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 Packages Projects Releases Wiki Activity

Gather mm new kernel and small refactoring (#2040)

Browse Source
This commit is contained in:
Angelos Katharopoulos
2025-04-14 16:37:36 -07:00
committed by GitHub
parent e9e268336b
commit 99eefd2ec0
23 changed files with 1260 additions and 378 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
mlx/backend/common/CMakeLists.txt
View File

@@ -1,6 +1,7 @@
target_sources(
mlx
PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/compiled.cpp
PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/broadcasting.cpp
${CMAKE_CURRENT_SOURCE_DIR}/compiled.cpp
${CMAKE_CURRENT_SOURCE_DIR}/common.cpp
${CMAKE_CURRENT_SOURCE_DIR}/load.cpp
${CMAKE_CURRENT_SOURCE_DIR}/reduce.cpp

24
mlx/backend/common/broadcasting.cpp Normal file
View File

@@ -0,0 +1,24 @@
// Copyright © 2024 Apple Inc.
#include "mlx/backend/common/utils.h"
namespace mlx::core {
void broadcast(const array& in, array& out) {
if (out.size() == 0) {
out.set_data(nullptr);
return;
}
Strides strides(out.ndim(), 0);
int diff = out.ndim() - in.ndim();
for (int i = in.ndim() - 1; i >= 0; --i) {
strides[i + diff] = (in.shape()[i] == 1) ? 0 : in.strides()[i];
}
auto flags = in.flags();
if (out.size() > in.size()) {
flags.row_contiguous = flags.col_contiguous = false;
}
out.copy_shared_buffer(in, strides, flags, in.data_size());
}
} // namespace mlx::core

11
mlx/backend/common/broadcasting.h Normal file
View File

@@ -0,0 +1,11 @@
// Copyright © 2024 Apple Inc.
#pragma once
#include "mlx/array.h"
namespace mlx::core {
void broadcast(const array& in, array& out);
} // namespace mlx::core

18
mlx/backend/common/common.cpp
View File

@@ -1,6 +1,7 @@
// Copyright © 2024 Apple Inc.
#include <cassert>
#include "mlx/backend/common/broadcasting.h"
#include "mlx/backend/common/utils.h"
#include "mlx/primitives.h"
@@ -42,23 +43,6 @@ void AsStrided::eval(const std::vector<array>& inputs, array& out) {
return out.copy_shared_buffer(in, strides_, flags, data_size, offset_);
}
void broadcast(const array& in, array& out) {
if (out.size() == 0) {
out.set_data(nullptr);
return;
}
Strides strides(out.ndim(), 0);
int diff = out.ndim() - in.ndim();
for (int i = in.ndim() - 1; i >= 0; --i) {
strides[i + diff] = (in.shape()[i] == 1) ? 0 : in.strides()[i];
}
auto flags = in.flags();
if (out.size() > in.size()) {
flags.row_contiguous = flags.col_contiguous = false;
}
out.copy_shared_buffer(in, strides, flags, in.data_size());
}
void Broadcast::eval(const std::vector<array>& inputs, array& out) {
broadcast(inputs[0], out);
}