[fpc-devel] Sucesfull (mostly) experiment to build CHROMuLAN application with FPC main (3.3.1-git) and Lazarus GIT on RISC-V 64 target

Pavel Pisa ppisa at pikron.com
Sat Aug 3 18:09:01 CEST 2024 

Dear Florian and others,

we have developed open-open source chromatography system
control, acquisition and processing system many years ago

  https://sourceforge.net/projects/chromulan/

but there are some users still. You can see references form
research publications for example

  https://scholar.google.com/scholar?q=%22CHROMuLAN%22

The project has been started by prof Jindrich and funded
by PiKRON company when we have delivered instrument
development and licenses to other/production company.
But program is used by many to evaluate data even from
other instruments setups. The design has been done initially
in some ancient Delphi version and then in Delphi 6 which
is used for production build/seldom maintenance till now.
The users are mostly Windows oriented but I prefer
and use GNU/Linux from middle nighties and have it
as main/only desktop/development system for years.

I look on free-pascal progress from time to time,
I have even tested some old TurboVision project
on it decade(s) ago etc. And I gradually update CHROMuLAN
in some spare time in a year and it runs with Lazarus
including instruments control on GNU/Linux x86
32 and 64-bits already. I have tested even Raspberry Pi
build and 64-bit Windows etc. and external projects state
and my changes slowly converge to state that one day
FPC build can reach quality to offer it to our users
as standard.

But my main daily focus is motion-control, CAN communication,
FPGA and SoC peripherals for Linux, RTEMS, NuttX as
well as teaching of computer architectures teaching
at Czech Technical University and even abroad.

We have been donated by two Milk-V Pioneer 64-core,
RISC-V boxes by RISC-V international based on our
other activities

  https://www.crowdsupply.com/milk-v/milk-v-pioneer/

and because they came at the same day as I have helped
to prof. Jindrich with revival of HPLC instruments
from our design at their university, I have come to
wicked idea to test CHROMuLAN on that box.

The build of our communication driver (uLAN) run
smoothly on that box and device connected on the
first go. I have tested building of our Qt based
RISC-V simulator (QtRvSim) as the next step.
Again smooth.

So I have looked for RISC-V native compiler
binary and I have stubled over

  http://fedora.riscv.rocks/koji/buildinfo?buildID=86978

But that is probably some old experiment form 2019
and cannot be used to build actual mainline.

I have had to go home after more alternative
unsuccessful attempts so I have setup RISC-V 64-bit
Debian unstable chroot with qemu-static, bootstrapped
FPC from mainline on x86 Debian, build FPC cross
for RISC-V by it and then build riscv64 to riscv64
compiler binary. That build has been partial
but allowed to combine that binary with cross
build units at the end .

I have continued with attempts to boostrap FPC
with that compiler in RISC-V chroot. It has more
problems, I have even tried something similar
with FPC-3.2.4 branch - that failed completely
on unrecognized riscv64-linux target and attempt
t regenerate makefiles did not succeed either.
So I have returned to the FPC mainline. The main
problems has been in ./fpmake for packages and utils.

With use of qemu-static strace feature and even
debugging stub and gdb-multiarch I have identified
two strange problems when boostraping native
on RISC-V (patch/hack to move forward attached)

 1) TProcess call to GetCompilerInfo in fpmkunit.pp failed
    with bad address. Same problem has been in ExecuteFPC.
    When I have explicitly reprocessed list of environment
    defines, it solved the problem. It looks like nill
    or unassigned state is not propagated correctly into
    TProcess.SysExecute and execve fails with errno = 14
    EFAULT (bad address)

 2) another problem is that attempt to write to standard
    output in fpmkunit.pp results in exception because
    descriptor or StdOut setup seems to be closed/invalid.

Boostraps and cross compiler build from same sources
have no problems on x86. So may it be that could
help somebody more knowable to identify actual problem
source on native RISC-V build.

