view TOOLS/fastmem.sh @ 10578:b9d289fd8a57

10000l, the old code was slow as hell, copying stuff extra times and actually broken -- blanking the whole screen at each 'page flip' with -dr enabled. benchmarks: before: 56% cpu for decode 56% cpu for vo with no -dr 25% cpu for vo with -dr after: 56% cpu for decode 25% cpu for vo without -dr 0% cpu for vo with -dr if vo_fbdev is going to do pageflip, it needs to do it for REAL, using vertical scroll registers (like g2), not copying a temp buffer (which will shear anyway and is super-slow).

sync
sleep 2
./fastmem-k6
sleep 2
./fastmem-k7
sleep 2
./fastmem-mmx
sleep 2
./fastmem-sse
sleep 2
./fastmem2-k6
sleep 2
./fastmem2-k7
sleep 2
./fastmem2-mmx
sleep 2
./fastmem2-sse
sleep 2