Hi.

Why on Win64 on AMD64 is Extended type 64bit same as double and on Win32, Linux32 and Linux64 is 80bit. It breaks for example this construction:

procedure Swap(var X, Y: Double); overload;
procedure Swap(var X, Y: Extended); overload;

(I know that extended type is defined as float with the best precision on the given platform.)

Petr

Why on Win64 on AMD64 is Extended type 64bit same as double and on Win32, Linux32 and Linux64 is 80bit.

Because the original win64 releases did not support the use of the 80x87 floating point unit, and it's still considered to be deprecated. The SSE/SSE2/SSE3/SSE4 units do not support the 80 bit floating point type.

It breaks for example this construction: procedure Swap(var X, Y: Double); overload; procedure Swap(var X, Y: Extended); overload;

This construction also doesn't work on all non-x86 platforms. The proper way is to use the following macros to conditionalise your code:

FPC_HAS_TYPE_SINGLE
FPC_HAS_TYPE_DOUBLE
FPC_HAS_TYPE_EXTENDED

Jonas

I now found some explanations:

And explanation that 80-bit precision on AMD64 is depracated and should be replaced by SIMD.
http://developer.amd.com/documentation/articles/Pages/62720069_4.aspx

Petr

Op Tue, 6 May 2008, schreef Jonas Maebe:

Actually this is a myth, originating from Microsoft itself. Win64 has always supported the x87, but early Microsoft documentation said it did not. This confused Microsofts' MASM coders which disabled the x87 instructions in MASM.

Daniël