113 lines
4.6 KiB
Markdown
113 lines
4.6 KiB
Markdown
|
# ArrayToFP8
|
|||
|
|
|
|||
|
|
Copies an array of type float or double into an array of type [uchar](/en/docs/basis/types/integer/integertypes#uchar) with the given format.
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
bool ArrayToFP8(
|
|||
|
|
const uchar& dst_array[], // copy to
|
|||
|
|
const float& src_array[], // copy from
|
|||
|
|
ENUM_FLOAT8_FORMAT fmt // format
|
|||
|
|
);
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
Overloading for the double type
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
bool ArrayToFP8(
|
|||
|
|
const uchar& dst_array[], // copy to
|
|||
|
|
const double& src_array[], // copy from
|
|||
|
|
ENUM_FLOAT8_FORMAT fmt // format
|
|||
|
|
);
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
Parameters
|
|||
|
|
|
|||
|
|
dst_array[]
|
|||
|
|
|
|||
|
|
[out] Receiver array or type uchar.
|
|||
|
|
|
|||
|
|
src_array[]
|
|||
|
|
|
|||
|
|
[in] Source array of type float or double.
|
|||
|
|
|
|||
|
|
fmt
|
|||
|
|
|
|||
|
|
[in] Copying format from the [ENUM_FLOAT8_FORMAT](/en/docs/onnx/onnx_structures#enum_float8_format) enumeration.
|
|||
|
|
|
|||
|
|
Return Value
|
|||
|
|
|
|||
|
|
Returns true if successful or false otherwise.
|
|||
|
|
|
|||
|
|
Note
|
|||
|
|
|
|||
|
|
All kinds of FP8 format are defined in the [ENUM_FLOAT8_FORMAT](/en/docs/onnx/onnx_structures#enum_float8_format) enumeration and are used in MQL5 only for operations with [ONNX models](/en/docs/onnx).
|
|||
|
|
|
|||
|
|
The function converts input parameters of type float or double into one of FP8 types. These input parameters are then used in the [OnnxRun](/en/docs/onnx/onnxrun) function.
|
|||
|
|
|
|||
|
|
FP8 (8-bit floating point) is one of the data types used to represent floating point numbers. In FP8, each number is represented by 8 data bits, typically divided into three components: sign, exponent and mantissa. This format offers a balance between accuracy and storage efficiency, making it attractive for applications that require memory and computational efficiency.
|
|||
|
|
|
|||
|
|
By employing compact number representation, FP8 reduces memory requirements and accelerates calculations. In addition, FP8 can be useful for implementing low-level operations such as arithmetic calculations and signal processing.
|
|||
|
|
|
|||
|
|
Example: function from the article [Working with ONNX models in float16 and float8 formats ](https://www.mql5.com/ru/articles/14330)
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
//+------------------------------------------------------------------+
|
|||
|
|
//| RunCastFloat8Float |
|
|||
|
|
//+------------------------------------------------------------------+
|
|||
|
|
bool RunCastFloat8ToFloat(long model_handle,const ENUM_FLOAT8_FORMAT fmt)
|
|||
|
|
{
|
|||
|
|
PrintFormat("TEST: %s(%s)",__FUNCTION__,EnumToString(fmt));
|
|||
|
|
//---
|
|||
|
|
float test_data[15] = {1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
|
|||
|
|
uchar data_float8[15] = {};
|
|||
|
|
if(!ArrayToFP8(data_float8,test_data,fmt))
|
|||
|
|
{
|
|||
|
|
Print("error in ArrayToFP8. error code=",GetLastError());
|
|||
|
|
OnnxRelease(model_handle);
|
|||
|
|
return(false);
|
|||
|
|
}
|
|||
|
|
U<uchar> input_float8_values[3*5];
|
|||
|
|
U<float> output_float_values[3*5];
|
|||
|
|
float test_data_float[];
|
|||
|
|
//--- convert float8 to float
|
|||
|
|
if(!ArrayFromFP8(test_data_float,data_float8,fmt))
|
|||
|
|
{
|
|||
|
|
Print("error in ArrayFromFP8. error code=",GetLastError());
|
|||
|
|
OnnxRelease(model_handle);
|
|||
|
|
return(false);
|
|||
|
|
}
|
|||
|
|
for(uint i=0; i<data_float8.Size(); i++)
|
|||
|
|
{
|
|||
|
|
input_float8_values[i].value=data_float8[i];
|
|||
|
|
PrintFormat("%d input value =%f Hex float8 = %s ushort value=%d",i,test_data_float[i],ArrayToHexString(input_float8_values[i].uc),input_float8_values[i].value);
|
|||
|
|
}
|
|||
|
|
Print("ONNX input array: ",ArrayToString(input_float8_values));
|
|||
|
|
//--- execute model (convert float8 to float using ONNX)
|
|||
|
|
if(!OnnxRun(model_handle,ONNX_NO_CONVERSION,input_float8_values,output_float_values))
|
|||
|
|
{
|
|||
|
|
PrintFormat("error in OnnxRun. error code=%d",GetLastError());
|
|||
|
|
OnnxRelease(model_handle);
|
|||
|
|
return(false);
|
|||
|
|
}
|
|||
|
|
Print("ONNX output array: ",ArrayToString(output_float_values));
|
|||
|
|
//--- calculate error (compare ONNX and ArrayFromFP8 results)
|
|||
|
|
double sum_error=0.0;
|
|||
|
|
for(uint i=0; i<test_data.Size(); i++)
|
|||
|
|
{
|
|||
|
|
double delta=test_data_float[i]-(double)output_float_values[i].value;
|
|||
|
|
sum_error+=MathAbs(delta);
|
|||
|
|
PrintFormat("%d output float %f = %s difference=%f",i,output_float_values[i].value,ArrayToHexString(output_float_values[i].uc),delta);
|
|||
|
|
}
|
|||
|
|
//---
|
|||
|
|
PrintFormat("%s(%s): sum_error=%f\n",__FUNCTION__,EnumToString(fmt),sum_error);
|
|||
|
|
return(true);
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
See also
|
|||
|
|
|
|||
|
|
[ArrayFromFP8](/en/docs/array/arrayfromfp8), [ArrayCopy](/en/docs/array/arraycopy)
|