zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-10-22 02:58:16 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

LogCumSumExp (#2069)

Browse Source
This commit is contained in:
Yury Popov
2025-04-13 11:27:29 +03:00
committed by GitHub
parent 7275ac7523
commit e9e268336b
15 changed files with 209 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
mlx/backend/metal/kernels/scan.h
View File

@@ -2,6 +2,8 @@
#pragma once
#include "mlx/backend/metal/kernels/binary_ops.h"
#define DEFINE_SIMD_SCAN() \
template <typename T, metal::enable_if_t<sizeof(T) < 8, bool> = true> \
T simd_scan(T val) { \
@@ -139,6 +141,29 @@ struct CumMin {
}
};
template <typename U>
struct CumLogaddexp {
static constexpr constant U init = Limits<U>::min;
template <typename T>
U operator()(U a, T b) {
return LogAddExp{}(a, static_cast<U>(b));
}
U simd_scan(U x) {
for (int i = 1; i <= 16; i *= 2) {
U other = simd_shuffle_and_fill_up(x, init, i);
x = LogAddExp{}(x, other);
}
return x;
}
U simd_exclusive_scan(U x) {
x = simd_scan(x);
return simd_shuffle_and_fill_up(x, init, 1);
}
};
template <typename T, typename U, int N_READS, bool reverse>
inline void load_unsafe(U values[N_READS], const device T* input) {
if (reverse) {

5
mlx/backend/metal/kernels/scan.metal
View File

@@ -101,4 +101,7 @@ instantiate_scan_helper(min_int64_int64, int64_t, int64_t, CumMi
instantiate_scan_helper(min_float16_float16, half, half, CumMin, 4)
instantiate_scan_helper(min_float32_float32, float, float, CumMin, 4)
instantiate_scan_helper(min_bfloat16_bfloat16, bfloat16_t, bfloat16_t, CumMin, 4)
instantiate_scan_helper(min_complex64_complex64, complex64_t, complex64_t, CumMin, 2) // clang-format on
instantiate_scan_helper(min_complex64_complex64, complex64_t, complex64_t, CumMin, 2)
instantiate_scan_helper(logaddexp_float16_float16, half, half, CumLogaddexp, 4)
instantiate_scan_helper(logaddexp_float32_float32, float, float, CumLogaddexp, 4)
instantiate_scan_helper(logaddexp_bfloat16_bfloat16, bfloat16_t, bfloat16_t, CumLogaddexp, 4) // clang-format on

3
mlx/backend/metal/scan.cpp
View File

@@ -60,6 +60,9 @@ void Scan::eval_gpu(const std::vector<array>& inputs, array& out) {
case Scan::Min:
reduce_type = "min";
break;
case Scan::LogAddExp:
reduce_type = "logaddexp";
break;
}
kname << reduce_type << "_" << type_to_name(in) << "_" << type_to_name(out);
auto kernel = get_scan_kernel(