Skip to content

softmax: adjust vectorization length according to shape #1781

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

softmax: adjust vectorization length according to shape #1781

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ddb9267
softmax: adjust vectorization length according to shape
jianyizh Jun 26, 2025
0d4a4f9
update heristic for spatial softmax
weishi-deng Jul 9, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 31 additions & 2 deletions src/ATen/native/xpu/sycl/SoftMaxKernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -199,6 +199,32 @@ static inline void get_wgroup_size_spatial(
GroupRow = std::min(GroupRow, int(dim_size));
}

static bool inline get_spatial_vec_choice(
int bs,
int dim_size,
int inner_size,
int scalar_size,
int max_vec_size) {
if (dim_size % max_vec_size != 0)
return false;
int total_resource = syclMaxWorkItemsPerTile();
int maxWGSize = syclDeviceMaxWorkGroupSize();
int group_col_size = std::min(inner_size, SIMD32);
auto local_group_num = (inner_size + group_col_size - 1) / group_col_size;
int group_row_size = 1;
while (bs *group_row_size * local_group_num * group_col_size <
total_resource ) {
group_row_size = group_row_size << 1;
if (group_row_size * SIMD32 == maxWGSize)
break;
}
group_row_size = std::min(group_row_size, int(dim_size));
if(bs *group_row_size * local_group_num * group_col_size <=
total_resource)
return false;
return true;
}

template <
int INNER_LOOP,
int vec_size,
Expand Down Expand Up @@ -1544,6 +1570,7 @@ void spatial_softmax_forward(

constexpr int float4_size = sizeof(float) * 4;
constexpr int max_vec_size = float4_size / sizeof(inscalar_t);
constexpr int scalar_size = sizeof(inscalar_t);
constexpr int INNER_LOOP = max_vec_size * 2;

// decide vec_size: max_vec_size or 1
Expand Down Expand Up @@ -1695,14 +1722,16 @@ void spatial_softmax_forward(
}
} else {
if (can_use_32bit_index) {
if (input_start == output_start && inner_size % max_vec_size == 0) {
if (input_start == output_start &&
get_spatial_vec_choice(outer_size, dim_size, inner_size, scalar_size, max_vec_size)) {
SPATIAL_SOFTMAX_FORWARD_IMPL(
/*vec_size*/ max_vec_size, /*IndexType*/ uint32_t);
} else {
SPATIAL_SOFTMAX_FORWARD_IMPL(/*vec_size*/ 1, /*IndexType*/ uint32_t);
}
} else {
if (input_start == output_start && inner_size % max_vec_size == 0) {
if (input_start == output_start &&
get_spatial_vec_choice(outer_size, dim_size, inner_size, scalar_size, max_vec_size)) {
SPATIAL_SOFTMAX_FORWARD_IMPL(
/*vec_size*/ max_vec_size, /*IndexType*/ uint64_t);
} else {
Expand Down
Loading