Skip to content

GitLab

Commit 9bf7ed8b authored by Paul's avatar Paul
Browse files

Fix

parent 4064d804
Show whitespace changes
Inline Side-by-side
Showing with 7 additions and 8 deletions
src/targets/gpu/jit/pointwise.cpp
View file @ 9bf7ed8b
...@@ -73,15 +73,14 @@ struct pointwise_compiler : compiler<pointwise_compiler> ...@@ -73,15 +73,14 @@ struct pointwise_compiler : compiler<pointwise_compiler>
options.virtual_inputs = reduce_dims(inputs); options.virtual_inputs = reduce_dims(inputs);
options.params = "-Wno-float-equal"; options.params = "-Wno-float-equal";
auto axis = find_fast_axis(options.virtual_inputs); auto axis = find_fast_axis(options.virtual_inputs);
auto vec_size = vectorize_elements(axis, options.virtual_inputs); auto vec = vectorize::elements(axis, options.virtual_inputs);
auto preloads = preload(axis, options.virtual_inputs); auto preloads = preload::broadcasts(axis, options.virtual_inputs);
auto is_preloading =
std::accumulate(preloads.begin(), preloads.end(), false, std::logical_or<>{});
options.kernel_name = v.get("kernel", "kernel"); options.kernel_name = v.get("kernel", "kernel");
options.set_launch_params(v, options.set_launch_params(
v,
compute_global_for(ctx, compute_global_for(ctx,
options.output.elements() / vec_size, options.output.elements() / vec.size,
oversubscribe_if(not is_preloading))); oversubscribe_if(not preloads.is_preloading())));
auto src = interpolate_string(pointwise_kernel, auto src = interpolate_string(pointwise_kernel,
{{"kernel", options.kernel_name}, {{"kernel", options.kernel_name},
{"params", enum_params(inputs.size(), "void * private_p")}, {"params", enum_params(inputs.size(), "void * private_p")},
......
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