ARM GCC bug [Was: Squeak on WinCE HPC MIPS?]
Tommy Thorn
tt1729 at yahoo.com
Tue Jul 16 19:59:03 UTC 2002
Kevin Fisher wrote:
> Well I just took a look around the Squeak-list archives and came across
> my old message from the Squeak 2.9 days...this is the error I was seeing
> with GCC 2.95.1:
>
> gcc -I/usr/X11R6/include -g -O2 -DHAVE_CONFIG_H -DUNIX
> -DSQ_LIBDIR=\"/usr/local/lib/squeak2.9a\" -I. -I../src/generated -I../src/unix
> -c -o SoundCodecPlugin.o ../src/generated/SoundCodecPlugin.c
> /tmp/ccFTDeUu.s: Assembler messages:
> /tmp/ccFTDeUu.s:1394: Error: Can not represent OFFSET_IMM8 relocation in this
> object file format (1)
> /tmp/ccFTDeUu.s:1396: Error: Can not represent OFFSET_IMM8 relocation in this
> object file format (1)
Thanks.
I use the standard cross compiler for Zaurus from Embeddix which is
2.95.2. When building Squeak I encountered the bug, but only for
sqGSMCodecPlugin.c. The rest, notably interp.o, was compiled with -O2
and works fine so it's not this GCC bug that holds back Squeaks
performance on ARM.
I get 12_326_656 bytecodes/sec; 413_533 sends/sec with that VM (some
3.1current, I should try a recent VM) and Dynapad 0.01 (not 0.1). Given
Ned's 101_506_740 bytecodes/sec; 3_490_879 sends/sec numbers (same
frequency? memory interface? etc) it looks like I should be getting
more.
I wouldn't be too surprised if Sharp had opted for a cheaper
(= narrower) and slower memory interface than Corel.
> Unfortunately, I found that using GCC 2.95.3 and up made the situation worse--
> you wouldn't get this compile-time error, you'd just get a VM executable
> that would segfault on start. :(
That's much worse, I agree
/Tommy
More information about the Squeak-dev
mailing list
|