size in bytes, returns GB/s

Synchronize all streams on all GPUs

Create all needed streams one each GPU

* In this example, we: * 1) construct an nvcompManager * 2) compress the input data * 3) decompress the input data */

* In this example, we: * 1) construct an nvcompManager with checksum support enabled * 2) compress the input data * 3) decompress the input dat

* In this example, we: * 1) compress the input data * 2) construct a new manager using the input data for demonstration purposes * 3) decompress

generate uniformly random data random bytes random chunk sizes returns max chunk size

returns nano-seconds

* Additionally, we can use the same manager to execute multiple streamed compressions / decompressions * In this example we configure the multiple de

* Additionally, we can use the same manager to execute multiple streamed compressions / decompressions * In this example we configure the multiple de

Benchmark the performance of the All-gather operation using LZ4 compression/decompression to reduce data transfers

size in bytes, returns GB/s