zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Explicit barriers with concurrent dispatch (#977)

Browse Source
This commit is contained in:
Awni Hannun
2024-04-10 21:45:31 -07:00
committed by GitHub
parent 8580d997ff
commit 12d4507ee3
21 changed files with 326 additions and 267 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
mlx/backend/metal/scan.cpp
View File

@@ -52,10 +52,10 @@ void Scan::eval_gpu(const std::vector<array>& inputs, array& out) {
kname << type_to_name(in) << "_" << type_to_name(out);
auto kernel = d.get_kernel(kname.str());
auto compute_encoder = d.get_command_encoder(s.index);
auto& compute_encoder = d.get_command_encoder(s.index);
compute_encoder->setComputePipelineState(kernel);
set_array_buffer(compute_encoder, in, 0);
set_array_buffer(compute_encoder, out, 1);
compute_encoder.set_input_array(in, 0);
compute_encoder.set_output_array(out, 1);
size_t size = in.shape(axis_);
compute_encoder->setBytes(&size, sizeof(size_t), 2);
@@ -101,10 +101,10 @@ void Scan::eval_gpu(const std::vector<array>& inputs, array& out) {
kname << type_to_name(in) << "_" << type_to_name(out);
auto kernel = d.get_kernel(kname.str());
auto compute_encoder = d.get_command_encoder(s.index);
auto& compute_encoder = d.get_command_encoder(s.index);
compute_encoder->setComputePipelineState(kernel);
set_array_buffer(compute_encoder, in, 0);
set_array_buffer(compute_encoder, out, 1);
compute_encoder.set_input_array(in, 0);
compute_encoder.set_output_array(out, 1);
size_t size = in.shape(axis_);
size_t stride = in.strides()[axis_];
compute_encoder->setBytes(&size, sizeof(size_t), 2);