Add matrix cores test to perf suite

[CRobeck]

AMD gfx90a architecture has built in hardware support (matrix cores) for dense matrix operations. The two that are of interest to RAJA are: __builtin_amdgcn_mfma_f64_4x4x4f64 __builtin_amdgcn_mfma_f64_16x16x4f64

Done criteria:

• Add a test to the perf suite that makes use of each of these intrinsics
• Confirm performance compared to hand coded matrix matrix multiply