static ucs_status_t uct_gdr_copy_query_md_resources(uct_md_resource_desc_t **resources_p,
unsigned *num_resources_p)
{
int num_gpus;
gdr_t ctx;
cudaError_t cudaErr;
cudaErr = cudaGetDeviceCount(&num_gpus);
if ((cudaErr != cudaSuccess) || (num_gpus == 0)) {
ucs_debug("not found cuda devices");
*resources_p = NULL;
*num_resources_p = 0;
return UCS_OK;
}
ctx = gdr_open();
if (ctx == NULL) {
ucs_debug("could not open gdr copy. disabling gdr copy resource");
*resources_p = NULL;
*num_resources_p = 0;
return UCS_OK;
}
gdr_close(ctx);
return uct_single_md_resource(&uct_gdr_copy_md_component, resources_p,
num_resources_p);
}
/* For Cray devices we have only one MD */
static ucs_status_t uct_ugni_query_md_resources(uct_md_resource_desc_t **resources_p,
unsigned *num_resources_p)
{
if (getenv("PMI_GNI_PTAG") != NULL) {
return uct_single_md_resource(&uct_ugni_md_component, resources_p, num_resources_p);
} else {
*resources_p = NULL;
*num_resources_p = 0;
return UCS_OK;
}
}
static ucs_status_t uct_knem_query_md_resources(uct_md_resource_desc_t **resources_p,
unsigned *num_resources_p)
{
int fd;
int rc;
struct knem_cmd_info info;
memset(&info, 0, sizeof(struct knem_cmd_info));
fd = open("/dev/knem", O_RDWR);
if (fd < 0) {
ucs_debug("Could not open the KNEM device file at /dev/knem: %m. Disabling knem resource");
*resources_p = NULL;
*num_resources_p = 0;
return UCS_OK;
}
rc = ioctl(fd, KNEM_CMD_GET_INFO, &info);
if (rc < 0) {
*resources_p = NULL;
*num_resources_p = 0;
close(fd);
ucs_debug("KNEM get info failed. not using knem, err = %d %m", rc);
return UCS_OK;
}
if (KNEM_ABI_VERSION != info.abi) {
*resources_p = NULL;
*num_resources_p = 0;
close(fd);
ucs_error("KNEM ABI mismatch: KNEM_ABI_VERSION: %d, Driver binary interface version: %d",
KNEM_ABI_VERSION, info.abi);
return UCS_OK;
}
/* We have to close it since it is not clear
* if it will be selected in future */
close(fd);
return uct_single_md_resource(&uct_knem_md_component, resources_p,
num_resources_p);
}
static ucs_status_t uct_self_query_md_resources(uct_md_resource_desc_t **resources_p,
unsigned *num_resources_p)
{
return uct_single_md_resource(&uct_self_md, resources_p, num_resources_p);
}
