CORTEX BASIC

[senior_falcon]

i have been playing around with Cortex BASIC to get an idea of what it is capable of doing. Here is a video showing it running a LINES program. It appears that the Cortex interpreter is contained within expansion ram. If so, it may be possible to save a Cortex BASIC program as an EA5 program. If that is possible, then it may be possible to save a Cortex program as a cartridge. But that is a lot of ifs and only time will tell.

Edited August 2, 2023 by senior_falcon

[+OLD CS1]

Impressive.  This BASIC implementation of Lines appears around only half as fast the assembler version which runs in the MiniMemory.

[Tursi]

I believe there is also a ROM version of Cortex BASIC, frees up all that memory for code

 

[Asmusr]

1 hour ago, Tursi said:

I believe there is also a ROM version of Cortex BASIC, frees up all that memory for code

Unfortunately it only frees up 14K.

[apersson850]

You shouldn't have shown this. Now I wonder how it would run in Pascal?

[Reciprocating Bill]

4 hours ago, Asmusr said:

Unfortunately it only frees up 14K.

Cortex BASIC includes a system command ("NEW sys(7)") that frees an additional 1K of RAM, leaving a total of 15,446 bytes free when running from FinalGROM, 6358 bytes when running entirely from RAM (very desirable vis performance when running on a console equipped with 16-bit RAM). Combining the latter with a SuperCart for 8K of fast integer storage opens up a lot of possibilities - successfully running the full BYTE sieve using an 8192 element array, for example, which can't be done in Extended BASIC even with expansion RAM. 

 

Also - Running Cortex is probably the fastest way to do floating point math on the TI - much faster than calling console routines in Forth or assembly. 

[+TheBF]

48 minutes ago, Reciprocating Bill said:

Also - Running Cortex is probably the fastest way to do floating point math on the TI - much faster than calling console routines in Forth or assembly. 

Is it using a much smaller floating point number format Bill? 

[Reciprocating Bill]

Here's what the source for Cortex BASIC says...

 

 

That makes the format six bytes, rather than the eight bytes of the TI format.

Edited August 2, 2023 by Reciprocating Bill

[senior_falcon]

I cannot view that image.

 

[Reciprocating Bill]

This any better?

 

Cortex FP format pdf.pdf

[senior_falcon]

2 minutes ago, Reciprocating Bill said:

This any better?

 

Cortex FP format pdf.pdf 181.82 kB · 2 downloads

Can't open file.

 

[Reciprocating Bill]

[senior_falcon]

Three times is a charm!

When I looked at it with the debugger, it looked like a normal TI format number, but of course, I wasn't paying attention to how many bytes were used. Is this Radix 100 like the TI uses, but with 6 bytes total instead of 8?

 

Edited August 2, 2023 by senior_falcon

[Reciprocating Bill]

From the manual:

So it's a binary format (right?)

Edited August 2, 2023 by Reciprocating Bill

[Stuart]

I actually found the 'maths' a bit surprising. If you print powers of 2 you get:

2

4

8.0000000001

16

32

64

128

256

512.00000001

1024

2048

4096.0000001

8192.0000001

16384

32768.000001

65535.999999

 

Although the errors are very small, I'm just used to the values being correct as per TI BASIC.

[+FarmerPotato]

25 minutes ago, Stuart said:

I actually found the 'maths' a bit surprising. If you print powers of 2 you get:

 

 

Are you multiplying by constant 2, or taking 2^n?

 

2 has an exact representation: it's just exponent 1 and mantissa 1.
 

But 2^n is likely done by a power series approximation of log2. Can you show the source for ^ / pow /exp?

It may be the decimal format: if it prints  more decimal digits than the mantissa implies. It happens when the last remainder that  is partly beyond the precision. . 

 

[Stuart]

13 minutes ago, FarmerPotato said:

Are you multiplying by constant 2, or taking 2^n?

 

2 has an exact representation: it's just exponent 1 and mantissa 1.
 

But 2^n is likely done by a power series approximation of log2. Can you show the source for ^ / pow /exp?

It may be the decimal format: if it prints  more decimal digits than the mantissa implies. It happens when the last remainder that  is partly beyond the precision. . 

 

I was doing 2^n. Source says "Raise (R1) to the power (R2) using log and exponential functions" and I can see a hard-coded value for log2 in the source. So would TI BASIC do it the same way?

[senior_falcon]

One of the great things about BASIC on the TI99 is the numerical accuracy.

10 X=0

20 X=X+.1

30 PRINT X

40 IF X=1 THEN 60

50 GOTO 20

60 END

On the TI this stops when X=1. On the TRS80 this does not stop because X never equals 1 exactly. I think this is true for Apple 2 and the Commodores as well.

I haven't tested Cortex yet to see how it behaves.

 

[RXB]

XB does exactly the same thing as TI Basic and does better in some math.

[+mizapf]

25 minutes ago, senior_falcon said:

One of the great things about BASIC on the TI99 is the numerical accuracy.

...

On the TI this stops when X=1. On the TRS80 this does not stop because X never equals 1 exactly. I think this is true for Apple 2 and the Commodores as well.

 

This is the usual IEEE 754 disease. People are often surprised to hear that all the single-decimal numbers  0.1, ... 0.9 are approximations with the exception of 0.5 which is precisely representable ... and that our TI, in contrast, can encode all of them exactly.

 

While the radix 100 format is utterly slow, it allows us to encode all interesting numbers without approximation - interesting in the sense that I'd like to use them as a numerical constant in the program. I'm not interested in 0.02734375, although this can be done with IEEE, but rather in 0.1 or 3.21 or the like.

[senior_falcon]

In my post above, the program stops when X=1, just like in TI BASIC/XB.

However, if you remove line 40 so the program does not stop, by the time you get to 2, the value printed is 1.999999999

Interesting...

 

[+OLD CS1]

7 minutes ago, senior_falcon said:

In my post above, the program stops when X=1, just like in TI BASIC/XB.

However, if you remove line 40 so the program does not stop, by the time you get to 2, the value printed is 1.999999999

Interesting...

How did that happen?

[senior_falcon]

Looks like you are running in TI BASIC or XB. The reference was to Cortex BASIC.

Edited August 3, 2023 by senior_falcon

[senior_falcon]

Here is the listing for LINES. Running on Cortex, it is much faster than the same program running under The Missing Link

Edited August 3, 2023 by senior_falcon

[+OLD CS1]

1 hour ago, senior_falcon said:

The reference was to Cortex BASIC.

DOH!  I read it as testing in XB.