You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Dynamic-compiler format, that whips up a semi optimized OpenCL kernel and runs it.
You could give this a try:
Take the existing dynamic_compiler_fmt.c and make a new one, dynamic_compiler_opencl_fmt.c. Within this format you could 'obtain' a valid working dynamic script, which is basically a step by step explanation of what needs to be done. Then generate your kernel based upon those instructions. I can easily provide a hook into the dyna-compiler to simply return the script. I had a function at one time (but removed it), where you could call the compiler (gives you a handle), and then pass that handle back to a function to get the script. I think I removed that code (just stub functions) here: 329c4ab
But adding things like this should be very easy, or even just a mode of the compiler to build the script only, and not link into the dynamic engine.
This is what I meant. We actually know how to do it, we just lack the real incentive. We could easily use your existing dynamic functions more or less as-is, and in OpenCL they'd be functions or macros or what not and it wouldn't matter. It would rock.
Given we do a run-time compile of OpenCL kernels, "how hard can it be" to do things like
It's just a matter of really wanting to do it. Both ideas have tumbled around in my head for quite a while.
The text was updated successfully, but these errors were encountered: