This (distilled) kernel produces {9.0, 8.0, 6.0, garbage, 9.0, 8.0, 6.0, garbage, ... repeats} like it's using a double4 with "w" coordinate uninitialized... Am I doing something wrong? This is causing premature hair loss...
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
__kernel void
rk4_test( __global double3* out)
{
int obj1 = get_global_id(0);
obj_acceleration_exp[obj1] = (double3)(9.0, 8.0, 6.0);
}
My system is Linux with a Radeon 6970 running OpenCL 1.1 AMD-APP (898.1).
Thanks for your time.
- Beau V.C. Bellamy
Solved! Go to Solution.
Please refer to the folllowing post:
http://devgurus.amd.com/message/1279719#1279719
In general, when using double3 the memory access is aligned as double4, this is not a bug.
Sorry, That was a miswrite..
obj_acceleration_exp[obj1] = (double3)(9.0, 8.0, 6.0);
should read as
out[obj1] = (double3)(9.0, 8.0, 6.0);
Please refer to the folllowing post:
http://devgurus.amd.com/message/1279719#1279719
In general, when using double3 the memory access is aligned as double4, this is not a bug.
Thank you very much for clarifying that.