Thursday, 2021-10-07

lkclGuest55_, hii again mepy :) thanks for sending that message to the list the other day.13:44
lkcltoshywoshy, thank you for yesterday. i need to do a write-up for you about the test API.  kylel helped write it, and will be documenting it.13:45
lkclone "ooooOoo" moment i didn't raise yesterday: the python-based unit tests could actually be *used* to generate c programs!13:45
lkclconforming to the OpenPOWER Validation Test Suite13:45
lkclso whilst there's one option to run them directly (by gdbmi "remote" gdb single-stepping, which will be horribly slow)13:46
lkclthe other option is to use some sort of mixed c/assembler templating where the exact same unit test written into the openpower-isa test suites in python will *auto-generate* an *actual* c/assembler source code program13:47
lkclwhere that program, when compiled and run, actually does the exact same job: runs the instruction and tests against the expected results13:48
lkclkylel, if you'd be interested to do that, the idea was floated during the meeting to put in *another* NLnet Grant request, this time perhaps with some of the other OpenPOWER Foundation Members joining in13:49
lkcl(IBM, other)13:49
kylelThat is an interesting idea14:38
kylela heckuva lot more robust14:48
kylelWould be happy to assist15:05
Guest55_ lkcl thank you ! I quite always read – and try to understand – your emails!15:45
lkclkylel, yeah it's just "yet another of the possibilities"16:02
lkcland makes the Test API a potential candidate for at the very least a de-facto standard "you're on your own, it does the job, but it's your responsibility to check" OpenPOWER Compliance Test Suite16:03
lkcli can't quite believe how little public cooperation there is right now on this.  from 20+ years of Libre/Open development it's really quite taking me aback that every single OpenPOWER Foundation Member Company basically duplicates and maintains their own personal Compliance Test Suite.16:04
kylelit is interesting because you start to wonder sometimes just what is "correct" sometimes with so many moving parts and so many different projects/organizations sort of doing their own thing.16:25
programmerjakelkcl, note that generating assembler to test native power instructions is exactly what I'm planning on doing for the python rewrite of power-instruction-analyzer19:38
programmerjakeso, that overlaps quite a bit19:38
lkclprogrammerjake, ah not quite. that's a dynamic task, that actively runs *from* python to execute assembler (after first compiling it to a binary)20:23
lkclthe above idea is a *static* task20:23
lkclzero actual execution (well, maybe)20:23
lkclwhere it pretty much literally takes the unit test case assembly listings, drops them into a .s file (or potentially an inline asm block inside a c source file)20:24
lkclblats out a standard template Makefile20:24
lkclblats out a c template which outputs the pre-run reg inputs, runs the .s file and picks up the regfile contents afterwards20:25
lkcland errr that's it20:25
lkclit becomes the user's responsibility to type "make"20:25
lkclat which point they end up with a binary executable that *they* run20:26
lkcland it reports "OK" or "fail" as appropriate20:26
lkclexactly like riscv-isa-tests does right now20:26
lkclmore sophisticated versions of that would output *different* templates20:26
lkcl(exactly like riscv-isa-tests does right now)20:27
lkclone of those templates would be "physical bare metal machine"20:27
lkclanother would be "Virtual Memory is initialised before running the test"20:27
lkcland so on and so forth20:27
lkclall of which is completely different from *dynamically* - and from *python* - actually *executing* compiled assembler20:28
programmerjakeah, ok. so they could possibly share code20:56
Las[m]_Is there any plan to migrate away from alliance for the soclayout stuff?21:35
*** Las[m]_ is now known as Las[m]__21:39
*** Las[m]__ is now known as Las[m]_21:39
*** Las[m]_ is now known as Las[m]21:39

Generated by 2.17.1 by Marius Gedminas - find it at!