See related issues here, including info about my hardware.

See related issues here, including info about my hardware.

Cross-post from Stack Overflow since this is very code intensive

Problem

gcc -std=c99 -O3 -msse3 -ffast-math -march=nocona -mtune=nocona -funroll-loops -fomit-frame-pointer -masm=intel

Hopefully is understandable. Please ask if not. I set up the macro structure (for loops) as described in the 2nd paper above. I pack the matrices as discussed in either paper. My inner kernel computes 2x8 blocks, as this seems to be the optimal computation for Nehalem architecture (see GotoBLAS source code - kernels). The inner kernel is based on the concept of calculating rank-1 updates as described here

Problem

gcc -std=c99 -O3 -msse3 -ffast-math -march=nocona -mtune=nocona -funroll-loops -fomit-frame-pointer -masm=intel

I set up the macro structure (for loops) as described in the 2nd paper above. I pack the matrices as discussed in either paper. My inner kernel computes 2x8 blocks, as this seems to be the optimal computation for Nehalem architecture (see GotoBLAS source code - kernels). The inner kernel is based on the concept of calculating rank-1 updates as described here.