FPU 80 bit extended precision

[mav]

Hello:
Hello, lately every time I update the trunk (I use a stable version and a trunk normally) with fpcup, I see the message: "Adding -dFPC_SOFT_FPUX80
to compiler option to enable 80bit (soft)  float support (trunk only) "...and I ask myself the following questions:
How do you use that "-dFPC_SOFT_FPUX80", is there an example somewhere?
Can it be used in 64 bit applications, in Windows ? (...I am a Windows user).
I usually start by writing code and then when it works,
I do the design of the program... this time I'm doing it backwards.
 Greetings to all!

It takes time and work to learn...!!    

[Thaddy]

The windows 64 ABI by itself does not support this (extended is double) but FPC supports this anyway, but not as standard and requires recompilation. NOT recommended. If you are really and I mean really experienced, I can give you a clue.
Otherwise it is pointless.

[PascalDragon]

Hello, lately every time I update the trunk (I use a stable version and a trunk normally) with fpcup, I see the message: "Adding -dFPC_SOFT_FPUX80
to compiler option to enable 80bit (soft)  float support (trunk only) "...and I ask myself the following questions:
How do you use that "-dFPC_SOFT_FPUX80", is there an example somewhere?

This is a define that is only respected by the compiler itself and controls whether it uses a software based 80-bit floating point type for cross compiling from a target that does not support 80-bit floating point types (e.g. Win64 on x86_64 or any non-x86 target) to a target that does (e.g. any x86 target that is not x86_64-win64). This define will not enable Extended support on x86_64-win64 and it's also not yet recommended to use that define anyway, because generated binaries will differ from one done with a compiler that has native 80-bit floating point support.

If you want to use 80-bit floating point in your own program then use the sfpux80 and ufloatx80 units. However these are compiled by default only on 3.3.1, not yet on 3.2.2, so you need to compile them yourself.

[mav]

Thanks for answering me.
  Thaddy, I'm not really experienced...but I'll try, I'll post the code I'm doing...I've been looking at the units
  sfpux80 and ufloatx80 and I already started to do something... let's see what comes out...

[photor]

Hello, lately every time I update the trunk (I use a stable version and a trunk normally) with fpcup, I see the message: "Adding -dFPC_SOFT_FPUX80
to compiler option to enable 80bit (soft)  float support (trunk only) "...and I ask myself the following questions:
How do you use that "-dFPC_SOFT_FPUX80", is there an example somewhere?

This is a define that is only respected by the compiler itself and controls whether it uses a software based 80-bit floating point type for cross compiling from a target that does not support 80-bit floating point types (e.g. Win64 on x86_64 or any non-x86 target) to a target that does (e.g. any x86 target that is not x86_64-win64). This define will not enable Extended support on x86_64-win64 and it's also not yet recommended to use that define anyway, because generated binaries will differ from one done with a compiler that has native 80-bit floating point support.

If you want to use 80-bit floating point in your own program then use the sfpux80 and ufloatx80 units. However these are compiled by default only on 3.3.1, not yet on 3.2.2, so you need to compile them yourself.

So where can I get these units for 3.2.2? And are there any docs about how to use these units? Thanks.

[DragoRosso]

So where can I get these units for 3.2.2? And are there any docs about how to use these units? Thanks.

You asked in other thread for extended use on 64 bit application, but since is not a standard why you don't use other libraries ?

One very famous lib about big numbers is Rudy Velthuis Big Numbers; this is the link of the new maintainer (he passed aways some years ago): https://github.com/TurboPack/RudysBigNumbers

They don't speak about FPC compatibility, but may be it is.

Take a look.

[srvaldez]

I tried RudysBigNumbers with FPC, no luck
but depending on mav's needs a big number library may be overkill
if all you need is 80 or 128-bit floats then with some tedious work on the bindings you could use softfloat http://www.jhauser.us/arithmetic/SoftFloat.html

[PascalDragon]

If you want to use 80-bit floating point in your own program then use the sfpux80 and ufloatx80 units. However these are compiled by default only on 3.3.1, not yet on 3.2.2, so you need to compile them yourself.

So where can I get these units for 3.2.2? And are there any docs about how to use these units? Thanks.

Download the source for 3.2.2 and compile them manually. Regarding use simply add the ufloatx80 unit to your uses-clause and instead of Extended use floatx80. Some additional functions are available in unit sfpux80, but some others are missing (like ln or conversion to string) which is the main reason why 80-bit software floating point can't yet be fully implemented in the compiler.

