[Mesa-dev] TGSI declarations missing type info

[Jose Fonseca]

----- Original Message -----
> On 14.11.2011 12:28, Jose Fonseca wrote:
> >
> > ----- Original Message -----
> >> Indirect addressing on TEMP should be forbidden and only allowed
> >> for
> >> TEMPORARY_ARRAY, I'd very much like to know which values I can
> >> keep
> >> in
> >> registers and which ones I have to store to thread-local memory (I
> >> can't
> >> do things like MOV %e{%ecx}x but I have 128 registers).
> > Yes, separating out arrays is light years more important that
> > typing, and on its own right, as it seriously impairs
> > optimization.
> Totally agree on that, so far every attempt to create an optimized
> register allocator for r600 hardware was  torpedoed by the fact that
> you
> don't really know at which index an temporary array starts and ends.
> 
> >> Then there should be a declaration per array, and we can have our
> >> type.
> > It's still unclear to me what typed registers buy over typed
> > operations.
> >
> > When one translates a floating point addition operation, one
> > immediately knows that the inputs are floating points and the
> > outputs as well. Ditto for integers.
> >
> > And type is not truly a property of the temporaries anyway, but of
> > the intermediate values. And if you're doing any decent
> > optimization, you'll transform the IR to SSA, and you'll annotate
> > the types on the SSA values, and not the temporary storage, so
> > that you can handle unions and bitcasts efficiently.
> >
> > Passes like trying to minimize the number of temporaries would be
> > come much harder/inefficiency with typed registers, as they would
> > need to be type aware.
> 
> Maybe we should define a bit more clear what typed vs. untyped means
> here. When we speak about typed does that include bit width?

It's a convention, but in my mind if we add type info to TGSI declarations, it would make no sense to include the float/int/uint info and exclude the bitwidth.

> My current understanding is that all TGSI registers are 32bit and
> either
> threated as floats or ints depending on the operation, that is also
> more
> or less what we currently have on radeon hardware. So as long as the
> operations stay 32bit I doubt that we really need more type info
> inside
> the register.

My understanding is that TGSI registers are actually 128bits, which could be interpreted as 4 x floats, 4 x [u]int32s, 2 x doubles, 2 x int64s .
 
> But what should we do about 64bit operations?

It boils down to the question: does any hardware has a different register namespace and/or memory bank for doubles/int64s or not?

If yes, then having typed declarations might simplify TGSI translation for such hardware. If not, then typed registers are nothing more than a time-consuming distraction, as typed operations would fit perfectly.

I think this question was brought up before, in a thread about double opcodes in TGSI, but I do not recall the which/if any hardware had it.

I known that sse/avx has a single set of registers for all int32/float/int64/double types. dx10/dx11 shaders too, make no distinction on registers.

Jose