Sha256: f3b548575ee3e8c5998068606ba4d1971c17ad05b2d20c8e06648f2c5ea4a04d
Contents?: true
Size: 787 Bytes
Versions: 10
Compression:
Stored size: 787 Bytes
Contents
% c_dtype = dtype_to_c_type(dtype)
__kernel void mean_<%= dtype %>(__global const <%= c_dtype %> *A, __global <%= c_dtype %> *C) {
// Get the index of the current element to be processed
const int id = get_global_id(0);
int offset = (id + <%= index %>) * <%= w %>;
<%= c_dtype %> sum = 0;
<% if n > 4 %>
for(int i = 0; i < <%= n/4 %> ; i++) {
<% sums = 4.times.map do |i|
"A[offset + #{i}]"
end %>
sum += <%= sums.join(' + ') %>;
offset += 4;
}
<% if n%4!=0 %>
<% (n % 4).times do |i| %>
sum += A[offset + <%= i %>];
<% end %>
<% end %>
<% else %>
<% n.times do |i| %>
sum += A[offset + <%= i %>];
<% end %>
<% end %>
C[id] = sum / <%= n %>;
}
Version data entries
10 entries across 10 versions & 1 rubygems