Atari OS with Fastchip?

[Maury Markowitz]

Does anyone have a pointer to an 800XL-compatible Atari ROM that has Fastchip ROMs in place of the original Atari FP?

 

I thought this would be easy to find but my google-fu is failing me.

[+Nezgar]

See the last post #52 in this thread:

 

- XL rev 3 base OS ROM (Self test can test XE extended memory)

- Hias' hispeed patch - includes reversed Option for BASIC and shift-reset for coldstart

- Charles Marslett fastchip math pack routines

- Self test patched to properly test all XE extended RAM, not just half of it twice

- Updated checksums so OS startup and selftest rom checks pass

[Faicuai]

5 minutes ago, Nezgar said:

Charles Marslett fastchip math pack routines

It will not pass AcidOS math-pack test, FYI.

[+Larry]

17 minutes ago, Faicuai said:

It will not pass AcidOS math-pack test, FYI.

 

Curious -- what is the significance of that?  (And do you know or suspect why it fails?)  If it fails from the Reverse BASIC, yes, that can be a problem in a fair number of situations.

 

[Faicuai]

1 hour ago, Larry said:

 

Curious -- what is the significance of that?  (And do you know or suspect why it fails?)  

Relates to number representation / conversion / rounding / precision at the limits. 

 

See (and test) four yourself:

 

OS/n, Newell FP pack (Marslett), latest known version:

 

 

 

Altirra FP pack, very precise (and XL extended FP pack, taking international char. set, the FASTEST pack I've ever found and built on my OS's):

 

(NOTE: when running Basic XL (in FAST mode), I have noticed fast results when using in combination of Newell / Marslett pack, in Colleen mode). Then the next level of performance is achieved with XL extended-FP pack, which works also well with Basic-XL, in XL-mode)  

[+Nezgar]

Is Turbo Basic compatible with ROM replacement Floating point routines? I can't recall .. ie mostly using it's own routines, but some calls still made to ROM may be a problem...

[Faicuai]

Just now, Nezgar said:

Is Turbo Basic compatible with replacement Floating point routines?

For the most part, except one function / constant in Altirra's FP-pack.

 

But TurboBasic loads and runs its own math-pack.

[Faicuai]

2 hours ago, Maury Markowitz said:

800XL-compatible Atari ROM that has Fastchip ROMs

Below are the overall best, fastest and/or most-precise pair (that I use as daily drivers), which you already have in soft-load format on Basic-benchmarks thread.

 

1. Best overall balance of Precision and Speed (includes Altirra's FP pack, which will work especially well with Altirra Basic):
   • For XL/XE (+Ultimate1MB): AtariOS-800XLXE-Rev03-ULT1M-F1_F4-FP-E_p1-v1.rom
   • For XL/XE (on INCOGNITO): AtariOS-800XLXE-Rev03-INCOG-F1_F4_H-FP-E_p1-v1.rom 
2. Bias towards highest FP performance, with standard (original) precision levels:
   • For XL/XE (+Ultimate1MB): AtariOS-800XLXE-Rev03-ULT1M-F1_F4-FPx-v1.rom (FP-pack uses Int'l Char. set space)
   • For XL/XE (on INCOGNITO):AtariOS-800XLXE-Rev03-INCOG-F1_F4_H-FPx.rom (FP-pack uses Int'l Char. set space)

 

All loads above:

• are correctly checksum-signed,
• they all test extended / banked 64K of 130XE correctly,
• all resurrect dormant 1200XL embedded-code for DMA control, Key-click sound, keyboard-lock and char-set switching,
• special loads for Incognito version that enables HELP-key as well so you can run XL/XE mode as compatibly  as possible.
• In addition, Loads in group #1 above include highly-accelerated E: handler routines that will free base ram for this purpose, and once you see them in action, you will probably never let go. (SDX, Atari DOS, Basic, Basic XL, Basic XE, Altirra basic, Turbo Basic, Fast Basic, EYE2, and other apps. will greatly benefit from this).

 

They are meant to operate on a PBI-rich environment, so no low-level device-dependent code is embedded on them (other than E: acceleration). SIO-acceleration is left for (say) Ultimate or Incognito BIOS extensions.

[phaeron]

6 hours ago, Larry said:

Curious -- what is the significance of that?  (And do you know or suspect why it fails?)  If it fails from the Reverse BASIC, yes, that can be a problem in a fair number of situations.

 

To expand on Faicuai's test results, the majority of the AcidOS failures with the Marslett math pack relate to differing overflow/underflow behavior. The Atari math pack issues an error for overflows >= 10^98 and underflows to zero <10^-98, but the Marslett math pack doesn't in cases as you can see in the results. The other issues are formatting mismatches in corner cases. The impact of this is low, but it can lead to BASIC acting unexpectedly instead of issuing an Error 3, or other compatibility issues. For instance, PRINT 9E+200 returns an error with the Atari math pack and 9E-56 with the Marslett pack. But as for non-test programs that are affected by these differences, I don't know of any specific cases.