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

Add contiguous_copy_gpu util for copying array (#2379)

Browse Source
This commit is contained in:
Cheng
2025-07-18 22:44:25 +09:00
committed by GitHub
parent 31fc530c76
commit 45adec102c
20 changed files with 40 additions and 67 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
mlx/backend/metal/quantized.cpp
View File

@@ -20,8 +20,7 @@ namespace {
inline array
ensure_row_contiguous(const array& x, metal::Device& d, const Stream& s) {
if (!x.flags().row_contiguous) {
array x_copy(x.shape(), x.dtype(), nullptr, {});
copy_gpu(x, x_copy, CopyType::General, s);
array x_copy = contiguous_copy_gpu(x, s);
d.add_temporary(x_copy, s.index);
return x_copy;
} else {
@@ -38,8 +37,7 @@ inline array ensure_row_contiguous_matrix(
if (stride_0 == x.shape(-1) && stride_1 == 1) {
return x;
} else {
array x_copy(x.shape(), x.dtype(), nullptr, {});
copy_gpu(x, x_copy, CopyType::General, s);
array x_copy = contiguous_copy_gpu(x, s);
d.add_temporary(x_copy, s.index);
return x_copy;
}