Skip to content

use dmabuf option when the ibperf binaries have them configured #235

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ahskabir wants to merge 1 commit into
base: main
Choose a base branch
from
Open

use dmabuf option when the ibperf binaries have them configured #235

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
112 changes: 108 additions & 4 deletions cvs/lib/ibperf_lib.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,78 @@ def detect_rocm_path(phdl, config_rocm_path):
return '/opt/rocm'


def check_perftest_dmabuf_support(phdl, binary_path):
"""
Check if the given perftest binary on the cluster node supports the
--use_rocm_dmabuf flag by executing the binary's help output and
grepping for the flag.

Args:
phdl: Pssh handle used to execute commands on one or more cluster nodes.
binary_path (str): Full path to the perftest binary on the node.

Returns:
bool: True if the binary supports --use_rocm_dmabuf on any host,
False otherwise.
"""
cmd = f"{binary_path} --help 2>&1 | grep use_rocm_dmabuf"
log.info("Checking dmabuf support: %s", cmd)

try:
# detailed=True gives us exit_code per host
host_results = phdl.exec(cmd, detailed=True, print_console=False)

@cijohnson cijohnson Jul 6, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It assumes a homogeneous cluster — all nodes have the same binary at the same app_path. That's a safe assumption in the ibperf context (the binary path comes from a shared config and is the same for all nodes), and it's consistent with how the rest of the test already treats the cluster.

So we can just use shdl.exec instead of phdl.exec, shdl.exec will test the binary only in the head node(which is sufficent)

# host_results: {host: {"output": str, "exit_code": int}}

dmabuf_supported_any = False

for host, result in host_results.items():

@cijohnson cijohnson Jul 6, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

when shdl is used this whole for loop to set "dmabuf_supported_any " becomes unnecessary.

# Some safety in case of unexpected shape
if not isinstance(result, dict):
log.warning("Unexpected exec result type for host %s: %r", host, result)
continue

exit_code = result.get("exit_code", -1)
output = result.get("output", "") or ""

log.debug(
"dmabuf check host=%s exit_code=%s, output=%r",
host,
exit_code,
output,
)

# Mirror your manual test: grep returns 0 on match
if "use_rocm_dmabuf" in output:
if exit_code != 0:
log.warning(
"Host %s: found 'use_rocm_dmabuf' in output but exit_code=%s; treating as supported.",
host,
exit_code,
)
log.info(
"Binary %s on host %s supports --use_rocm_dmabuf",
binary_path,
host,
)
dmabuf_supported_any = True
else:
log.info(
"Binary %s on host %s does NOT support --use_rocm_dmabuf (no match in output).",
binary_path,
host,
)

return dmabuf_supported_any

except Exception as e: # pylint: disable=broad-except
log.warning(
"Failed to check dmabuf support for %s: %s. Assuming no dmabuf support.",
binary_path,
str(e),
)
return False


def get_ib_bw_pps(phdl, msg_size, cmd):
res_dict = {}

Expand Down Expand Up @@ -201,6 +273,7 @@ def run_ib_perf_bw_test(
log.info(f'Setting LD_LIBRARY_PATH to {rocm_path}/lib for perftest binaries')
phdl.exec(f'echo "export LD_LIBRARY_PATH={rocm_path}/lib:$LD_LIBRARY_PATH" >> /tmp/ib_cmds_file.txt')
server_addr = None
dmabuf_supported = check_perftest_dmabuf_support(phdl, f'{app_path}/{bw_test}')
for node in bck_nic_dict.keys():
result_dict[node] = {}
cmd_dict[node] = []
Expand All @@ -215,7 +288,15 @@ def run_ib_perf_bw_test(
for gpu_no in range(0, 8):
card_no = 'card' + str(gpu_no)
rdma_dev = gpu_nic_dict[node][card_no]['rdma_dev']
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} -d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} -s {msg_size} -q {qp_count} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
if dmabuf_supported:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} \
-d {rdma_dev} --use_rocm={gpu_no} --use_rocm_dmabuf -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} \
-s {msg_size} -q {qp_count} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
else:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} \
-d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} \
-s {msg_size} -q {qp_count} > /tmp/ib_perf_{inst_count}_logs & 2>&1'

cmd_dict[node].append(f'echo "{cmd}" >> /tmp/ib_cmds_file.txt')
inst_count = inst_count + 1
port_no = port_no + 1
Expand All @@ -228,7 +309,14 @@ def run_ib_perf_bw_test(
for gpu_no in range(0, 8):
card_no = 'card' + str(gpu_no)
rdma_dev = gpu_nic_dict[node][card_no]['rdma_dev']
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} -d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} -s {msg_size} -q {qp_count} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
if dmabuf_supported:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} \
-d {rdma_dev} --use_rocm={gpu_no} --use_rocm_dmabuf -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} \
-s {msg_size} -q {qp_count} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
else:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{bw_test} \
-d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -b -F -D {duration} -p {port_no} \
-s {msg_size} -q {qp_count} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
cmd_dict[node].append(f'echo "{cmd}" >> /tmp/ib_cmds_file.txt')
inst_count = inst_count + 1
port_no = port_no + 1
Expand Down Expand Up @@ -282,6 +370,7 @@ def run_ib_perf_lat_test(
log.info(f'Setting LD_LIBRARY_PATH to {rocm_path}/lib for perftest binaries')
phdl.exec(f'echo "export LD_LIBRARY_PATH={rocm_path}/lib:$LD_LIBRARY_PATH" >> /tmp/ib_cmds_file.txt')
server_addr = None
dmabuf_supported = check_perftest_dmabuf_support(phdl, f'{app_path}/{lat_test}')
for node in bck_nic_dict.keys():
result_dict[node] = {}
cmd_dict[node] = []
Expand All @@ -296,7 +385,15 @@ def run_ib_perf_lat_test(
for gpu_no in range(0, 8):
card_no = 'card' + str(gpu_no)
rdma_dev = gpu_nic_dict[node][card_no]['rdma_dev']
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} -d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -F -p {port_no} -s {msg_size} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
if dmabuf_supported:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} \
-d {rdma_dev} --use_rocm={gpu_no} --use_rocm_dmabuf -x {gid_index} --report_gbits -F -p {port_no} \
-s {msg_size} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
else:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} \
-d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -F -p {port_no} \
-s {msg_size} > /tmp/ib_perf_{inst_count}_logs & 2>&1'

cmd_dict[node].append(f'echo "{cmd}" >> /tmp/ib_cmds_file.txt')
inst_count = inst_count + 1
port_no = port_no + 1
Expand All @@ -309,7 +406,14 @@ def run_ib_perf_lat_test(
for gpu_no in range(0, 8):
card_no = 'card' + str(gpu_no)
rdma_dev = gpu_nic_dict[node][card_no]['rdma_dev']
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} -d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -F -p {port_no} -s {msg_size} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
if dmabuf_supported:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} \
-d {rdma_dev} --use_rocm={gpu_no} --use_rocm_dmabuf -x {gid_index} --report_gbits -F -p {port_no} \
-s {msg_size} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
else:
cmd = f'numactl --physcpubind={gpu_numa_dict[node][card_no]["local_cpulist"]} --localalloc {app_path}/{lat_test} \
-d {rdma_dev} --use_rocm={gpu_no} -x {gid_index} --report_gbits -F -p {port_no} \
-s {msg_size} {server_addr} > /tmp/ib_perf_{inst_count}_logs & 2>&1'
cmd_dict[node].append(f'echo "{cmd}" >> /tmp/ib_cmds_file.txt')
inst_count = inst_count + 1
port_no = port_no + 1
Expand Down
Loading