On Sat, Jul 29, 2017 at 10:39 AM, Luc Van Oostenryck <luc.vanoostenryck@xxxxxxxxx> wrote: > No it's fine. It's the way of displaying that the element have been removed. > >> it should be: >> >> phi.32 %r1(ansi) <- 1, %phi4(ansi) > > No, not really. > L1 has now only a single parent, the phi-node is now a trivial one, > not really a join anymore. > An explicit way of displaying it should simply be: > phi.32 %r1(ansi) <- %phi4(ansi) OK, Great. It seems my decode attempt did not go far at all :( Nice try. . > > Everything is correct until: > test_menu_iteminfo: > .L0: > <entry-point> > br .L4 > .L1: > phi.32 %r1(ansi) <- VOID, %phi4(ansi) > cbr %r1(ansi), .L4, .L5 > .L4: > cast.64 %r3 <- (64) stringA > br .L5 > .L5: > cbr %r1(ansi), .L6, .L3 > .L6: > ptrcast.64 %r6 <- (64) %r3 > ptrcast.64 %r8 <- (64) VOID <== wrong but unrelated > call.64 %r9 <- strcpy, %r6, %r8 > br .L2 > .L2: > seteq.32 %r11 <- %r1(ansi), $0 > phisrc.32 %phi4(ansi) <- %r11 > cbr %r1(ansi), .L1, .L3 > .L3: > ret > > Then simplify_branch_branch() is called on L2 -> L1 > and wrongly succeed giving: My original thinking was .L1: Ophi.32 %r1(ansi) <- VOID, %phi4(ansi) feeding the .L2: cbr %r1(ansi), .L1, .L3 Got the wrong result so I suspect some thing was wrong with %r1(ansi). I admit I don't know what exactly simplify_branch_branch() was doing there. .L5: also have cbr %r1(ansi), .L6, .L3 I don't know why this one is fine. > seteq.32 %r11 <- %r1(ansi), $0 > phisrc.32 %phi4(ansi) <- %r11 > cbr %r1(ansi), .L4, .L3 <== bad change > .L3: > ret > > This will then trigger the wrong simplifications: > %phi4 <- %r11 > and > %r1 <- %r11 > giving finally: > setne.32 %r11 <- %r11, $0 > > As far as I can see, simplify_branch_branch() needs something > like the bb_defines_phi() I had to add to try_to_simplify_bb() > (basically it's bb_depends_on() which is not 'strong' enough). Thanks you so much for the excellent diagnose. > At this point I don't think anymore that the revert is the most > appropriate thing to do. That is what I am looking for. I want to find out if there is any thing else like this on -rc4+. Thanks again. Chris -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
