Page 2 of 2
Re: BSwap64/32/16
Posted: Sun Apr 16, 2023 6:03 pm
by mk-soft
There are not builtin ...
Code: Select all
Procedure.a bswap8(value.a)
CompilerIf #PB_Compiler_Backend = #PB_Backend_C
!return (v_value << 4 | v_value >> 4);
CompilerElse
!mov al, byte [p.v_value]
!rol al, 4
ProcedureReturn
CompilerEndIf
EndProcedure
a = $9E
b = bswap8(a)
Debug Hex(b, #PB_Byte)
C-Code declare
Code: Select all
// Procedure.a bswap8(value.a)
static unsigned char f_bswap8(unsigned char v_value) {
...
}
But actually there is something like bswap8. It should be NippleSwap.
Where do you need something like that
Re: BSwap64/32/16
Posted: Sun Apr 16, 2023 8:15 pm
by Lunasole
mk-soft wrote: Sun Apr 16, 2023 6:03 pm
But actually there is something like bswap8. It should be NibbleSwap.
Why it doesn't works like this?
I also tried that when doing others bswaps, wondered why failed and wonder if same works in C (not tried too).
Code: Select all
Procedure.b t(v_value.b)
ProcedureReturn (v_value >> 4 | v_value << 4)
EndProcedure
Define e.b = %10001000
Debug Bin(e, #PB_Byte)
Debug Bin(t(e), #PB_Byte)
Debug Bin(t(t(e)), #PB_Byte)
; =
; 10001000
; 11111000
; 11111111
Where do you need something like that
Ha, there are some ideas even for this extremely funny, but probably will not use soon.
Re: BSwap64/32/16
Posted: Sun Apr 16, 2023 9:08 pm
by idle
It fails as input is signed.
Byte swap is a stage in swar parallel bitreverse see here
https://graphics.stanford.edu/~seander/bithacks.html
Re: BSwap64/32/16
Posted: Sun Apr 16, 2023 9:22 pm
by Lunasole
I though he may explain his code, but well
Re: BSwap64/32/16
Posted: Sun Apr 16, 2023 10:07 pm
by mk-soft
The type byte is signed (-128 .. 127)
You must take type ascii (0 ..255)
However, I prefer the type unsigned char for C-backend as in my example, because PB internally converts everything back to quad before shifting bits.
Code: Select all
Procedure.a t(v_value.a)
ProcedureReturn (v_value >> 4 | v_value << 4)
EndProcedure
Define e.b = %10001001
Debug Bin(e, #PB_Byte)
Debug Bin(t(e), #PB_Byte)
Debug Bin(t(t(e)), #PB_Byte)
Re: BSwap64/32/16
Posted: Mon Apr 17, 2023 2:54 am
by Lunasole
mk-soft wrote: Sun Apr 16, 2023 10:07 pm
However, I prefer the type unsigned char for C-backend as in my example, because PB internally converts everything back to quad before shifting bits.
No any theoretically possible fix for that with 1-byte? Like this I tried with masking:
Code: Select all
Procedure.b t(v_value.b)
ProcedureReturn %00001111&v_value<<4|v_value&%11110000>>4
EndProcedure
It's far of course to dig to anything behind that, unless I really need it, but curious to ask if you know that all already so good.
Re: BSwap64/32/16
Posted: Mon Apr 17, 2023 4:38 am
by idle
Do you understand it looking at the assembly?
Code: Select all
Procedure t(v_value.b)
ProcedureReturn ((v_value & $FF) >> 4 | (v_value & $FF) << 4)
;MOVSX r15,byte [rsp+PS0+0]
;And r15,255
;SAR r15,4
;MOVSX r14,byte [rsp+PS0+0]
;And r14,255
;SAL r14,4
;Or r15,r14
;MOV rax,r15
;ProcedureReturn
EndProcedure
a.b = $F1
a = t(a)
Debug Hex(a,#PB_Byte)
This thread should be moved to general discussions I think.
Re: BSwap64/32/16
Posted: Mon Apr 17, 2023 11:59 am
by Lunasole
idle wrote: Mon Apr 17, 2023 4:38 am
Do you understand it looking at the assembly?
Yes, at least see key ops, even though I don't want to go to ASM now and didn't looked at that <<4 also. Just curious as I said, asked without doing anything myself until really will need that, as soon at there are talks around that.
This thread should be moved to general discussions I think.
