[PATCH] Teach instcombine how to combine multiply of constant splat of known power-of-2

[Andrea_DiBiagio at sn.scee.net]

Hi Rafael,

Rafael Espíndola <rafael.espindola at gmail.com> wrote on 23/05/2013 
18:13:29:

> you explicitly test that things like
> 
> define i32 @MulTest3_i8(i32 %InVec)  {
> entry:
>   %0 = bitcast i32 %InVec to <4 x i8>
>   %mul = mul <4 x i8> %0, <i8 4, i8 4, i8 2, i8 2>
>   %1 = bitcast <4 x i8> %mul to i32
>   ret i32 %1
> }
> 
> are not optimized. Why is it not profitable to use
> 
> shl <4 x i8> %0, <i8 2, i8 2, i8 1, i8 1>
> 
> ?

My original test simply tried to verify that multiply by constant splats 
of known 
power-of-2 were correctly optimized into shifts.
However, I think that converting a mul into a shift is profitable even in 
the
case you reported and probably my patch should be rewritten to address the 
general case
(i.e. not only the case of constant splats).

I will update this thread as soon as I have a new patch available.

Thanks for the feedback!
-- Andrea Di Biagio
SN Systems - Sony Computer Entertainment Group


**********************************************************************
This email and any files transmitted with it are confidential and intended 
solely for the use of the individual or entity to whom they are addressed. 
If you have received this email in error please notify postmaster at scee.net
This footnote also confirms that this email message has been checked for 
all known viruses.
Sony Computer Entertainment Europe Limited
Registered Office: 10 Great Marlborough Street, London W1F 7LP, United 
Kingdom
Registered in England: 3277793
**********************************************************************

P Please consider the environment before printing this e-mail