RFA: pervasive SSE codegen inefficiency

[Dale Johannesen]

On Sep 14, 2005, at 9:50 PM, Andrew Pinski wrote:
> On Sep 14, 2005, at 9:21 PM, Dale Johannesen wrote:
>> Consider the following SSE code
>> (-march=pentium4 -mtune=prescott -O2 -mfpmath=sse -msse2)
>> <4256776a.c>
>>
>> The first inner loop compiles to
>>
>>         paddq   %xmm0, %xmm1
>>
>> Good.  The second compiles to
>>
>>         movdqa  %xmm2, %xmm0
>>         paddw   %xmm1, %xmm0
>>         movdqa  %xmm0, %xmm1
>>
>> when it could be using a single paddw.  The basic problem is that
>> our approach defines __m128i to be V2DI even though all the operations
>> on the object are V4SI, so there are a lot of subreg's that don't need
>> to generate code.  I'd like to fix this, but am not sure how to go 
>> about it.
>
> From real looks of this looks more like a register allocation issue and
> nothing to do with subregs at all, except subregs being there.

That's kind of an overstatement; obviously getting rid of the subregs 
would
solve the problem as you can see from the first function.  I think 
you're right that

> If we allocated 64 and 63 as the same register, it would have worked 
> correctly.

(you mean 64 and 66) would fix this example; I'll look at that.  Having 
a more
uniform representation for operations on __m128i objects would simplify 
things
all over the place, though.