As was discussed [here](https://github.com/intel/intel-xpu-backend-for-triton/pull/1860#discussion_r1720079513), there is a proposal to implement true fp32 matmul, however it won't be using DPAS.
