[Libre-soc-isa] [Bug 701] document Matrix REMAP in SVP64

[bugzilla-daemon at libre-soc.org]

https://bugs.libre-soc.org/show_bug.cgi?id=701

--- Comment #6 from Andrey Miroshnikov <andrey at technepisteme.xyz> ---
(In reply to Luke Kenneth Casson Leighton from comment #5)
> (for the next para, read page vii onwards carefully).
> 
> it is permitted to implement any additional instructions over-and-above
> what is in any given subset. however *some* instructions (such as the
> MMA group) are a "all or nothing".
Thanks, I found the optional features listed on pages x-xiii.

> 
> maddld is not one such instruction.
> 
> therefore we can add it.
Maddld belongs to the 3.3.9 Fixed-Point Arithmetic Instructions Category, more
specifically the *3.3.9.1 64-bit Fixed-Point Arithmetic Instructions* (both on
3.0c and 3.1b).

On page xiii, the optional feature:
"Fixed-point instructions that modify OV to
indicate whether overflow occurred (OV)
(addex and instructions with OE=1 such
as addo, subfo, etc.)"

With the reference pointing to 3.3.9.

Do we need to implement the whole set of Fixed-Point Arithmetic Instructions
under section 3.3.9 to be compliant (be easily supported by the compiler)?

It seems far-fetched to me that just maddld by itself would be acceptable,
unless we would also have provide compiler modifications ourselves (i.e. a flag
to say this instruction is present).

> 000100 ..... 110000I ..XX maddhd v3.0 p88 Multiply-Add High Doubleword
> VA-form
> 
> so again, v3.0.
> 
> let's have a look at the "key" to the table, find out what "..XX" means.
> 
> OpenPOWER Compliancy Subsets
> X... Instruction included in the Scalar Fixed-Point Compliancy subset
> .X.. Instruction included in the Scalar Fixed-Point + Floating-Point
> Compliancy
> ..X. Instruction included in the Linux Compliancy subset
> ...X Instruction included in the AIX Compliancy subset.
> 
> that's plain wrong.  let's check addi.
Not sure what you meant by "that's plain wrong".
The Compliancy Subset column looks to be correct, however I'm probably not
understanding something.

> 111111 ..... ///// ..... 11010 01110.I .XXX fcfid
> 
> ok better. ok so the coding of X is correct.
Yep, part of SFFS and above.

-- 
You are receiving this mail because:
You are on the CC list for the bug.