Teaching and learning resources • Re: Advent of Code 2023

From a software development point of view I wonder how the make -j4 build-world compilation speeds compare. Are the RK3588 and 3588S single-board computers faster on the parallel make?

For large parallel builds, like GCC or Linux kernel builds, make -j8 on the Rockchip boards is certainly much faster than make -j4 on the Pi5. Cmake also does a nice job of multi-threading Pico app builds on all cores. I've never tried -j4 on Rockchip restricting it to the A76 cores. I'll give it a try, but we already know that memory bandwidth is greater on Rockchip.

For a full GCC 14 4 thread compile with this config

Code:

../gcc/configure \    --prefix=/usr \    --program-suffix=-14 \    --enable-languages=c,c++ \    --disable-libquadmath \    --disable-libquadmath-support \    --disable-werror \    --disable-bootstrap \    --disable-multilib \    --enable-gold

I get

Code:

pi@rock5b:/gcc/build$ time taskset -c 4-7 make -j4...real    44m53.155suser    153m53.939ssys     4m14.516s

and

Code:

pi@pi5:/gcc/build $ time make -j4...real    61m49.524suser    225m59.444ssys     3m48.176s

I didn't expect that much difference!

Unrestricted on Rockchip this job takes about 35 minutes.

Statistics: Posted by ejolson — Tue Jan 02, 2024 1:23 am