nextpnr-ecp5 build failed during chipdb generation

[solverstin]

I've attempt to build nextpnr from current github repository, and it fails to build ECP5 target. My setup is Debian-12, yosys installed from debian package and prjtrellis installed from its current github repository. The result is: ~/work/fpga/ecp/nextpnr$ cmake . -DARCH=ecp5 -- Building with IPO -- Found Python3: /usr/include/python3.11 (found suitable version "3.11.2", minimum required is "3.5") found components: Development Development.Module Development.Embed -- Found Boost: /usr/include (found version "1.74.0") found components: filesystem program_options iostreams system thread regex chrono date_time atomic -- Found Boost: /usr/include (found version "1.74.0") found components: program_options filesystem system -- Configuring architecture: ecp5 -- Enabled ECP5 devices: 25k;45k;85k -- Found Python3: /usr/bin/python3 (found suitable version "3.11.2", minimum required is "3.5") found components: Interpreter -- Trellis install prefix: /usr/local -- Trellis library directory: /usr/local/lib/trellis -- Trellis data directory: /usr/local/share/trellis -- Using ECP5 chipdb: ecp5/chipdb -- Configuring done -- Generating done -- Build files have been written to: . ~/work/fpga/ecp/nextpnr$ make [ 1%] Building CXX object bba/CMakeFiles/bbasm.dir/main.cc.o [ 3%] Linking CXX executable bbasm [ 3%] Built target bbasm [ 4%] Generating chipdb/chipdb-25k.bba Traceback (most recent call last): File "ecp5/trellis_import.py", line 709, in <module> main() File "ecp5/trellis_import.py", line 692, in main chip = pytrellis.Chip(dev_names[args.device]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: No such node (row_bias) make[2]: *** [ecp5/CMakeFiles/chipdb-ecp5-bbas.dir/build.make:77: ecp5/chipdb/chipdb-25k.bba] Błąd 1 make[1]: *** [CMakeFiles/Makefile2:255: ecp5/CMakeFiles/chipdb-ecp5-bbas.dir/all] Błąd 2 make: *** [Makefile:136: all] Błąd 2

Other tarets (ice40, generic and himbaechel) build correctly.

There must be something obvious I did wrong, but I cannot figure it out. Any help appreciated.

[mmicko]

Hi. Please do check your local checkout of prjtrellis and make sure that database submodule is also properly updated. My assumption is that issue could be there. Also after any change make sure to go into libtrellis and execute make install frin there.

[solverstin]

OK, better, but failed nonetheless a little bit later.

$ make [ 3%] Built target bbasm [ 4%] Generating chipdb/chipdb-25k.bba Traceback (most recent call last): File "ecp5/trellis_import.py", line 749, in main() File "ecp5/trellis_import.py", line 742, in main process_devices_db(chip.info.family, [ chip.info.name, chip.info.name.replace("LFE5UM","LFE5U") ]) File "ecp5/trellis_import.py", line 260, in process_devices_db for varname, vardata in sorted(devicedb["families"][family]["devices"][dev]["variants"].items()):

KeyError: 'variants'
make[2]: *** [ecp5/CMakeFiles/chipdb-ecp5-bbas.dir/build.make:77: ecp5/chipdb/chipdb-25k.bba] Błąd 1
make[1]: *** [CMakeFiles/Makefile2:255: ecp5/CMakeFiles/chipdb-ecp5-bbas.dir/all] Błąd 2
make: *** [Makefile:136: all] Błąd 2

[mmicko]

Just tested with latest and got now issues. Log points that your database.json is not right. Please go to your local prjtrellis/database and run git log -n 1 from there, it should point to version 4dda149b9e4f1753ebc8b011ece2fe794be1281a if it is not that version make sure to update submodule.

[solverstin]

I have the same version, here:

$ git log -n 1 commit 4dda149b9e4f1753ebc8b011ece2fe794be1281a (HEAD, origin/master, origin/HEAD, master) Author: Miodrag Milanovic mmicko@gmail.com Date: Fri Sep 22 11:20:25 2023 +0200

Fuzzed data for rest of SLICE connections

I also did the following "update" in the database submodule:

$ git pull Nie jesteś obecnie w gałęzi. Podaj, z jaką gałęzią scalić. Więcej szczegółów w git-pull(1).

git pull <gałąź>

$ git branch * (HEAD odłączone na 4dda149) master $ git checkout master Przełączono na gałąź „master” Twoja gałąź jest na bieżąco z „origin/master”. $ git branch * master $ git branch -a * master remotes/origin/HEAD -> origin/master remotes/origin/efb_support remotes/origin/gh-pages remotes/origin/master remotes/origin/micko/ecp5_update

But nothing needed any pdate, when I did another "make install" in the libtrellis subdir. The problem must be somwhere else...

[mmicko]

Default location for install, is in /usr/local so file at /usr/local/share/trellis/database/devices.json should match device.json from database sub directory. Just to be sure you are using Trellis from proper location, run cat CMakeCache.txt | grep TRELLIS inside nextpnr directory it should give you:

TRELLIS_INSTALL_PREFIX:STRING=/usr/local
TRELLIS_LIBDIR:PATH=/usr/local/lib/trellis
TRELLIS_PROGRAM_PREFIX:STRING=

[solverstin]

I get exactly the same values.

But I've looked into the devices.json, and not all the defices do have a variant keyword within it. Could that be the reason *import.py script fails here:

File "ecp5/trellis_import.py", line 260, in process_devices_db for varname, vardata in sorted(devicedb["families"][family]["devices"][dev]["variants"].items()): KeyError: 'variants'

[mmicko]

Ah, it looks like you are actually on branch, or have applied some of commits from micko/ecp5_update branch. On current master there is no use of "variants" for ECP5 https://github.com/YosysHQ/nextpnr/blob/master/ecp5/trellis_import.py#L260

[solverstin]

Oh, this is it!

I've switched to it in a desperate search for reasons why my build failed.... and the actual reason was the missing sobmodule in my prjtrellis build.

Thenx for your help!

BTW: would you think, that there should be some guard in the prjtrellis for such misshaps? Like make install checking-if/depensing--on the devices.json file, and doing apropriate git submodule init if its missing?

[mmicko]

Glad it is solved. Not sure if this would actually help, doing such checks would complicate maintaining and still there will be room for user error.