cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

corry
Adept III

Is it broken or taken away?

GPU_DUMP_DEVICE_KERNEL=3  does not seem to do anything in 11.10 preview 2.  Is it broken, or was it intenionally removed?

I took out my scratch registers, because in some of my initial tests, the ISA code using A0 was slow.  As a hack, I had used x0 to fill my buffer in pieces, and when I got enough data to process, moved it into non-scratch registers with pretty good performace.  Now that I have arranged my code to what I think is optimal for the wacko "optimizations" the IL compiler is doing (see my other thread about it http://forums.amd.com/devforum/messageview.cfm?catid=390&threadid=155639&enterthread=y I figured I'd just write the big switch statements, and rid myself of the extra registers entirely.  However, SKA says it compiles fine, shows ISA code, but shows NA for all statistics.  Since it only can use 11.7, I wanted to see the ISA 11.10 produced, as my cache hit rate is now an abysmal 2.8808904e-007, and I saw a 33% decrease in performance...not quite what I was expecting...

So which is it, broken or removed?  If broken, can it be fixed for 11.10 when its released?  I don't mind downgrading, but don't want to make a habit of that.  How about posting in my other thread too!

0 Likes
2 Replies
corry
Adept III

Maybe I'm going insane, I thought I used this for IL kernels before...but now I'm starting to question that....is it supposed to work for IL calls, or just opencl calls?

Yeah, its already been a long day...

 

0 Likes

Yeah, I just had a temporary lapse in sanity...it never worked for CAL, you have the option to disassemble the IL code when you compile it...Sanity temporarily returned...

0 Likes