mfkiwl / btcc Goto Github PK

btcc

A Bifrost wrapper for the Tensor-Core Correlator.

Compiling BTCC

To build BTCC:

1. Install appropriate dependancies
   1. Bifrost (https://github.com/telegraphic/bifrost)
   2. Tensor-Core Correlator (https://git.astron.nl/RD/tensor-core-correlator)
   3. Meson build system (https://github.com/mesonbuild/meson)
2. Setup your build environment (on topaz, run source setup_env.sh).
3. Ensure correct paths to dependancies are setup in meson.build
4. Compile with meson by running:

meson setup build
cd build
meson compile

There are several conditional compliation constants included in btcc.cu:

• DCP_DEBUG: enables full debugging output
• NO_CHECKS: removes all sanity checks
• TIME_CORR: write all correlation times (ms) to cuda_results.csv

Using BTCC in Python

from btcc import Btcc

tcc = Btcc()
tcc.init(nbits, ntime, nchan, nant, npol)
tcc.execute(tcc_input, tcc_output, True)

See validate.py, benchmark.py and eda.py for example usages of BTCC.

Example Performance (Tesla V100)	Example correlation matrix	Example image