/sci/ - Science & Math » Thread #7619731

[3 / 1 / ?]

cache misses

Anonymous Tue 27 Oct 2015 17:11:40 No.7619731 View Reply Report Delete

Quoted By:

what is the difference between a cache miss and a branch miss? I have this one snippet of code that is run very frequently, something like this

void f( type** x )
{
do stuff with (*x)
}

which resulted in a 40% cache miss according to perf, with 0.5% branch misses. I changed it slightly to

void f( type** x )
{
type* temp = *x;
do stuff with temp;
}

which resulted in a drop to 16% cache miss but the branch misses when up to 5%..

What is the difference, and why did the branch misses increase?

Anonymous

Anonymous Tue 27 Oct 2015 17:39:35 No.7619788 Report Delete

Quoted By:

I'm no expert but from the sound of the definitions these occur when the data being held for processing is allocated in fragments. So you allocate data to a certain area but had cleared only parts of the data that was allocared prior to the current instruction.

1 1 1
1 1 1
1 1 1
1 1 1

if this were your data, then a cache miss might occur when your commands set the four 4 lines to execute, and then upon return, you set the second line to something new but still use the same instruction pattern but you only needed the lines with full 1 1 1 branches.

1 1 1
1 0 1 <-- meaningless change, btw
1 1 1
1 1 1

What happens here then is that the processor fetches the data at 1, 2, 3, 4 but only needed 1, 3, 4. This then means that upon reaching two and returning, the processor realizes that the data is wrong, stores 1 and then processes new instructions to fetch 3, 4 and append it to the source of 1, which essentially adds to the time complexity.

The branch misses could then have to do with the expected change in line 2 as it's expected to be the first instruction when fetching from the cache. So it fetches 1, 2, 3, 4 from the cache to save on computing time but realizes, once it reaches 2 in the cache, that it didn't need 1 and discards that data which in turn adds to your cache miss.

I think, Anon. Don't take my word as bond. This is just a guess.

Anonymous

Anonymous Tue 27 Oct 2015 17:47:15 No.7619801 Report Delete

Quoted By:

thanks, and that makes some sense. I know very little and thought i might try my hand at some mild optimization, but i think that will have to wait since doing what i did, the 5% branch misses seems to overshadow the reduction in cache misses and so the code takes longer to perform!

thanks for the info

Subject

Name

E-mail

Password

Capcode	All Only User Posts Only Verified Posts Only Moderator Posts Only Manager Posts Only Admin Posts Only Developer Posts Only Founder Posts
Show Posts	All Only With Images Only Without Images Only Spoiler Images Only Non-Spoiler Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

On these boards

Your latest searches

cache misses