Merge wernsaar's Haswell gemm kernels.

1. +5404

   -0
     kernel/x86/dgemm_kernel_16x2_haswell.S
2. +3167

   -0
     kernel/x86/sgemm_kernel_16x4_haswell.S
3. +2318

   -0
     kernel/x86_64/cgemm_kernel_8x2_haswell.S
4. +5386

   -0
     kernel/x86_64/dgemm_kernel_16x2_haswell.S
5. +3151

   -0
     kernel/x86_64/sgemm_kernel_16x4_haswell.S
6. +1848

   -0
     kernel/x86_64/zgemm_kernel_4x2_haswell.S