Inastemp: A Novel Intrinsics-as-Template Library for Portable SIMD-Vectorization

Figure 4

Gigaflop per second to compute a general square matrix-matrix product, where the average was taken from three executions. Matrix dimension is in Double and in Float. () These executions use a simpler blocking scheme that shows better performance for the respective configurations (Xl-P8-OP Figure 4(f)).

(a) Gcc-I3-PC

(b) Clang-I3-PC

(c) Gcc-IX-HPC

(d) Intel-IX-HPC

(e) Gcc-P8-OP

(f) Xl-P8-OP

(g) Gcc-KNL

(h) Intel-KNL