static ucs_status_t uct_perf_create_md(ucx_perf_context_t *perf)
{
uct_md_resource_desc_t *md_resources;
uct_tl_resource_desc_t *tl_resources;
unsigned i, num_md_resources;
unsigned j, num_tl_resources;
ucs_status_t status;
uct_md_h md;
uct_md_config_t *md_config;
status = uct_query_md_resources(&md_resources, &num_md_resources);
if (status != UCS_OK) {
goto out;
}
for (i = 0; i < num_md_resources; ++i) {
status = uct_md_config_read(md_resources[i].md_name, NULL, NULL, &md_config);
if (status != UCS_OK) {
goto out_release_md_resources;
}
status = uct_md_open(md_resources[i].md_name, md_config, &md);
uct_config_release(md_config);
if (status != UCS_OK) {
goto out_release_md_resources;
}
status = uct_md_query_tl_resources(md, &tl_resources, &num_tl_resources);
if (status != UCS_OK) {
uct_md_close(md);
goto out_release_md_resources;
}
for (j = 0; j < num_tl_resources; ++j) {
if (!strcmp(perf->params.uct.tl_name, tl_resources[j].tl_name) &&
!strcmp(perf->params.uct.dev_name, tl_resources[j].dev_name))
{
uct_release_tl_resource_list(tl_resources);
perf->uct.md = md;
status = UCS_OK;
goto out_release_md_resources;
}
}
uct_md_close(md);
uct_release_tl_resource_list(tl_resources);
}
ucs_error("Cannot use transport %s on device %s", perf->params.uct.tl_name,
perf->params.uct.dev_name);
status = UCS_ERR_NO_DEVICE;
out_release_md_resources:
uct_release_md_resource_list(md_resources);
out:
return status;
}
/* Device and transport to be used are determined by minimum latency */
static ucs_status_t dev_tl_lookup(const char *dev_name, const char *tl_name, struct iface_info *iface_p)
{
int i;
int j;
ucs_status_t status;
uct_md_resource_desc_t *md_resources; /* Memory domain resource descriptor */
uct_tl_resource_desc_t *tl_resources; /*Communication resource descriptor */
unsigned num_md_resources; /* Number of protected domain */
unsigned num_tl_resources; /* Number of transport resources resource objects created */
uct_md_config_t *md_config;
status = uct_query_md_resources(&md_resources, &num_md_resources);
CHKERR_JUMP(UCS_OK != status, "query for protected domain resources", error_ret);
/* Iterate through protected domain resources */
for (i = 0; i < num_md_resources; ++i) {
status = uct_md_config_read(md_resources[i].md_name, NULL, NULL, &md_config);
CHKERR_JUMP(UCS_OK != status, "read PD config", release_pd);
status = uct_md_open(md_resources[i].md_name, md_config, &iface_p->pd);
uct_config_release(md_config);
CHKERR_JUMP(UCS_OK != status, "open protected domains", release_pd);
status = uct_md_query_tl_resources(iface_p->pd, &tl_resources, &num_tl_resources);
CHKERR_JUMP(UCS_OK != status, "query transport resources", close_pd);
/* Go through each available transport and find the proper name */
for (j = 0; j < num_tl_resources; ++j) {
if (!strcmp(dev_name, tl_resources[j].dev_name) &&
!strcmp(tl_name, tl_resources[j].tl_name)) {
status = init_iface(tl_resources[j].dev_name, tl_resources[j].tl_name, iface_p);
if (UCS_OK == status) {
printf("Using %s with %s.\n", tl_resources[j].dev_name, tl_resources[j].tl_name);
fflush(stdout);
uct_release_tl_resource_list(tl_resources);
goto release_pd;
}
}
}
uct_release_tl_resource_list(tl_resources);
uct_md_close(iface_p->pd);
}
fprintf(stderr, "No supported (dev/tl) found (%s/%s)\n", dev_name, tl_name);
status = UCS_ERR_UNSUPPORTED;
release_pd:
uct_release_md_resource_list(md_resources);
error_ret:
return status;
close_pd:
uct_md_close(iface_p->pd);
goto release_pd;
}
static ucs_status_t ucp_fill_resources(ucp_context_h context,
const ucp_config_t *config)
{
unsigned num_tl_resources;
unsigned num_md_resources;
uct_md_resource_desc_t *md_rscs;
ucs_status_t status;
ucp_rsc_index_t i;
unsigned md_index;
uct_md_h md;
uct_md_config_t *md_config;
uint64_t masks[UCT_DEVICE_TYPE_LAST] = {0};
/* if we got here then num_resources > 0.
* if the user's device list is empty, there is no match */
if ((0 == config->devices[UCT_DEVICE_TYPE_NET].count) &&
(0 == config->devices[UCT_DEVICE_TYPE_SHM].count) &&
(0 == config->devices[UCT_DEVICE_TYPE_ACC].count) &&
(0 == config->devices[UCT_DEVICE_TYPE_SELF].count)) {
ucs_error("The device lists are empty. Please specify the devices you would like to use "
"or omit the UCX_*_DEVICES so that the default will be used.");
status = UCS_ERR_NO_ELEM;
goto err;
}
/* if we got here then num_resources > 0.
