quad-mul Operation Development

[gkielian]

Curious about the current plans for the quad-mul instruction, namely, if there a good branch where this instruction is poised for development?

https://github.com/riscv/riscv-v-spec/blob/master/quad-mul.adoc

Thanks : )

[freemin7]

This would make risc-v the first ISA with support for quadprecision vectors i am aware of. The NEC Vector Engine supports quad precision numbers only for scalar operations.

As such there might be no existing applications waiting for this capability. What could be interesting is support for accumulation of double precision vectors into quad precision numbers. Implementations of the dual simplex method relies on the 80 bit Intel accumulator for nummerically difficult problems. This could be useful for a variety of applications such as better averaging of numbers and reduce the impact of differences in accumulation order which happen between vector architectures.