[Bug target/35018] [m68k-elf] Gcc ouputs invalid asm when compiling with -O2 or higher
--- Comment #3 from nizze86 at hotmail dot com 2008-10-26 11:25 --- Created an attachment (id=16551) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16551&action=view) Reduced testcase A reduced testcase derived from the preprocessed sources. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35018
[Bug target/35018] [m68k-elf] Gcc ouputs invalid asm when compiling with -O2 or higher
--- Comment #4 from nizze86 at hotmail dot com 2008-10-26 11:28 --- (In reply to comment #3) > Created an attachment (id=16551) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16551&action=view) [edit] > Reduced testcase > > A reduced testcase derived from the preprocessed sources. > compiled with: m68k-elf-gcc -mcpu=5249 testcase.c -O2 -o testcase.o using gcc 4.3.1 with configuration: --target=m68k-elf --prefix=/usr/local/m68k-elf --enable-languages=c --disable-ibssp --with-arch=cf -- nizze86 at hotmail dot com changed: What|Removed |Added CC| |nizze86 at hotmail dot com http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35018
[Bug c/35018] New: [m68k-elf] Gcc ouputs invalid asm when compiling with -O2 or higher
Using gcc version 4.3.0 20080125 (experimental) (GCC) to crosscompile the attached file for m68k-elf, I get the following line of asm: move.l (%a2,%a0.l*4),76(%sp) which is invalid for the target cpu I have specified. (-mcpu=5249) This is my compiler command line: /rockbox/apps/codecs/Tremor$ /usr/bin/ccache m68k-elf-gcc -I/home/nils/rockbox/apps -I.. -I. -I/home/nils/rockbox/firmware/include -I/home/nils/rockbox/firmware/export -I/home/nils/rockbox/firmware/common -I/home/nils/rockbox/firmware/drivers -I/home/nils/rockbox/gcctest -I/home/nils/rockbox/apps/recorder -I/home/nils/rockbox/apps/gui -W -Wall -Wundef -Os -nostdlib -ffreestanding -Wstrict-prototypes -mcpu=5249 -malign-int -mstrict-align -fomit-frame-pointer -I/home/nils/rockbox/firmware/target/coldfire/iriver/h300 -I/home/nils/rockbox/firmware/target/coldfire/iriver -I/home/nils/rockbox/firmware/target/coldfire -O2 -DIRIVER_H300 -DMEM=32 -save-temps block.c -o /home/nils/rockbox/gcctest/apps/codecs/Tremor/block.o I used the following switches to configure gcc: --target=m68k-elf --prefix=/usr/local/m68k-elf --enable-languages=c --disable-libssp --with-arch=cf on my amd64 linux system -- Summary: [m68k-elf] Gcc ouputs invalid asm when compiling with - O2 or higher Product: gcc Version: 4.3.0 Status: UNCONFIRMED Severity: major Priority: P3 Component: c AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: nizze86 at hotmail dot com GCC host triplet: Linux amd 64 GCC target triplet: m68k-elf http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35018
[Bug c/35018] [m68k-elf] Gcc ouputs invalid asm when compiling with -O2 or higher
--- Comment #1 from nizze86 at hotmail dot com 2008-01-29 15:36 --- Created an attachment (id=15048) --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15048&action=view) Preprocessed sources -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35018
[Bug target/43675] New: [m68k] Wrong code due to missing sign extension
When compiling the following function for a coldfire target, such as mcf5249:
(using the following command line: m68k-elf-gcc saturate.c -mcpu=5249 -O1 -S)
int saturate(int x)
{
return (x == (short)x ? x : (x >> 31) ^ 0x7FFF);
}
gcc produces:
saturate:
link.w %fp,#0
move.l 8(%fp),%d0
move.w %d0,%a0
cmp.l %a0,%d0
jeq .L2
add.l %d0,%d0
subx.l %d0,%d0
eor.l #32767,%d0
.L2:
unlk %fp
rts
Where a0 is compared to d0 without a sign extension so the upper half of a0
remtains whatever was already there.
This is with gcc:
4.5.0 20100401 (experimental) (GCC)
configured with:
--target=m68k-elf --prefix=/usr/local --enable-languages=c --disable-libssp
--with-arch=cf --with-cpu=5249
Also observed with gcc 3.4.6 so likely not a regression.
--
Summary: [m68k] Wrong code due to missing sign extension
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: nizze86 at hotmail dot com
GCC target triplet: m68k-elf-*
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43675
[Bug target/43676] New: [m68k] Wrong code due to missing sign extension
When compiling the following function for a coldfire target, such as mcf5249:
(using the following command line: m68k-elf-gcc saturate.c -mcpu=5249 -O1 -S)
int saturate(int x)
{
return (x == (short)x ? x : (x >> 31) ^ 0x7FFF);
}
gcc produces:
saturate:
link.w %fp,#0
move.l 8(%fp),%d0
move.w %d0,%a0
cmp.l %a0,%d0
jeq .L2
add.l %d0,%d0
subx.l %d0,%d0
eor.l #32767,%d0
.L2:
unlk %fp
rts
Where a0 is compared to d0 without a sign extension so the upper half of a0
remtains whatever was already there.
This is with gcc:
4.5.0 20100401 (experimental) (GCC)
configured with:
--target=m68k-elf --prefix=/usr/local --enable-languages=c --disable-libssp
--with-arch=cf --with-cpu=5249
Also observed with gcc 3.4.6 so likely not a regression.
--
Summary: [m68k] Wrong code due to missing sign extension
Product: gcc
Version: 4.5.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: nizze86 at hotmail dot com
GCC target triplet: m68k-elf-*
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43676
[Bug target/43675] [m68k] Wrong code due to missing sign extension
--- Comment #3 from nizze86 at hotmail dot com 2010-04-07 15:42 --- I am sorry, it appears i am mistaken, since the dest is an address reg, movea is used which apparently *does* sign extend word size source to 32 bits while a move to a d reg does not. Again sorry for the noise. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43675
[Bug target/42516] New: [m68k] Suboptimal halfword swap on coldfire
When compiling the following function for a coldfire target, such as mcf5249:
(using the following command line m68k-elf-gcc hswap.c -mcpu=5249 -S -O2)
unsigned hswap(unsigned x)
{
return (x << 16) | (x >> 16);
}
gcc produces:
hswap:
link.w %fp,#0
move.l 8(%fp),%d1
move.l %d1,%d0
clr.w %d0
swap %d0
swap %d1
clr.w %d1
or.l %d1,%d0
unlk %fp
rts
Optimization level affects only instr ordering.
the same function compiled for 68020
(using the following command line m68k-elf-gcc hswap.c -mcpu=68020 -S -O2)
hswap:
link.w %fp,#0
move.l 8(%fp),%d0
swap %d0
unlk %fp
rts
which is valid code for the coldfire target too and much smaller and faster
this is using gcc 4.4.2 configured with
--enable-languages=c --disable-libssp --with-arch=cf --with-cpu=5249
the same behaviour is present in gcc 3.4.6 too.
--
Summary: [m68k] Suboptimal halfword swap on coldfire
Product: gcc
Version: 4.4.2
Status: UNCONFIRMED
Severity: enhancement
Priority: P3
Component: target
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: nizze86 at hotmail dot com
GCC host triplet: amd64-unknown-linux
GCC target triplet: m68k-elf
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42516
