On 7/14/26 9:34 AM, Kishan Parmar wrote: > A future PowerPC processor may provide the Dense Math Facility (DMF), > where accumulators are backed by dedicated Dense Math Registers (DMRs) > instead of aliased VSX/FPR registers. Since DMRs do not require > priming or depriming, xxmtacc/xxmfacc become no-op operations.
I don't like saying xxmtacc/xxmfacc are nops on DMR enabled cpus, since they are most certainly NOT nops. They are 100% required when executing MMA (-mmma -mno-dense-math) binaries on DMF enabled cores. I think what you meant to say is that you don't use xxmtacc/xxmfacc to initialize the accumulators/dmr registers when compiling with -mdense-math since DMF has new dmr insert and extract insns to initialize the accumulators /dmr regs. Ie, they are (better) replacements for xxmtacc/xxmfacc. Maybe cut your git log entry to something like the following? A future PowerPC processor may provide the Dense Math Facility (DMF), where accumulators are backed by dedicated Dense Math Registers (DMRs). Update accumulator assembly to build accumulators in DMRs using the new DMR insert and extract operations rather than using xxmtacc/xxmfacc. Enable the required OOmode VSX pair patterns under TARGET_DMF and avoid emitting xxmtacc/xxmfacc during internal accumulator moves and GIMPLE folding when DMF is enabled. Peter
