UASM 2.35 Relase

[jj2007]

It is not right to complain, movups = Move Unaligned Packed Single-Precision Floating-Point Values
It expects 4 floating point values.  :P

Right :t

Intel and AMD are to blame here, as they created a lot of confusion with the equivalent movups, movdqu, lddqu etc instructions. Like many other people, I use movups to load integers, e.g. GUIDs, or FILETIMEs with movlps. Timings are identical, size is shorter 8)

OWORD = 16byte, so that would be declaring (effectively) an array of 4 16byte(OWORD) values, for which a single float initializer isn't going to make any sense hence it just producing zero.
ML and UASM will both allow you to declare this, but the result is garbage in both cases.

I take the point of 16byte float initialiser not making sense, but it should trigger at least a warning then.

[johnsa]

Ok, right something for you all to try out:

www.terraspace.co.uk/uasm.zip (Windows UASM32, UASM64 2.36pre)

Size based checking for SIMD is in and I had another idea which i've put in too, automatic checking of alignment.

If you use movaps, movapd, movdqa, vmovaps, vmovapd, vmovdqa in conjunction with a variable, it will automatically warn you if the variable is not aligned to 16/32 as per the type of register you are using it with.
IE:

Code Select Expand



.data

dummy db 1
;.align 16
myVariable dd 1.0, 2.0, 3.0, 4.0

.code

movaps xmm0,myVariable     ;emit warning about alignment



The checking with regard to size now doesn't care if you use byte,word,dword,real as long as the total size of the symbol is compatible it will allow it. Un-typed memory references will work as before.

Regards,
John

[jj2007]

Ok, right something for you all to try out:

www.terraspace.co.uk/uasm.zip (Windows UASM32, UASM64 2.36pre

OK for my usual tests :t

[aw27]

Size based checking for SIMD is in and I had another idea which i've put in too, automatic checking of alignment.
dummy db 1
;.align 16
myVariable dd 1.0, 2.0, 3.0, 4.0
movaps xmm0,myVariable     ;emit warning about alignment

I can't test today, I just wonder what will be the outcome of:
dummy db 1
.align 16
myVariable1 dd 1.0, 2.0, 3.0, 4.0
myVariable2 dd 1.0, 2.0, 3.0, 4.0
movaps xmm0,myVariable2

[johnsa]

Update at same url: http://www.terraspace.co.uk/uasm.zip

The outcome of that would be fine, if myVariable1 is aligned 16, and is 16 bytes myVariable2 will be aligned too. The warning for alignment is derived from the symbols calculated offset (assuming the start of the section is aligned). I'm not checking locals for alignment only items declared in .const/.data and .data? (any global).

The size checking is applied to both globals and locals instead of relying on the type.