Warning: cast from 'const __global float *' to 'const __global float2 *'
newling opened this issue · 3 comments
newling commented
Some kernels give a cast-align warning,
cast from 'const __global float *' to 'const __global float2 *'
increases required alignment from 4 to 8 [-Werror,-Wcast-align]
const __global TVFLOATA * a_vec = (const __global TVFLOATA * )a;
Is this something to be concerned about? I see in MIOpen that this warning is suppressed (line).
Is this only relevant for non-homogeneous datatypes? I don't know, could someone with more experience please comment here.
dagamayank commented
@alyashev - Can you comment here.
alyashev commented
It's OK. our HW does not need a particular alignment to handle a double float. the only concern is with out a care it might step OOB.