r4600 flag
in arch/mips/Makefile, we use the -mcpu=r4600 flag for MIPS32. What's interesting is that the use of this flag can cause the toolchain to generate incorrect code. For example:
la k0, 1f
nop
1: nop
The la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
Is there a truly correct -mcpu option for a mips32 cpu?
Pete
Quoted 3 lines · Pete Popov
in
arch/mips/Makefile, we use the -mcpu=r4600 flag for MIPS32. What's interesting is that the use of this flag can cause the toolchain to generate incorrect code.
Which toolchain versions, what patches applied to it?
Quoted 10 lines
For example:
la k0, 1f
nop
1: nopThe la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
This is IIRC a known bug in at least current binutils CVS, a patch to fix it really was already discussed.
Quoted 1 line
Is there a truly correct -mcpu option for a mips32 cpu?
In theory, no -mcpu option (which is to be deprecated in favor of -march/-mtune) and -mips32 as ISA flag _should_ work.
In practice, the patch which adds this to gas was discussed on the binutils list the last days and is likely to go in CVS today or tomorrow.
Thiemo
Quoted 4 lines
in
arch/mips/Makefile, we use the -mcpu=r4600 flag for MIPS32. What's interesting is that the use of this flag can cause the toolchain to generate incorrect code. For example:
Quoted 8 lines
la k0, 1f
nop
1: nopThe la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
Using mips-linux-gcc from egcs-2.91.66, I don't see exactly this behavior in the test case above. I *do* see that *if* I have -mcpu=4600 set *and* I have not otherwise set the ISA level to be MIPS I or MIPS II (-mips1, -mips2), 64-bit instructions will be emitted. But that's to be expected. To generate 32-bit code for an R4600-like platform, you need to specify both the ISA level (to deal with issues like the above) and the R4600 pipeline (to get the MAD instruction). One place where care must
be exercised is in MIPS32 exception handlers, where eret must be used. Prior to MIPS32, eret implied MIPS III which implied a 64-bit CPU, and the Linux compilers still see things that way. If the whole module is built with -mips3, one does risk seeing some cursed macro expansions.
Until we get proper support for MIPS32 integrated into
the standard Linux tool chain (see below), the most
correct option would be to build with -mips2 and to use explicit ".set mips3/.set mips0" directives.
Quoted 1 line
Is there a truly correct -mcpu option for a mips32 cpu?
It's "-mips32", which is sort of a -mips option and a -mcpu option rolled into one. It's supported by several gnu distributions, notably those of Algorithmics and Cygnus/Red Hat. I believe that someone at MIPS or Algorithmics succeeded in building a Linux kernel of some description using the Algorithmics SDE, but I don't know the details.
Regards,
Kevin K.
Quoted 6 lines · Thiemo Seufer
Quoted 3 lines
The la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
This is IIRC a known bug in at least current binutils CVS, a patch to fix it really was already discussed.
Is this macro expaned by the compiler or assembler? Just -mcpu=r4600 should not make cc1 generate any instructions beyond MIPS I.
In the past we already had inline assembler fragments that left the assembler in .misp3 mode thus resulting the rest of the file bein assembled in mips3 mode.
Quoted 1 line
Quoted 1 line
Is there a truly correct -mcpu option for a mips32 cpu?
None is really good, none is really bad ...
Quoted 6 lines
In theory, no -mcpu option (which is to be deprecated in favor of -march/-mtune) and -mips32 as ISA flag _should_ work.
In practice, the patch which adds this to gas was discussed on the binutils list the last days and is likely to go in CVS today or tomorrow.
Ralf
Quoted 5 lines · Kevin D. Kissell
Using mips-linux-gcc from egcs-2.91.66, I don't see exactly this behavior in the test case above. I *do* see that *if* I have -mcpu=4600 set *and* I have not otherwise set the ISA level to be MIPS I or MIPS II (-mips1, -mips2), 64-bit instructions will be emitted. But that's to be expected.
No, it contradict the GCC documentation:
`-mcpu=CPU TYPE'
Assume the defaults for the machine type CPU TYPE when scheduling instructions. The choices for CPU TYPE are r2000', r3000', r3900', r4000', r4100', r4300', r4400', r4600', r4650', r5000', r6000', r8000', and orion'. Additionally, the r2000', r3000', r4000', r5000', and r6000' can be abbreviated as r2k' (or r2K'), `r3k', etc. While picking a specific CPU TYPE will schedule things appropriately for that particular chip, the compiler will not generate any code that does not meet level 1 of the MIPS ISA (instruction set architecture) without a -mipsX' or -mabi' switch being used.
