Commit 7af75b2e authored by Paul's avatar Paul
Browse files

Try to fix the tests

parent 59c4cdb0
...@@ -39,6 +39,7 @@ auto nary_nonstandard(argument result, Arguments... args) ...@@ -39,6 +39,7 @@ auto nary_nonstandard(argument result, Arguments... args)
inline auto binary_broadcast(argument result, argument arg1, argument arg2) inline auto binary_broadcast(argument result, argument arg1, argument arg2)
{ {
return [=](auto f) { return [=](auto f) {
// const auto& output_shape = result.get_shape();
const auto& b_shape = arg2.get_shape(); const auto& b_shape = arg2.get_shape();
auto bdim = std::distance(b_shape.strides().begin(), auto bdim = std::distance(b_shape.strides().begin(),
std::find_if(b_shape.strides().begin(), std::find_if(b_shape.strides().begin(),
...@@ -63,13 +64,11 @@ inline auto binary_broadcast(argument result, argument arg1, argument arg2) ...@@ -63,13 +64,11 @@ inline auto binary_broadcast(argument result, argument arg1, argument arg2)
buffer[i] = yp[i]; buffer[i] = yp[i];
} }
__syncthreads(); __syncthreads();
for(size_t i = idx.local; i < bdim_len; i += nlocal) for(size_t i = idx.global; i < n; i += nglobal)
{ {
auto b = buffer[i]; auto bidx = i % bdim_len;
for(size_t j = idx.global; j < n; j += nglobal) auto b = buffer[bidx];
{ outp[i] = f(xp[i], b);
outp[j] = f(xp[j], b);
}
} }
}); });
}); });
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment