single bit test in one instruction

[notzed]

[Gravis]

[EggBaconAndSpam]

do not forget pipelining, IALU operations do not cause stalls (in this scenario)...

-test, bit is constant
lsl // 1 cycle
blt // 1 if not taken, 4 if taken
b // 4, use plain branch instead!

overall we are talking about either 5 or 6 cycles, inlining would make sense


-bit stored in register
mov // 1
sub // 1
lsl // 1
blt // 1 or 4
b // 4

overall either 7 or 8 cycles

// write this in C instead, the compiler might be able to do (31-param) at compile time!
// also, inlining!

[timpart]

For other playing around with bits tips, take a look at . It is in C code but using operators that easily translate to assembler in most cases.

I ignore the first 3 cycles of an instruction when thinking about timings. They get hidden in the pipeline overlap.

Tim

[EggBaconAndSpam]

those 3 particular pipeline stages are responsible for the 3 cycle branch penalty

[Gravis]

[notzed]

[hewsmike]

[Gravis]

[notzed]