A few days ago Intel compiler expert H.J. Lu landed an FMA-optimized log2 function for the GNU C Library that could yield up to a 69% performance improvement on tested Intel Skylake processors. Merged today to Glibc Git was another FMA-optimized function.
H.J. Lu has continued his recent FMA optimization push for the GNU C Library by optimizing expm1, the function to return a value of an exponential minus 1.
While the expm1 function may not be as common, testing again on Intel Skylake showed some nice improvements. The mean improvement was 24% better than the prior Glibc behavior before introducing this Fused Multiply Add (FMA) optimized code for benefiting modern Intel and AMD CPUs.
This commit lands the code and not a bad improvement for x86_64 CPUs with the next Glibc release. Great as always to Intel and their relentless open-source contributions for further tuning the performance potential of Linux x86_64 systems.
