mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-27 20:02:15 +00:00
disable half2 kernel by dfault (#9034)
This commit is contained in:
parent
8eb6546e8e
commit
289999af35
3 changed files with 6 additions and 6 deletions
|
|
@ -34,7 +34,7 @@ using namespace ONNX_NAMESPACE;
|
|||
template <typename T>
|
||||
FastGelu<T>::FastGelu(const OpKernelInfo& op_kernel_info) : CudaKernel(op_kernel_info) {
|
||||
const TransformerOptions* options = TransformerOptions::GetInstance();
|
||||
use_half2_ = !options->DisableHalf2();
|
||||
use_half2_ = options->EnableHalf2();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
|
|
|
|||
|
|
@ -25,7 +25,7 @@ const TransformerOptions* TransformerOptions::GetInstance() {
|
|||
if (value > 0)
|
||||
std::cout << "ORT_TRANSFORMER_OPTIONS: IsPrecisionMode=" << instance.IsPrecisionMode()
|
||||
<< ",DisablePersistentSoftmax=" << instance.DisablePersistentSoftmax()
|
||||
<< ",DisableHalf2=" << instance.DisableHalf2()
|
||||
<< ",EnableHalf2=" << instance.EnableHalf2()
|
||||
<< std::endl;
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -15,12 +15,12 @@ class TransformerOptions {
|
|||
|
||||
bool DisablePersistentSoftmax() const { return disable_persistent_softmax_; }
|
||||
|
||||
bool DisableHalf2() const { return disable_half2_; }
|
||||
bool EnableHalf2() const { return enable_half2_; }
|
||||
|
||||
void Initialize(int value) {
|
||||
is_precision_mode_ = (value & 0x01) > 0;
|
||||
disable_persistent_softmax_ = (value & 0x02) > 0;
|
||||
disable_half2_ = (value & 0x04) > 0;
|
||||
enable_half2_ = (value & 0x04) > 0;
|
||||
initialized_ = true;
|
||||
}
|
||||
|
||||
|
|
@ -31,8 +31,8 @@ class TransformerOptions {
|
|||
// Disable persistent softmax.
|
||||
bool disable_persistent_softmax_{false};
|
||||
|
||||
// Disable half2 kernel.
|
||||
bool disable_half2_{false};
|
||||
// Enable half2 kernel.
|
||||
bool enable_half2_{false};
|
||||
|
||||
bool initialized_{false};
|
||||
|
||||
|
|
|
|||
Loading…
Reference in a new issue