Emulated FPU precision

michaljarz · 09 November 2016, 23:12

Hi.
Last time I run Mand2000:
http://aminet.net/gfx/fract/Mand2000demo.lha

When I started to zoom fractal, I saw some distortion. I compared it with my Amiga, where it turned out that everything was OK. The problem arises only when emulation is enabled the FPU. When the WinUAE does not use the FPU then everything is OK.

Screenshots show what's going on.
Problems arise when zoomed in row x 44 and larger.

EDIT: I tested different settings (JIT on/off, 040/881/882 fpu, more compatible, unimplemented fpu instructions), each time the same effect.
(sorry for my english, I often use the translator)

PeterK · 10 November 2016, 00:33

The difference between floatingpoint calculations in WinUAE and on a real Amiga with an FPU is that the FPU emulation in WinUAE is restricted to DoublePrecision (64 bit) and a real FPU can perform ExtendedPrecision (80 bit) too. This limitation is caused by the VC-compilers. VisualC can not handle ExtendedPrecision.

Toni Wilen · 10 November 2016, 10:33

It gets updated if someone writes assembly wrappers for all required FPU functions..

Multiprecision math libraries are not the answer because they are slower.

09 November 2016, 23:12	#1
michaljarz Registered User Join Date: Jul 2014 Location: Katowice/Poland Posts: 51	Emulated FPU precision Hi. Last time I run Mand2000: http://aminet.net/gfx/fract/Mand2000demo.lha When I started to zoom fractal, I saw some distortion. I compared it with my Amiga, where it turned out that everything was OK. The problem arises only when emulation is enabled the FPU. When the WinUAE does not use the FPU then everything is OK. Screenshots show what's going on. Problems arise when zoomed in row x 44 and larger. EDIT: I tested different settings (JIT on/off, 040/881/882 fpu, more compatible, unimplemented fpu instructions), each time the same effect. (sorry for my english, I often use the translator) Attached Thumbnails Last edited by michaljarz; 09 November 2016 at 23:22.

10 November 2016, 00:33	#2
PeterK Registered User Join Date: Apr 2005 Location: digital hell, Germany, after 1984, but worse Posts: 3,365	The difference between floatingpoint calculations in WinUAE and on a real Amiga with an FPU is that the FPU emulation in WinUAE is restricted to DoublePrecision (64 bit) and a real FPU can perform ExtendedPrecision (80 bit) too. This limitation is caused by the VC-compilers. VisualC can not handle ExtendedPrecision. Last edited by PeterK; 10 November 2016 at 10:49.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Lua interpreter with arbitrary precision math.	Thorham	Coders. Releases	19	06 January 2020 23:13
Chrome gets emulated.	amiga_Forever	Retrogaming General Discussion	2	30 November 2014 13:05
Logitech Precision gamepad hasn't worked since 2.3.3	Sigma	support.WinUAE	13	20 March 2013 09:29
File timestamps lose precision in shared folders	mark_k	support.WinUAE	9	06 February 2013 16:25
System you would like to see emulated next	Ian	Retrogaming General Discussion	18	08 November 2001 08:16

10 November 2016, 10:33	#3
Toni Wilen WinUAE developer Join Date: Aug 2001 Location: Hämeenlinna/Finland Age: 49 Posts: 26,506	It gets updated if someone writes assembly wrappers for all required FPU functions.. Multiprecision math libraries are not the answer because they are slower.

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)