Quoted 4 lines
To generate 32-bit code for an
R4600-like platform, you need to specify both the ISA level (to deal with issues like the above) and the R4600 pipeline (to get the MAD instruction).
No MAD on R4600. Again it would be in contradiction with above document- ation. Mad you get with:
`-mmad'
`-mno-mad'
Permit use of the mad', madu' and mul' instructions, as on the r4650' chip.
Quoted 8 lines
Quoted 1 line
Is there a truly correct -mcpu option for a mips32 cpu?
It's "-mips32", which is sort of a -mips option and a -mcpu option rolled into one. It's supported by several gnu distributions, notably those of Algorithmics and Cygnus/Red Hat. I believe that someone at MIPS or Algorithmics succeeded in building a Linux kernel of some description using the Algorithmics SDE, but I don't know the details.
Ralf
Quoted 8 lines
Quoted 5 lines
Using mips-linux-gcc from egcs-2.91.66, I don't see exactly this behavior in the test case above. I *do* see that *if* I have -mcpu=4600 set *and* I have not otherwise set the ISA level to be MIPS I or MIPS II (-mips1, -mips2), 64-bit instructions will be emitted. But that's to be expected.
No, it contradict the GCC documentation:
Quoted 11 lines
`-mcpu=CPU TYPE'
Assume the defaults for the machine type CPU TYPE when scheduling instructions. The choices for CPU TYPE arer2000',r3000',r3900',r4000',r4100',r4300',r4400',r4600',r4650',r5000',r6000',r8000', andorion'. Additionally, ther2000',r3000',r4000',r5000', andr6000' can be abbreviated asr2k' (orr2K'), `r3k', etc. While picking a specific CPU TYPE will schedule things appropriately for that particular chip, the compiler will not generate any code that does not meet level 1 of the MIPS ISA (instruction set architecture) without a-mipsX' or-mabi' switch being used.
In that case, the tools that I've been using are technically
broken. Surprise surprise. Because -mcpu=4600 is
most assuredly setting the ISA level, even if it doesn't override one explicitly set!
Quoted 7 lines
Quoted 4 lines
To generate 32-bit code for an
R4600-like platform, you need to specify both the ISA level (to deal with issues like the above) and the R4600 pipeline (to get the MAD instruction).No MAD on R4600. Again it would be in contradiction with above document- ation. Mad you get with:
Right. Sorry. I got the 4600 and 4650 confused. I no longer understand why "4600" and not "4650" is the model for MIPS32.
Kevin K.
Quoted 11 lines · Ralf Baechle
Quoted 6 lines · Thiemo Seufer
Quoted 3 lines
The la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
This is IIRC a known bug in at least current binutils CVS, a patch to fix it really was already discussed.
Is this macro expaned by the compiler or assembler? Just -mcpu=r4600 should not make cc1 generate any instructions beyond MIPS I.
It's the assembler, -mcpu does not only affect scheduling in gas. To clean this up -march and -tune were introduced recently to obsolete -mcpu and -m<cpu> (in both binutils and gcc).
Thiemo
Kevin D. Kissell wrote:
[snip]
Quoted 12 lines
Quoted 11 lines
`-mcpu=CPU TYPE'
Assume the defaults for the machine type CPU TYPE when scheduling instructions. The choices for CPU TYPE arer2000',r3000',r3900',r4000',r4100',r4300',r4400',r4600',r4650',r5000',r6000',r8000', andorion'. Additionally, ther2000',r3000',r4000',r5000', andr6000' can be abbreviated asr2k' (orr2K'), `r3k', etc. While picking a specific CPU TYPE will schedule things appropriately for that particular chip, the compiler will not generate any code that does not meet level 1 of the MIPS ISA (instruction set architecture) without a-mipsX' or-mabi' switch being used.
Quoted 4 lines
In that case, the tools that I've been using are technically
broken. Surprise surprise. Because -mcpu=4600 is
most assuredly setting the ISA level, even if it doesn't override one explicitly set!
gas and gcc have different meanings for this option, gas uses the cpu's default ISA (if none specified) while gcc uses MIPS I. In current binutils/gcc CVS, there was -march and -mtune introduced as a replacement for -mcpu and -m<cpu>, which were kept for backward compatibility only.
