## Where communities thrive

• Join over 1.5M+ people
• Join over 100K+ communities
• Free without limits
##### Activity
sylefeb
@sylefeb
@lawrie I had made this introduction video that might help clarify: https://www.youtube.com/watch?v=_OhxEY72qxI ; your feedback made me realize it may indeed be confusing for those familiar with Verilog to transition to Silice, will add notes on that in the repo. Basically everything you write is combinatorial (ends up in a Verilog always@* with = assignments) but variables are turned into flip-flops as needed. So if you write a = a + 1 it becomes a_d = a_q + 1 with a_q <= a_d in a posedge block. Code is split into FSM states as required (with predictable rules so you can try to minimize/avoid that).
Silice always blocks are statements applied at every clock, before any of the other states. This is useful for instance to monitor an input pulsing high, while the rest of the logic/FSM is being run. Also, the idea is to help progressively go from a FSM style (with while / ++: / subroutines ) to a more compact 'always' style.

When I tried changing while (1) to always in my led_glow example, I got an error message about in_run

Ah, you've just hit a bug: sylefeb/Silice#113 This would work in any algorithm other than main ... will fix!

sylefeb
@sylefeb
@emard @lawrie US2: thanks, will push a first attempt on 'wip'. inout are still a bit difficult to use in Silice (requires a small Verilog glue), also something to be improved (sylefeb/Silice#19), so I'll probably specialize to PS/2 for now.

Incidentally when I tried your uart echo example on the Ulx3s, it appeared to lose characters.

Oh no, I thought this one was stable! will investigate sylefeb/Silice#117

sylefeb
@sylefeb

What relationship does that have to the assignments of uart_tx within the block?

The always block is specified before any other code. So besides the always block this uart_tx=1 is the only thing performed by the algorithm, and only once (while the always block remains always active). This was used to initialize the output to 1 -- no longer necessary, Silice now supports the syntax output uint1 uart_tx = 1.

