altera-single-work-item-barrier¶
Finds OpenCL kernel functions that call a barrier function but do not call
an ID function (get_local_id, get_local_id, get_group_id, or
get_local_linear_id).
These kernels may be viable single work-item kernels, but will be forced to execute as NDRange kernels if using a newer version of the Altera Offline Compiler (>= v17.01).
If using an older version of the Altera Offline Compiler, these kernel functions will be treated as single work-item kernels, which could be inefficient or lead to errors if NDRange semantics were intended.
Based on the Altera SDK for OpenCL: Best Practices Guide.
Examples:
// error: function calls barrier but does not call an ID function.
void __kernel barrier_no_id(__global int * foo, int size) {
for (int i = 0; i < 100; i++) {
foo[i] += 5;
}
barrier(CLK_GLOBAL_MEM_FENCE);
}
// ok: function calls barrier and an ID function.
void __kernel barrier_with_id(__global int * foo, int size) {
for (int i = 0; i < 100; i++) {
int tid = get_global_id(0);
foo[tid] += 5;
}
barrier(CLK_GLOBAL_MEM_FENCE);
}
// ok with AOC Version 17.01: the reqd_work_group_size turns this into
// an NDRange.
__attribute__((reqd_work_group_size(2,2,2)))
void __kernel barrier_with_id(__global int * foo, int size) {
for (int i = 0; i < 100; i++) {
foo[tid] += 5;
}
barrier(CLK_GLOBAL_MEM_FENCE);
}