Disable the generational translation cache - I've got no evidence that it
actually helps performance, and it simplifies things to get rid of it (in
particular, translated code doesn't have to worry about being moved now).
src/dreamcast.h
src/sh4/xltcache.c
src/test/testxlt.c

Only call finish_rendering() for texture renders - workaround bug in apple/intel drivers
src/pvr2/pvr2.c

Update to print symbols for matching operands
src/x86dasm/i386-dis.c
src/x86dasm/x86dasm.c

Fixup testsh4x86 to actually work (as opposed to just barely compiling)
src/test/testsh4x86.c

Fix compile of testsh4x86
src/test/testsh4x86.c

Fix "0" being confused with "not defined"
src/drivers/genkeymap.pl

Fix incorrect calculation of width and size (darn missing parentheses)
src/pvr2/pvr2.c

Implement a sorted TLB lookup table (big improvement over the linear table scan)
Optimize out the 1C000000 -> FC000000 check at the end of the lookup functions
src/sh4/mmu.c

Fix (extremely boneheaded) failure to convert pc to physical address before
storing in the translation cache (in other words, the translation cache was
effectively disabled for MMU code). MMU code is now about 3 times faster...
src/sh4/sh4trans.c

Enable sse-math for all platforms
configure
configure.in