In our many-core co-processor, there are 64 registers from R0 to R63. Since we do not have much space in the FPGA, we need to use the registers in an efficient way to run more threads. If it is possible to specify the number of registers that will be used by llc, it makes our device more useful.
Does LLVM have the aforementioned feature? If it doesn’t, I wonder if it is likely or not to implement this feature. What do you think?