On 5/6/19 8:24 AM, Jiufu Guo wrote: > Hi, > > This patch implements the optimization in PR77820. The optimization > eliminates phi and phi's basic block, if the phi is used only by > condition branch, and the phi's incoming value in the result of a > CMP result. > > This optimization eliminates: > 1. saving CMP result: p0 = a CMP b. > 2. additional CMP on branch: if (phi != 0). > 3. converting CMP result if there is phi = (INT_CONV) p0 if there is. > > <P0> > p0 = a CMP b > goto <X>; > > <P1> > p1 = c CMP d > goto <X>; > > <X> > # phi = PHI <p0 (P0), p1 (P1)> > if (phi != 0) goto <Y>; else goto <Z>; > > Transform to: > > <P0> > p0 = a CMP b > if (p0 != 0) goto <Y>; else goto <Z>; > > <P1> > p1 = c CMP d > if (p1 != 0) goto <Y>; else goto <Z>; > > Bootstrapped and tested on powerpc64le with no regressions, and testcases were > saved. Is this ok for trunk? So there's been talk of somehow integrating with jump threading. The big problem with that idea is jump threading is only concerned with rearranging the CFG when doing so allows it to determine the result of a conditional branch.
This is actually a generalized form of path splitting that we currently do for loop latches. You may find you can twiddle that code to do what you want. One of the things that falls out of that realization is that you have to be very careful that this doesn't muck up if-conversion. Jeff
