--- Log opened Sat Mar 25 00:00:58 2017 | ||
shorne | mithro: not sure what happened its repo is here | 00:35 |
---|---|---|
shorne | https://github.com/openrisc/newlib/tree/gh-pages | 00:35 |
shorne | wallento: mentioned the site moved to new hardware, maybe the redirect is broekn? | 00:36 |
mithro | Has anyone done work for gcc or1k 6.3.0 ? | 05:05 |
shorne | mithro: richard henderson sent some patches to the list last month for gcc 7, you might want to check that out | 10:04 |
stekern | shorne: yes, that's what should be done... | 11:05 |
stekern | I just never got to it | 11:05 |
stekern | not a really big deal | 11:05 |
stekern | I can't remember, but I think one problem was that we don't have a way to disable the "builtin" timer | 11:05 |
stekern | so, that would need to be done too | 11:06 |
promach | Manual floating point calculation http://i.imgur.com/ZQUBuiz.png does not seem to match with the multiplication result as in https://www.edaplayground.com/x/59Rs | 11:24 |
promach | Am I missing something ? | 11:24 |
bandvig | promach: what does mean "manual"? | 12:56 |
promach | hand calculation | 12:56 |
promach | I must be making some mistakes somewhere which I still had not noticed. I have to go now. | 12:57 |
bandvig | promach: Oh! I tested FPU with SoftFloat library (could be easily found in Internet). Each time I made a modification I run SoftFloat -based testing tool to check correctness of FPU operation. | 12:58 |
bandvig | promach: In particular the tool performs 46464 tests for each operation mul/add/sub/div for each rounding mode. All of them passes successfully. | 12:59 |
promach | I am just trying to think more mathematically | 12:59 |
bandvig | promach: I would recommend you to use SoftFloat library to generate test vectors. | 12:59 |
bandvig | promach: By the way, all commented examples in my test bench were gathered as error from the SoftFloat -based testing tool. | 13:00 |
promach | I found that this floating point unit testing is not trivial. I need some time to use softfloat. | 13:01 |
promach | in the meanwhile, would you mind confirming my hand calculation, just for this time ? | 13:02 |
bandvig | promach: I'm not sure. First, as you at last found, manipulations with floating point numbers are not trivial. | 13:13 |
bandvig | promach: Second I just don't remember now all peculiarities, especially about rounding, because I made last FPU modification long time ago. | 13:14 |
bandvig | promach: By the way, using SoftFloat library you should remember that there are at least two approaches for rounding. | 13:14 |
bandvig | promach: One "rounding before tininess" and "rounding after tininess". | 13:14 |
bandvig | promach: In my FPU the "rounding before tininess" is implemented. You should tune SoftFloat library appropriately. | 13:15 |
shorne | stekern: got smp patches booting on or1ksim and qemu, (disabilitng SMP), had to add shadow gprs to qemu. So everything seems sane | 19:31 |
shorne | Ill have a go running on de0 nano with SMP | 19:31 |
--- Log closed Sun Mar 26 00:00:00 2017 |
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!