* if the user's tls list is empty, there is no match */
if (0 == config->tls.count) {
ucs_error("The TLs list is empty. Please specify the transports you would like to use "
"or omit the UCX_TLS so that the default will be used.");
status = UCS_ERR_NO_ELEM;
goto err;
}
/* List memory domain resources */
status = uct_query_md_resources(&md_rscs, &num_md_resources);
if (status != UCS_OK) {
goto err;
}
/* Sort md's by name, to increase the likelihood of reusing the same ep
* configuration (since remote md map is part of the key).
*/
qsort(md_rscs, num_md_resources, sizeof(*md_rscs), ucp_md_rsc_compare_name);
/* Error check: Make sure there is at least one MD */
if (num_md_resources == 0) {
ucs_error("No md resources found");
status = UCS_ERR_NO_DEVICE;
goto err_release_md_resources;
}
context->num_mds = 0;
context->md_rscs = NULL;
context->mds = NULL;
context->md_attrs = NULL;
context->num_tls = 0;
context->tl_rscs = NULL;
/* Allocate array of MD resources we would actually use */
context->md_rscs = ucs_calloc(num_md_resources, sizeof(*context->md_rscs),
"ucp_md_resources");
if (context->md_rscs == NULL) {
status = UCS_ERR_NO_MEMORY;
goto err_free_context_resources;
}
/* Allocate array of memory domains */
context->mds = ucs_calloc(num_md_resources, sizeof(*context->mds), "ucp_mds");
if (context->mds == NULL) {
status = UCS_ERR_NO_MEMORY;
goto err_free_context_resources;
}
/* Allocate array of memory domains attributes */
context->md_attrs = ucs_calloc(num_md_resources, sizeof(*context->md_attrs),
"ucp_md_attrs");
if (context->md_attrs == NULL) {
status = UCS_ERR_NO_MEMORY;
goto err_free_context_resources;
}
/* Open all memory domains, keep only those which have at least one TL
* resources selected on them.
*/
md_index = 0;
for (i = 0; i < num_md_resources; ++i) {
status = uct_md_config_read(md_rscs[i].md_name, NULL, NULL, &md_config);
if (status != UCS_OK) {
goto err_free_context_resources;
}
status = uct_md_open(md_rscs[i].md_name, md_config, &md);
uct_config_release(md_config);
if (status != UCS_OK) {
goto err_free_context_resources;
}
context->md_rscs[md_index] = md_rscs[i];
context->mds[md_index] = md;
/* Save MD attributes */
status = uct_md_query(md, &context->md_attrs[md_index]);
if (status != UCS_OK) {
goto err_free_context_resources;
}
/* Add communication resources of each MD */
status = ucp_add_tl_resources(context, md, md_index, config,
&num_tl_resources, masks);
if (status != UCS_OK) {
goto err_free_context_resources;
}
/* If the MD does not have transport resources, don't use it */
if (num_tl_resources > 0) {
++md_index;
++context->num_mds;
} else {
ucs_debug("closing md %s because it has no selected transport resources",
md_rscs[i].md_name);
uct_md_close(md);
}
}
/* Error check: Make sure there is at least one transport */
if (0 == context->num_tls) {
ucs_error("There are no available resources matching the configured criteria");
status = UCS_ERR_NO_DEVICE;
goto err_free_context_resources;
}
if (context->num_mds > UCP_MD_INDEX_BITS) {
ucs_error("Only up to %d memory domains are supported (have: %d)",
UCP_MD_INDEX_BITS, context->num_mds);
status = UCS_ERR_EXCEEDS_LIMIT;
goto err_release_md_resources;
}
/* Notify the user if there are devices from the command line that are not available */
ucp_check_unavailable_devices(config->devices, masks);
/* Error check: Make sure there are not too many transports */
if (context->num_tls >= UCP_MAX_RESOURCES) {
ucs_error("Exceeded resources limit (%u requested, up to %d are supported)",
context->num_tls, UCP_MAX_RESOURCES);
status = UCS_ERR_EXCEEDS_LIMIT;
goto err_free_context_resources;
}
status = ucp_check_tl_names(context);
if (status != UCS_OK) {
goto err_free_context_resources;
}
uct_release_md_resource_list(md_rscs);
return UCS_OK;
err_free_context_resources:
ucp_free_resources(context);
err_release_md_resources:
uct_release_md_resource_list(md_rscs);
err:
return status;
}
void print_uct_config(ucs_config_print_flags_t print_flags, const char *tl_name)
{
uct_md_resource_desc_t *md_resources;
unsigned md_rsc_index, num_md_resources;
uct_tl_resource_desc_t *tl_resources;
unsigned tl_rsc_index, num_tl_resources;
uct_iface_config_t *config;
char tl_names[UINT8_MAX][UCT_TL_NAME_MAX];
char cfg_title[UCT_TL_NAME_MAX + 128];
unsigned i, num_tls;
ucs_status_t status;
uct_md_h md;
uct_md_config_t *md_config;
status = uct_query_md_resources(&md_resources, &num_md_resources);
if (status != UCS_OK) {
return;
}
uct_md_component_config_print(print_flags);
num_tls = 0;
for (md_rsc_index = 0; md_rsc_index < num_md_resources; ++md_rsc_index) {
status = uct_md_config_read(md_resources[md_rsc_index].md_name, NULL,
NULL, &md_config);
if (status != UCS_OK) {
continue;
}
status = uct_md_open(md_resources[md_rsc_index].md_name, md_config, &md);
uct_config_release(md_config);
if (status != UCS_OK) {
continue;
}
status = uct_md_query_tl_resources(md, &tl_resources, &num_tl_resources);
if (status != UCS_OK) {
uct_md_close(md);
continue;
}
for (tl_rsc_index = 0; tl_rsc_index < num_tl_resources; ++tl_rsc_index) {
i = 0;
while (i < num_tls) {
if (!strcmp(tl_names[i], tl_resources[tl_rsc_index].tl_name)) {
break;
}
++i;
}
/* Make sure this transport is not inserted to the array before, and
* if user selects a specific transport - also make sure this is it.
