And did you see any improvement? I saw some code size improvement when compiling scummvm caused by -mcpu=arm926ej-s. I don't know about speed. Here are the sizes
-rwxrwxr-x 1 fanoush fanoush 3234488 Feb 13 21:08 scummvm.334
-rwxrwxr-x 1 fanoush fanoush 3119452 Feb 13 20:30 scummvm.342.arm926ej-s
-rwxrwxr-x 1 fanoush fanoush 3299676 Feb 13 21:15 scummvm.342.armv4
the .342 ones contain libstdc++ linked statically (otherwise it would not run on the device) but the .334 has it linked dynamically so the 342.armv4 is a bit bigger than .334. But you can still see that the -mcpu=arm926ej-s shrunk code by ~200KB which is pretty good on 3MB code when some of it are static libraries (libmad,tremor) that are not influenced by the flag.
here is test with dynamically linked stdlibc++
-rwxrwxr-x 1 fanoush fanoush 3080308 Feb 15 08:20 scummvm342.arm926ej-s.dynamic
so the static one adds ~40KB to the executable.
This stdlibc++ thing is all because scummv is written in c++. I suppose there are no such troubles with libraries with plain C code when mixing compilers.
-rwxrwxr-x 1 fanoush fanoush 3234488 Feb 13 21:08 scummvm.334
-rwxrwxr-x 1 fanoush fanoush 3119452 Feb 13 20:30 scummvm.342.arm926ej-s
-rwxrwxr-x 1 fanoush fanoush 3299676 Feb 13 21:15 scummvm.342.armv4
the .342 ones contain libstdc++ linked statically (otherwise it would not run on the device) but the .334 has it linked dynamically so the 342.armv4 is a bit bigger than .334. But you can still see that the -mcpu=arm926ej-s shrunk code by ~200KB which is pretty good on 3MB code when some of it are static libraries (libmad,tremor) that are not influenced by the flag.
here is test with dynamically linked stdlibc++
-rwxrwxr-x 1 fanoush fanoush 3080308 Feb 15 08:20 scummvm342.arm926ej-s.dynamic
so the static one adds ~40KB to the executable.
This stdlibc++ thing is all because scummv is written in c++. I suppose there are no such troubles with libraries with plain C code when mixing compilers.