[U-Boot-Users] Compiler optimizations and debugging

[U-Boot-Users] Compiler optimizations and debugging

[llandre]

To make the step-by-step debugging easier with GDB/DDD, I removed the 
optimization flag "-Os" from the config.mk file. In fact, with the 
optimization turned on, the code seems not to be executed sequentially 
inside code window. However, without the -Os flag, the linker is unable to 
complete the build:

ppc-linux-ld -Bstatic -T 
/proj/ppc405ep/sw/u-boot/board/bubinga405ep/u-boot.lds -Ttext 0xFFF80000 -n 
cpu/ppc4xx/start.o cpu/ppc4xx/resetvec.o \
                 --start-group board/bubinga405ep/libbubinga405ep.a 
cpu/ppc4xx/libppc4xx.a lib_ppc/libppc.a fs/jffs2/libjffs2.a 
fs/fdos/libfdos.a net/libnet.a disk/libdisk.a rtc/librtc.a dtt/libdtt.a 
drivers/libdrivers.a post/libpost.a post/cpu/libcpu.a common/libcommon.a 
lib_generic/libgeneric.a --end-group \
                 -Map u-boot.map -o u-boot
cpu/ppc4xx/libppc4xx.a(405gp_pci.o): In function `pci_405gp_init':
/proj/ppc405ep/sw/u-boot/cpu/ppc4xx/405gp_pci.c:125: undefined reference to 
`pci_set_region'
/proj/ppc405ep/sw/u-boot/cpu/ppc4xx/405gp_pci.c:132: undefined reference to 
`pci_set_region'
/proj/ppc405ep/sw/u-boot/cpu/ppc4xx/405gp_pci.c:146: undefined reference to 
`pci_set_region'
/proj/ppc405ep/sw/u-boot/cpu/ppc4xx/405gp_pci.c:157: undefined reference to 
`pci_set_region'
drivers/libdrivers.a(pci_indirect.o): In function `indirect_read_config_byte':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:48: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:48: undefined reference to 
`in_8'
drivers/libdrivers.a(pci_indirect.o): In function `indirect_read_config_word':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:49: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:49: undefined reference to 
`in_le16'
drivers/libdrivers.a(pci_indirect.o): In function `indirect_read_config_dword':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:50: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:50: undefined reference to 
`in_le32'
drivers/libdrivers.a(pci_indirect.o): In function `indirect_write_config_byte':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:56: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:56: undefined reference to 
`out_8'
drivers/libdrivers.a(pci_indirect.o): In function `indirect_write_config_word':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:57: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:57: undefined reference to 
`out_le16'
drivers/libdrivers.a(pci_indirect.o): In function 
`indirect_write_config_dword':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:58: undefined reference to 
`out_le32'
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:58: undefined reference to 
`out_le32'
drivers/libdrivers.a(pci_indirect.o): In function `pci_setup_indirect':
/proj/ppc405ep/sw/u-boot/drivers/pci_indirect.c:63: undefined reference to 
`pci_set_ops'
make: *** [u-boot] Error 1

Why is the -Os flag absolutely necessary?


llandre

[U-Boot-Users] Compiler optimizations and debugging

[Wolfgang Denk]

In message <5.1.1.6.0.20030320104504.00a81840@dns.struinfo.it> you wrote:
> To make the step-by-step debugging easier with GDB/DDD, I removed the 
> optimization flag "-Os" from the config.mk file. In fact, with the 
> optimization turned on, the code seems not to be executed sequentially 
> inside code window. However, without the -Os flag, the linker is unable to 

Well, the _optimized_ code _is_ exectued sequentially. But you cannot
expect that a sequence of lines of  C  code  results  in  a  strictly
sequential  sequence  of  machine  instructions. The potimizer will -
ummm - optimize the code.

> Why is the -Os flag absolutely necessary?

Because  otherwise  inline  functions  are  not  expanded.  See   the
documentation for GCC.

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd at denx.de
Given a choice between two theories, take the one which is funnier.

[U-Boot-Users] Compiler optimizations and debugging

[llandre]

> > To make the step-by-step debugging easier with GDB/DDD, I removed the
> > optimization flag "-Os" from the config.mk file. In fact, with the
> > optimization turned on, the code seems not to be executed sequentially
> > inside code window. However, without the -Os flag, the linker is unable to
>
>Well, the _optimized_ code _is_ exectued sequentially. But you cannot
>expect that a sequence of lines of  C  code  results  in  a  strictly
>sequential  sequence  of  machine  instructions. The potimizer will -
>ummm - optimize the code.

I meant that the code is apparently non-sequential.
Recently I worked with the GCC cross compiler for ARM (arm-elf-gcc) and I 
encountered the same problem. In that case the default compilation is 
usually performed with the -O2 flag but, during the debugging, we are used 
to build with the -O0 flag. This way, when you debug step-by-step, the C 
lines are executed sequentially, too. Thus, even if the code is obviously 
much larger and much slower, it is easier follow the flow of the code.
So I was wondering if it is possible to do the same thing with the 
ppc-linux-gcc compiler. Is it?


llandre

[U-Boot-Users] Compiler optimizations and debugging

[Vladimir Gurevich]

Hello llandre,

llandre wrote:
> To make the step-by-step debugging easier with GDB/DDD, I removed the 
> optimization flag "-Os" from the config.mk file. In fact, with the 
> optimization turned on, the code seems not to be executed sequentially 
> inside code window. However, without the -Os flag, the linker is unable 
> to complete the build:
> 
> Why is the -Os flag absolutely necessary?

Because the functions you are missing now (like out_8, in_32, etc.)
are, in fact, inline functions. You need -O{something} to enable them.

As far as "non-sequential" code execution in the debugger, you
can try to ADD the following flags to gcc command line:

     -fno-schedule-insns -fno-schedule-insns2

This should disable instruction reordering (as well as slow down
the execution quite noticeably since your pipeline will be stalled
too often).

Oh, and in any case (with or without -fno-schedule-insns) your
code is executed correctly.

Best regards,
Vladimir

[U-Boot-Users] Compiler optimizations and debugging

[Wolfgang Denk]

In message <3E79FBA1.2090707@paulidav.org> Vladimir Gurevich wrote:
> 
> As far as "non-sequential" code execution in the debugger, you
> can try to ADD the following flags to gcc command line:
> 
>      -fno-schedule-insns -fno-schedule-insns2
> 
> This should disable instruction reordering (as well as slow down
> the execution quite noticeably since your pipeline will be stalled
> too often).

100% correct. See also section "8.2.5. Tips and Tricks" in
http://www.denx.de/re/DPLG.html

Best regards,

Wolfgang Denk

-- 
Software Engineering:  Embedded and Realtime Systems,  Embedded Linux
Phone: (+49)-8142-4596-87  Fax: (+49)-8142-4596-88  Email: wd at denx.de
Do you suppose the reason the ends of the `Intel Inside'  logo  don't
match up is that it was drawn on a Pentium?