For me as it works and produces correct and predictable results (16-32-64), it's still good and usable variant, maybe even more goes to "tricks".
The other about 8 more a discussion.
Re: BSwap64/32/16
Posted: Mon Apr 17, 2023 1:12 pm
by mk-soft
You're making it too complicated for yourself.
In Purebasic, all comparison, shift and math of byte, word, longs, integers, quads are processed signed.
With the introduction of Unicode, the vartype var.a (ascii) was added as an unsigned byte and the vartype var.u (unicode) as an unsigned word.
I hope you know how negative integers are represented and what happens when bits are shifted to avoid losing the sign.
Only with the use of the correct vartype it also works optimised.
There are already many examples from us ...
Re: BSwap64/32/16
Posted: Mon Apr 17, 2023 3:37 pm
by Lunasole
mk-soft wrote: Mon Apr 17, 2023 1:12 pm
You're making it too complicated for yourself.
In Purebasic, all comparison, shift and math of byte, word, longs, integers, quads are processed signed.
With the introduction of Unicode, the vartype var.a (ascii) was added as an unsigned byte and the vartype var.u (unicode) as an unsigned word.
I hope you know how negative integers are represented and what happens when bits are shifted to avoid losing the sign.
Only with the use of the correct vartype it also works optimised.
There are already many examples from us ...
Well I'd send you a silent "like" for your replies if this forum had them, but it doesn't so there is textural.
> I hope you know how negative integers are represented and what happens when bits are shifted to avoid losing the sign.
Remembering that single bit.
Re: BSwap64/32/16
Posted: Tue Apr 18, 2023 4:58 pm
by RichAlgeni
infratec wrote: Thu Apr 13, 2023 9:27 am
The main problem is, that PB does not have unsigned long and quad.
I would so love unsigned integers.
Re: BSwap64/32/16
Posted: Tue Apr 18, 2023 9:28 pm
by idle
RichAlgeni wrote: Tue Apr 18, 2023 4:58 pm
infratec wrote: Thu Apr 13, 2023 9:27 am
The main problem is, that PB does not have unsigned long and quad.
I would so love unsigned integers.
Yes and add type alias as int32 or uint32 it's clear for everyone with no ambiguity.
c backend abuse variable a is treated as unsigned!
Code: Select all
CompilerIf #PB_Compiler_Backend = #PB_Backend_C
!unsigned long long v_a =0;
Macro BSWAP64(v)
v = ((v >> 8) & $00FF00FF00FF00FF) | ((v & $00FF00FF00FF00FF) << 8) ;flip adjacent bytes
v = ((v >> 16) & $0000ffff0000ffff) | ((v & $0000ffff0000ffff) << 16) ; flip words
v = (v >> 32) | (v << 32) ;flip dwords
EndMacro
Macro BSWAP32(v)
v = ((v >> 8) & $00FF00FF) | ((v & $00FF00FF) << 8) ;flip adjacent bytes
v = (v >> 16) | v << 16 ;flip words
v & $FFFFFFFF
EndMacro
Macro BSWAP16(v)
v = ((v >> 8) & $00FF) | ((v & $00FF) << 8) ;flip adjacent bytes
v & $FFFF
EndMacro
Global a.q ;a is unsigned
Global b.q ; b is signed
a = $FEDCBA9876543210
b = a
bswap64(a)
bswap64(b)
Debug Hex(a)
Debug Hex(b)
a = $FEDCBA98
bswap32(a)
Debug Hex(a)
a = $FEDC
bswap16(a)
Debug Hex(a)
CompilerEndIf
Re: BSwap64/32/16
Posted: Wed Apr 19, 2023 6:48 pm
by Lunasole
RichAlgeni wrote: Tue Apr 18, 2023 4:58 pm
infratec wrote: Thu Apr 13, 2023 9:27 am
The main problem is, that PB does not have unsigned long and quad.
I would so love unsigned integers.
Nice.
Re: BSwap64/32/16
Posted: Wed Apr 19, 2023 7:01 pm
by Lunasole
idle wrote: Tue Apr 18, 2023 9:28 pm
RichAlgeni wrote: Tue Apr 18, 2023 4:58 pm
infratec wrote: Thu Apr 13, 2023 9:27 am
The main problem is, that PB does not have unsigned long and quad.
I would so love unsigned integers.
Yes and add type alias as int32 or uint32 it's clear for everyone with no ambiguity.
c backend abuse variable a is treated as unsigned!
[/code]
Too.