Skip to content

Fix GPU Softmax NaN propagation for mixed infinite inputs #33498

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
AyusKumarPathak wants to merge 8 commits into
base: master
Choose a base branch
from
+130 −1
Open

Fix GPU Softmax NaN propagation for mixed infinite inputs #33498

Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
a44b14a
Fix GPU Softmax NaN propagation for mixed infinite inputs
AyusKumarPathak Jan 7, 2026
eb3e17b
Add comprehensive GPU Softmax edge-case tests for INF and NaN handling
AyusKumarPathak Jan 8, 2026
ce97a90
Move GPU-specific Softmax edge-case tests into GPU plugin
AyusKumarPathak Jan 13, 2026
6a912b6
Address reviewer feedback on GPU softmax tests
AyusKumarPathak Jan 16, 2026
9c3ae64
Update src/plugins/intel_gpu/tests/functional/single_layer_tests/soft…
AyusKumarPathak Jan 16, 2026
f561033
updated
AyusKumarPathak Jan 16, 2026
c46613a
Apply reviewer suggestions to GPU softmax tests
AyusKumarPathak Jan 16, 2026
2e50813
Apply reviewer suggestions to GPU softmax tests
AyusKumarPathak Jan 16, 2026
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
25 changes: 25 additions & 0 deletions src/plugins/intel_gpu/src/kernel_selector/cl_kernels/softmax_gpu_ref.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,31 @@ KERNEL(softmax)(
max_value = max(max_value, in);
data[cls*TMP_CLASS_PITCH] = in;
}
// Handle IEEE-754 case when max_value is INF
if (isinf((float)max_value)) {
for (cls = 0; cls < class_num; ++cls) {
ACCUMULATOR_TYPE v = data[cls * TMP_CLASS_PITCH];
if (v == max_value)
data[cls * TMP_CLASS_PITCH] = (ACCUMULATOR_TYPE)NAN;
else
data[cls * TMP_CLASS_PITCH] = (ACCUMULATOR_TYPE)0.0f;
}

// Write results and exit
for (cls = 0; cls < class_num; ++cls) {
#if INPUT0_SIMPLE == 1
const uint output_idx = out_depth_offset + cls*OUTPUT_CLASS_PITCH;
#else
#if INPUT0_DIMS == 5
const uint output_idx = OUTPUT_GET_INDEX(b + *b_offset, f + *f_offset, z + *z_offset, y + *y_offset, x + *x_offset);
#else
const uint output_idx = OUTPUT_GET_INDEX(b + *b_offset, f + *f_offset, y + *y_offset, x + *x_offset);
#endif
#endif
output[output_idx] = data[cls * TMP_CLASS_PITCH];
}
return;
}
Comment on lines +109 to +133
Copy link
Contributor

@e-ddykim e-ddykim Jan 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we need to apply fused ops for the issued case too.
Please check my suggestion below.

Suggested change
// Handle IEEE-754 case when max_value is INF
if (isinf((float)max_value)) {
for (cls = 0; cls < class_num; ++cls) {
ACCUMULATOR_TYPE v = data[cls * TMP_CLASS_PITCH];
if (v == max_value)
data[cls * TMP_CLASS_PITCH] = (ACCUMULATOR_TYPE)NAN;
else
data[cls * TMP_CLASS_PITCH] = (ACCUMULATOR_TYPE)0.0f;
}
// Write results and exit
for (cls = 0; cls < class_num; ++cls) {
#if INPUT0_SIMPLE == 1
const uint output_idx = out_depth_offset + cls*OUTPUT_CLASS_PITCH;
#else
#if INPUT0_DIMS == 5
const uint output_idx = OUTPUT_GET_INDEX(b + *b_offset, f + *f_offset, z + *z_offset, y + *y_offset, x + *x_offset);
#else
const uint output_idx = OUTPUT_GET_INDEX(b + *b_offset, f + *f_offset, y + *y_offset, x + *x_offset);
#endif
#endif
output[output_idx] = data[cls * TMP_CLASS_PITCH];
}
return;
}
for (cls = 0; cls < class_num; ++cls) {
// Handle IEEE-754 case when max_value is INF
if (isinf((float)max_value)) {
if (data[cls*TMP_CLASS_PITCH] == max_value)
data[cls*TMP_CLASS_PITCH] = TO_ACCUMULATOR_TYPE(NAN);
else
data[cls*TMP_CLASS_PITCH] = TO_ACCUMULATOR_TYPE(0.0f);
} else {
ACCUMULATOR_TYPE t = native_exp(data[cls*TMP_CLASS_PITCH] - max_value);
denominator += t;
data[cls*TMP_CLASS_PITCH] = t;
}
}
....
for (cls = 0; cls < class_num; ++cls) {
ACCUMULATOR_TYPE res = data[cls*TMP_CLASS_PITCH];
if (!isinf((float)max_value)) {
res = res / denominator;
}


// TODO: currently we calculate on float32 because it's lot of "add" operation and it stuck on the value "8192.0f"
ACCUMULATOR_TYPE denominator = 0.0;
Expand Down
Loading