[email protected] wrote:
Hi Sergey,
I am trying to build the new mspgcc tools. I compiled the binutils and the
msp430-gcc with the sources from the repository (new branch).
My problem now is when I try to compile libc I get the following error during
the compile:
[...]
msp430-gcc -c -I./../include -I. -Wall -O2 -g -mmcu=msp1 -c stdlib/abs.c -o
msp1/abs.o
seems like you got libc sources from trunk, not from MSP430X branch.
try cvs up -r MSP430X
Do you have an idea, why my new msp430-gcc doesn't accept msp1 as argument for
mmcu?
it doesn't need anymore. See comments in gcc/congig/msp430/t-msp430
Thanks in advance
-------- Original-Nachricht --------
Datum: Mon, 26 Jan 2009 22:55:48 +0200
Von: "Sergey A. Borshch" <[email protected]>
An: "\'GCC for MSP430 - http://mspgcc.sf.net\'"
<[email protected]>
Betreff: [Mspgcc-users] 430X support branch in cvs repo
Hi everyone.
I created branch in mspgcc repository (tag MSP430X_fartext) an submitted
my additions:
- calla/reta/bra instructions generated for 430X core
- pushm[.a]/popm[.a] instructions in prologues/epilogues for 430X
- __attribute__((__far__)) for placing data (const) and functions into
high memory.
- -mdata-64k option for disabling far data support. Now it disables
attribute((far)) for const
data, replaces pushm.a/popm.a to pushm.w/popm.w (and movx.a/mova to mov.w,
pushx.a/popx.a to
push.w/pop.w in libraries compiled from assembly), in future it should
generate all 430 data
processing instructions instead of 430X and reduce pointer size from 32
bits to 16 bit. Code will
became faster and smaller, less stack usage in price of no far data and no
pointers to far functions.
- -mcode-64k option for disabling far code support. This option disables
__attribute__((__far__))
for functions and generates call/ret/br instead of calla/reta/br. Faster
and smaller code in price
of no far functions.
- .farrodata section added to 430X ld scripts.
- multilib support:
* separate libraries for 430/430X cores without hwmul/with HWMUL/with
HWMUL32 (need to write
routines for HWMUL32, they are the same as for HWMUL now).
* separate lib for 430 cores and -msave-prologues option.
* separate libs for 430 cores with HWMUL/HWMUL32 and -mdisable-hwmul
option
* separate lib for 430 cores and -msave-prologues and -mdisable-hwmul
option
* separate libs for 430X cores and all combinations of
-mdisable-hwmul, -mdata-64k, -mcode-64k
options
* separate libs for 430 cores with CPU4 bug (so mcu's without CPU4
bug should use more fast and
short library functions without CPU4 workaround).
* 36 library sets in total :)
one disadvantage: improper option (-msave-prologues with 430X mcu or
-mdata-64k with 430 mcu) leads
to linking default library (library for 430 mcu without HWMUL). The only
workarounds I see:
1) edit specs file - remove unmeaning options from multilib selection
lines. Need to write sed(?)
script. Who helps?
2) or build fake default library and remove it after building. This
leads to unresolved symbols
and forces user to check options.
-------
todo:
- full far data support (including 32bits pointers and table jumps in
far code) - need to
rewrite almost all msp430.md file.
- full HWMUL32 support - need to add 32*16, 32*32, 32*64, 64*64 to
libgcc.S and calls to msp430.md.
feedbacks, patches and help are welcome.
--
Regards,
Sergey A. Borshch mailto: [email protected]
SB ELDI ltd. Riga, Latvia
--
Regards,
Sergey A. Borshch mailto: [email protected]
SB ELDI ltd. Riga, Latvia