[snip]
Quoted 5 lines
Quoted 2 lines
No MAD on R4600. Again it would be in contradiction with above document- ation. Mad you get with:
Right. Sorry. I got the 4600 and 4650 confused. I no longer understand why "4600" and not "4650" is the model for MIPS32.
At least for the CVS binutils gas it is Generic-MIPS32, not 4600.
Thiemo
Quoted 9 lines · Ralf Baechle
Quoted 6 lines · Thiemo Seufer
Quoted 3 lines
The la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
This is IIRC a known bug in at least current binutils CVS, a patch to fix it really was already discussed.
Quoted 6 lines
Is this macro expaned by the compiler or assembler? Just -mcpu=r4600 should not make cc1 generate any instructions beyond MIPS I.
In the past we already had inline assembler fragments that left the assembler in .misp3 mode thus resulting the rest of the file bein assembled in mips3 mode.
Yes, and I'm sure I fixed that so it worked on MIPS32 CPUs, only leaving the "eret" instructions.
Quoted 13 lines
Quoted 2 lines
Quoted 2 lines
Is there a truly correct -mcpu option for a mips32 cpu?
None is really good, none is really bad ...
Quoted 6 lines
In theory, no -mcpu option (which is to be deprecated in favor of -march/-mtune) and -mips32 as ISA flag _should_ work.
In practice, the patch which adds this to gas was discussed on the binutils list the last days and is likely to go in CVS today or tomorrow.
Ralf
Looking at the system map from the generic build of 2.4.3, it looks like the code is 64 bit. The upper 32 bits are "f" instead of 0. I built it using the R4600 flag. This differs from the system map for 2.4.0-test9 where the upper 32 bits are 0. For the indy, do I specify mips2 to generate 32 bit code? objdump says it is ELF32, but it doesn't look like that. I would like to generate 32bit.
thanks,
john
Quoted 38 lines · Carsten Langgaard
Quoted 9 lines · Ralf Baechle
Quoted 6 lines · Thiemo Seufer
Quoted 3 lines
The la macro is split into a lui and a daddiu. The daddiu is not correct for a mips32 cpu. Getting rid of the -mcpu=4600 fixes the problem and the daddiu is then changed addiu.
This is IIRC a known bug in at least current binutils CVS, a patch to fix it really was already discussed.
Quoted 6 lines
Is this macro expaned by the compiler or assembler? Just -mcpu=r4600 should not make cc1 generate any instructions beyond MIPS I.
In the past we already had inline assembler fragments that left the assembler in .misp3 mode thus resulting the rest of the file bein assembled in mips3 mode.
Yes, and I'm sure I fixed that so it worked on MIPS32 CPUs, only leaving the "eret" instructions.
Quoted 13 lines
Quoted 2 lines
Quoted 2 lines
Is there a truly correct -mcpu option for a mips32 cpu?
None is really good, none is really bad ...
Quoted 6 lines
In theory, no -mcpu option (which is to be deprecated in favor of -march/-mtune) and -mips32 as ISA flag _should_ work.
In practice, the patch which adds this to gas was discussed on the binutils list the last days and is likely to go in CVS today or tomorrow.
Ralf
Quoted 2 lines · John D. Davis
Looking at the system map from the generic build of 2.4.3, it looks like the code is 64 bit. The upper 32 bits are "f" instead of 0.
No. Sign extended, that is bit 31 is copied into bits 32 to 63.
Quoted 2 lines
I built it > using the R4600 flag. This differs from the system map for 2.4.0-test9 where the upper 32 bits are 0.
No. Different binutils. Older binutils did zero extend 32-bit addresses to 64-bit addresses in the objdump output which is wrong.
Quoted 1 line
For the indy, do I specify mips2 to generate 32 bit code?
-mips2 :-)
For the kernel we use a few 64-bit instructions on the Indy though. These are carefully chosen such that nothing go wrong like exceptions corrupting the upper 32-bit of a register.
Quoted 2 lines
objdump says it is ELF32, but it doesn't look like that. I would like to generate 32bit.
ELF is an object format. Nothing prevents you from putting 64-bit code into a 32-bit ELF file and vice versa. You just need to be careful.
Ralf