saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-06-29 03:30:52 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merged PR 4687622: ORT DML EP Pads-11 does not support float16 (falls back to CPU)

Browse source 
Pad-11 float16 wasn't registered because there was no way to read constant CPU tensor inputs of that data type, without a dependency on half or ORT's MLFloat16 type. Since the half package is WindowsAI specific (not used in ORT), since OperatorHelper avoids dependencies on ORT-specific data types like MLFloat16, and since we only need a small aspect of it, just add a helper function.

Related work items: #26007728
This commit is contained in:
Dwayne Robinson 2020-05-20 00:11:19 +00:00
parent 4e84df25e7
commit 34cdd9ab0e
2 changed files with 32 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
onnxruntime/core/providers/dml/DmlExecutionProvider/src/Operators/OperatorRegistration.cpp
View file

@ -275,7 +275,6 @@ const static SupportedTensorDataTypes supportedTypeListLogicalComparison9[2] = /
const static SupportedTensorDataTypes supportedTypeListSigned[1] = { SupportedTensorDataTypes::Float16to32 | SupportedTensorDataTypes::Int32 | SupportedTensorDataTypes::Int16 | SupportedTensorDataTypes::Int8 };
const static SupportedTensorDataTypes supportedTypeListRange[1] = {SupportedTensorDataTypes::Int16|SupportedTensorDataTypes::Int32|SupportedTensorDataTypes::Float32};
const static SupportedTensorDataTypes supportedTypeListInteger[3] = {SupportedTensorDataTypes::Int8|SupportedTensorDataTypes::UInt8, SupportedTensorDataTypes::Int8|SupportedTensorDataTypes::UInt8, SupportedTensorDataTypes::Int32 };
const static SupportedTensorDataTypes supportedTypeListPadWithoutFloat16[1] = { SupportedTensorDataTypes::Int8to32 | SupportedTensorDataTypes::Float32 };
const static SupportedTensorDataTypes supportedTypeListQLinearMatMul[3] = {
SupportedTensorDataTypes::Int8|SupportedTensorDataTypes::UInt8,
SupportedTensorDataTypes::Int8|SupportedTensorDataTypes::UInt8,
@ -358,7 +357,7 @@ const static OperatorRegistrationInformation operatorRegistrationInformationTabl
{REG_INFO_VER( 10, Slice, typeNameListSlice10, supportedTypeListSlice10, DmGraphSupport::Supported, {1, 2, 3, 4}, std::nullopt, QuerySlice)}, // Adds negative axes.
{REG_INFO_VER( 11, Slice, typeNameListSlice10, supportedTypeListSlice10, DmGraphSupport::Supported, {1, 2, 3, 4}, std::nullopt, QuerySlice)},
{REG_INFO_VER( 7, Pad, typeNameListDefault, supportedTypeListNumericDefault, DmGraphSupport::Supported)},
{REG_INFO_VER( 11, Pad, typeNameListDefault, supportedTypeListPadWithoutFloat16, DmGraphSupport::Supported, {1, 2} /*pads, value*/)}, // https://microsoft.visualstudio.com/OS/_workitems/edit/26007728
{REG_INFO_VER( 11, Pad, typeNameListDefault, supportedTypeListNumericDefault, DmGraphSupport::Supported, {1, 2} /*pads, value*/)}, // https://microsoft.visualstudio.com/OS/_workitems/edit/26007728
{REG_INFO( 7, SpaceToDepth, typeNameListDefault, supportedTypeListScalars8to32, DmGraphSupport::Supported)},
{REG_INFO( 7, DepthToSpace, typeNameListDefault, supportedTypeListScalars8to32, DmGraphSupport::Supported)},
{REG_INFO( 11, DepthToSpace, typeNameListDefault, supportedTypeListScalars8to32, DmGraphSupport::Supported)},

32
onnxruntime/core/providers/dml/OperatorAuthorHelper/OperatorHelper.cpp
View file

@ -113,6 +113,36 @@ namespace OperatorHelper
}
}
float CastFloat16ToFloat32(uint16_t input)
{
// Promote float16m10e5s1 to float32m23e8s1.
// Note this works on machines of both ascending and descending byte
// endianness, so long as float32 and uint32 endianness match.
// It does not work for a few abberant architectures which store
// float32 and uint32 with opposite endianness.
const uint32_t float16unsignedValueMask = 0x7FFF;
const uint32_t float16signMask = 0x8000;
const uint32_t float16exponentMask = 0x7C00;
const uint32_t float32exponentMask = 0x7F800000;
uint32_t float16unsignedValue = input & float16unsignedValueMask;
uint32_t float16sign = input & float16signMask;
uint32_t float16exponent = input & float16exponentMask;
// Shift mantissa bits left (23 - 10 = 13).
// Adjust exponent bias (127 - 15 = 112, 112 << 23 == 0x38000000).
// Move sign bit to float32 MSB (32 - 16 = 16).
uint32_t float32unsignedValue = (float16unsignedValue << 13) + 0x38000000;
uint32_t float32sign = float16sign << 16;
uint32_t result = (float16exponent == 0) ? (float32unsignedValue & ~float32exponentMask) : // Denormal
(float16exponent == float16exponentMask) ? (float32unsignedValue | float32exponentMask) : // Infinity
float32unsignedValue; // Any other normal value
result |= float32sign;
return reinterpret_cast<float&>(result);
}
int64_t CastToInt64(MLOperatorTensorDataType tensorDataType, const void* p)
{
switch (tensorDataType)
@ -150,7 +180,7 @@ namespace OperatorHelper
case MLOperatorTensorDataType::Int64: return static_cast<double>(*reinterpret_cast<const int64_t*>(p));
case MLOperatorTensorDataType::String: ML_INVALID_ARGUMENT("MLOperatorTensorDataType::String type is unsupported for reading as an integer.");
case MLOperatorTensorDataType::Bool: return static_cast<double>(*reinterpret_cast<const uint8_t*>(p));
case MLOperatorTensorDataType::Float16: ML_INVALID_ARGUMENT("MLOperatorTensorDataType::Float16 type is unsupported for reading as an integer.");
case MLOperatorTensorDataType::Float16: return static_cast<double>(CastFloat16ToFloat32(*reinterpret_cast<const uint16_t*>(p)));
case MLOperatorTensorDataType::Double: return static_cast<double>(*reinterpret_cast<const double*>(p));
case MLOperatorTensorDataType::UInt32: return static_cast<double>(*reinterpret_cast<const uint32_t*>(p));
case MLOperatorTensorDataType::UInt64: return static_cast<double>(*reinterpret_cast<const uint64_t*>(p));