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

Fix compilation with CUDA 11 (#2331)

Browse Source
This commit is contained in:
Cheng
2025-07-08 12:00:43 +09:00
committed by GitHub
parent 4a9b29a875
commit 2ca533b279
11 changed files with 115 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
mlx/backend/cuda/rms_norm.cu
View File

@@ -74,7 +74,7 @@ __global__ void rms_norm(
for (int r = 0; r < cuda::ceil_div(axis_size, BLOCK_DIM * N_READS); ++r) {
auto index = r * BLOCK_DIM + block.thread_rank();
T xn[N_READS];
cub::LoadDirectBlocked(index, x, xn, axis_size, 0);
cub::LoadDirectBlocked(index, x, xn, axis_size, cast_to<T>(0));
for (int i = 0; i < N_READS; ++i) {
float t = static_cast<float>(xn[i]);
normalizer += t * t;
@@ -130,7 +130,7 @@ __global__ void rms_norm_vjp(
T wn[N_READS] = {};
T gn[N_READS] = {};
auto index = r * BLOCK_DIM + block.thread_rank();
cub::LoadDirectBlocked(index, x, xn, axis_size, 0);
cub::LoadDirectBlocked(index, x, xn, axis_size, cast_to<T>(0));
cub::LoadDirectBlocked(index, g, gn, axis_size);
cub::LoadDirectBlocked(index, strided_iterator(w, w_stride), wn, axis_size);
for (int i = 0; i < N_READS; i++) {