UT CHPC--10,000-Digit Pi Computation
This C code exercised the Cray's high-speed vector pipeline architecture
to compute 10,000 digits of pi. The code took about 3 minutes to
execute on the Cray X/MP supercomputer. The first page of code is
shown below.
previous
next up
The chicken scratches below diagram arithmetic stages of the vector
pipelines in several routines. These need to be analyzed to make
efficient use of the Cray's architecture. The multiprecision division
routine contains an unfortunate scalar bottleneck; a Fourier transform
is perhaps a better approach there.
Here are the first 2500 digits of output.
previous
next up
