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

divmod, partition, sort fixes (#2302)

Browse Source
This commit is contained in:
Awni Hannun
2025-06-16 18:49:32 -07:00
committed by GitHub
parent bc53f8293f
commit b8022c578a
8 changed files with 271 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
mlx/backend/cuda/binary.cu
View File

@@ -125,13 +125,12 @@ constexpr bool supports_binary_op() {
template <typename Op>
void binary_op_gpu_inplace(
const std::vector<array>& inputs,
std::vector<array>& outputs,
array& out,
std::string_view op,
const Stream& s) {
assert(inputs.size() > 1);
const auto& a = inputs[0];
const auto& b = inputs[1];
auto& out = outputs[0];
if (out.size() == 0) {
return;
}
@@ -146,7 +145,6 @@ void binary_op_gpu_inplace(
if constexpr (cu::supports_binary_op<Op, CTYPE_IN, CTYPE_OUT>()) {
using InType = cuda_type_t<CTYPE_IN>;
using OutType = cuda_type_t<CTYPE_OUT>;
auto bopt = get_binary_op_type(a, b);
if (bopt == BinaryOpType::General) {
auto [shape, strides] = collapse_contiguous_dims(a, b, out);
@@ -219,20 +217,6 @@ void binary_op_gpu_inplace(
});
}
template <typename Op>
void binary_op_gpu(
const std::vector<array>& inputs,
std::vector<array>& outputs,
std::string_view op,
const Stream& s) {
auto& a = inputs[0];
auto& b = inputs[1];
auto bopt = get_binary_op_type(a, b);
set_binary_op_output_data(a, b, outputs[0], bopt);
set_binary_op_output_data(a, b, outputs[1], bopt);
binary_op_gpu_inplace<Op>(inputs, outputs, op, s);
}
template <typename Op>
void binary_op_gpu(
const std::vector<array>& inputs,
@@ -243,8 +227,7 @@ void binary_op_gpu(
auto& b = inputs[1];
auto bopt = get_binary_op_type(a, b);
set_binary_op_output_data(a, b, out, bopt);
std::vector<array> outputs{out};
binary_op_gpu_inplace<Op>(inputs, outputs, op, s);
binary_op_gpu_inplace<Op>(inputs, out, op, s);
}
#define BINARY_GPU(func) \
@@ -254,14 +237,6 @@ void binary_op_gpu(
binary_op_gpu<cu::func>(inputs, out, get_primitive_string(this), s); \
}
#define BINARY_GPU_MULTI(func) \
void func::eval_gpu( \
const std::vector<array>& inputs, std::vector<array>& outputs) { \
nvtx3::scoped_range r(#func "::eval_gpu"); \
auto& s = outputs[0].primitive().stream(); \
binary_op_gpu<cu::func>(inputs, outputs, get_primitive_string(this), s); \
}
BINARY_GPU(Add)
BINARY_GPU(ArcTan2)
BINARY_GPU(Divide)