Data Overflow During the Conversion from the Double-Precision Floating-Point Type to the Integer Type

Symptom

When a C or C++ double-precision floating-point value is converted to an integer, the performance of the Kunpeng platform is different from that of the x86 platform if the integer is out of the integer range.

long aa = (long)0x7FFFFFFFFFFFFFFF; 
long bb; 
bb = (long)(aa*(double)10); // long->double->long 
//x86: aa=9223372036854775807, bb=-9223372036854775808 
//arm64:aa=9223372036854775807, bb=9223372036854775807

Cause

In the two platforms, there are two CPU architectures. The ways of implementing the arithmetic logic unit, OS, and compilers ay be different. The instruction for converting a floating-point number to an integer in x86 defines an indefinite integer value (64-bit 0x8000000000000000). In most cases, the x86 platform actually complies with this principle. However, when the double data is converted to an unsigned integer, different results occur. The processing of Kunpeng is very clear and simple. When overflow or underflow occurs, the maximum or minimum value that can be represented by an integer is reserved. Developers do not face uncertain or unexpected results.

Procedure

Adjust the code implementation based on the following data conversion table.

Conversion from double data to long data:

CPU	double Value	long Value	Description
x86	The positive value is out of the long range.	0x8000000000000000	indefinite integer value
x86	The negative value is out of the long range.	0x8000000000000000	indefinite integer value
Kunpeng	The positive value is out of the long range.	0x7FFFFFFFFFFFFFFF	Kunpeng assigns the maximum positive value to the long variable.
Kunpeng	The negative value is out of the long range.	0x8000000000000000	Kunpeng assigns the minimum negative value to the long variable.

Conversion from double data to unsigned long data:

CPU	double Value	unsigned long Value	Description
x86	The positive value is out of the long range.	0x0000000000000000	x86 assigns the minimum value 0 to the long variable.
x86	The negative value is out of the long range.	0x8000000000000000	indefinite integer value
Kunpeng	The positive value is out of the long range.	0xFFFFFFFFFFFFFFFF	Kunpeng assigns the maximum value to the unsigned long variable.
Kunpeng	The negative value is out of the long range.	0x0000000000000000	Kunpeng assigns the minimum value to the unsigned long variable.

Conversion from double data to int data:

CPU	double Value	int Value	Description
x86	The positive value is out of the int range.	0x80000000	indefinite integer value
x86	The negative value is out of the int range.	0x80000000	indefinite integer value
Kunpeng	The positive value is out of the int range.	0x7FFFFFFF	Kunpeng assigns the maximum positive value to the int variable.
Kunpeng	The negative value is out of the int range.	0x80000000	Kunpeng assigns the minimum negative number to the int variable.

Conversion from double data to unsigned int data:

CPU	double Value	unsigned int Value	Description
x86	The positive value is out of the unsigned int range.	The integer part of double modulo 2^32	x86 assigns the minimum negative value to the unsigned int variable.
x86	The negative value is out of the unsigned int range.	The integer part of double modulo 2^32	x86 assigns the minimum negative value to the unsigned int variable.
Kunpeng	The positive value is out of the unsigned int range.	0xFFFFFFFF	Kunpeng assigns the maximum positive number to the unsigned int variable.
Kunpeng	The negative value is out of the unsigned int range.	0x00000000	Kunpeng assigns the minimum negative value to the unsigned int variable.

Parent topic: Source Code Modification