On 05/19/2010 08:16 PM, Ian Lance Taylor wrote:
"Johan Schuld"<Johan.Schuld@xxxxxxxxxxxxxxxxxxxxx> writes:
My first post in this group, so I hope it's the correct one.
I'm using a MIPS32 base processor (4KSd) which also has the MIPS16e extensions.
First of all, some of the default compiler flags I have use the
-mips16e compiler flag, even though I can only find the -mips16
(so without the 'e') flag in the GNU GCC documentation
online. (http://gcc.gnu.org/onlinedocs/gcc-4.4.4/gcc/MIPS-Options.html#MIPS-Options). I can
not find any -mips16e specific documentation.)
So for my project, I use both the -mips23r2 and -mips16e
compiler flags. Some of my files (startup files) contain assembly
code, and do not get compiled with the above settings (it does not
recognize 'di' 'sync' and 'ei' as valid opcodes). These opcodes
exist in the MIPS32r2 instruction set, but not in the MIPS16e
extensions.
My assumption was that using the -mips16e compiler flag, would
only enable the use of the mips16e specific extensions, but it seems
like it will only use the compact instruction set? Can I use
-mips16e and -mips32r2 simultaneously, or does it force the compiler
to ONLY use the compact instruction set?
gcc will generate MIPS16e code if you use the -mips16 option in
combination with the -mips32 or -mips32r2 option. So you should
indeed get MIPS16e code with the options you are using.
However, you are correct that using -mips16 will cause the compiler to
only use the compact instruction set. There is no way to flip back
and forth in a single source file. Files which contain non-MIPS16e
assembler code must be compiled without the -mips16 option.
The GCC-4.5 manual indicates that this may not be the case. The mips16
and nomips16 function attributes appear to allow you to control this on
a function by function basis.
David Daney
