This architecture is tricky as the instruction window has to have an optimal mix of scalar and vector instructions to take advantage of this core. Else, the vector SIMD lanes might be starved leading to sub optimal speedups.
Please log in to leave a comment.
This architecture is tricky as the instruction window has to have an optimal mix of scalar and vector instructions to take advantage of this core. Else, the vector SIMD lanes might be starved leading to sub optimal speedups.