The full Lazarus build fails on  

9009) Assembling formeditingintf
units/riscv64-linux/nogui/formeditingintf.s: Assembler messages:
units/riscv64-linux/nogui/formeditingintf.s:3366: Error: illegal operands `lui ireg32,1'
units/riscv64-linux/nogui/formeditingintf.s:3367: Error: illegal operands `addiw ireg32,ireg32,-1736'
units/riscv64-linux/nogui/formeditingintf.s:3368: Error: illegal operands `add ireg33,x5,ireg32'
units/riscv64-linux/nogui/formeditingintf.s:3369: Error: illegal operands `ld x5,0(ireg33)'
units/riscv64-linux/nogui/formeditingintf.s:3453: Error: illegal operands `lui ireg32,1'
units/riscv64-linux/nogui/formeditingintf.s:3454: Error: illegal operands `addiw ireg32,ireg32,-1736'
units/riscv64-linux/nogui/formeditingintf.s:3455: Error: illegal operands `add ireg33,x5,ireg32'
units/riscv64-linux/nogui/formeditingintf.s:3456: Error: illegal operands `ld x5,0(ireg33)'
units/riscv64-linux/nogui/formeditingintf.s:3540: Error: illegal operands `lui ireg32,1'
units/riscv64-linux/nogui/formeditingintf.s:3541: Error: illegal operands `addiw ireg32,ireg32,-1736'
units/riscv64-linux/nogui/formeditingintf.s:3542: Error: illegal operands `add ireg33,x5,ireg32'
units/riscv64-linux/nogui/formeditingintf.s:3543: Error: illegal operands `ld x5,0(ireg33)'
units/riscv64-linux/nogui/formeditingintf.s:3557: Error: illegal operands `lui ireg32,1'
units/riscv64-linux/nogui/formeditingintf.s:3558: Error: illegal operands `addiw ireg32,ireg32,-1432'
units/riscv64-linux/nogui/formeditingintf.s:3559: Error: illegal operands `add ireg33,x5,ireg32'
units/riscv64-linux/nogui/formeditingintf.s:3560: Error: illegal operands `ld x5,0(ireg33)'
/home/pi/repo/fpc/lazarus-git/components/ideintf/formeditingintf.pas(847) Error: (9007) Error while assembling exitcode 1
/home/pi/repo/fpc/lazarus-git/components/ideintf/formeditingintf.pas(847) Fatal: (10026) There were 2 errors compiling module, stopping
Fatal: (1018) Compilation aborted

but lcl and other components can be build and when CHROMuLAN
is build from commandline it seems to run and process data
well when run under RISC-V qemu-static.

I plan to do testing on real RISC-V system next week
or after I return from vacations.

Generally I congratulate to enormous amount of work done
and ability to support new targets as well as keep
support for old ones.

If you have interrest, I can try some other build, experiments
and benchmarks when I have Milk-V boxes setup and stable running. 

Best wishes, 

                Pavel
--
                Pavel Pisa

    phone:      +420 603531357
    e-mail:     pisa at cmp.felk.cvut.cz
    Department of Control Engineering FEE CVUT
    Karlovo namesti 13, 121 35, Prague 2
    university: http://control.fel.cvut.cz/
    personal:   http://cmp.felk.cvut.cz/~pisa
    company:    https://pikron.com/ PiKRON s.r.o.
    Kankovskeho 1235, 182 00 Praha 8, Czech Republic
    projects:   https://www.openhub.net/accounts/ppisa
    social:     https://social.kernel.org/ppisa
    CAN related:http://canbus.pages.fel.cvut.cz/
    RISC-V education: https://comparch.edu.cvut.cz/
    Open Technologies Research Education and Exchange Services
    https://gitlab.fel.cvut.cz/otrees/org/-/wikis/home
-------------- next part --------------
A non-text attachment was scrubbed...
Name: fpc-native-riscv64.diff
Type: text/x-diff
Size: 2122 bytes
Desc: not available
URL: <http://lists.freepascal.org/pipermail/fpc-devel/attachments/20240803/7019b878/attachment.diff>

More information about the fpc-devel mailing list