Create a trace to test BP about mips-traces HOT 9 CLOSED

@NullByte4532 Is bubble sort a nice basis for this, how do you think?

from mips-traces.

i have just tried to write a program that would jump around a tree of conditional jumps based on value of $t0, and change that value after almost every jump, so in the end every condition has been true and false at least one time. In the end $t0 has a value of 42. But instructions BLTZ and BGTZ do not seem to work in simulator.

from mips-traces.

Alexey, I think you didn’t get the goal of the trace correctly (that’s not your fault but mine, as it is not described clearly here). Let’s postpone the discussion until branch prediction lecture is delivered.

from mips-traces.

@NullByte4532 So, I think you get the idea for such trace. Random branches are not good as they are not predictable by any hardware. We need something regular but tricky, that's why bubble sort looks promising here.

from mips-traces.

I cannot write an interesting branch prediction test when we have only one branching instruction implemented. Do i just write something using only beq or do i need to implement bgtz and bltz in the simulator first?

from mips-traces.

Why only one? We have four branches: beq, bgtz, blez, bne + all the jumps (see here: Supported MIPS instructions). bltz could be implemented via slt:

slt $at, $X, $zero
bne $at, $zero, target

from mips-traces.

But if you really want to use bltz and other instructions, you should ask @inedostoev who is implementing them at the moment.

from mips-traces.

@NullByte4532 bltz is implemented.

from mips-traces.

@NullByte4532 Any news about bubble sort?

from mips-traces.