World: r3wp

i could handle 16 bit registers directly in binary series

Yeah, that is likely slower. You are likely to have to do some bytewise 
reads and writes internally. You would be faster if you had seperate 
read/write macros for the 16-bit load/store instructions.

hu ?

what are you saying ?

i said that the faster way is to write 2 times 8bit instead of writing 
16 bit in one time

I was agreeing.

so there is no optimization to have different macro for access/writing 
16bit or 8bits registers

ah ok

'scuse me

I wonder if there would be a fast way to cache the 16-bit values 
in _HL, _BC and such, and writing them quickly.

that could be faster if 16 bits register was handled internally as 
binary series

but that will be a problem to perform math operation on them

It occurs to me that overflow on these inserts isn't handled the 
way you think.

?

explain

If you poke a value over 255 into a string/binary, the upper bytes 
may be ignored rather that triggering an overflow error. Test that.

ok

It might save on some AND x 255 statements.

could be a usefull trick if it works

It doesn't work in REBOL code, but it might in rebcode.

you're right

the upper byte of 16bit value is not kept

hum

so i could remove some and var 255 from my code

That provides a way to optimize the 16-bit load/store ops, particularly 
if there are 16-bit registers.

hmm...

You could invalidate the affected 16-bit register if you use an 8-bit 
op, or perhaps combine 2 8-bit load/stores that would add up to a 
16-bit load/store into their 16-bit operation.

perhaps...

This would be so much easier without self-modifying code - then we 
could just compile the code and be done with it.

BTW, don't forget the ROTL and ROTR opcodes. They may help with 16-bit 
combination registers.

hum, i think that write/read the 16bits register into/from memory 
need more instructions if i follow you

because we perform 2 read/write ,even for 16bits registers, right 
?

currently to write _bc register in memory  , i write:
pokez mem adr _c 
add.i adr 1 
pokez mem adr _b

if _bc is handle as a 16 bit value, i need to do:
pokez mem adr _bc
add.i adr 1
rotr _bc 8
pokez mem adr _bc
rotl _bc 8

right ?

I don't know, I haven't found the stores in your code yet.

Is the Z80 a strict load/store architecture, or do they have other 
operations that can reference memory?

i don't understand what u mean, Z80 is not concerned, that is how 
we perfom updates in memory with rebcode

I am asking if operations like ADD can add values directly to/from 
memory like it can on x86, or does it have to load first and store 
after?

no, only registers can read/wites memory

Z80 can only perform operations on registers

OK, strict load/store.

anyway that is not the problem, i just said, that handling 16bits 
registers in 16 bits values instead to have 2 register of 8 bits 
is much slower when we have to read/write into memory

plus lost of time due to the syncronization between 16 and 8 bits 
registers

finally i'm not sure that it's a good idea to separate them

Well, I would have to read more of that document to agree with you 
or not :(

but we would have a gain when performing operations between 16 bits 
registers

quite balanced

in C it will not be a problem, 8 bits registers and 16 bits registers 
will share the same space adressing.

it's missing in rebol