The sample code (below) when compiled with Pelles v10 produces the following output.
The same code run through MSVC, Clang, Mingw, and LccWin32 does not loose precision.
Pelles
1.110999999999
2.221999999999
3.333
4.443999999999
5.554999999999
Others
1.111
2.222
3.333
4.444
5.555
#include <stdarg.h>
#include <stdio.h>
double ChooseNumber(long nIndex, ...)
{
double Choice = { 0 };
va_list(args);
va_start(args, nIndex);
{
long i;
for (i = 1; i <= nIndex; i += 1)
{
Choice = va_arg(args, double);
if (Choice == 0)
{
goto L1002;
}
if (i == nIndex)
{
goto L1002;
}
}
}
L1002:;
va_end(args);
return Choice;
}
int main(int argc, char *argv[])
{
{
int Index;
for (Index = 1; Index <= 5; Index += 1)
{
printf ("% .15G\n", ChooseNumber (Index, 1.111, 2.222, 3.333, 4.444, 5.555, 0));
}
}
return 0;
}
If I change
printf("% .15G\n", .....
to
printf("% .12G\n",..... ( .12 or lower )
then the results match the other compilers output but I think that is a band-aid solution.
So basically it's Microsoft runtime (of some version) vs Pelles C runtime.
Maybe there is a bug, maybe there isn't. I hardly ever use binary floating-point, but in my experience rounding is just one of many problems with that format.
I will look at this, but it will likely take a while to analyze and reach some kind of conclusion...
Quote from: Pelle on August 23, 2020, 09:35:52 PM
So basically it's Microsoft runtime (of some version) vs Pelles C runtime.
Maybe there is a bug, maybe there isn't. I hardly ever use binary floating-point, but in my experience rounding is just one of many problems with that format.
I will look at this, but it will likely take a while to analyze and reach some kind of conclusion...
Thanks Pelle ...
This issue is a little fish in a big pond, so I'm not going to loose any sleep over it.
That was a terribly constructed, mixed-up metaphor.
Confirmed - indeed a rounding error somewhere, apparently in the representation of the numbers (printf?):
TestP10.exe
1.110999999999
2.221999999999
3.333
4.443999999999
5.554999999999
TestP9
1.111
2.222
3.333
4.444
5.555
With an int 3 inserted here...
_asm int 3;
Choice = va_arg(args, double);
_asm nop;
... I get identical results some steps later:
P9:
CPU Disasm
Address Hex dump Command Comments
00071076 ³> Ú6A 00 Úpush 0
00071078 ³. ³DD05 20700700 ³fld qword ptr [77020] ; float 5.555000000000000
0007107E ³. ³83EC 08 ³sub esp, 8
00071081 ³. ³DD1C24 ³fstp qword ptr [esp]
00071084 ³. ³DD05 18700700 ³fld qword ptr [77018] ; float 4.444000000000000
0007108A ³. ³83EC 08 ³sub esp, 8
0007108D ³. ³DD1C24 ³fstp qword ptr [esp]
00071090 ³. ³DD05 10700700 ³fld qword ptr [77010] ; float 3.333000000000000
00071096 ³. ³83EC 08 ³sub esp, 8
00071099 ³. ³DD1C24 ³fstp qword ptr [esp]
0007109C ³. ³DD05 08700700 ³fld qword ptr [77008] ; float 2.222000000000000
000710A2 ³. ³83EC 08 ³sub esp, 8
000710A5 ³. ³DD1C24 ³fstp qword ptr [esp]
000710A8 ³. ³DD05 00700700 ³fld qword ptr [77000] ; float 1.111000000000000
000710AE ³. ³83EC 08 ³sub esp, 8
000710B1 ³. ³DD1C24 ³fstp qword ptr [esp]
000710B4 ³. ³FF75 FC ³push dword ptr [ebp-4]
000710B7 ³. ³E8 44FFFFFF ³call 00071000
000710BC ³. ³83C4 30 ³add esp, 30
000710BF ³. ³DD5D F0 ³fstp qword ptr [ebp-10]
000710C2 ³. ³DD45 F0 ³fld qword ptr [ebp-10]
000710C5 ³. ³83EC 08 ³sub esp, 8
000710C8 ³. ³DD1C24 ³fstp qword ptr [esp]
000710CB ³. ³68 28700700 ³push offset 00077028 ; ASCII "% .15G",LF
000710D0 ³. ³E8 4B000000 ³call 00071120
P10:
CPU Disasm
Address Hex dump Command Comments
01391076 ³> Ú6A 00 Úpush 0
01391078 ³. ³DD05 20803901 ³fld qword ptr [1398020] ; float 5.555000000000000
0139107E ³. ³83EC 08 ³sub esp, 8
01391081 ³. ³DD1C24 ³fstp qword ptr [esp]
01391084 ³. ³DD05 18803901 ³fld qword ptr [1398018] ; float 4.444000000000000
0139108A ³. ³83EC 08 ³sub esp, 8
0139108D ³. ³DD1C24 ³fstp qword ptr [esp]
01391090 ³. ³DD05 10803901 ³fld qword ptr [1398010] ; float 3.333000000000000
01391096 ³. ³83EC 08 ³sub esp, 8
01391099 ³. ³DD1C24 ³fstp qword ptr [esp]
0139109C ³. ³DD05 08803901 ³fld qword ptr [1398008] ; float 2.222000000000000
013910A2 ³. ³83EC 08 ³sub esp, 8
013910A5 ³. ³DD1C24 ³fstp qword ptr [esp]
013910A8 ³. ³DD05 00803901 ³fld qword ptr [1398000] ; float 1.111000000000000
013910AE ³. ³83EC 08 ³sub esp, 8
013910B1 ³. ³DD1C24 ³fstp qword ptr [esp]
013910B4 ³. ³FF75 FC ³push dword ptr [ebp-4]
013910B7 ³. ³E8 44FFFFFF ³call 01391000
013910BC ³. ³83C4 30 ³add esp, 30
013910BF ³. ³DD5D F0 ³fstp qword ptr [ebp-10]
013910C2 ³. ³DD45 F0 ³fld qword ptr [ebp-10]
013910C5 ³. ³83EC 08 ³sub esp, 8
013910C8 ³. ³DD1C24 ³fstp qword ptr [esp]
013910CB ³. ³68 28803901 ³push offset 01398028 ; ASCII "% .15G",LF
013910D0 ³. ³E8 4B000000 ³call 01391120