[Hardware] Replace torch.cuda.device_count/current_device/set_device API (#36145)

Signed-off-by: Kunshang Ji <jikunshang95@gmail.com>
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
This commit is contained in:
Kunshang Ji
2026-03-12 07:57:47 -07:00
committed by GitHub
parent 2e693f48e7
commit 53ec16a705
89 changed files with 254 additions and 219 deletions
@@ -64,7 +64,7 @@ def bench_run(
per_out_ch: bool,
mkn: tuple[int, int, int],
):
init_workspace_manager(torch.cuda.current_device())
init_workspace_manager(torch.accelerator.current_device_index())
(m, k, n) = mkn
dtype = torch.half
@@ -495,7 +495,7 @@ def main():
# Set device
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
torch.accelerator.set_device_index(device)
# Get CPU process group
cpu_group = dist.new_group(backend="gloo")
@@ -392,7 +392,7 @@ def benchmark_operation(
num_op_per_cudagraph = 10
# Use vLLM's graph_capture to make tensor_model_parallel_all_reduce graph-safe
device = torch.device(f"cuda:{torch.cuda.current_device()}")
device = torch.device(f"cuda:{torch.accelerator.current_device_index()}")
with graph_capture(device=device), torch.cuda.graph(graph):
for _ in range(num_op_per_cudagraph):
operation_func(*args, **kwargs)
@@ -984,7 +984,7 @@ def main():
world_size = int(os.environ["WORLD_SIZE"])
device = torch.device(f"cuda:{rank}")
torch.cuda.set_device(device)
torch.accelerator.set_device_index(device)
torch.set_default_device(device)
init_distributed_environment()
@@ -50,7 +50,7 @@ def bench_run(
per_out_ch: bool,
mkn: tuple[int, int, int],
):
init_workspace_manager(torch.cuda.current_device())
init_workspace_manager(torch.accelerator.current_device_index())
label = "Quant Matmul"
sub_label = (
@@ -285,7 +285,7 @@ def tune_on_gpu(args_dict):
weight_shapes = args_dict["weight_shapes"]
args = args_dict["args"]
torch.cuda.set_device(gpu_id)
torch.accelerator.set_device_index(gpu_id)
print(f"Starting tuning on GPU {gpu_id} with batch sizes {batch_sizes}")
block_n = args.block_n
@@ -334,7 +334,7 @@ def distribute_batch_sizes(batch_sizes, num_gpus):
def main(args):
print(args)
num_gpus = torch.cuda.device_count()
num_gpus = torch.accelerator.device_count()
if num_gpus == 0:
raise RuntimeError("No GPU available for tuning")
print(f"Found {num_gpus} GPUs for parallel tuning")