03:40imanho: is JMX R2 0x30, jump to R2+0x30 ? or jump to {R2,R3}+0x30?
03:46imirkin: just R2
03:47imirkin: i guess i'm not _100%_ sure, but like ... 99.9999784%?
03:56imanho: hmm, does this mean anything? : /*0390*/ LDL.64 R2, [R19]; /* 0xef45000000071302 */
03:56imanho: /*0398*/ PRET 0x3b0; /* 0xe270000001000040 */
03:56imanho: /* 0x003fc400164107fd */
03:56imanho: /*03a8*/ BRX R2 -0x3b0;
03:56imanho: the LDL.64 R2, [R19] makes me wonder..
03:57imirkin: hrmph
03:57imirkin: i guess i dunno. i'm also weak on all the nvidia semantic names for the branch things
03:57imirkin: like which one is JMX? who knows
03:58imirkin: PRET is probably "preret" in envyas terminology?
03:58imirkin: still ... weird that the PRET would be forward from the current position
03:59imirkin: otoh i might not remmber how it works
04:08imanho: btw good point, what the hell is a preret?
04:08imirkin: it's a thing which makes 'RET' work :)
04:08imirkin: i think preret _is_ going to jump forward though
04:09imirkin: it's just a thing to indicate which address RET should jump to
04:09imanho: oh.. it's a way to keep track of where to come back to?
04:09imirkin: basically yeah
04:10imanho: I tried a JMX, hooked a debugger into it.. and BAM: "Error: Failed to read the syscall call depth (dev=0, sm=0, wp=0, ln=0), error=CUDBG_ERROR_UNKNOWN(0x1)."
04:12imanho: with BRX though we get another error: "Error: Failed to get warp state (dev=0, sm=0, wp=0), error=CUDBG_ERROR_UNKNOWN_FUNCTION(0x3)."
04:13imanho: I wish, like _normal_ gdb it would at least tell me _what_ instruction it tried to jump to.