LoadNone cannot initialize n-bytes of the texture where n is the size of a
pointer on an architecture. It is most likely too small to initialize a big
texture but it can be too big and overwrite unrelated parts of the memory.
n.pepinpe requested in #540 to allow overwriting the automatically searched
CFLAGS and LIBS for used libraries. This should allow distributors to set the
build configuration without patching the makefile. This is important for
systems were pkg-config or sdl-config aren't normally used.
n.pepinpe requested in #540 to allow overwriting the automatically searched
CFLAGS and LIBS for used libraries. This should allow distributors to set the
build configuration without patching the makefile. This is important for
systems were pkg-config or sdl-config aren't normally used.
The GCC introduced Link-time optimization in GCC 4.5 (2010-04-14). This should
be long enough available that interested users have upgraded to a compiler
supporting it.
The MSVC project already enabled WholeProgramOptimization since a long time.
Enabling it by default in GCC seems to be equally valid.
The GCC manual states for different parameters that the options for compilation
must also be used when linking. The options for compilation are stored in
CXXFLAGS and added to LINK.o to fix the behavior.
Option which need this are for example -fPIC/-fPIE or -flto.
Some linker on different platforms don't handle the garbage collection
correctly and create extreme bloated binaries. Therefore, leave it to the user
to enable this feature or not.
The byte swapping was realized using inline assembler code using the bswap
instruction. This was less portable and even resulted in build failures when
compiling for OSX. It also reduced the effectivness of the optimizer slightly.
Instead the builtin functions of the compiler can be used to do the same.
MacOSX and MinGW don't provide gl.pc or glu.pc to find the cflags or libs using
pkg-config. Removing this search and replacing it with an hardcoded value is
easier on these platforms.
The POSTFIX make option is useful for distributions to compile different
versions of the plugin in parallel. The object files will be stored in a
directory with the postfix appended and the linker result will also have this
postfix appended.
The CROSS_COMPILE make option can be used to automatically prepend the prefix
to all build relevant tools to seamlessly allow cross compilation without
setting each tool name separately.
Cross compiling for MinGW32 would can be done using
$ make -C projects/unix/ CROSS_COMPILE=i686-pc-mingw32- HOST_CPU=i686 UNAME=MINGW
nogagplz tested mupen64plus on 32-bit ppc and didn't detect any big show
stoppers with interpreter cores and the mupen64plus example rom. This makes the
PowerPC an interesting target for further tests and may reveal other endianness
problems.
nogagplz noticed on ppc32 that PIC is necessary to compile everything as shared
object. Therefore, it seems to be a better idea to have architecture specific
PIC default settings instead of checking only whether it is a 32 or 64 bit
architecture.
