Skip to content

fix blha_get_max_len big tensor #73712

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 2, 2025
Merged

fix blha_get_max_len big tensor #73712

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion paddle/phi/kernels/fusion/gpu/blha_get_max_len.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@ void GetMaxLenTensor(const phi::GPUContext& dev_ctx,
max_len_tensor.Resize({{1}});
auto* max_len_tensor_data = dev_ctx.template Alloc<int>(
&max_len_tensor, max_len_tensor.numel() * sizeof(int));
const int bsz = batch_size.dims()[0];
const int64_t bsz = batch_size.dims()[0];
constexpr int blockSize = 128;
int max_len_cpu = 0;
GetMaxLenKernel<blockSize><<<1, blockSize, 0, dev_ctx.stream()>>>(
Expand Down
6 changes: 3 additions & 3 deletions paddle/phi/kernels/fusion/gpu/block_attn.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4342,14 +4342,14 @@ struct MaxOp {
template <int THREADBLOCK_SIZE>
__global__ void GetMaxLenKernel(const int *seq_lens,
int *max_len,
const int batch_size) {
const int tid = threadIdx.x;
const int64_t batch_size) {
const int64_t tid = threadIdx.x;

typedef cub::BlockReduce<int, THREADBLOCK_SIZE> BlockReduce;
__shared__ typename BlockReduce::TempStorage temp_storage;

int max_len_this_thread = 0;
for (int i = tid; i < batch_size; i += blockDim.x) {
for (int64_t i = tid; i < batch_size; i += blockDim.x) {
max_len_this_thread = max(seq_lens[i], max_len_this_thread);
}
int total =
Expand Down
Loading