PASM: d,s field specified by negative number causes overflow and produces unexpcted op-code

yisiguro

Perhaps it is useless to discuss so much but I feel it is not natural behavior.

Code below is valid and does not cause any compile error,

con
  rate = 115200
obj
  t:"Parallax Serial Terminal"
pub main
  t.Start(rate)
  t.Hex(x,8) ' "A0FFFE00" expected
  t.NewLine
  t.Hex(y,8) ' "A0FC01FF" expected
  t.NewLine
  t.Hex(z,8) ' "A0FC0000" expected
  t.NewLine
  repeat
dat
  org 0
x
  mov -1,#0
y
  mov 0,#-1
z
  mov 0,#0

but results

FFFFFE00
FFFFFFFF
A0FC0000

I'm using PropellerTool PropellerIDE. This occurs when d and/or s field is specified by negative number ( decimal or not ).

(edit:)
Thanks for comments. I see this issue is caused with OpenSpin invoked by PropellerIDE.
I read source code of OpenSpin on github repo, and found lines related to this issue.

Actually, I've mistakenly believed that source operand specified by immediate value is sign-extended for some instructions, such as MOV, ADD, etc.
Of course sign is not extended, but filled by zero, I know now.

Wuerfel_21

Strange that it compiles for you. I get errors when trying to assemble such instructions. What version of Propeller Tool are you using? (I have 1.3.2)

yisiguro

Wuerfel_21 wrote: »

Strange that it compiles for you. I get errors when trying to assemble such instructions. What version of Propeller Tool are you using? (I have 1.3.2)

Sorry, I mistook. I'm using PropellerIDE, not PropellerTool.

(edit)
I was very careless that I did not test this issue on any other pasm compilers:

PropellerTool (1.3.2) : raise error
PropellerIDE (0.33.3) : pass but unexpected output
Propellent (1.6) : raise error

localroger

Negative numbers in PASM are 2's complement 32-bit integers; S and D are 9-bit unsigned integers. Normal behavior if assigning an out of range number to S and D does not throw an error would be to take the least significant 9 bits. For -1, which is all bits 1, that would result in all 9 bits of S or D being 1, or a value of 511 or $1FF. And this is exactly what happened in your test.

To get the bit pattern of a negative number, complement the bit pattern of the absolute value (all 1's to 0 and all 0's to 1) and then add 1. For example, if you do this with 1, you go from all zeroes except 1 in the lsb, to all 1's except 0 in the lsb, and when you add 1 you get all bits 1 for minus 1.

yisiguro

localroger wrote: »

For -1, which is all bits 1, that would result in all 9 bits of S or D being 1, or a value of 511 or $1FF. And this is exactly what happened in your test.

What you say is exactly what I expected, but what happened is not: that produces wrong instruction "WAITVID" from any instruction.

Anyway, that seems to be a bug of PropellerIDE, and I'll try newer version from github repo.

Thanks.

localroger

Ah, PropIDE is probably not masking the out of range S or D value before adding or ORing it into the opcode.

yeti

I got no PropellerIDE here. The 4 compilers which might be launched by PropellerIDE on Linux vote 3:1 that this code is not ok.

$ cat main.spin 
con
  rate = 115200
obj
  t:"Parallax Serial Terminal"
pub main
  t.Start(rate)
  t.Hex(x,8) ' "A0FFFE00" expected
  t.NewLine
  t.Hex(y,8) ' "A0FC01FF" expected
  t.NewLine
  t.Hex(z,8) ' "A0FC0000" expected
  t.NewLine
  repeat
dat
  org 0
x
  mov -1,#0
y
  mov 0,#-1
z
  mov 0,#0

BSTC, HomeSpun and FastSpin complain about the operands:

$ bstc -L /opt/parallax.spin.src/spin main.spin
Brads Spin Tool Compiler v0.15.3 - Copyright 2008,2009 All rights reserved
Compiled for i386 Linux at 08:17:46 on 2009/07/20
Loading Object main
Loading Object Parallax Serial Terminal

main(17,7) Error : Parameter exceeds $1FF
  mov -1,#0
______^

Compiled 285 Lines of Code in 0.009 Seconds

$ fastspin -L /opt/parallax.spin.src/spin main.spin
Propeller Spin/PASM Compiler 'FastSpin' (c) 2011-2019 Total Spectrum Software Inc.
Version 3.9.24 Compiled on: Mar 24 2019
main.spin
|-Parallax Serial Terminal.spin
main.spin(17) error: Destination operand too big for mov
main.spin(19) error: Source operand too big for mov

$ homespun -L /opt/parallax.spin.src/spin main.spin
Homespun Spin Compiler 0.32 - Batang Build
parsing main.spin
parsing /opt/parallax.spin.src/spin/Parallax Serial Terminal.spin
compiling main.spin
Error: main.spin (17, 8): Destination register cannot exceed $1ff
  mov -1,#0
       ^

OpenSpin does not.

$ openspin -L /opt/parallax.spin.src/spin main.spin
Propeller Spin/PASM Compiler 'OpenSpin' (c)2012-2018 Parallax Inc. DBA Parallax Semiconductor.
Version 1.00.81 Compiled on Dec 18 2018 13:23:20
Compiling...
main.spin
|-Parallax Serial Terminal.spin
Done.
Program size is 1176 bytes

Maybe OpenSpin just should complain about this like the others already do?

Wuerfel_21

Gotta love how all the spin compilers are a little bit garbo.
PropTool is GUI only and has the least features, but is the reference compiler.
OpenSpin has this funky bug and lacks some features (assembler listings, adding a #define from the command line).
BSTC is closed source but has lots of features, but causes strange bugs sometimes.
Fastspin uses a lot of memory.
Homespun, I haven't used yet due to not having found a binary yet (and being too lazy to download a C# compiler).

Rayman

It's really only a bug if the instruction changes to WAITVID, right?
Does this happen with OpenSpin?

Actually, I guess it is a bug ether way...