[LV]

Arbitrary float point precision (80, 128, 256 bits...)
https://github.com/VadimAnIsaev/GNU-MP-for-FreePascal 

[srvaldez]

on Windows you can use libmingwex from the mingw package
libmingwex comes as a static library but it's tricky to link with FPC due to dependencies, so I created a dll which you can get from https://u.pcloud.link/publink/show?code=XZQUNR5ZWS6dIkjFsOVOxDmtYUgd3XaeFwy7
for the basic arithmetic functions you use inline asm and although the FPU has trig and other functions I would choose to use the ones from the libmingwex, because I don't trust that FPU will handle value range ok

Code: Pascal  [Select][+][-]

1.         {$MODE OBJFPC}{$H+}
2.         {$MINFPCONSTPREC 64}
3.         {$MODESWITCH ADVANCEDRECORDS}
4.         //{$LINKLIB libmingwex}
5.        
6.         program libmingwex-test1;
7.         uses sysutils, windows, math;
8.        
9.         type extfloat = record
10.                 ext_number : array[0..15] of uint8;
11.         end;
12.        
13.         type extfloatp = ^extfloat;
14.        
15.         function snprintf ( s:pchar; n:uint32; format:pchar;args: extfloatp):longint;cdecl; external 'libmingwex' name '__mingw_snprintf';
16.         function strtold (s:pchar;var nd:pchar):extfloat;cdecl; external 'libmingwex' name '__mingw_strtold';
17.         function scanfld (s:pchar;frmt:pchar;x:extfloatp):int32;cdecl; external 'libmingwex' name '__mingw_sscanf';
18.        
19.         function sinl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'sinl';
20.         function asinl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'asinl';
21.         function cosl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'cosl';
22.         function acosl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'acosl';
23.         function tanl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'tanl';
24.         function atanl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'atanl';
25.         function logl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'logl';
26.         function expl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'expl';
27.        
28.         function extsqrt(x: extfloat):extfloat;
29.         var
30.           y:extfloat;
31.         begin
32.           {$ASMMODE att}
33.                 asm
34.                   fldt     x
35.                   fsqrt
36.                   fstpt y
37.                 end;
38.                 extsqrt:=y;
39.         end;
40.  
41.         function extadd(a: extfloat; b: extfloat):extfloat;
42.  
43.         var
44.           y:extfloat;
45.         begin
46.           {$ASMMODE att}
47.                 asm
48.                   fldt     a
49.                   fldt     b
50.                   faddp   %st, %st(1)
51.                   fstpt y
52.                 end;
53.                 extadd:=y;
54.         end;
55.        
56.         function extsub(a: extfloat; b: extfloat):extfloat;
57.  
58.         var
59.           y:extfloat;
60.         begin
61.           {$ASMMODE att}
62.                 asm
63.                   fldt     a
64.                   fldt     b
65.                   fsubrp   %st, %st(1)
66.                   fstpt y
67.                 end;
68.                 extsub:=y;
69.         end;
70.  
71.         function extmul(a: extfloat; b: extfloat):extfloat;
72.  
73.         var
74.           y:extfloat;
75.         begin
76.           {$ASMMODE att}
77.                 asm
78.                   fldt     a
79.                   fldt     b
80.                   fmulp   %st, %st(1)
81.                   fstpt y
82.                 end;
83.                 extmul:=y;
84.         end;
85.        
86.         function extdiv(a: extfloat; b: extfloat):extfloat;
87.  
88.         var
89.           y:extfloat;
90.         begin
91.           {$ASMMODE att}
92.                 asm
93.                   fldt     a
94.                   fldt     b
95.                   fdivrp   %st, %st(1)
96.                   fstpt y
97.                 end;
98.                 extdiv:=y;
99.         end;
100.  
101.         function extval( s:ansistring):extfloat;
102.         var
103.                 pc, frmt:pchar;
104.                 y:extfloat;
105.         begin
106.                 pc:=pchar(s+#0);
107.                 frmt:='%Lf'+#0;
108.                 scanfld(pc, frmt, @y);
109.                 extval:=y;
110.         end;
111.  
112.         function extstr( x:extfloat; frmt:ansistring):ansistring;
113.         var
114.                 s, frm:pchar;
115.                 res:ansistring;
116.         begin
117.                 frm:=pchar(frmt+#0);
118.                 res:='                                                                        ';
119.                 s:=pchar(res+#0);
120.                 snprintf(s, 64, frm, @x);
121.                 extstr:=s;
122.         end;
123.        
124.         const frmt=' %.20Lg';
125.         var x, y, z:extfloat;
126.  
127.         begin
128.                 x:=extval('2');
129.                 y:=extval('3');
130.                 z:=extadd(x, y);
131.                 writeln(extstr(z, ' %.20Lg'));
132.                
133.                 z:=extsub(x, y);
134.                 writeln(extstr(z, ' %.20Lg'));
135.                
136.                 z:=extmul(x, y);
137.                 writeln(extstr(z, ' %.20Lg'));
138.                
139.                 z:=extdiv(x, y);
140.                 writeln(extstr(z, ' %.20Lg'));
141.                
142.                 z:=extsqrt(x);
143.                 writeln(extstr(z, ' %.20Lg'));
144.         end.
145.  

