Richard Henderson wrote: > In the same vein as your CAS boolean output patch, if we rearrange the > copies here we can get the combined compare-and-branch insn for the z10. > I see that the z196 prefers not to use those, but the number of insns > in that case remains the same, merely in a different order.
Yes, on z196 the compare-and-branch insns are discouraged since they're "cracked" (i.e. split into two micro-ops, just the same as if we'd have two original insns, but in effect a bit worse for the pipeline). In any case, the patch certainly makes sense. > Can you please test with --with-arch=z10? Tested with no regressions. > * config/s390/s390.c (s390_expand_cs_hqi): Copy val to a temp before > performing the compare for the restart loop. OK. Thanks, Ulrich -- Dr. Ulrich Weigand GNU Toolchain for Linux on System z and Cell BE ulrich.weig...@de.ibm.com
