Re: [hatari-devel] FPU update

[Andreas Grabher <andreas.grabher@xxxxxxxxxxxx>]

Douglas, thank you very much for your efforts! I think this will be very useful. But it would make things much easier, if on error it would also print the operands. Then i could reproduce the bug using the operands and step through the code until finding the failing part. I'd not need to do that from the guest system, but just call the failing routines directly.

Am 25.01.2017 um 00:10 schrieb Douglas Little <doug694@xxxxxxxxxxxxxx>:

Hi,

Here is a link to the diagnostic version of the fpu test program:

https://www.dropbox.com/s/gp3jix256olay9i/fputestv1.1d_src.zip?dl=0

Unfortunately it will not fit on a floppy image. It does generate too much data for that. The recording is around 6mb (3.5mb zipped). You'll need another way to get the data mounted for tests!

Here is some sample output from Hatari 2.00 using expectations recorded from a real 68882.

....for the 32bit/float tests...

68881/2 FPU functional test v1.1

*** diagnostic build ***

by dml/2016

loading expectations...

fmoves:#imm:reg pass

fmoves:#spc:reg pass

fmoved:#imm:reg pass

fmoved:#spc:reg pass

fmovecr:#i:reg  pass

fmovel:rnd:impl fp result fault @ pc:   $010106f6

expected:               $00000000

calculated:             $ffffffff

fp result fault @ pc:   $0101070e

expected:               $00000000

calculated:             $00000001

fp result fault @ pc:   $010106f6

expected:               $00000000

calculated:             $ffffffff

fp result fault @ pc:   $0101070e

expected:               $00000000

calculated:             $00000001

fp result fault @ pc:   $010106f6

expected:               $00000000

calculated:             $ffffffff

fp result fault @ pc:   $0101070e

expected:               $00000000

calculated:             $00000001

fp result fault @ pc:   $010106f6

expected:               $00000000

calculated:             $ffffffff

fp result fault @ pc:   $0101070e

expected:               $00000000

calculated:             $00000001

pass

fmovel:rnd:expl fp result fault @ pc:   $010107b2

expected:               $00000000

calculated:             $ffffffff

fp result fault @ pc:   $010107ce

expected:               $00000000

...and for some of the 64bit/double tests

fmovel:rnd:zero pass

fadd/fsub:brand fp result fault @ pc:   $01010914

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $0101093a

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $01010914

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $0101093a

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $01010914

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $0101093a

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $01010914

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

fp result fault @ pc:   $0101093a

expected:               $ffe5056aa5cd7886

calculated:             $ffe5056aa5cd7887

On 24 January 2017 at 18:25, Toni Wilen <twilen@xxxxxxxxxx> wrote:

> I'll post the modified code and binaries later on. The expectd.dat file
> will need to be recorded on real HW though - I have currently only
> generated one for Hatari 2.00...

Thanks, I'll wait for real hw data files... (and hopefully it still runs
under Amiga emutos and fits on floppy image)