Some time ago I had to do a few hw-independent experiments with floating point, and faced with "./fpc/rtl/inc/softfpu.pp" unit. Its' current state is far from finishing of porting stage, so I had to make some changes to get it up and compile with originally foreseen functionality.

This is roughly a draft patch -- just conversion of unfinished porting of C-syntax; it omits some details like 'public'/'alias:'/'compilerproc' placing (since I have no idea about these modifiers). In addition, fixed several lost 'exit's, 'var'-modifier for 'shortShift128Left', some range-warnings..

Ambiguities were checked against "John R. Hauser's SoftFloat Release 2b". Missing parts were also ported from there. But almost no serious testing was performed yet by now due to lack of time.

Could this patch be reviewed and accepted/adapted/rejected?

---
WBR,
Max

Would you please open an issue on bugs.freepascal.org for this and preferrably add an uncompressed .patch file. This way your patch won't be forgotten that easily. Thank you.

Regards,
Sven

Hi Max,

thank you for your contribution...
While testing it, I managed to fix another independent longstanding bug in softfpu.pp
(rev 24864) which fixes a int64<->single conversion error.

I must confess that it's been a long time since I last looked at the software fpu support code... thus I cannot make any prediction on the integration of your code in the compiler.

Pierre Muller

is

far from finishing of porting stage, so I had to make some changes to get

it

up and compile with originally foreseen functionality. This is roughly a draft patch -- just conversion of unfinished porting of

C-

syntax; it omits some details like 'public'/'alias:'/'compilerproc'

placing

(since I have no idea about these modifiers). In addition, fixed several lost 'exit's, 'var'-modifier for 'shortShift128Left', some

range-warnings..

Ambiguities were checked against "John R. Hauser's SoftFloat Release 2b". Missing parts were also ported from there. But almost no serious testing

was

Max Nazhalov wrote

Could this patch be reviewed and accepted/adapted/rejected?

Thanks, I applied it and fixed a few more things.

Wednesday, June 12, 2013, 4:00:52 AM, Florian wrote:

FK> Am 09.06.2013 18:22, schrieb Max Nazhalov:

> > Could this patch be reviewed and accepted/adapted/rejected?

FK> Thanks, I applied it and fixed a few more things.

Thanks a lot! Just two addition notes:

1. Catched another 4 missing exits

Index: softfpu.pp
===================================================================
--- softfpu.pp  (revision 24880)
+++ softfpu.pp  (working copy)
@@ -3468,6 +3468,7 @@
             End;
         end;
         float32_round_to_int.float32 := packFloat32( aSign, 0, 0 );
+        exit;
       End;
     lastBitMask := 1;
     {_____________________________!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!}
@@ -3774,6 +3775,7 @@
         if ( (aSig0) OR ( ( bExp = $FF ) AND  (bSig0) ) ) then
         Begin
             float32_mul.float32 := propagateFloat32NaN( a.float32, b.float32 );
+            exit;
         End;
         if ( ( bExp OR  bSig ) = 0 ) then
         Begin
@@ -3863,6 +3865,7 @@
             if ( bSig  0) then
             Begin
               float32_div.float32 := propagateFloat32NaN( a.float32, b.float32 );
+              exit;
             End;
             float_raise( float_flag_invalid );
             float32_div.float32 := float32_default_nan;
@@ -4849,6 +4852,7 @@
                exit;
             End;
             packFloat64( zSign, $7FF, 0, 0, out );
+            exit;
         End;
         if ( aExp = 0 ) then
         Begin

2. r24872 seems to revert r24864 by Pierre Muller (see above in mail list) -- was this intended?..

Thanks, applied.

2. r24872 seems to revert r24864 by Pierre Muller (see above in mail list) -- was this intended?..

This was a mistake, I fixed it.