list(struct job_record *job_ptr,
list_next(config_iterator))) {
config_filter = 0;
- if ((detail_ptr->pn_min_cpus > config_ptr->cpus ) ||
+ if ((detail_ptr->pn_min_cpus > _adjust_cpus_nppcu(_get_ntasks_per_core(detail_ptr), config_ptr->threads, config_ptr->cpus)) ||
((detail_ptr->pn_min_memory & (~MEM_PER_CPU)) >
config_ptr->real_memory) ||
(detail_ptr->pn_min_tmp_disk > config_ptr->tmp_disk))
@@ -2234,7 +2250,7 @@ static void _filter_nodes_in_set(struct node_set *node_set_ptr,
continue;
node_con = node_record_table_ptr[i].config_ptr;
- if ((job_con->pn_min_cpus <= node_con->cpus) &&
+ if ((job_con->pn_min_cpus <= _adjust_cpus_nppcu(_get_ntasks_per_core(job_con), node_con->threads, node_con->cpus)) &&
((job_con->pn_min_memory & (~MEM_PER_CPU)) <=
node_con->real_memory) &&
(job_con->pn_min_tmp_disk <= node_con->tmp_disk))
@@ -2263,7 +2279,7 @@ static void _filter_nodes_in_set(struct node_set *node_set_ptr,
continue;
node_ptr = &node_record_table_ptr[i];
- if ((job_con->pn_min_cpus <= node_ptr->cpus) &&
+ if ((job_con->pn_min_cpus <= _adjust_cpus_nppcu(_get_ntasks_per_core(job_con), node_ptr->threads, node_ptr->cpus)) &&
((job_con->pn_min_memory & (~MEM_PER_CPU)) <=
node_ptr->real_memory) &&
(job_con->pn_min_tmp_disk <= node_ptr->tmp_disk))