Skip to content

[SYCL][libdevice] return qNaN for (float/double)sNaN to bfloat16 conversion #23084

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
KornevNikita merged 1 commit into from
Sep 4, 2026
+8 −7
Merged

[SYCL][libdevice] return qNaN for (float/double)sNaN to bfloat16 conversion #23084

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 4 additions & 3 deletions libdevice/imf_bf16.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,8 @@ __float2bfloat16(float f, __iml_rounding_mode rounding_mode) {
return bf16_sign ? 0xFF80 : 0x7F80;
else {
// input fp32 val is Nan.
bf16_mant = (f_mant & 0x400000) ? 0x40 : 0x3F;
// returns a quiet NaN
bf16_mant = 0x40;
return (bf16_sign << 15) | (bf16_exp << 7) | bf16_mant;
}
}
Expand Down Expand Up @@ -90,8 +91,8 @@ static _iml_bf16_internal __double2bfloat16(double d) {
if (!fp64_mant) {
return bf16_sign ? 0xFF80 : 0x7F80;
} else {
// returns a signaling or quiet Nan
return (fp64_mant & 0x8'0000'0000'0000) ? 0x7FC0 : 0x7F81;
// returns a quiet Nan
return 0x7FC0;
}
}

Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/DeviceLib/imf/double2bfloat16.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,11 +10,11 @@ int check_nan_convert(std::initializer_list<double> Inputs,
const std::vector<uint16_t> &Outputs) {
assert(Inputs.size() == Outputs.size());
size_t Idx = 0;
bool Signaling = false;
bool Signaling = true;
Comment thread
KseniyaTikhomirova marked this conversation as resolved.
for (const double *It = Inputs.begin(); It != Inputs.end(); ++It) {
uint16_t bf16_bits = Outputs[Idx];
if (is_bfloat16_nan(bf16_bits, Signaling)) {
if (Signaling != is_signaling_nan(*It))
if (Signaling)
return 1;
} else
return 1;
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/DeviceLib/imf/float2bfloat16.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,11 +10,11 @@ int check_nan_convert(std::initializer_list<float> Inputs,
const std::vector<uint16_t> &Outputs) {
assert(Inputs.size() == Outputs.size());
size_t Idx = 0;
bool Signaling = false;
bool Signaling = true;
for (const float *It = Inputs.begin(); It != Inputs.end(); ++It) {
uint16_t bf16_bits = Outputs[Idx];
if (is_bfloat16_nan(bf16_bits, Signaling)) {
if (Signaling != is_signaling_nan(*It))
if (Signaling)
return 1;
} else
return 1;
Expand Down
Loading