| 03:38.08 | *** join/#debianppc peterrooney (~peter@S0106001ff341c9a8.gv.shawcable.net) |
| 09:33.39 | *** join/#debianppc thierry-ibm (tfauck@nat/ibm/x-hztahchcdwhjwasm) |
| 10:55.12 | *** join/#debianppc mauricfo (~mauricfo@177.44.49.168) |
| 11:23.24 | *** join/#debianppc pfsmorigo (~pfsmorigo@177.91.79.176) |
| 12:36.31 | *** join/#debianppc klebers (Kleber@nat/ibm/x-kjenakqdcbrhtpzg) |
| 12:56.49 | *** join/#debianppc mauricfo (~mauricfo@177.44.49.168) |
| 13:19.00 | *** join/#debianppc smow (~pfsmorigo@177.115.11.176) |
| 13:32.59 | *** join/#debianppc tdaitx__ (~tdaitx@179.181.68.113) |
| 14:58.11 | *** join/#debianppc jelly (~jelly@pdpc/supporter/active/jelly) |
| 14:58.59 | *** join/#debianppc synmuffin (~yah@S01069094e4c9bb67.ls.shawcable.net) |
| 15:00.17 | *** join/#debianppc camm` (~user@nat-tvwna-outside-visitornet2-c-170.princeton.org) |
| 15:03.19 | camm` | bergner: I'm double checking a linker trampoline to avoid posting an unnecessary bug. Can you point me to an instruction reference/manual? I've just found summaries or 32bit docs online, and am working from reverse engineering the gcc produced assembly. All I need is the docs on the branch instructions, link registers, conditional registers, etc. |
| 15:06.30 | bergner | camm`: you mean a PPC/PPC64 ISA doc? Sure, let me find a URL... Do you need the latest (ie, POWER8) or will any do? |
| 15:12.25 | camm` | Probably anything reasonably recent will do, but I'm after the ppc64le Debian target, the new Elfv2 abi, though that's likely irrelevant |
| 15:26.40 | bergner | camm`: and here is a URL to a freely downloadbale ISA 2.07 doc: http://ibm.biz/powerisa |
| 15:26.57 | bergner | let me know if you need the ELFv2 ABI doc and I'll try and find a URL for that too |
| 16:36.19 | *** part/#debianppc jelly (~jelly@pdpc/supporter/active/jelly) |
| 17:21.40 | *** join/#debianppc pankid (~pan@76-243-106-69.lightspeed.livnmi.sbcglobal.net) |
| 17:37.05 | *** join/#debianppc pfsmorigo (~pfsmorigo@177.91.79.176) |
| 18:58.21 | camm` | If anyone has a thought on just filed https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63360, (f31 not restored across function call at -O0), I'd be most interested. |
| 19:20.06 | bergner | camm`: ok, I'll have a look |
| 19:21.48 | bergner | this is a powerpc64le-linux compile, correct? |
| 19:29.11 | camm` | yes, pastel.debian.net |
| 19:30.09 | camm` | (and thanks!) |
| 19:35.18 | *** join/#debianppc mauricfo (~mauricfo@177.44.49.168) |
| 20:00.12 | bergner | camm`: looking at your gdb disassembly output, I see: |
| 20:00.13 | bergner | ... |
| 20:00.15 | bergner | <PROTECTED> |
| 20:00.16 | bergner | <PROTECTED> |
| 20:00.16 | bergner | <PROTECTED> |
| 20:00.29 | bergner | that looks like a restore of f31 to me |
| 20:00.42 | bergner | :1 |
| 20:01.03 | bergner | exposes he vim usage tp the world :-) |
| 20:02.17 | camm` | yes, at the end of the function L2, but not after each call through a pointer in the body. |
| 20:02.36 | camm` | (see gdb session) |
| 20:03.05 | camm` | (back in a bit...) |
| 20:08.39 | bergner | f31 is a non-volatile register, so function L@ doesn't have to worry about saving/restoring it across calls L2 makes, because those callees will save/restore it. L2 only has to worry about saving/restoring f31 for uses of f31 that it makes...meaning, saving f31 in the prologue and restoring it in the epilogue is all it needs to do. ..and it is doing that. So I don't see any bug here. Now if the functions we're calling throug |
| 20:08.39 | bergner | h the bctrl clobber f31 without restoring it, then that would be a bug. |
| 20:10.07 | bergner | s/L@/L2/ |
| 20:23.37 | bergner | It seems you're saying that one of the functions L2 calls is clobbering f31 without restoring it. Can you set a break point at each call and find out which function is clobbering f31 and then attach the disassembly of that function? |
| 20:26.44 | bergner | camm`: ^^^ |
| 20:28.01 | bergner | compiling the preprossed source you attached with my gcc 4.8 LE compiler, all of the functions that use f31 seem to restore it |
| 21:38.54 | *** join/#debianppc peterrooney (~peter@S0106001ff341c9a8.gv.shawcable.net) |
| 22:09.50 | camm` | bergner: thanks so much for the clear explanation! Tomorrow I will resume and find out which callee does the clobber. It is at least a child pointed to by Lnk23(). |
| 23:22.38 | *** join/#debianppc tdaitx_ (~tdaitx@189.27.173.118.dynamic.adsl.gvt.net.br) |