| --- Log opened Fri Mar 24 00:00:57 2017 | ||
| bandvig | promach: Ah! You are right. I deleted it from repo because it hasn’t been used anymore. | 02:50 |
|---|---|---|
| bandvig | promach: I’m not access to my home PC with it, so I’ll be able to add it somewhere in my personal repository on GitHub in about 10-12 hours. | 02:50 |
| bandvig | promach: However, if I understand correctly the FPU variant is not mandatory for your needs. So you can just comment it for a while. | 02:50 |
| psychotrope | how does openRISC generate random numbers for /dev/random? | 04:45 |
| psychotrope | prng? | 04:45 |
| psychotrope | or are there any hardware sources | 04:46 |
| promach | bandvig: strange, https://www.edaplayground.com/x/59Rs only addition works, multiply and division do not give correct result | 05:59 |
| bandvig | promach: In your example you are doing (32’h2 + 32’h3) as line #136 is uncomment . I suppose expected value is 32’h5. To do division, line #136 should be connected and line #83 should be uncommented. | 06:53 |
| bandvig | promach: By the way which result you expect after (32’h2 / 32’h3) taking into account that both of them are denormalized? | 06:53 |
| promach | denormalized by what factor ? | 06:55 |
| promach | bandvig: you were saying line#136 should be commented, right ? | 06:56 |
| promach | I am getting http://i.imgur.com/Lx5cdKQ.png for multiply | 06:58 |
| promach | sorry, typo | 06:59 |
| promach | should be division | 06:59 |
| bandvig | promach: any line with “fpu_op = “ is command for FPU. | 07:01 |
| promach | yup | 07:01 |
| promach | I am getting http://i.imgur.com/T07jAOs.png for multiply | 07:04 |
| bandvig | promach: “denormalized” means “exponent” is zero, “mantissa” is not zero. | 07:05 |
| promach | I am still not sure why 0 for pfpu32_result | 07:05 |
| bandvig | promach: It looks you don’t understand FP-standard. | 07:13 |
| bandvig | promach: Actually you multiply (6*2^(-126) x 3*2^(-126)), intermediate value is (18 * 2 ^ (-252)) could not be packed in single precision FP. | 07:14 |
| bandvig | promach: So you receive zero and flags “inexact” and “underflow” | 07:14 |
| promach | bandvig: Sorry, let me study further. | 07:15 |
| -!- shorne_ is now known as shorne | 10:10 | |
| shorne | stekern: hows it going? I boiled down the smp patcches to this | 10:12 |
| shorne | https://github.com/stffrdhrn/linux/commits/openrisc-4.12-smp-qspinlock | 10:12 |
| shorne | https://github.com/stffrdhrn/linux/commits/openrisc-4.12-smp (before sqaushing & applying the qspinlocks suggested by peter) | 10:12 |
| shorne | any thoughts? I still need to fix a few things like | 10:13 |
| shorne | 1. commit descriptions | 10:13 |
| shorne | 2. you have a comment about hacks in the timer sync | 10:13 |
| shorne | 3. possible IPI handling of cpu stop | 10:13 |
| shorne | 4. make it boot (currently it crashes in or1ksim, and doesnt even start on qemu) | 10:14 |
| * shorne heads off | 10:15 | |
| stekern | shorne: I'll take a look, but just a heads up, I've only tested the SMP code on real targets | 11:31 |
| shorne | stekern: I think its failing at this timer_init part (the number 2 I listed) | 18:38 |
| shorne | tc_base = ioremap_nocache(0x99000000, 4); | 18:38 |
| shorne | what is supposed to be at 0x99000000? The sim doesnt have anything, but I didnt see much in hardware... double checking | 18:39 |
| shorne | [slave tc] | 18:40 |
| shorne | offset=0x99000000 | 18:40 |
| shorne | size=4 | 18:40 |
| shorne | I see... | 18:40 |
| shorne | https://github.com/skristiansson/orpsoc-cores/blob/multicore/cores/timer/rtl/verilog/timer.v | 18:45 |
| shorne | I guess thats a bit of a hack, I dont see why we cant just move that global timer out to dtb and create a driver | 18:46 |
| mithro | So - anyone know where the instructions for building gcc at https://openrisc.io/newlib when away? | 21:24 |
| --- Log closed Sat Mar 25 00:00:58 2017 | ||
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!