sylefeb
@sylefeb

@lawrie pushed the us2_ps2 addition, Makefile change: silice-make.py -s blinky.ice -b $@ -p basic,us2_ps2 ...main becomes: algorithm main( output uint$NUM_LEDS$leds, input uint1 us2_bd_dp, input uint1 us2_bd_dn, usb_fpga_pu_dp/usb_fpga_pu_dn are assigned 1. Hope this works! Rob S @rob-ng15 Hi Sylvain, I've been waiting for you to do that! I have those set in my ulx3s.v for a while during testing :) I'm just trying to wire in the ps2.ice from @lawrie to see how it goes. I'm wiring it to the USB pins. I'll let you know. sylefeb @sylefeb Hi Rob - thanks! looking forward to the PS/2 keyboard working, will be very useful (plus we can put all these old keyboards to good use!). Rob S @rob-ng15 Well the first step is to get it to give the same output 0xaa as I get using the keyboard I have. I've got to get a working PS2 keyboard for the ulx3s. Just catching up with the changes that have been made in Silice since I last updated. Mainly, changes to the bram interface passing at the moment. Just looking at the audo_pwm.ice... THAT I definitely have a use for :) Rob S @rob-ng15 Excellent news! The modified PS2 code from Lawrie that I integrated into PAWS gives the same 0xaa code as his original verilog code, so I take that as a good sign :) Lawrie Griffiths @lawrie @sylefeb us2_ps2 worked for me as well. I can see PS/2 scan codes from a keyboard connected to us2. I checked the uart_echo test on the Wip branch as it might have been the main branch I tried it on before, but I still see the problem. I run stty -F /dev/ttyUSB0 raw -echo 115200; cat /dev/ttyUSB0 on one terminal and echo Hello >/dev/ttyUSB0 on another, and every other character is lost. Lawrie Griffiths @lawrie @sylefeb I am familiar with languages that have different semantics to Verilog. In migen and nmigen you have to be explicit about whether every statement in synchronous or combinatorial (by adding them to the appropriate domain). In SpinalHDL you have to be explicit about which signals are flip-flops (Reg) and it then generates synchronous code for the registers. But Silice has different semantics that doesn't really conform to the combinatorial/synchronous distinction, so I think some of the references to combinatorial logic in your manual are confusing. Rob S @rob-ng15 @lawrie @sylefeb The modified ps2.ice that I integrated into PAWS is here: https://github.com/rob-ng15/Silice-Playground/blob/master/PAWS/common/ps2.ice the main change being to remove from main and input the data and clock lines rather than the whole gpio. PAWS is now caught up with the Silice WIP branch :) Lawrie Griffiths @lawrie @rob-ng15 Yes, those are sensible changes. My version was just to test if it works. That PS/2 algorithm is from @hoglet67 (David Banks) and his Ice40Beeb and Ice40Atom projects. A version that produces an 11-bit ps2_key value in the Mist/Mister format might also be useful. That has extra bits to say whether an event has occurred, whether it is a press or release and whether it is an extended scan code, so requires processing the F0 and E0 values from the keyboard. Rob S @rob-ng15 That's good. I'll put appropriate credits in the file. Lawrie Griffiths @lawrie It gets hard to know how to credit this stuff when it gets modified and rewritten in different languages. emard @emard https://github.com/emard/oberon/blob/master/hdl/PS2.v oberon has also some PS/2 core with only few lines of verilog sylefeb @sylefeb @lawrie thanks for the feedback on the documentation - I'll revise and improve! great news for PS/2 Rob S @rob-ng15 @sylefeb do you have an example of goto in Silice? I may have a use for it... I'll bleach the keyboard if I do! (Sadly, the softfloat c code uses gotos, and it is hard enough converting it as it is, without having to change the control flow logic). sylefeb @sylefeb Rob S @rob-ng15 Cheers, exactly what I am unfortunately looking for! Lawrie Griffiths @lawrie @sylefeb One thing that I notice that you don't have with Silice is reading and writing flash memory. It can be more convenient to load risc-v programs into flash memory rather than having to write raw images to the sd card. Most risc-v implementations include the ability to run code in-place from flash memory (XIP). Another feature which would be useful is a read-write spi memory slave that worked with @emard's micropython code for an on-screen display (OSD) and for remote ram access. That can be very convenient for loading programs from a formatted SD card. It is used by many of the retro computer implementations on the Ulx3s. emard @emard  assign sd_d = R_prog_release[C_prog_release_timeout] ? 4'hz : { 3'b101, S_prog_out[0] }; // does NOT work assign sd_d = R_prog_release[C_prog_release_timeout] == 0 ? { 3'b101, S_prog_out[0] } : 4'hz; // works I have above strange issue with yosys splinedrive @splinedrive @emard and this?  assign sd_d = &R_prog_release[C_prog_release_timeout] ? 4'hz : { 3'b101, S_prog_out[0] }; Radu Stoichita @radu_stoichita_gitlab Good evening guys, is there any trick to speed up the yosys/nextpnr synthesis for the SaxonSoc project ? Everytime making a small change it takes a few minutes to complete. Do you iterate often or is there any performance tweak i am missing somewhere. Thanks and have a great day, Radu Btw, i have notice the whole process uses only one CPU thread :-( splinedrive @splinedrive @radu_stoichita_gitlab Vivado has the feature synthesized Out-Of-Context (OOC) but it seems yosys hasn't such feature implemented. More infos under https://forums.xilinx.com/t5/Adaptable-Advantage-Blog/High-Impedance-and-Out-of-Context-OOC-Synthesis/ba-p/664110 Radu Stoichita @radu_stoichita_gitlab @splinedrive thanks for your feedback, i am experimenting with Lattice and ULX3S therefore i think Vivado is not compatible with Lattice is it? splinedrive @splinedrive You have to check diamond lattice has something similar - it is free. I have no idea, I have only a little bit experience with yosys.http://www.latticesemi.com/en/Products/FPGAandCPLD/ECP5 emard @emard After changing other unrelated lines this yosys bug disappears and now works what before didnt work Radu Stoichita @radu_stoichita_gitlab @splinedrive i come from Diamond and i used it for a few days but it really lacks ooen source tooling for building SOCs. Their IP cores and LatticeMico SOC doesn't fit with the board splinedrive @splinedrive Which ulx3s you have? Radu Stoichita @radu_stoichita_gitlab Euh... The 12F version from Mouser I wanted the 85F but made a mistake when ordering Anyway I love thie board With all the colourful LEDs it's so cute Just expected a little more RAM onboard like 128MB But for my lack of skills it is already a masterpiece emard @emard For practically all projects 12F is enough and it even accepts slightly faster clocks than 85F because of less routing, also projects compiles faster for 12F than 85F RAM had to be 32MB for cost about 3$. There's pin compatible 64MB but for around 12\$
sylefeb
@sylefeb
@lawrie agreed on spi-flash ; I have a bit banging version to boot my risc-v cores (https://github.com/sylefeb/Silice/blob/wip/projects/fire-v/smoke/mylibc/spiflash.c) and it is very convenient indeed. Doing a hardware version is on my TODO list! (also would like to investigate QSPI ...). Was not aware of @emard loader - sounds great - any recommended doc/example?