This is a CPU implementation of miniBUDE using OpenMP 3.
Select the compiler you want to use and pass it to Make:
make COMPILER=GNU
The supported compilers names are: ARM, CLANG, CRAY, GNU, INTEL.
By default, the native architecture is targetted, but this can be changed by setting the ARCH parameter to the name of the target processor.
For example, the following are both valid:
make COMPILER=GNU ARCH=thunderx2t99
make COMPILER=GNU ARCH=skylake-avx512
### Block Sizes
This implementation includes a tunable block size similar to OpenCL workgroups.
The default value is 64, which is suitable for 512-bit vectors.
Intel platforms benefit from more unrolling, so the default on x86 is 256.
For 128-bit vectors, e.g. in ThunderX2, 16 is a good choice.
This parameter can be set using the WGSIZE parameter, as follows:
make COMPILER=GNU ARCH=thunderx2t99 WGSIZE=16
For AVX-512 targets, the 512-bit registers (zmm) are used by default, because this increases performance.
To disable this and fall back to the compiler's default, which is 256-bit vectors as of Cascade Lake, set AVX512 to the empty string:
make COMPILER=GNU ARCH=skylake-avx512 AVX512=''
This implementation has no special run-time options.
The -n and -i parameters are available, and the number of threads should be set through the OMP_NUM_THREADS environment variable.
Run bude -h for a help message.