*/
if ((i == num_tls) &&
((tl_name == NULL) || !strcmp(tl_name, tl_resources[tl_rsc_index].tl_name)))
{
strncpy(tl_names[num_tls], tl_resources[tl_rsc_index].tl_name,
UCT_TL_NAME_MAX);
++num_tls;
}
}
uct_release_tl_resource_list(tl_resources);
uct_md_close(md);
}
uct_release_md_resource_list(md_resources);
for (i = 0; i < num_tls; ++i) {
snprintf(cfg_title, sizeof(cfg_title), "%s transport configuration",
tl_names[i]);
status = uct_iface_config_read(tl_names[i], NULL, NULL, &config);
if (status != UCS_OK) {
printf("# < Failed to read configuration >\n");
continue;
}
uct_config_print(config, stdout, cfg_title, print_flags);
uct_config_release(config);
}
}
static void print_md_info(const char *md_name, int print_opts,
ucs_config_print_flags_t print_flags,
const char *req_tl_name)
{
uct_tl_resource_desc_t *resources, tmp;
unsigned resource_index, j, num_resources, count;
ucs_status_t status;
const char *tl_name;
uct_md_config_t *md_config;
uct_md_attr_t md_attr;
uct_md_h md;
status = uct_md_config_read(md_name, NULL, NULL, &md_config);
if (status != UCS_OK) {
goto out;
}
status = uct_md_open(md_name, md_config, &md);
uct_config_release(md_config);
if (status != UCS_OK) {
printf("# < failed to open memory domain %s >\n", md_name);
goto out;
}
status = uct_md_query_tl_resources(md, &resources, &num_resources);
if (status != UCS_OK) {
printf("# < failed to query memory domain resources >\n");
goto out_close_md;
}
if (req_tl_name != NULL) {
resource_index = 0;
while (resource_index < num_resources) {
if (!strcmp(resources[resource_index].tl_name, req_tl_name)) {
break;
}
++resource_index;
}
if (resource_index == num_resources) {
/* no selected transport on the MD */
goto out_free_list;
}
}
status = uct_md_query(md, &md_attr);
if (status != UCS_OK) {
printf("# < failed to query memory domain >\n");
goto out_free_list;
} else {
printf("#\n");
printf("# Memory domain: %s\n", md_name);
printf("# component: %s\n", md_attr.component_name);
if (md_attr.cap.flags & UCT_MD_FLAG_ALLOC) {
printf("# allocate: %s\n",
size_limit_to_str(md_attr.cap.max_alloc));
}
if (md_attr.cap.flags & UCT_MD_FLAG_REG) {
printf("# register: %s, cost: %.0f",
size_limit_to_str(md_attr.cap.max_reg),
md_attr.reg_cost.overhead * 1e9);
if (md_attr.reg_cost.growth * 1e9 > 1e-3) {
printf("+(%.3f*<SIZE>)", md_attr.reg_cost.growth * 1e9);
}
printf(" nsec\n");
}
printf("# remote key: %zu bytes\n", md_attr.rkey_packed_size);
}
if (num_resources == 0) {
printf("# < no supported devices found >\n");
goto out_free_list;
}
resource_index = 0;
while (resource_index < num_resources) {
/* Gather all resources for this transport */
tl_name = resources[resource_index].tl_name;
count = 1;
for (j = resource_index + 1; j < num_resources; ++j) {
if (!strcmp(tl_name, resources[j].tl_name)) {
tmp = resources[count + resource_index];
resources[count + resource_index] = resources[j];
resources[j] = tmp;
++count;
}
}
if ((req_tl_name == NULL) || !strcmp(tl_name, req_tl_name)) {
print_tl_info(md, tl_name, &resources[resource_index], count,
print_opts, print_flags);
}
resource_index += count;
}
out_free_list:
uct_release_tl_resource_list(resources);
out_close_md:
uct_md_close(md);
out:
;
}
