NVIDIA/nvbench
1
0
Fork 0
mirror of https://github.com/NVIDIA/nvbench.git synced 2026-04-19 22:38:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Change test and examples from using camelCase to using snake_case as implementation changed

Browse Source
This commit is contained in:
Oleksandr Pavlyk
2025-07-21 12:50:28 -05:00
parent 81fff085b9
commit e589518376
9 changed files with 101 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
python/examples/throughput.py
View File

@@ -41,33 +41,27 @@ def make_kernel(items_per_thread: int) -> Callable:
def throughput_bench(state: nvbench.State) -> None:
stride = state.getInt64("Stride")
ipt = state.getInt64("ItemsPerThread")
stride = state.get_int64("Stride")
ipt = state.get_int64("ItemsPerThread")
nbytes = 128 * 1024 * 1024
elements = nbytes // np.dtype(np.int32).itemsize
alloc_stream = as_cuda_Stream(state.getStream())
alloc_stream = as_cuda_Stream(state.get_stream())
inp_arr = cuda.device_array(elements, dtype=np.int32, stream=alloc_stream)
out_arr = cuda.device_array(elements * ipt, dtype=np.int32, stream=alloc_stream)
state.addElementCount(elements, column_name="Elements")
state.addGlobalMemoryReads(inp_arr.nbytes, column_name="Datasize")
state.addGlobalMemoryWrites(inp_arr.nbytes)
state.add_element_count(elements, column_name="Elements")
state.add_global_memory_reads(inp_arr.nbytes, column_name="Datasize")
state.add_global_memory_writes(inp_arr.nbytes)
threads_per_block = 256
blocks_in_grid = (elements + threads_per_block - 1) // threads_per_block
krn = make_kernel(ipt)
# warm-up call ensures that kernel is loaded into context
# before blocking kernel is launched
krn[blocks_in_grid, threads_per_block, alloc_stream, 0](
stride, elements, inp_arr, out_arr
)
def launcher(launch: nvbench.Launch):
exec_stream = as_cuda_Stream(launch.getStream())
exec_stream = as_cuda_Stream(launch.get_stream())
krn[blocks_in_grid, threads_per_block, exec_stream, 0](
stride, elements, inp_arr, out_arr
)
@@ -77,7 +71,7 @@ def throughput_bench(state: nvbench.State) -> None:
if __name__ == "__main__":
b = nvbench.register(throughput_bench)
b.addInt64Axis("Stride", [1, 2, 4])
b.addInt64Axis("ItemsPerThread", [1, 2, 3, 4])
b.add_int64_axis("Stride", [1, 2, 4])
b.add_int64_axis("ItemsPerThread", [1, 2, 3, 4])
nvbench.run_all_benchmarks(sys.argv)