5
 -1
 6
 0.66666666666666666668
 1.4142135623730950488

[srvaldez]

by the way, I found that I had to add #0 to the strings for the functions in libmingwex otherwise it would crash

[photor]

on Windows you can use libmingwex from the mingw package
libmingwex comes as a static library but it's tricky to link with FPC due to dependencies, so I created a dll which you can get from https://u.pcloud.link/publink/show?code=XZQUNR5ZWS6dIkjFsOVOxDmtYUgd3XaeFwy7
for the basic arithmetic functions you use inline asm and although the FPU has trig and other functions I would choose to use the ones from the libmingwex, because I don't trust that FPU will handle value range ok

Code: Pascal  [Select][+][-]

1.         {$MODE OBJFPC}{$H+}
2.         {$MINFPCONSTPREC 64}
3.         {$MODESWITCH ADVANCEDRECORDS}
4.         //{$LINKLIB libmingwex}
5.        
6.         program libmingwex-test1;
7.         uses sysutils, windows, math;
8.        
9.         type extfloat = record
10.                 ext_number : array[0..15] of uint8;
11.         end;
12.        
13.         type extfloatp = ^extfloat;
14.        
15.         function snprintf ( s:pchar; n:uint32; format:pchar;args: extfloatp):longint;cdecl; external 'libmingwex' name '__mingw_snprintf';
16.         function strtold (s:pchar;var nd:pchar):extfloat;cdecl; external 'libmingwex' name '__mingw_strtold';
17.         function scanfld (s:pchar;frmt:pchar;x:extfloatp):int32;cdecl; external 'libmingwex' name '__mingw_sscanf';
18.        
19.         function sinl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'sinl';
20.         function asinl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'asinl';
21.         function cosl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'cosl';
22.         function acosl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'acosl';
23.         function tanl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'tanl';
24.         function atanl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'atanl';
25.         function logl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'logl';
26.         function expl (x:extfloatp):extfloat;cdecl; external 'libmingwex' name 'expl';
27.        
28.         function extsqrt(x: extfloat):extfloat;
29.         var
30.           y:extfloat;
31.         begin
32.           {$ASMMODE att}
33.                 asm
34.                   fldt     x
35.                   fsqrt
36.                   fstpt y
37.                 end;
38.                 extsqrt:=y;
39.         end;
40.  
41.         function extadd(a: extfloat; b: extfloat):extfloat;
42.  
43.         var
44.           y:extfloat;
45.         begin
46.           {$ASMMODE att}
47.                 asm
48.                   fldt     a
49.                   fldt     b
50.                   faddp   %st, %st(1)
51.                   fstpt y
52.                 end;
53.                 extadd:=y;
54.         end;
55.        
56.         function extsub(a: extfloat; b: extfloat):extfloat;
57.  
58.         var
59.           y:extfloat;
60.         begin
61.           {$ASMMODE att}
62.                 asm
63.                   fldt     a
64.                   fldt     b
65.                   fsubrp   %st, %st(1)
66.                   fstpt y
67.                 end;
68.                 extsub:=y;
69.         end;
70.  
71.         function extmul(a: extfloat; b: extfloat):extfloat;
72.  
73.         var
74.           y:extfloat;
75.         begin
76.           {$ASMMODE att}
77.                 asm
78.                   fldt     a
79.                   fldt     b
80.                   fmulp   %st, %st(1)
81.                   fstpt y
82.                 end;
83.                 extmul:=y;
84.         end;
85.        
86.         function extdiv(a: extfloat; b: extfloat):extfloat;
87.  
88.         var
89.           y:extfloat;
90.         begin
91.           {$ASMMODE att}
92.                 asm
93.                   fldt     a
94.                   fldt     b
95.                   fdivrp   %st, %st(1)
96.                   fstpt y
97.                 end;
98.                 extdiv:=y;
99.         end;
100.  
101.         function extval( s:ansistring):extfloat;
102.         var
103.                 pc, frmt:pchar;
104.                 y:extfloat;
105.         begin
106.                 pc:=pchar(s+#0);
107.                 frmt:='%Lf'+#0;
108.                 scanfld(pc, frmt, @y);
109.                 extval:=y;
110.         end;
111.  
112.         function extstr( x:extfloat; frmt:ansistring):ansistring;
113.         var
114.                 s, frm:pchar;
115.                 res:ansistring;
116.         begin
117.                 frm:=pchar(frmt+#0);
118.                 res:='                                                                        ';
119.                 s:=pchar(res+#0);
120.                 snprintf(s, 64, frm, @x);
121.                 extstr:=s;
122.         end;
123.        
124.         const frmt=' %.20Lg';
125.         var x, y, z:extfloat;
126.  
127.         begin
128.                 x:=extval('2');
129.                 y:=extval('3');
130.                 z:=extadd(x, y);
131.                 writeln(extstr(z, ' %.20Lg'));
132.                
133.                 z:=extsub(x, y);
134.                 writeln(extstr(z, ' %.20Lg'));
135.                
136.                 z:=extmul(x, y);
137.                 writeln(extstr(z, ' %.20Lg'));
138.                
139.                 z:=extdiv(x, y);
140.                 writeln(extstr(z, ' %.20Lg'));
141.                
142.                 z:=extsqrt(x);
143.                 writeln(extstr(z, ' %.20Lg'));
144.         end.
145.  

Sorry, what is this for? Hardware F80 with FPU for Free Pascal on Win64 or just software simulation?