Floating Point Operation Exception Masks

From RAD Studio Athens (R120), all of the floating-point-related exception masks will be turned ON. The applications will no longer raise Floating Point exceptions on all platforms as default.

The default behavior of Delphi RTL is to disable all floating-point-related exceptions.

Platform	Control variable name in System	since 11.3	after 12.0
Win/x32	Default8087CW	$1332	$033F
Win/x64	DefaultMXCSR	$1900	$1F80
macOS/x64	DefaultFPEnv	$0032	$003F
macOS/Arm64	DefaultFPSCR	$0000_0700	$0000_0000
iOS/Arm64	DefaultFPSCR	$0000_0700	$0000_0000
iOS-Sim/Arm64	DefaultFPSCR	$0000_0700	$0000_0000
Android/Arm32	DefaultFPSCR	$0000_0700	$0000_0000
Android/Arm64	DefaultFPSCR	$0000_0700	$0000_0000
Linux/x64	DefaultFPEnv	$0032	$003F

Old and New behavior table

Platform	Program Type	Old Behavior		New Behavior
		1/0	sqrt (-1)	1/0	sqrt (-1)
Win/x32 Win/x64	Default / Console / VCL	EZeroDivide exception The result is undefined.	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
Win/x32 Win/x64	FMX	No exception The result is Inf	No exception The result is NAN	No exception The result is Inf	No exception NAN
macOS/x64	FMX/Console	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
macOS/Arm64	FMX/Console	No exception The result is Inf	No exception The result is NAN	No exception The result is Inf	No exception The result is NAN
iOS/Arm64	FMX	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
iOS-Sim/Arm64	FMX	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
Android/Arm32	FMX	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
Android/Arm64	FMX	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN
Linux/x64	Console	No exception The result is Inf	EInvalidOp exception The result is undefined.	No exception The result is Inf	No exception The result is NAN

How to Restore Old Behavior

If you want to revert the exception behavior to the previous one, call SetExceptionMask in System.Math with the following options into the application's main source block ( Usually, .DPR file )

System.math.SetExceptionMask( [exPrecision, exUnderflow, exDenormalized]); // Enable exInvalidOp, exZeroDivide, and exOverflow exceptions

For example:

uses
  System.Math, // Add System.Math unit
  <span class="glossaryTerm">Vcl<span class="tooltip"><p>Visual Component Library</p></span></span>.Forms,
  Unit2 in 'Unit2.pas' {Form2};
{$R *.res}
begin
  System.math.SetExceptionMask( [exPrecision, exUnderflow, exDenormalized]); // Add this line.
  Application.Initialize;.

Topics

• Floating-Point Exception: Divide by 0, Domain or Overflow
• Floating-Point Exception: Partial Loss of Precision or Underflow
• Floating-Point Exception: Stack Fault

See Also

• Floating-Point Exceptions
• System.Math.SetExceptionMask
• System.Math.GetExceptionMask
• About Floating-Point Arithmetic
• Exceptions (Delphi)