Skip to content

Commit 0fdf0f1

Browse files
authored
[PHI] Fix contiguous kernel for big tensor (#72705)
1 parent 5edb171 commit 0fdf0f1

File tree

1 file changed

+2
-2
lines changed

1 file changed

+2
-2
lines changed

paddle/phi/kernels/gpu/contiguous_kernel.cu

+2-2
Original file line numberDiff line numberDiff line change
@@ -287,7 +287,7 @@ bool LaunchContiguousCazeOneKernel(
287287
const phi::Array<int64_t, phi::DDim::kMaxRank + 1>& input_stride,
288288
const phi::Array<int64_t, phi::DDim::kMaxRank + 1>& input_dims,
289289
int rank,
290-
int numel,
290+
int64_t numel,
291291
T* output_data) {
292292
dim3 grid(1, 1, 1), block(1, 1, 1);
293293
phi::Array<int64_t, 6> cur_input_dims;
@@ -427,7 +427,7 @@ void LaunchContiguousDefaultKernel(
427427
const phi::Array<int64_t, phi::DDim::kMaxRank + 1>& input_stride,
428428
const phi::Array<int64_t, phi::DDim::kMaxRank + 1>& input_dims,
429429
int rank,
430-
int numel,
430+
int64_t numel,
431431
T* output_data) {
432432
int64_t block = 512;
433433
int64_t grid = (numel + block - 1) / block;

0 commit comments

Comments
 (0)