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 some internal GPU apis (#1177)

Browse Source 
* Add unary/binary/ternay/slice/concat internal GPU ops

* add pad internal op

* formatting + no_cpu fix
This commit is contained in:
Alex Barron
2024-06-04 09:24:26 -07:00
committed by GitHub
parent ea9090bbc4
commit 375a8bbdcc
17 changed files with 449 additions and 203 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
mlx/backend/common/common.cpp
View File

@@ -250,49 +250,6 @@ void Split::eval(
}
}
std::tuple<bool, int64_t, std::vector<int64_t>> Slice::prepare_slice(
const array& in) {
int64_t data_offset = 0;
bool copy_needed = false;
std::vector<int64_t> inp_strides(in.ndim(), 0);
for (int i = 0; i < in.ndim(); ++i) {
data_offset += start_indices_[i] * in.strides()[i];
inp_strides[i] = in.strides()[i] * strides_[i];
copy_needed |= strides_[i] < 0;
}
return std::make_tuple(copy_needed, data_offset, inp_strides);
}
void Slice::shared_buffer_slice(
const array& in,
const std::vector<size_t>& out_strides,
size_t data_offset,
array& out) {
// Compute row/col contiguity
auto [data_size, is_row_contiguous, is_col_contiguous] =
check_contiguity(out.shape(), out_strides);
auto flags = in.flags();
flags.row_contiguous = is_row_contiguous;
flags.col_contiguous = is_col_contiguous;
if (data_size == 1) {
// Broadcasted scalar array is contiguous.
flags.contiguous = true;
} else if (data_size == in.data_size()) {
// Means we sliced a broadcasted dimension so leave the "no holes" flag
// alone.
} else {
// We sliced something. So either we are row or col contiguous or we
// punched a hole.
flags.contiguous &= flags.row_contiguous || flags.col_contiguous;
}
out.copy_shared_buffer(in, out_strides, flags, data_size, data_offset);
}
std::tuple<int64_t, std::vector<int64_t>> SliceUpdate::prepare_slice(
const array& in) {
int64_t data_offset = 0;